test: trigger after reopen

Two llm_as_judge evaluators score the review generation directly: a
NUMERIC 0-1 on finding actionability, a BOOLEAN on whether the summary
agrees with the findings. Both run on every observation whose trace
name is pr-review or opencode-review.

The judge is kimi-k2.7-code through the headroom hub. Local Ollama
returns Anthropic-format responses but the thinking blocks lack the
signature field Langfuse Zod schema requires; the evaluator preflight
fails as Invalid JSON response. A small judge-proxy pod on 8802
forwards to the hub and patches every thinking block with a synthetic
signature before returning.

Trace + generation output now includes the findings themselves
(capped at 25) rather than just the count, so a judge has something
to grade. generation input/output mirrors the trace so an
observation-level evaluator can read them.

Idempotent: existing evaluators and rules are skipped on re-run,
not duplicated. The connection is upserted on provider.

.gitignore

@@ -8,3 +8,5 @@ dist/
 __pycache__/
 *.pyc
 .worktrees/
+.claude/
+.opencode/package-lock.json

.pr-review.json

@@ -0,0 +1,5 @@
+{
+  "enabled": true,
+  "model": "headroom/MiniMax-M2.7",
+  "static_message": "PR-Agent pilot on this repo. Comments are LLM-generated; treat as suggestions, not mandates."
+}

README.md

@@ -94,6 +94,11 @@ path is in [`pilot/README.md`](pilot/README.md).
 The model endpoint is supplied at runtime via `PRAGENT_MODEL_BASE_URL`; the
 committed `opencode.json` carries a placeholder.
 
+Per-review token spend, latency and equivalent cost are shipped to a
+self-hosted Langfuse, split into `ollama` and `claude` environments so the two
+spend stories stay separate: [`pilot/README-langfuse.md`](pilot/README-langfuse.md).
+Emission is a silent no-op unless `LANGFUSE_HOST` and the key pair are set.
+
 ## Extending it
 
 The review "factory" is [`.opencode/`](.opencode/README.md) — agent definitions

k8s/judge-proxy.yaml

@@ -0,0 +1,82 @@
+# Judge-side think-block patcher. Stands between Langfuse evaluators and the
+# headroom-ollama hub (port 8790). Local Ollama does not emit the `signature`
+# field that Langfuse's Anthropic adapter's Zod schema requires on every
+# `thinking` content block — without it, the evaluator preflight fails as
+# "Invalid JSON response". The proxy forwards /v1/* verbatim and adds a dummy
+# signature to each thinking block before returning.
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: judge-proxy
+  namespace: pragent
+data:
+  proxy.py: |
+    #!/usr/bin/env python3
+    """Judge proxy: forward to headroom-ollama, fix thinking blocks."""
+    import json, sys, urllib.request, urllib.error
+    from http.server import BaseHTTPRequestHandler, HTTPServer
+    from socketserver import ThreadingMixIn
+    UPSTREAM = "http://100.74.17.70:8790"
+    DUMMY_SIG = "kimi-local-judge-no-signature"
+    class H(BaseHTTPRequestHandler):
+        def _proxy(self):
+            n = int(self.headers.get("Content-Length", 0))
+            body = self.rfile.read(n) if n else b""
+            h = {k: v for k, v in self.headers.items() if k.lower() not in ("host", "content-length")}
+            req = urllib.request.Request(UPSTREAM + self.path, data=body, headers=h, method=self.command)
+            try:
+                with urllib.request.urlopen(req, timeout=120) as r:
+                    resp_body = r.read(); status = r.status; rh = dict(r.headers)
+            except urllib.error.HTTPError as e:
+                resp_body = e.read(); status = e.code; rh = dict(e.headers)
+            ct = rh.get("content-type", "")
+            if status == 200 and "application/json" in ct and self.path.startswith("/v1/messages"):
+                try:
+                    obj = json.loads(resp_body)
+                    patched = 0
+                    for blk in obj.get("content") or []:
+                        if isinstance(blk, dict) and blk.get("type") == "thinking" and "signature" not in blk:
+                            blk["signature"] = DUMMY_SIG; patched += 1
+                    if patched:
+                        resp_body = json.dumps(obj).encode("utf-8")
+                        rh["content-length"] = str(len(resp_body))
+                        print(f"judge-proxy: patched {patched} thinking block(s)", file=sys.stderr, flush=True)
+                except Exception as e:
+                    print(f"judge-proxy: patch failed: {e}", file=sys.stderr, flush=True)
+            self.send_response(status)
+            for k, v in rh.items():
+                if k.lower() not in ("transfer-encoding", "content-length", "connection"):
+                    self.send_header(k, v)
+            self.send_header("Content-Length", str(len(resp_body)))
+            self.end_headers(); self.wfile.write(resp_body)
+        def do_POST(self): self._proxy()
+        def do_GET(self): self._proxy()
+        def log_message(self, *a, **k): pass
+    class S(ThreadingMixIn, HTTPServer): daemon_threads = True
+    S(("0.0.0.0", 8802), H).serve_forever()
+---
+apiVersion: v1
+kind: Pod
+metadata:
+  name: judge-proxy
+  namespace: pragent
+  labels:
+    app: judge-proxy
+spec:
+  nodeSelector:
+    kubernetes.io/hostname: kubernets
+  hostNetwork: true
+  dnsPolicy: ClusterFirstWithHostNet
+  restartPolicy: Always
+  containers:
+  - name: p
+    image: python:3.12-alpine
+    command: ["sh","-c","apk add --no-cache ca-certificates >/dev/null && python3 -u /etc/cfg/proxy.py"]
+    volumeMounts:
+    - {name: cfg, mountPath: /etc/cfg}
+    ports:
+    - {containerPort: 8802, hostPort: 8802}
+  volumes:
+  - name: cfg
+    configMap:
+      name: judge-proxy

opencode.json

@@ -21,19 +21,32 @@
         }
       }
     },
-    "local": {
+    "vllm-qwen38": {
       "npm": "@ai-sdk/openai-compatible",
-      "name": "Local AI workstation (qwen3.8-27b)",
+      "name": "Qwen3.8-27B vLLM (RTX 3090, MTP spec-decode)",
       "options": {
         "baseURL": "http://192.168.1.79:18020/v1",
-        "apiKey": "PLACEHOLDER_REPLACED_AT_RUNTIME"
+        "apiKey": "PLACEHOLDER_REPLACED_AT_RUNTIME",
+        "timeout": 300000,
+        "chunkTimeout": 30000
       },
       "models": {
         "qwen3.8-27b": {
-          "name": "Qwen 3.8 27B (local)",
+          "name": "Qwen3.8-27B (vLLM, MTP, 150k ctx)",
+          "tools": true,
+          "thinking": true,
+          "attachments": false,
           "limit": {
-            "context": 57344,
+            "context": 150000,
             "output": 8192
+          },
+          "options": {
+            "temperature": 0.3,
+            "topP": 0.8,
+            "topK": 20,
+            "repetitionPenalty": 1.05,
+            "frequencyPenalty": 0,
+            "presencePenalty": 0
           }
         }
       }

pilot/README-eval.md

@@ -0,0 +1,224 @@
+# Evaluation — scorers, ground truth, and the dataset
+
+Langfuse already receives one trace per review (`README-langfuse.md`). This is
+the layer on top: numbers attached to those traces that say how the reviewer
+*behaved*, and the beginnings of a ground-truth signal that says whether it was
+*right*.
+
+Those two things are deliberately kept apart, because only one of them exists
+yet.
+
+## What could and could not be built
+
+`feedback.db` has recorded 113 reviews across 4 repos. It has recorded **zero**
+reactions, zero thread resolutions and zero replies. The harvester, the schema
+and the daily analyzer are all working; nobody has ever reacted to a bot
+comment.
+
+That rules out an accuracy metric today. Correctness needs labels, and a
+judge scored against no labels is theatre. So the scorers here measure
+behaviour, which is computable from data already in hand, and a separate
+bridge exists to turn human reactions into scores the moment any arrive.
+
+## The five behavioural scores
+
+Emitted with every review by `eval_scores.py`, folded into the same ingestion
+batch as the trace so they cost no extra request.
+
+| score | type | what a change in it means |
+|---|---|---|
+| `finding_rate` | NUMERIC | Findings posted. 0 is the restraint case — good on clean code, a failure when the run degraded. Only the rate over time separates those. |
+| `severity_info_ratio` | NUMERIC 0–1 | Share of findings the model rated `info`/`trivial`. Rising = the model is hedging rather than committing. `None` when the review was silent: a ratio over an empty set is undefined, and charting it as 0 would read as perfect calibration. |
+| `severity_max` | CATEGORICAL | Highest severity surfaced, `none` when silent. Categorical because "did this ever surface something serious" is the real question, and a mean of severity ranks answers nothing. |
+| `dropped_findings` | NUMERIC | Findings the model emitted that the parser rejected for an unusable `path`/`line`. This is the only score here that measures the model's raw output. |
+| `cost_per_finding` | NUMERIC | Equivalent USD per finding. A cheaper model that finds nothing is not cheaper. |
+
+### Why `dropped_findings` needed a change to the parser
+
+`parse_findings` and `parse_review_output` discard any finding with a missing or
+unusable location. That happens silently, so a model emitting ten findings at
+invalid locations was indistinguishable from a model that found nothing — both
+produce an empty list. `ai_review.last_parse_dropped()` exposes the delta,
+recorded at parse time.
+
+It must be read at parse time specifically: by the time findings reach
+`_emit_langfuse`, `apply_repo_config` has already filtered them by
+`severity_threshold` and `max_findings`, and those drops are the config working
+as intended, not the model misbehaving.
+
+## Ground truth: `feedback_scores.py`
+
+Turns `feedback.db` into two session-level scores, keyed on `"{repo}#{pr}"`
+(which is what `langfuse_trace` already sets as `sessionId`).
+
+| score | meaning |
+|---|---|
+| `review_engagement` | Share of a PR's findings that drew any human reaction, resolution or reply. **Watch this first** — every quality number is vapour until it moves off 0. |
+| `review_acceptance` | Net verdict over engaged findings, −1 to +1. Absent, not 0, when nothing was engaged: zero would claim humans judged the review neutral, when the truth is nobody looked. |
+
+Session-level rather than trace-level because feedback arrives days later
+against a PR, and nothing in `feedback.db` records which re-run of the reviewer
+produced which comment. The session is both the available join and the honest
+granularity.
+
+Score ids are `uuid5(namespace, repo#pr#name)`, so the daily backfill updates
+rather than duplicates.
+
+## The dataset
+
+`pragent-reviews`, one item per PR the reviewer has run on, seeded by
+`eval_bootstrap.py` from `feedback.db`.
+
+`expectedOutput` is **the reviewer's own prior output**, not human-verified
+truth — every item carries `metadata.labelled_by_human: false`. Read it as a
+regression baseline: re-run a candidate model over these PRs and the diff
+against this column is the behaviour change. Promoting an item to real ground
+truth means a human editing it in the dataset view after re-reading the PR.
+
+### Item ids
+
+`{owner}__{repo}__pr{n}`. The obvious `{repo}#{pr}` cannot be used: items are
+routed as `/datasets/{id}/items/{item_id}`, so the `/` in `owner/repo` splits
+into extra path segments and everything after `#` is a fragment the browser
+never sends — the item is created fine by the API and then 404s when opened.
+Session ids elsewhere keep `{repo}#{pr}`; those are never path segments.
+
+### Filterable metadata
+
+The filter bar matches on `metadata` only — not on `input`, and not on the item
+id — so every facet worth slicing on is a flat, primitive key in `metadata`
+even where it duplicates `input`:
+
+| key | why it is there |
+| --- | --- |
+| `repo`, `owner`, `repo_name` | `owner` exists because a filter on the joined `repo` matches one repo, never a whole org |
+| `pr`, `head_sha` | jump from a filtered row back to the actual PR |
+| `finding_count`, `has_findings` | isolate the silent reviews, which are the interesting negatives |
+| `max_severity` | `"none"` rather than absent — an absent key matches no filter |
+| `reviews_run` | how churny the PR was; high values skew per-item averages |
+| `last_reviewed_at` / `_iso` | epoch sorts, ISO reads |
+| `labelled_by_human` | `false` everywhere today; the flag to filter on before trusting any of it |
+
+Nested objects and lists are deliberately absent: the filter bar cannot reach
+into them.
+
+`max_severity` is derived from `feedback.db`, whose `severity` column is
+re-parsed out of the rendered comment by `feedback_harvest._parse_severity` and
+defaults to `INFO` when its regex misses the badge. Trust the `severity_max`
+**score** (read from the model's structured output) over this facet.
+
+## Experiments
+
+`eval_experiment.py` links reviews that already ran into a dataset run, so the
+Experiments tab is populated without re-running anything. Runs are grouped by
+model — the comparison the pilot actually needs is the same PRs under a
+candidate model with `finding_rate` and `cost_per_finding` side by side. A new
+model produces a new run automatically on the next invocation.
+
+One trace per (run, item), the most recent: a PR re-reviewed on every push has
+many traces, and a run is one output per input.
+
+It uses `POST /api/public/dataset-run-items`, which is deprecated in favour of
+the SDK experiment runner and disappears in Langfuse v4. The deprecation notice
+exempts self-hosted v3 from the cutoff date, and this pilot is stdlib-only by
+design. Revisit when this deployment moves to v4.
+
+Coverage is bounded by the dataset, not by the traces: items only exist for PRs
+with a row in `feedback.db`, and a review that posted no comment leaves a trace
+but no row. That is why a run links fewer items than there are traces.
+
+## Evaluators: `eval_judges.py`
+
+Behaviour scores answer "how many, how severe, how much" — computable from data
+already in hand. Two things they cannot answer:
+
+- **Was the finding any good?** Specificity vs. hedge, generic advice vs.
+  fix-it-now advice — the difference between a useful review and one a
+  developer scrolls past.
+- **Did the summary match the findings?** Claiming "no issues" above two
+  criticals, or describing a problem in prose that never became a finding.
+
+These need a judge. `eval_judges.py` registers two `llm_as_judge` evaluators
+against the trace names this project emits (`pr-review`, `opencode-review`)
+and wires a sampling=1 rule per evaluator. Both run on every observation in a
+matching trace; the only observations in those traces are the review itself.
+
+| evaluator | output | what it answers |
+|---|---|---|
+| `finding_actionability` | NUMERIC 0–1 | How specific and fixable is each finding? |
+| `review_self_consistency` | BOOLEAN | Does the summary agree with the findings? |
+
+The judge is a different model from the reviewer (`kimi-k2.7-code` through the
+headroom hub). A model grading its own output agrees with itself for reasons
+that have nothing to do with quality. The judges are also asked only what they
+can answer from the review itself — never whether a finding is correct, since
+that needs the diff the trace does not carry.
+
+### Why the judge goes through `judge-proxy` (port 8802)
+
+The headroom hub in front of local Ollama returns Anthropic-format responses,
+but every `thinking` content block is missing the `signature` field real
+Claude emits. Langfuse's Zod schema requires it; the omission fails the
+evaluator preflight as `Invalid JSON response`. The `judge-proxy` pod sits in
+front of the hub on `100.74.17.70:8802` and patches every thinking block with
+a synthetic signature before forwarding the response. The model is unchanged;
+only the wire shape is fixed.
+
+```bash
+python3 pilot/eval_judges.py --dry-run   # show what would be created
+python3 pilot/eval_judges.py             # create the LLM connection, evaluators, rules
+```
+
+Idempotent: existing evaluators and rules are skipped, not duplicated. The
+connection is upserted on `provider` so re-runs return the same record.
+
+## Running it
+
+```bash
+# once per project: score configs + dataset (+ score historical traces)
+python3 pilot/eval_bootstrap.py --db /data/feedback.db --backfill-traces
+
+# ship feedback verdicts (runs daily from the feedback CronJob)
+python3 pilot/feedback_scores.py --db /data/feedback.db
+
+# link already-traced reviews into a dataset run per model
+python3 pilot/eval_experiment.py --dry-run
+python3 pilot/eval_experiment.py
+```
+
+Both need `LANGFUSE_HOST`, `LANGFUSE_PUBLIC_KEY`, `LANGFUSE_SECRET_KEY`. In
+cluster they come from the `pragent-langfuse` Secret and point at the ClusterIP
+— never the NodePort, whose oauth2-proxy 302s ingestion to Logto and drops it.
+
+## Gotcha: HTTP 207 is not success
+
+The ingestion endpoint answers `207 Multi-Status` when *some* events failed, so
+a batch where **every** event was rejected still returns 207. An early version
+of these scorers omitted the required per-event `timestamp` and silently
+ingested nothing while reporting success. `langfuse_trace._warn_on_rejected_events`
+now logs the per-event errors under `LANGFUSE_DEBUG=1`. If scores are missing,
+check that before anything else.
+
+## What the first run showed
+
+Backfilled over 42 existing traces and 13 PRs:
+
+```
+cost_per_finding     n=42  mean=0.3133  min=0.0880  max=0.9042
+finding_rate         n=42  mean=0.4762  min=0.0000  max=4.0000
+severity_info_ratio  n=14  mean=0.0000
+review_engagement    n=14  mean=0.0000
+severity_max         {none: 28, medium: 11, high: 1, critical: 2}
+```
+
+Two things worth keeping:
+
+- **The reviewer is not info-heavy.** `feedback.db` shows 61 of 62 findings at
+  `INFO`, which looked like a badly calibrated model. It is not: `severity_max`
+  reads `medium`/`high`/`critical` on every trace that found anything, and
+  `severity_info_ratio` is flat 0. The `INFO` in the DB comes from
+  `feedback_harvest._parse_severity`, which defaults to `INFO` when its regex
+  misses the severity badge in the rendered comment. The DB severity is a
+  re-parse artifact; the score reads the model's structured output directly.
+- **28 of 42 reviews found nothing** (67%), and **engagement is flat zero**. The
+  first is not yet interpretable without the second.

pilot/README-langfuse.md

@@ -0,0 +1,122 @@
+# pragent → Langfuse
+
+Every review the pilot runs ships one **trace** to a self-hosted Langfuse. The
+review body already prints a usage table, but that table lives and dies inside
+one Gitea PR. Langfuse is where the same numbers become a trend: tokens per
+review, latency per model, equivalent cost per repo, and how those move when
+the model or the tiering changes.
+
+## The ollama / claude split
+
+Both paths route through the same headroom proxy, so the provider prefix does
+not distinguish them — `headroom/claude-sonnet-5` is Claude spend,
+`headroom/glm-5.2:cloud` is not. The split is keyed off the **bare model name**
+and lands on the trace's `environment`:
+
+| resolved model              | environment |
+| --------------------------- | ----------- |
+| `headroom/claude-sonnet-5`  | `claude`    |
+| `claude-opus-5`             | `claude`    |
+| `headroom/glm-5.2:cloud`    | `ollama`    |
+| `headroom/MiniMax-M2.7`     | `ollama`    |
+| `vllm-qwen38/qwen3.8-27b`   | `ollama`    |
+
+Langfuse takes an environment selector on every dashboard, filter and cost
+breakdown, so the two spend stories stay separate inside one project — one key
+pair to rotate instead of two. Tags carry the finer cut:
+`provider:headroom`, `model:<bare>`, `engine:opencode`, `repo:<owner/name>`,
+`lens:<id>` per fan-out lens.
+
+To split into two *projects* later, point `LANGFUSE_PUBLIC_KEY` /
+`LANGFUSE_SECRET_KEY` at the second project on whichever deployment runs the
+Claude path. Nothing in the code needs to change.
+
+## What a trace carries
+
+- **trace** `pr-review` — `sessionId` = `owner/repo#index`, so every push to one
+  PR groups together. Input is the PR identity; output is the summary + finding
+  count; metadata carries steps, duration, severity counts and the provider's
+  own reported cost.
+- **generation** `opencode-review` — `model`, `usageDetails`, `costDetails`.
+
+`usageDetails.input` is the **uncached** input. opencode reports `cache_read`
+*inside* `input`, and Langfuse sums the keys it is given, so passing both
+verbatim would bill the resent prefix twice.
+
+### How cost is priced
+
+Langfuse has no price table of its own here — we compute the number and ship it
+as `costDetails.total`, so what Langfuse charts is exactly what
+`cost_model.PRICES` says.
+
+A model that genuinely bills (`claude-*`, `gpt-*`, `gemini-*`, `grok-*`) is
+priced **as itself**: basis `actual`.
+
+A model that costs nothing through the headroom proxy is priced against a
+**comparison target** instead: basis `equivalent:<target>`. That covers the
+models absent from `PRICES` (`MiniMax-M2.7` — which is what the webhook
+actually runs — and `glm-5.2:cloud`) as well as entries priced at all zeros
+(the self-hosted vLLM `qwen3.8-27b`). Without this the dashboard would be a
+flat $0.00 line, since the pilot's own path is free.
+
+The target follows the same precedence as the review body, so the PR and the
+dashboard never disagree:
+
+    .pr-review.json:cost_target  >  PRAGENT_PRICE_TARGET  >  claude-sonnet-5
+
+An equivalent cost is a hypothetical, not money spent, so every trace is tagged
+`cost:actual` or `cost:equivalent:<target>` and the generation metadata carries
+`cost_basis`. Filter on it before reading any cost chart as spend.
+
+If the comparison target itself is unknown, the trace ships usage with **no**
+cost block — better no number than a wrong one.
+
+Anthropic prices in `cost_model.PRICES` were fetched 2026-08-18; re-check them
+before quoting anything externally.
+
+## Configuration
+
+| env                   | meaning                                                   |
+| --------------------- | --------------------------------------------------------- |
+| `LANGFUSE_HOST`       | `http://langfuse-web.langfuse.svc.cluster.local:3000`      |
+| `LANGFUSE_PUBLIC_KEY` | `pk-lf-…`                                                  |
+| `LANGFUSE_SECRET_KEY` | `sk-lf-…`                                                  |
+| `LANGFUSE_TIMEOUT`    | seconds, default `5`                                       |
+| `LANGFUSE_DEBUG`      | `1` to log ingestion failures to stderr                    |
+
+Unset host or either key ⇒ emission is a silent no-op. That is the default, so
+a checkout without Langfuse behaves exactly as before.
+
+## Fail-open
+
+`langfuse_trace` is stdlib-only (`urllib`) and every entry point swallows its
+own exceptions; `_emit_langfuse` in `ai_review.py` wraps even the import. A
+Langfuse outage cannot fail, delay past `LANGFUSE_TIMEOUT`, or alter a review.
+
+Both token-spending exit paths emit — the normal post **and** the salvage path
+where the agent produced unparseable output. That run cost the same as a clean
+one, and is precisely the failure worth trending.
+
+## Deployment
+
+Cluster side lives outside this repo: `~/k8s/langfuse.yaml` (ClickHouse +
+web + worker, reusing the gitea postgres, gitea valkey and minio),
+`~/k8s/oauth2-proxy-langfuse.yaml` (the Logto gate), and
+`~/k8s/langfuse-setup.sh`, which provisions the database, the bucket, the
+secrets, and wires `pragent-webhook` with the three env vars above.
+
+The UI is at **https://langfuse.marcospaulo.dev.br**:
+
+    browser -> Caddy (VPS, TLS, DNS-01) -> tailscale
+            -> 100.74.17.70:30361 -> oauth2-proxy (Logto, email allowlist)
+            -> langfuse-web (ClusterIP)
+
+Logto sits at *both* layers off one app (`langfuse`, two redirect URIs): the
+proxy gates the domain, and Langfuse's own NextAuth uses the same Logto as a
+custom OIDC provider, so the inner login is a silent redirect rather than a
+second password.
+
+pragent does **not** go through any of that. It posts to
+`langfuse-web.langfuse.svc.cluster.local:3000` from inside the cluster, on
+API-key auth — putting ingestion behind an interactive SSO gate would break it
+on the first review.

pilot/ai_review.py

@@ -489,7 +489,7 @@ def _resolve_display_model(base_model: str, config: dict | None) -> str:
          like `claude-sonnet-5` or `qwen3.8-27b` is safe. Re-prefixed with
          the model's `provider` field from `cost_model.Price` (default
          `headroom`) so the opencode subprocess routes correctly — e.g.
-         `qwen3.8-27b` → `local/qwen3.8-27b` (local AI workstation on
+         `qwen3.8-27b` → `vllm-qwen38/qwen3.8-27b` (vLLM on RTX 3090 at
          192.168.1.79:18020), `claude-sonnet-5` → `headroom/claude-sonnet-5`
          (Anthropic pricing proxy).
       3. Default — `f"headroom/{base_model}"` where `base_model` is the bare
@@ -505,8 +505,8 @@ def _resolve_display_model(base_model: str, config: dict | None) -> str:
     cfg_model = (config or {}).get("model")
     if isinstance(cfg_model, str) and cfg_model.strip():
         # Look up the provider from PRICES so the opencode subprocess routes
-        # through the right provider block (local vs headroom). Lazy import —
-        # the ollama path doesn't touch cost_model.
+        # through the right provider block (vllm-qwen38 vs headroom). Lazy
+        # import — the ollama path doesn't touch cost_model.
         from cost_model import PRICES
         provider = PRICES.get(cfg_model.strip())
         if provider is not None:
@@ -678,6 +678,19 @@ def _strip_path_prefix(p: str) -> str:
 # ---------------------------------------------------------------------------
 
 
+# How many raw findings the last `parse_review_output` / `parse_findings` call
+# rejected for an unusable path/line. A side channel rather than a return value
+# because both parsers already return fixed-width tuples that several callers
+# and their tests unpack positionally; widening them to carry a telemetry
+# number would be a breaking change for a fail-open signal.
+_LAST_PARSE_DROPPED: dict[str, int] = {"n": 0}
+
+
+def last_parse_dropped() -> int:
+    """Findings the last parse discarded. Read it immediately after parsing."""
+    return int(_LAST_PARSE_DROPPED.get("n") or 0)
+
+
 def _normalize_finding(f: dict) -> dict | None:
     """Validate + normalize one raw finding dict. Returns None if it's unusable
     (missing path/line). Normalises severity, keeps `reference` (default "")."""
@@ -754,6 +767,7 @@ def parse_findings(text: str) -> list[dict]:
     Also accepts a bare JSON array as the outer value: ``[{...}, {...}]`` —
     some agents skip the ``{"summary":..., "findings":[...]}`` wrapper.
     """
+    _LAST_PARSE_DROPPED["n"] = 0
     data = _parse_json_tolerant(text)
     if isinstance(data, dict):
         findings = data.get("findings")
@@ -768,6 +782,7 @@ def parse_findings(text: str) -> list[dict]:
         n = _normalize_finding(f)
         if n is not None:
             out.append(n)
+    _LAST_PARSE_DROPPED["n"] = len(findings) - len(out)
     return out
 
 
@@ -823,6 +838,7 @@ def parse_review_output(
     block), with a tolerant fallback that scans for the last balanced
     object/array in the prose tail. Never raises.
     """
+    _LAST_PARSE_DROPPED["n"] = 0
     blob = _last_json_block(text)
     if blob is None:
         return "", [], [], [], [], "", ""
@@ -856,6 +872,12 @@ def parse_review_output(
             n = _normalize_finding(f)
             if n is not None:
                 out.append(n)
+        # A model that emits findings at unusable locations is indistinguishable
+        # from one that found nothing, because both end up with an empty `out`.
+        # Stash the delta so the caller can score it (see `eval_scores`).
+        _LAST_PARSE_DROPPED["n"] = len(findings_raw) - len(out)
+    else:
+        _LAST_PARSE_DROPPED["n"] = 0
     return summary, out, summary_changes, risks, walkthrough, risk_verdict, test_coverage
 
 
@@ -2068,6 +2090,50 @@ def _need(name: str) -> str:
     return v
 
 
+def _emit_langfuse(
+    *,
+    repo: str,
+    index: str,
+    sha: str,
+    title: str,
+    model: str,
+    usage: dict | None,
+    findings: list[dict],
+    summary: str,
+    engine: str,
+    config: dict | None = None,
+    dropped_count: float | None = None,
+) -> None:
+    """Ship this review's usage to Langfuse, if one is configured.
+
+    Called on both exit paths that spent tokens — the normal post and the
+    salvage path — because an unparseable run costs the same as a clean one and
+    is exactly the kind of thing worth trending.
+
+    Local import + blanket except: `langfuse_trace` is stdlib-only but optional,
+    and telemetry is never allowed to fail a review (see the fail-open contract
+    in `review_pr`). The trace's `environment` is `claude` or `ollama`, so the
+    two spend stories stay separated in every Langfuse view.
+    """
+    try:
+        import langfuse_trace
+
+        # Same comparison model the review body prices against, so the number
+        # in Langfuse and the number in the PR agree. Free/unknown models
+        # (MiniMax, glm, self-hosted qwen) are priced against it; a paid model
+        # is priced as itself.
+        price_target, _err = _resolve_price_target(config)
+
+        langfuse_trace.emit_review_trace(
+            repo=repo, index=index, sha=sha, title=title, model=model,
+            usage=usage, findings=findings, summary=summary or "",
+            engine=engine, lenses=(usage or {}).get("lenses"),
+            price_target=price_target, dropped_count=dropped_count,
+        )
+    except Exception as e:
+        print(f"pragent: langfuse emit skipped: {e}", file=sys.stderr)
+
+
 def review_pr(
     api: str,
     repo: str,
@@ -2192,6 +2258,7 @@ def review_pr(
                     additional_context=additional_context,
                 )
             review_summary, findings, summary_changes, risks, _walkthrough, _risk_verdict, _test_coverage = parse_review_output(stdout)
+            parse_dropped = last_parse_dropped()
             if not findings and not review_summary:
                 # The findings JSON was missing or malformed. Don't discard the
                 # run: salvage the prose, keep the usage report (the tokens were
@@ -2208,11 +2275,18 @@ def review_pr(
                     salvaged or "AI review produced no parseable output.",
                     display_model, sha, usage_section=usage_section,
                     static_message=(config or {}).get("static_message", "")))
+                _emit_langfuse(
+                    repo=repo, index=index, sha=sha, title=title,
+                    model=display_model, usage=usage, findings=[],
+                    summary=salvaged, engine=engine, config=config,
+                    dropped_count=parse_dropped,
+                )
                 return True
         else:
             user_prompt = build_user_prompt(title, body + compression_note, diff, config, prior, additional_context)
             raw_findings = call_model(ollama_url, model, SYSTEM_PROMPT, user_prompt, max_tokens)
             findings = parse_findings(raw_findings)
+            parse_dropped = last_parse_dropped()
             usage = None
 
         # Filter / cap findings per `.pr-review.json` (style, threshold, max,
@@ -2291,6 +2365,12 @@ def review_pr(
         )
 
         post_inline_review(api, repo, index, token, summary_body, anchored)
+        _emit_langfuse(
+            repo=repo, index=index, sha=sha, title=title,
+            model=display_model, usage=usage, findings=findings,
+            summary=review_summary, engine=engine, config=config,
+            dropped_count=parse_dropped,
+        )
         print(
             f"pragent: reviewed {repo}#{index} sha={sha[:8]} "
             f"engine={engine} findings={len(findings)} inline={len(anchored)}",

pilot/cost_model.py

@@ -57,7 +57,7 @@ class Price:
     """Per-MTok prices. `cache_write` and `cache_read` are absolute rates, not
     multipliers, so providers with different cache economics stay comparable.
 
-    `provider` is the opencode provider name (`headroom`, `local`, ...). It
+    `provider` is the opencode provider name (`headroom`, `vllm-qwen38`, ...). It
     doubles as the dispatch key for `.pr-review.json:model` overrides — when
     a per-repo override is set, `_resolve_display_model` returns
     `f"{provider}/{key}"` so the opencode subprocess routes correctly.
@@ -98,12 +98,13 @@ PRICES: dict[str, Price] = {
     # xAI Grok — cache_write = input
     "grok-4.5": Price("Grok 4.5", 2.00, 6.00, 2.00, 0.30),
     "grok-4.3": Price("Grok 4.3", 1.25, 2.50, 1.25, 0.20),
-    # Self-hosted — local AI workstation, no per-token charge. provider="local"
-    # so the opencode subprocess routes via the `local` provider block in
-    # opencode.json (baseURL=http://192.168.1.79:18020/v1). Equivalent-cost
-    # column will read $0 — the cost-comparison signal is that the same work
-    # would bill $X on a paid model.
-    "qwen3.8-27b": Price("Qwen 3.8 27B (local)", 0.0, 0.0, 0.0, 0.0, provider="local"),
+    # Self-hosted — AI workstation RTX 3090, vLLM + DFlash2 spec-decode, no
+    # per-token charge. provider="vllm-qwen38" so the opencode subprocess
+    # routes via the matching provider block in opencode.json
+    # (baseURL=http://192.168.1.79:18020/v1). Equivalent-cost column reads $0
+    # — the cost-comparison signal is that the same work would bill $X on a
+    # paid model.
+    "qwen3.8-27b": Price("Qwen3.8-27B (vLLM, MTP, 150k ctx)", 0.0, 0.0, 0.0, 0.0, provider="vllm-qwen38"),
 }
 
 

pilot/eval_bootstrap.py

@@ -0,0 +1,350 @@
+#!/usr/bin/env python3
+"""pragent pilot — one-time Langfuse project setup for evaluation.
+
+Three jobs, each idempotent so it can be re-run after any change:
+
+  1. **Score configs.** Registers the schema for every score pragent emits
+     (`eval_scores.SCORE_CONFIGS` + `feedback_scores.SCORE_CONFIGS`). Without
+     these the scores still ingest, but nothing stops a later scorer writing
+     `severity_max="HIGH"` beside today's `"high"` and quietly splitting one
+     series into two. Configs are immutable in Langfuse — a name that already
+     exists is left alone rather than updated.
+
+  2. **Dataset.** Seeds `pragent-reviews` from `feedback.db`: one item per PR
+     the reviewer has actually run on, carrying the repo/PR/sha as input and
+     the findings it posted as `expectedOutput`.
+
+     Read `expectedOutput` here as "what the reviewer said last time", not "what
+     is correct" — no human has labelled any of it. It is a regression baseline:
+     re-run a candidate model over these PRs and the diff against this column is
+     the behaviour change. Promoting an item to real ground truth means a human
+     editing it after reviewing the PR, which is what the dataset view is for.
+
+  3. **Trace backfill** (`--backfill-traces`). Scores only ride along with new
+     reviews, so without this the charts stay empty until the next PR lands.
+     Every trace `langfuse_trace` has ever written already carries the finding
+     count, the severity histogram and the cost in its metadata, which is
+     everything four of the five scorers need. `dropped_findings` is absent from
+     historical traces and is left unscored rather than backfilled as zero.
+
+  4. **Reports** what it found, so the gap between "reviews recorded" and
+     "reviews with human feedback" is visible rather than assumed.
+
+Usage:
+    LANGFUSE_HOST=... LANGFUSE_PUBLIC_KEY=... LANGFUSE_SECRET_KEY=... \\
+        python3 eval_bootstrap.py --db /data/feedback.db
+"""
+from __future__ import annotations
+
+import argparse
+import base64
+import json
+import os
+import sqlite3
+import sys
+import urllib.error
+import urllib.request
+from datetime import datetime, timezone
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import eval_scores  # noqa: E402
+import feedback_scores  # noqa: E402
+
+DATASET_NAME = "pragent-reviews"
+
+
+def _conf() -> tuple[str, str, str]:
+    host = (os.environ.get("LANGFUSE_HOST") or "").strip().rstrip("/")
+    pk = (os.environ.get("LANGFUSE_PUBLIC_KEY") or "").strip()
+    sk = (os.environ.get("LANGFUSE_SECRET_KEY") or "").strip()
+    if not host or not pk or not sk:
+        raise SystemExit("LANGFUSE_HOST / LANGFUSE_PUBLIC_KEY / LANGFUSE_SECRET_KEY must be set")
+    return host, pk, sk
+
+
+def _call(method: str, path: str, body: dict | None = None, timeout: float = 20.0):
+    host, pk, sk = _conf()
+    auth = base64.b64encode(f"{pk}:{sk}".encode()).decode("ascii")
+    data = json.dumps(body).encode() if body is not None else None
+    req = urllib.request.Request(
+        host + path,
+        data=data,
+        headers={
+            "Content-Type": "application/json",
+            "Authorization": f"Basic {auth}",
+            "User-Agent": "pragent-pilot/1.0",
+        },
+        method=method,
+    )
+    try:
+        with urllib.request.urlopen(req, timeout=timeout) as resp:
+            raw = resp.read()
+            return resp.status, (json.loads(raw) if raw else None)
+    except urllib.error.HTTPError as e:
+        return e.code, e.read()[:400].decode("utf-8", "replace")
+
+
+# ---------------------------------------------------------------------------
+# 1. Score configs
+# ---------------------------------------------------------------------------
+
+def ensure_score_configs() -> dict:
+    status, existing = _call("GET", "/api/public/score-configs?limit=100")
+    have = set()
+    if status == 200 and isinstance(existing, dict):
+        have = {c.get("name") for c in existing.get("data", [])}
+
+    created, skipped, failed = [], [], []
+    for cfg in list(eval_scores.SCORE_CONFIGS) + list(feedback_scores.SCORE_CONFIGS):
+        if cfg["name"] in have:
+            skipped.append(cfg["name"])
+            continue
+        st, resp = _call("POST", "/api/public/score-configs", cfg)
+        if st in (200, 201):
+            created.append(cfg["name"])
+        else:
+            failed.append({"name": cfg["name"], "status": st, "error": resp})
+    return {"created": created, "already_present": skipped, "failed": failed}
+
+
+# ---------------------------------------------------------------------------
+# 2. Dataset from recorded reviews
+# ---------------------------------------------------------------------------
+
+def item_id(repo: str, pr) -> str:
+    """A dataset-item id that survives being put in a URL path.
+
+    The obvious `{repo}#{pr}` is unusable: the UI routes items as
+    `/datasets/{id}/items/{item_id}`, so the `/` in `owner/repo` splits into
+    extra path segments and everything after the `#` is a fragment the browser
+    never sends. The item is created fine and then 404s when opened.
+
+    Session ids elsewhere keep the `{repo}#{pr}` form — those are never path
+    segments, and `feedback_scores` depends on that shape.
+    """
+    return f"{repo.replace('/', '__')}__pr{pr}"
+
+
+def _item_metadata(*, repo, pr, head_sha, reviews_run, last_seen, findings) -> dict:
+    """Filterable facets for one dataset item.
+
+    Kept flat and primitive: the filter bar matches a metadata key against a
+    literal, so a nested object or a list is not reachable from the UI.
+    """
+    owner, _, repo_name = str(repo).partition("/")
+    sevs = [str(f["severity"] or "").lower() for f in findings]
+    ranked = [s for s in sevs if s in eval_scores.SEVERITY_RANK]
+    return {
+        "repo": repo,
+        "owner": owner or repo,
+        "repo_name": repo_name or repo,
+        "pr": int(pr),
+        "head_sha": head_sha,
+        "reviews_run": reviews_run,
+        "last_reviewed_at": last_seen,
+        "last_reviewed_iso": datetime.fromtimestamp(last_seen, timezone.utc).isoformat(),
+        "finding_count": len(findings),
+        "has_findings": bool(findings),
+        # "none" rather than omitting the key: a filter for silent reviews needs
+        # something to match, and an absent key matches nothing.
+        "max_severity": (
+            max(ranked, key=lambda s: eval_scores.SEVERITY_RANK[s]) if ranked else "none"
+        ),
+        # Flags that this row is the reviewer's own past output, not a human
+        # judgement. Filter on it before anyone treats the dataset as truth.
+        "labelled_by_human": False,
+    }
+
+
+def read_review_items(db_path: str) -> list[dict]:
+    """One dataset item per (repo, pr) the reviewer has run on.
+
+    Keyed on the PR rather than on each individual review row: the same PR is
+    re-reviewed on every push, and 113 rows over 26 PRs would make a benchmark
+    that is 4x redundant and weighted towards whichever PR churned most.
+    """
+    conn = sqlite3.connect(db_path)
+    conn.row_factory = sqlite3.Row
+    try:
+        prs = conn.execute(
+            """
+            SELECT repo, pr, MAX(posted_at) AS last_seen, COUNT(*) AS reviews,
+                   MAX(head_sha) AS head_sha
+            FROM review GROUP BY repo, pr ORDER BY repo, pr
+            """
+        ).fetchall()
+        items = []
+        for row in prs:
+            findings = conn.execute(
+                """
+                SELECT path, line, severity, problem, fix
+                FROM inline_finding WHERE repo = ? AND pr = ?
+                ORDER BY path, line
+                """,
+                (row["repo"], row["pr"]),
+            ).fetchall()
+            items.append(
+                {
+                    "id": item_id(row["repo"], row["pr"]),
+                    "input": {
+                        "repo": row["repo"],
+                        "pr": int(row["pr"]),
+                        "head_sha": row["head_sha"],
+                    },
+                    "expectedOutput": {
+                        "findings": [dict(f) for f in findings],
+                        "finding_count": len(findings),
+                    },
+                    # The UI's filter bar reads metadata and nothing else, so
+                    # anything worth slicing on is a top-level key here even
+                    # where it duplicates `input`. `owner` and `repo_name` are
+                    # split out because a filter on the joined `repo` can only
+                    # match one repo at a time, never a whole org.
+                    "metadata": _item_metadata(
+                        repo=row["repo"],
+                        pr=row["pr"],
+                        head_sha=row["head_sha"],
+                        reviews_run=int(row["reviews"]),
+                        last_seen=int(row["last_seen"]),
+                        findings=findings,
+                    ),
+                }
+            )
+        return items
+    finally:
+        conn.close()
+
+
+def ensure_dataset(items: list[dict], name: str = DATASET_NAME) -> dict:
+    st, _ = _call(
+        "POST",
+        "/api/public/datasets",
+        {
+            "name": name,
+            "description": (
+                "PRs the pragent pilot has reviewed, seeded from feedback.db. "
+                "expectedOutput is the reviewer's own prior output — a regression "
+                "baseline, not human-verified ground truth."
+            ),
+            "metadata": {"source": "feedback.db", "seeded_by": "eval_bootstrap.py"},
+        },
+    )
+    # A duplicate name is fine: the dataset already exists from an earlier run.
+    dataset_ok = st in (200, 201, 409)
+
+    created, failed = 0, []
+    for item in items:
+        body = {
+            "datasetName": name,
+            "id": item["id"],  # idempotent: same PR updates rather than duplicates
+            "input": item["input"],
+            "expectedOutput": item["expectedOutput"],
+            "metadata": item["metadata"],
+        }
+        ist, resp = _call("POST", "/api/public/dataset-items", body)
+        if ist in (200, 201):
+            created += 1
+        else:
+            failed.append({"item": item["id"], "status": ist, "error": resp})
+    return {"dataset": name, "dataset_created": dataset_ok, "items_upserted": created, "failed": failed}
+
+
+# ---------------------------------------------------------------------------
+# 3. Backfill scores onto traces that predate the scorers
+# ---------------------------------------------------------------------------
+
+def _synth_findings(severities: dict) -> list[dict]:
+    """Rebuild a findings list from a trace's severity histogram.
+
+    Only severity matters to the scorers, and that is all the histogram kept.
+    Reconstructing placeholders is honest here because every scorer being
+    backfilled reads nothing else off a finding.
+    """
+    out = []
+    for sev, count in (severities or {}).items():
+        out.extend({"severity": sev} for _ in range(int(count)))
+    return out
+
+
+def backfill_traces(limit_pages: int = 20) -> dict:
+    import eval_scores as es
+
+    scored, skipped, events = 0, 0, []
+    page = 1
+    while page <= limit_pages:
+        st, resp = _call("GET", f"/api/public/traces?limit=50&page={page}&name=pr-review")
+        if st != 200 or not isinstance(resp, dict):
+            break
+        rows = resp.get("data") or []
+        if not rows:
+            break
+        for tr in rows:
+            meta = tr.get("metadata") or {}
+            severities = meta.get("severities") or {}
+            count = meta.get("findings")
+            if count is None:
+                skipped += 1
+                continue
+            findings = _synth_findings(severities)
+            # The histogram is authoritative when present; a trace that recorded
+            # a count but no histogram still scores its rate.
+            if not findings and count:
+                findings = [{"severity": "medium"} for _ in range(int(count))]
+            batch = es.build_scores(
+                trace_id=tr["id"],
+                findings=findings,
+                environment=tr.get("environment") or "default",
+                cost_usd=(tr.get("totalCost") or meta.get("provider_cost_usd")),
+                timestamp=tr.get("timestamp"),
+                comment="backfilled from trace metadata",
+            )
+            events.extend(batch)
+            scored += 1
+        page += 1
+
+    posted = False
+    status = None
+    if events:
+        import langfuse_trace
+
+        host, pk, sk = _conf()
+        # Chunked: one 2000-event POST is refused, and a partial backfill that
+        # reports success is worse than a slow one.
+        for i in range(0, len(events), 200):
+            status = langfuse_trace._post(host, pk, sk, events[i:i + 200], 30.0)
+            posted = status in (200, 201, 207)
+            if not posted:
+                break
+    return {"traces_scored": scored, "traces_skipped": skipped, "scores": len(events),
+            "posted": posted, "http_status": status}
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description="Bootstrap Langfuse evaluation for the pragent pilot")
+    ap.add_argument("--db", default=os.environ.get("PRAGENT_FEEDBACK_DB", "/data/feedback.db"))
+    ap.add_argument("--skip-dataset", action="store_true")
+    ap.add_argument("--skip-configs", action="store_true")
+    ap.add_argument("--backfill-traces", action="store_true",
+                    help="score traces written before the scorers existed")
+    args = ap.parse_args()
+
+    out: dict = {}
+    if not args.skip_configs:
+        out["score_configs"] = ensure_score_configs()
+    if not args.skip_dataset:
+        items = read_review_items(args.db)
+        out["dataset"] = ensure_dataset(items)
+        out["dataset"]["items_read"] = len(items)
+    if args.backfill_traces:
+        out["trace_backfill"] = backfill_traces()
+    print(json.dumps(out, indent=2))
+
+    failed = (out.get("score_configs", {}).get("failed") or []) + (
+        out.get("dataset", {}).get("failed") or []
+    )
+    return 1 if failed else 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

pilot/eval_experiment.py

@@ -0,0 +1,212 @@
+#!/usr/bin/env python3
+"""pragent pilot — populate the Experiments tab from reviews already traced.
+
+An "experiment" in Langfuse is a dataset run: a set of (dataset item, trace)
+links under one run name. The Experiments tab then shows one row per item with
+its scores, and lets two runs be diffed side by side.
+
+Nothing here re-runs the reviewer. Every PR in `pragent-reviews` has already
+been reviewed, and each of those reviews left a trace carrying its findings,
+cost and scores. This links what exists, which is what makes the tab useful on
+day one instead of after the next N pushes.
+
+Runs are grouped by **model** by default, because that is the comparison the
+pilot actually needs to make: the same PRs reviewed by MiniMax vs whatever
+replaces it, with `finding_rate` and `cost_per_finding` side by side. Group by
+`none` for a single "all traces" run.
+
+One trace per (run, item) — the most recent. A PR re-reviewed on every push has
+many traces, and a dataset run is defined as one output per input; feeding it
+the other five would make the per-run averages meaningless.
+
+Note on the endpoint: `POST /api/public/dataset-run-items` is deprecated in
+favour of the SDK experiment runner / OTel ingestion, and disappears in
+Langfuse v4. This instance is self-hosted v3, which the deprecation notice
+explicitly exempts from the cutoff date, and the pilot is stdlib-only by
+design. Revisit when this deployment moves to v4.
+
+Usage:
+    LANGFUSE_HOST=... LANGFUSE_PUBLIC_KEY=... LANGFUSE_SECRET_KEY=... \\
+        python3 eval_experiment.py --dry-run
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+import urllib.parse
+from collections import defaultdict
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import eval_bootstrap as eb  # noqa: E402
+
+TRACE_NAME = "pr-review"
+
+
+# ---------------------------------------------------------------------------
+# Reading what already exists
+# ---------------------------------------------------------------------------
+
+def fetch_traces(name: str = TRACE_NAME, limit: int = 100, max_pages: int = 50) -> list[dict]:
+    """Every review trace, newest first."""
+    out: list[dict] = []
+    for page in range(1, max_pages + 1):
+        q = urllib.parse.urlencode({"name": name, "limit": limit, "page": page})
+        st, body = eb._call("GET", f"/api/public/traces?{q}")
+        if st != 200 or not isinstance(body, dict):
+            raise SystemExit(f"listing traces failed: {st} {body}")
+        data = body.get("data") or []
+        out.extend(data)
+        meta = body.get("meta") or {}
+        if page * meta.get("limit", limit) >= meta.get("totalItems", 0):
+            break
+    return out
+
+
+def fetch_item_ids(dataset: str) -> set[str]:
+    """Ids present in the dataset, so runs never reference a missing item."""
+    ids: set[str] = set()
+    for page in range(1, 51):
+        q = urllib.parse.urlencode({"datasetName": dataset, "limit": 100, "page": page})
+        st, body = eb._call("GET", f"/api/public/dataset-items?{q}")
+        if st != 200 or not isinstance(body, dict):
+            raise SystemExit(f"listing dataset items failed: {st} {body}")
+        ids.update(i["id"] for i in body.get("data") or [])
+        meta = body.get("meta") or {}
+        if page * meta.get("limit", 100) >= meta.get("totalItems", 0):
+            break
+    return ids
+
+
+# ---------------------------------------------------------------------------
+# Grouping traces into runs
+# ---------------------------------------------------------------------------
+
+def trace_model(trace: dict) -> str:
+    """The model that produced a review, from its `model:` tag."""
+    for tag in trace.get("tags") or []:
+        if tag.startswith("model:"):
+            return tag[len("model:"):] or "unknown"
+    return "unknown"
+
+
+def trace_item_id(trace: dict) -> str | None:
+    """The dataset item a trace belongs to, or None if it is not a PR review."""
+    md = trace.get("metadata") or {}
+    repo, pr = md.get("repo"), md.get("pr")
+    if not repo or pr in (None, ""):
+        return None
+    return eb.item_id(str(repo), pr)
+
+
+def _sort_key(trace: dict):
+    return (trace.get("timestamp") or "", trace.get("id") or "")
+
+
+def plan_runs(traces: list[dict], known_items: set[str], group_by: str = "model") -> dict:
+    """Map run name -> {item id: trace}, keeping only the newest trace per item.
+
+    Traces whose PR is not in the dataset are dropped: `feedback.db` is the
+    source for both, but a review can be traced without its row landing (the
+    posting step can fail after the model ran), and a run item pointing at a
+    non-existent dataset item is rejected.
+    """
+    runs: dict[str, dict[str, dict]] = defaultdict(dict)
+    skipped_no_item, skipped_unknown = 0, 0
+    for tr in traces:
+        iid = trace_item_id(tr)
+        if iid is None:
+            skipped_unknown += 1
+            continue
+        if iid not in known_items:
+            skipped_no_item += 1
+            continue
+        run = "all-traces" if group_by == "none" else trace_model(tr)
+        prev = runs[run].get(iid)
+        if prev is None or _sort_key(tr) > _sort_key(prev):
+            runs[run][iid] = tr
+    return {
+        "runs": dict(runs),
+        "skipped_not_in_dataset": skipped_no_item,
+        "skipped_not_a_review": skipped_unknown,
+    }
+
+
+def run_name(prefix: str, key: str) -> str:
+    return f"{prefix}-{key}" if prefix else key
+
+
+# ---------------------------------------------------------------------------
+# Writing the runs
+# ---------------------------------------------------------------------------
+
+def create_run(name: str, items: dict[str, dict], description: str = "") -> dict:
+    """Link each (item, trace) pair into the named run. Idempotent per pair."""
+    created, failed = 0, []
+    for iid, tr in sorted(items.items()):
+        md = tr.get("metadata") or {}
+        body = {
+            "runName": name,
+            "runDescription": description,
+            "datasetItemId": iid,
+            "traceId": tr["id"],
+            "metadata": {
+                "model": trace_model(tr),
+                "engine": md.get("engine"),
+                "findings": md.get("findings"),
+                "duration_s": md.get("duration_s"),
+                "cost_basis": md.get("cost_basis"),
+                "linked_by": "eval_experiment.py",
+            },
+        }
+        st, resp = eb._call("POST", "/api/public/dataset-run-items", body)
+        if st in (200, 201):
+            created += 1
+        else:
+            failed.append({"item": iid, "status": st, "error": resp})
+    return {"run": name, "items_linked": created, "failed": failed}
+
+
+def main(argv: list[str] | None = None) -> int:
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument("--dataset", default=eb.DATASET_NAME)
+    ap.add_argument("--group-by", choices=("model", "none"), default="model")
+    ap.add_argument("--prefix", default="baseline",
+                    help="run name prefix; '' for the bare group key")
+    ap.add_argument("--dry-run", action="store_true")
+    args = ap.parse_args(argv)
+
+    traces = fetch_traces()
+    items = fetch_item_ids(args.dataset)
+    plan = plan_runs(traces, items, group_by=args.group_by)
+
+    report = {
+        "traces_read": len(traces),
+        "dataset_items": len(items),
+        "skipped_not_in_dataset": plan["skipped_not_in_dataset"],
+        "skipped_not_a_review": plan["skipped_not_a_review"],
+        "runs": {},
+    }
+    for key, mapping in sorted(plan["runs"].items()):
+        name = run_name(args.prefix, key)
+        if args.dry_run:
+            report["runs"][name] = {"items_would_link": len(mapping)}
+            continue
+        report["runs"][name] = create_run(
+            name,
+            mapping,
+            description=(
+                "Reviews already run by the pilot, linked after the fact. "
+                "Scores come from the traces; expectedOutput is the reviewer's "
+                "own prior output, not human-verified ground truth."
+            ),
+        )
+    report["dry_run"] = args.dry_run
+    print(json.dumps(report, indent=2))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

pilot/eval_judges.py

@@ -0,0 +1,308 @@
+#!/usr/bin/env python3
+"""pragent pilot — LLM-as-a-judge evaluators for the reviewer.
+
+The deterministic scorers in `eval_scores.py` measure *behaviour*: how many
+findings, how severe, how much they cost. None of them can say whether a
+finding was any good. With no human labels in `feedback.db`, a judge is the
+only thing that can — so these two ask the questions that need no ground truth,
+only the review itself:
+
+  `finding_actionability` — is each finding concrete enough to act on? A
+    reviewer that says "consider improving error handling" at file level is
+    indistinguishable from a useful one by finding count alone. This is the
+    failure mode a cheap model degrades into first.
+
+  `review_self_consistency` — does the summary agree with the findings it
+    posted? Claiming "no issues found" above a list of two criticals, or
+    describing a problem in prose that never became a finding, is a defect the
+    reviewer can commit entirely on its own.
+
+Neither judge is asked whether a finding is *correct*. That needs the diff,
+which these traces do not carry, and a judge asked to rule on correctness from
+a summary alone will confabulate. Accuracy stays an open question until humans
+start labelling — which is what `feedback_scores.py` is there to capture.
+
+**The judge is a different model from the reviewer.** The reviewer runs
+MiniMax-M2.7; the judge runs kimi-k2.7-code through the same headroom hub. A
+model grading its own output agrees with itself for reasons that have nothing
+to do with quality.
+
+Evaluators score *observations*, and their variable mapping reads the
+observation's own input/output — which is why `langfuse_trace` now writes the
+review onto the generation and not just onto the trace.
+
+Usage:
+    LANGFUSE_HOST=... LANGFUSE_PUBLIC_KEY=... LANGFUSE_SECRET_KEY=... \\
+        python3 eval_judges.py --dry-run
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import eval_bootstrap as eb  # noqa: E402
+
+# The headroom hub in front of the local Ollama, plus a small pass-through
+# proxy (`judge-proxy` on 8802) that patches every `thinking` content block
+# to carry the `signature` field Langfuse's Anthropic adapter requires. The
+# underlying model is kimi-k2.7-code through the hub on 8790; the proxy fixes
+# the shape so Mastra's Zod parse stops failing.
+JUDGE_PROVIDER = "headroom-ollama"
+JUDGE_BASE_URL = os.environ.get("PRAGENT_JUDGE_BASE_URL", "http://100.74.17.70:8802")
+JUDGE_API_KEY = os.environ.get("PRAGENT_JUDGE_API_KEY", "ollama")
+JUDGE_MODEL = os.environ.get("PRAGENT_JUDGE_MODEL", "kimi-k2.7-code:cloud")
+
+# The trace names this project emits (`pr-review` on the trace, `opencode-review`
+# on the generation). Filter on `traceName` rather than observation `name` — the
+# observation-rule schema only exposes `traceName` as a stringOptions column, and
+# every observation inside these traces is the review itself, so the narrowness
+# is the same.
+REVIEW_TRACE_NAMES = ["pr-review", "opencode-review"]
+
+
+def _model_config() -> dict:
+    return {"provider": JUDGE_PROVIDER, "model": JUDGE_MODEL}
+
+
+JUDGES = [
+    {
+        "name": "finding_actionability",
+        "prompt": (
+            "You are auditing the output of an automated code reviewer.\n\n"
+            "PR under review:\n{{input}}\n\n"
+            "What the reviewer produced:\n{{output}}\n\n"
+            "Rate how ACTIONABLE the findings are, from 0 to 1. A finding is "
+            "actionable when a developer could act on it without asking a "
+            "follow-up question: it points at a specific location, names a "
+            "concrete problem, and proposes a fix that could be applied.\n\n"
+            "Score 1.0 when every finding is specific and fixable. Score around "
+            "0.5 when findings identify a real area but leave the developer to "
+            "work out what to change. Score near 0.0 when findings are generic "
+            "advice that would apply to almost any pull request.\n\n"
+            "Judge only specificity and actionability. You cannot see the diff, "
+            "so do NOT attempt to judge whether a finding is factually correct, "
+            "and do not penalise a finding for being one you cannot verify.\n\n"
+            "If the reviewer reported no findings at all, return 1.0 and say in "
+            "your reasoning that there was nothing to judge — a silent review is "
+            "measured by finding_rate, not here."
+        ),
+        "outputDefinition": {
+            "dataType": "NUMERIC",
+            "minValue": 0,
+            "maxValue": 1,
+            "reasoning": {
+                "description": (
+                    "Name the least actionable finding and say what it would "
+                    "need in order to be acted on."
+                )
+            },
+            "score": {"description": "0 = generic advice, 1 = every finding is specific and fixable."},
+        },
+    },
+    {
+        "name": "review_self_consistency",
+        "prompt": (
+            "You are auditing the output of an automated code reviewer.\n\n"
+            "PR under review:\n{{input}}\n\n"
+            "What the reviewer produced:\n{{output}}\n\n"
+            "The output contains a prose `summary` and a list of `findings`. "
+            "Decide whether the summary is CONSISTENT with the findings.\n\n"
+            "Inconsistent means, for example: the summary says no issues were "
+            "found while findings are listed; the summary describes a problem "
+            "that never became a finding; the summary characterises the severity "
+            "of the findings in a way the findings themselves contradict; or the "
+            "summary refers to files that appear in no finding and in no part of "
+            "the PR description.\n\n"
+            "A summary that adds context beyond the findings is NOT inconsistent "
+            "as long as nothing in it contradicts them. A review that found "
+            "nothing and says so is consistent.\n\n"
+            "You cannot see the diff. Judge the summary against the findings and "
+            "the PR title only — never against what you imagine the code does."
+        ),
+        "outputDefinition": {
+            "dataType": "BOOLEAN",
+            "reasoning": {
+                "description": "Quote the part of the summary that conflicts with the findings, if any."
+            },
+            "score": {"description": "true = summary agrees with the findings, false = it contradicts them."},
+        },
+    },
+]
+
+# Both judges read the observation's own input/output.
+MAPPING = [
+    {"variable": "input", "source": "input"},
+    {"variable": "output", "source": "output"},
+]
+
+
+# ---------------------------------------------------------------------------
+# LLM connection
+# ---------------------------------------------------------------------------
+
+def ensure_llm_connection() -> dict:
+    """Point the project at the judge model. Upserted on `provider`."""
+    body = {
+        "provider": JUDGE_PROVIDER,
+        "adapter": "anthropic",
+        "baseURL": JUDGE_BASE_URL,
+        "secretKey": JUDGE_API_KEY,
+        "customModels": [JUDGE_MODEL],
+        # The hub serves two local models and none of Anthropic's, so the
+        # default catalogue would be a list of models that all fail on use.
+        "withDefaultModels": False,
+    }
+    st, resp = eb._call("PUT", "/api/public/llm-connections", body)
+    return {"status": st, "ok": st in (200, 201), "provider": JUDGE_PROVIDER,
+            "error": None if st in (200, 201) else resp}
+
+
+# ---------------------------------------------------------------------------
+# Evaluators
+# ---------------------------------------------------------------------------
+
+def existing_evaluators() -> dict[str, str]:
+    """name -> id for evaluators already in the project."""
+    out: dict[str, str] = {}
+    st, body = eb._call("GET", "/api/public/unstable/evaluators?limit=100")
+    if st == 200 and isinstance(body, dict):
+        for ev in body.get("data") or []:
+            out[ev.get("name")] = ev.get("id")
+    return out
+
+
+def ensure_evaluators() -> dict:
+    """Create each judge if no version exists for the name yet.
+
+    POST /evaluators with a name that already exists creates a new version, not
+    a no-op — re-running this script would pile up versions until the page
+    listing them is unreadable. Skip when an evaluator of that name is present.
+    """
+    created, skipped, failed = {}, [], []
+    existing = set(existing_evaluators())
+    for judge in JUDGES:
+        if judge["name"] in existing:
+            skipped.append(judge["name"])
+            continue
+        body = {
+            "type": "llm_as_judge",
+            "name": judge["name"],
+            "prompt": judge["prompt"],
+            "outputDefinition": judge["outputDefinition"],
+            "modelConfig": _model_config(),
+        }
+        st, resp = eb._call("POST", "/api/public/unstable/evaluators", body, timeout=60.0)
+        if st in (200, 201) and isinstance(resp, dict):
+            created[judge["name"]] = resp.get("id")
+        else:
+            failed.append({"name": judge["name"], "status": st, "error": resp})
+    return {"created": created, "skipped": skipped, "failed": failed}
+
+
+# ---------------------------------------------------------------------------
+# Rules — what gets judged, and how often
+# ---------------------------------------------------------------------------
+
+def rule_body(name: str, judge_name: str, sampling: float) -> dict:
+    """POST /evaluation-rules shape for an LLM-as-judge observation rule.
+
+    The judge is referenced by `name`+`scope`, not by id — ids name specific
+    versions, names name the evaluator across versions. Mapping is required at
+    both the rule root (the server validates it there) and inside `evaluator`
+    (the API echoes it back). Filter is on `traceName` because that is the only
+    stringOptions column the observation-rule schema exposes.
+    """
+    return {
+        "name": name,
+        "enabled": True,
+        "target": "observation",
+        "sampling": sampling,
+        "filter": [
+            {"column": "traceName", "operator": "any of",
+             "value": REVIEW_TRACE_NAMES, "type": "stringOptions"},
+        ],
+        "evaluator": {
+            "name": judge_name,
+            "scope": "project",
+            "variableMapping": MAPPING,
+        },
+        "mapping": MAPPING,
+    }
+
+
+def ensure_rules(evaluator_ids: dict[str, str], sampling: float) -> dict:
+    """Idempotent: existing rules with the same name are skipped, not duplicated.
+
+    The API has no `name`-keyed upsert; the convention is to POST once and
+    re-run the script to verify the response. A duplicate POST raises 409.
+    """
+    created, failed, skipped = [], [], []
+    existing = existing_rule_names()
+    for name, eid in evaluator_ids.items():
+        if not eid:
+            continue
+        rule_name = f"{name}-on-reviews"
+        if rule_name in existing:
+            skipped.append(name)
+            continue
+        st, resp = eb._call(
+            "POST", "/api/public/unstable/evaluation-rules",
+            rule_body(rule_name, name, sampling), timeout=60.0,
+        )
+        if st in (200, 201):
+            created.append(name)
+        else:
+            failed.append({"rule": name, "status": st, "error": resp})
+    return {"created": created, "failed": failed, "skipped": skipped}
+
+
+def existing_rule_names() -> set[str]:
+    """Names of observation-target rules already in the project."""
+    out: set[str] = set()
+    st, body = eb._call("GET", "/api/public/unstable/evaluation-rules?limit=100")
+    if st == 200 and isinstance(body, dict):
+        for r in body.get("data") or []:
+            if r.get("target") == "observation":
+                out.add(r.get("name"))
+    return out
+
+
+def main(argv: list[str] | None = None) -> int:
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument("--sampling", type=float, default=1.0,
+                    help="fraction of matching observations to judge (default: all)")
+    ap.add_argument("--skip-connection", action="store_true")
+    ap.add_argument("--dry-run", action="store_true")
+    args = ap.parse_args(argv)
+
+    if args.dry_run:
+        print(json.dumps({
+            "would_connect": {"provider": JUDGE_PROVIDER, "baseURL": JUDGE_BASE_URL,
+                              "model": JUDGE_MODEL},
+            "would_create": [j["name"] for j in JUDGES],
+            "existing_evaluators": sorted(existing_evaluators()),
+            "sampling": args.sampling,
+        }, indent=2))
+        return 0
+
+    report = {}
+    if not args.skip_connection:
+        report["llm_connection"] = ensure_llm_connection()
+    report["evaluators"] = ensure_evaluators()
+    ids = dict(report["evaluators"]["created"])
+    # Fall back to whatever is already registered, so a re-run still wires rules.
+    for name, eid in existing_evaluators().items():
+        ids.setdefault(name, eid)
+    report["rules"] = ensure_rules(
+        {j["name"]: ids.get(j["name"]) for j in JUDGES}, args.sampling
+    )
+    print(json.dumps(report, indent=2))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

pilot/eval_scores.py

@@ -0,0 +1,233 @@
+#!/usr/bin/env python3
+"""pragent pilot — deterministic review scorers.
+
+Four numbers computed from a review that already happened, shipped to Langfuse
+as scores on the review's trace. All are derived from data the reviewer already
+has in hand: no LLM judge, no ground truth, no extra token spend.
+
+Why these four and not `helpfulness`/`quality`
+----------------------------------------------
+They come from what the recorded reviews actually did, not from a generic eval
+checklist:
+
+  * `severity_info_ratio` — of the findings ever posted to a PR, effectively all
+    landed at `info`. Either the model will not commit to a severity or the
+    per-repo `severity_threshold` is filtering the rest out. Trending the ratio
+    per model says which.
+  * `finding_rate` — most reviews post nothing at all. Silence on clean code is
+    the goal; silence because the run degraded is a failure. Same output, two
+    causes, and only the rate over time separates them.
+  * `dropped_findings` — `ai_review.parse_findings` discards any finding whose
+    `path`/`line` is unusable. That happens silently, so a model that emits ten
+    findings at invalid locations is indistinguishable from one that found
+    nothing. This is the only signal here that measures the *model's* output
+    rather than the review's.
+  * `cost_per_finding` — the equivalent-cost number is already trended per
+    review; per finding is what actually compares two models, since a cheaper
+    model that finds nothing is not cheaper.
+
+None of these say whether a finding was *correct*. That needs labels, and the
+labels come from `feedback_scores.py` once maintainers start reacting to review
+comments. Read these as behavioural drift detectors, not as accuracy.
+
+Fail-open, like every other telemetry path here: a scorer that raises returns no
+score rather than failing the review.
+"""
+from __future__ import annotations
+
+import uuid
+from datetime import datetime, timezone
+
+# Mirrors ai_review.SEVERITY_RANK. Duplicated rather than imported because this
+# module is also run standalone (backfill) where ai_review's import side effects
+# are unwanted.
+SEVERITY_RANK = {"info": -1, "trivial": 0, "low": 1, "medium": 2, "high": 3, "critical": 4}
+
+# Findings at or below this rank are "the model declined to commit". `trivial`
+# and `info` are advisory by the reviewer's own prompt contract.
+_ADVISORY_MAX_RANK = 0
+
+# Score names. Named for what is measured, not for the mechanism producing it —
+# these land on every trace and become the axis of every chart.
+FINDING_RATE = "finding_rate"
+SEVERITY_INFO_RATIO = "severity_info_ratio"
+SEVERITY_MAX = "severity_max"
+DROPPED_FINDINGS = "dropped_findings"
+COST_PER_FINDING = "cost_per_finding"
+
+
+def _sev(f: dict) -> str:
+    return str(f.get("severity") or "medium").strip().lower()
+
+
+def finding_rate(findings: list[dict] | None) -> float:
+    """How many findings this review posted. 0.0 is the restraint case."""
+    return float(len(findings or []))
+
+
+def severity_info_ratio(findings: list[dict] | None) -> float | None:
+    """Share of findings the model rated advisory (`info`/`trivial`).
+
+    `None` for a review with no findings — a ratio over an empty set is not 0,
+    it is undefined, and charting it as 0 would read as "perfectly calibrated".
+    """
+    fs = findings or []
+    if not fs:
+        return None
+    advisory = sum(1 for f in fs if SEVERITY_RANK.get(_sev(f), 2) <= _ADVISORY_MAX_RANK)
+    return round(advisory / len(fs), 4)
+
+
+def severity_max(findings: list[dict] | None) -> str:
+    """Highest severity present, or `none` when the review was silent.
+
+    Categorical on purpose: the useful question is "did this review ever surface
+    something serious", and an average of severity ranks answers nothing.
+    """
+    fs = findings or []
+    if not fs:
+        return "none"
+    top = max(fs, key=lambda f: SEVERITY_RANK.get(_sev(f), 2))
+    sev = _sev(top)
+    return sev if sev in SEVERITY_RANK else "medium"
+
+
+def dropped_findings(raw_count: int | None, kept_count: int | None) -> float | None:
+    """Findings the model emitted that the parser could not use.
+
+    `raw_count` is what came back in the JSON; `kept_count` is what survived
+    `_normalize_finding`. `None` when the caller could not determine the raw
+    count — better no score than a fabricated zero.
+    """
+    if raw_count is None or kept_count is None:
+        return None
+    return float(max(0, int(raw_count) - int(kept_count)))
+
+
+def cost_per_finding(cost_usd: float | None, findings: list[dict] | None) -> float | None:
+    """Equivalent USD spent per finding posted.
+
+    `None` when nothing could be priced. A silent review divides by one, not by
+    zero: the run still cost money, and attributing that whole cost to "found
+    nothing" is the honest reading.
+    """
+    if cost_usd is None:
+        return None
+    try:
+        c = float(cost_usd)
+    except (TypeError, ValueError):
+        return None
+    return round(c / max(1, len(findings or [])), 6)
+
+
+def build_scores(
+    *,
+    trace_id: str,
+    findings: list[dict] | None,
+    environment: str,
+    cost_usd: float | None = None,
+    dropped_count: float | None = None,
+    timestamp: str | None = None,
+    comment: str = "",
+) -> list[dict]:
+    """The `score-create` ingestion events for one review.
+
+    `dropped_count` must be measured at parse time, not here: by the time
+    `findings` reaches this function the per-repo config has already filtered it
+    by severity threshold and `max_findings`, and those drops are the config
+    working as intended, not the model emitting garbage.
+
+    Returns [] rather than raising if something is unscoreable — scores are
+    telemetry and must never cost a review.
+    """
+    # The ingestion envelope requires a timestamp on every event; omitting it
+    # gets the whole batch rejected with an HTTP 207 whose per-event 400s are
+    # easy to mistake for success.
+    ts = timestamp or datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
+    out: list[dict] = []
+
+    def add(name: str, value, data_type: str) -> None:
+        if value is None:
+            return
+        body = {
+            "id": str(uuid.uuid4()),
+            "traceId": trace_id,
+            "name": name,
+            "dataType": data_type,
+            "environment": environment,
+        }
+        if data_type == "CATEGORICAL":
+            body["value"] = str(value)
+        else:
+            body["value"] = float(value)
+        if comment:
+            body["comment"] = comment
+        out.append(
+            {
+                "id": str(uuid.uuid4()),
+                "type": "score-create",
+                "timestamp": ts,
+                "body": body,
+            }
+        )
+
+    try:
+        add(FINDING_RATE, finding_rate(findings), "NUMERIC")
+        add(SEVERITY_INFO_RATIO, severity_info_ratio(findings), "NUMERIC")
+        add(SEVERITY_MAX, severity_max(findings), "CATEGORICAL")
+        add(DROPPED_FINDINGS, dropped_count, "NUMERIC")
+        add(COST_PER_FINDING, cost_per_finding(cost_usd, findings), "NUMERIC")
+    except Exception:  # pragma: no cover - defensive
+        return out
+    return out
+
+
+# ---------------------------------------------------------------------------
+# Score configs — the schema these scores must comply with
+# ---------------------------------------------------------------------------
+
+# Registered once per project via `eval_bootstrap.py`. Without configs the
+# scores still ingest, but nothing constrains a future scorer from writing
+# `severity_max="HIGH"` next to today's `"high"` and silently splitting the
+# series in two.
+SCORE_CONFIGS = [
+    {
+        "name": FINDING_RATE,
+        "dataType": "NUMERIC",
+        "minValue": 0,
+        "description": "Findings posted by one review. 0 = the reviewer stayed silent.",
+    },
+    {
+        "name": SEVERITY_INFO_RATIO,
+        "dataType": "NUMERIC",
+        "minValue": 0,
+        "maxValue": 1,
+        "description": "Share of a review's findings rated info/trivial. High = the model is not committing to a severity.",
+    },
+    {
+        "name": SEVERITY_MAX,
+        "dataType": "CATEGORICAL",
+        "categories": [
+            {"label": "none", "value": 0},
+            {"label": "info", "value": 1},
+            {"label": "trivial", "value": 2},
+            {"label": "low", "value": 3},
+            {"label": "medium", "value": 4},
+            {"label": "high", "value": 5},
+            {"label": "critical", "value": 6},
+        ],
+        "description": "Highest severity surfaced by one review; 'none' when it posted nothing.",
+    },
+    {
+        "name": DROPPED_FINDINGS,
+        "dataType": "NUMERIC",
+        "minValue": 0,
+        "description": "Findings the model emitted that the parser rejected for an unusable path/line.",
+    },
+    {
+        "name": COST_PER_FINDING,
+        "dataType": "NUMERIC",
+        "minValue": 0,
+        "description": "Equivalent USD per finding posted. Silent reviews divide by 1, not 0.",
+    },
+]

pilot/feedback_scores.py

@@ -0,0 +1,247 @@
+#!/usr/bin/env python3
+"""pragent pilot — feedback DB to Langfuse scores.
+
+`feedback.db` already records every reaction, thread resolution and reply a
+maintainer leaves on a bot comment. That is the only ground truth pragent has
+about whether a finding was any good, and until now it went to a markdown report
+nobody reads and nowhere else. This ships it to Langfuse as session-level
+scores, so "was the reviewer right" sits on the same axis as "what did it cost".
+
+Session, not trace
+------------------
+`langfuse_trace` sets `sessionId` to `"{repo}#{pr}"` and lets the trace id be a
+fresh uuid per review. Feedback arrives days later against a PR, not against one
+particular re-run of the reviewer, and nothing in `feedback.db` records which
+trace produced which comment. Scoring the session is therefore both the
+available join and the honest granularity: this is feedback on the review of
+this PR, not on one invocation.
+
+Two scores, deliberately separated
+----------------------------------
+  * `review_engagement` — the share of a PR's findings that got any human
+    response at all. This is a signal about the *feedback loop*, not the
+    reviewer: at the time of writing it is 0.0 across all 113 recorded reviews,
+    which is exactly the fact that makes an accuracy metric impossible today.
+    It must be watched first, because every other quality number is vapour
+    until it moves.
+  * `review_acceptance` — net verdict over the findings that *did* get a
+    response: (upvotes + resolved) - (downvotes + negation replies), normalised
+    to -1..1. Computed only over engaged findings, so an ignored review scores
+    `None` rather than 0. Zero would read as "humans judged this exactly
+    neutral"; the truth is nobody looked.
+
+Fail-open and idempotent. Score ids are derived from (repo, pr, name) so a
+re-run overwrites rather than duplicates.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sqlite3
+import sys
+import uuid
+from datetime import datetime, timezone
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+from feedback_harvest import classify_reaction, _is_negation_reply  # noqa: E402
+
+REVIEW_ENGAGEMENT = "review_engagement"
+REVIEW_ACCEPTANCE = "review_acceptance"
+
+# Stable namespace so the same (repo, pr, score) always produces the same score
+# id — Langfuse treats a repeated id as an update, which is what a backfill of a
+# still-accumulating PR should do.
+_NS = uuid.UUID("6f1d9c2e-4a77-4f2a-9c1a-0d3b5e8a7c41")
+
+
+def _score_id(repo: str, pr: int, name: str) -> str:
+    return str(uuid.uuid5(_NS, f"{repo}#{pr}#{name}"))
+
+
+def collect_pr_feedback(conn: sqlite3.Connection, repo: str, pr: int) -> dict:
+    """Tally one PR's findings and the human responses attached to them.
+
+    Returns counts only — the scoring maths lives in `score_pr` so it can be
+    tested without a database.
+    """
+    rows = conn.execute(
+        "SELECT id, comment_id FROM inline_finding WHERE repo = ? AND pr = ?",
+        (repo, pr),
+    ).fetchall()
+    total = len(rows)
+    engaged = 0
+    positive = 0
+    negative = 0
+
+    for row in rows:
+        fid = row["id"] if isinstance(row, sqlite3.Row) else row[0]
+        cid = row["comment_id"] if isinstance(row, sqlite3.Row) else row[1]
+        pos = neg = 0
+
+        if cid is not None:
+            for r in conn.execute(
+                "SELECT content FROM reaction WHERE comment_id = ?", (cid,)
+            ):
+                kind = classify_reaction(r[0])
+                if kind == "positive":
+                    pos += 1
+                elif kind == "negative":
+                    neg += 1
+
+        for r in conn.execute(
+            "SELECT resolved FROM thread_state WHERE finding_id = ?", (fid,)
+        ):
+            # A resolved thread means the maintainer acted on the finding.
+            if r[0]:
+                pos += 1
+
+        # A reply counts as engagement either way; only a negation phrase makes
+        # it a vote against. A neutral reply ("done", "good catch, but…") is
+        # deliberately not a positive vote — it says someone looked, not that
+        # they agreed.
+        replied = 0
+        for r in conn.execute(
+            "SELECT body FROM reply WHERE finding_id = ?", (fid,)
+        ):
+            replied += 1
+            if _is_negation_reply(r[0]):
+                neg += 1
+
+        if pos or neg or replied:
+            engaged += 1
+        positive += pos
+        negative += neg
+
+    return {"total": total, "engaged": engaged, "positive": positive, "negative": negative}
+
+
+def score_pr(tally: dict) -> dict:
+    """Turn one PR's tally into score values.
+
+    `review_acceptance` is `None` when nothing was engaged — see the module
+    docstring on why that is not 0.
+    """
+    total = int(tally.get("total") or 0)
+    engaged = int(tally.get("engaged") or 0)
+    pos = int(tally.get("positive") or 0)
+    neg = int(tally.get("negative") or 0)
+
+    engagement = round(engaged / total, 4) if total else None
+    acceptance = None
+    if pos or neg:
+        acceptance = round((pos - neg) / (pos + neg), 4)
+    return {REVIEW_ENGAGEMENT: engagement, REVIEW_ACCEPTANCE: acceptance}
+
+
+def build_score_events(
+    repo: str, pr: int, values: dict, environment: str = "default",
+    timestamp: str | None = None,
+) -> list[dict]:
+    """`score-create` events for one PR's feedback.
+
+    Every event carries a timestamp: the ingestion endpoint rejects those that
+    do not, and it reports the rejection as a per-event 400 inside an HTTP 207,
+    which reads as success to a caller that only checks the status code.
+    """
+    ts = timestamp or datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
+    events = []
+    for name, value in values.items():
+        if value is None:
+            continue
+        events.append(
+            {
+                "id": str(uuid.uuid4()),
+                "type": "score-create",
+                "timestamp": ts,
+                "body": {
+                    "id": _score_id(repo, pr, name),
+                    "sessionId": f"{repo}#{pr}",
+                    "name": name,
+                    "value": float(value),
+                    "dataType": "NUMERIC",
+                    "environment": environment,
+                    "comment": f"from feedback.db · {repo}#{pr}",
+                },
+            }
+        )
+    return events
+
+
+SCORE_CONFIGS = [
+    {
+        "name": REVIEW_ENGAGEMENT,
+        "dataType": "NUMERIC",
+        "minValue": 0,
+        "maxValue": 1,
+        "description": "Share of a PR's findings that drew any human reaction, resolution or reply. 0 = nobody engaged with the review.",
+    },
+    {
+        "name": REVIEW_ACCEPTANCE,
+        "dataType": "NUMERIC",
+        "minValue": -1,
+        "maxValue": 1,
+        "description": "Net human verdict over engaged findings: +1 all accepted, -1 all rejected. Absent when nothing was engaged.",
+    },
+]
+
+
+def iter_prs(conn: sqlite3.Connection):
+    for row in conn.execute(
+        "SELECT DISTINCT repo, pr FROM inline_finding ORDER BY repo, pr"
+    ):
+        yield row[0], int(row[1])
+
+
+def backfill(db_path: str, *, environment: str = "default", dry_run: bool = False) -> dict:
+    """Score every PR in the feedback DB. Returns a summary dict."""
+    conn = sqlite3.connect(db_path)
+    conn.row_factory = sqlite3.Row
+    events: list[dict] = []
+    scanned = 0
+    engaged_prs = 0
+    try:
+        for repo, pr in iter_prs(conn):
+            scanned += 1
+            tally = collect_pr_feedback(conn, repo, pr)
+            values = score_pr(tally)
+            if (values.get(REVIEW_ENGAGEMENT) or 0) > 0:
+                engaged_prs += 1
+            events.extend(build_score_events(repo, pr, values, environment))
+    finally:
+        conn.close()
+
+    summary = {"prs_scanned": scanned, "prs_with_engagement": engaged_prs, "scores": len(events)}
+    if dry_run or not events:
+        summary["posted"] = False
+        return summary
+
+    import langfuse_trace
+
+    conf = langfuse_trace._enabled()
+    if conf is None:
+        summary["posted"] = False
+        summary["error"] = "Langfuse not configured (LANGFUSE_HOST / keys unset)"
+        return summary
+    host, pk, sk = conf
+    status = langfuse_trace._post(host, pk, sk, events, 15.0)
+    summary["posted"] = status in (200, 201, 207)
+    summary["http_status"] = status
+    return summary
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description="Ship feedback.db verdicts to Langfuse as scores")
+    ap.add_argument("--db", default=os.environ.get("PRAGENT_FEEDBACK_DB", "/data/feedback.db"))
+    ap.add_argument("--environment", default="default")
+    ap.add_argument("--dry-run", action="store_true")
+    args = ap.parse_args()
+
+    summary = backfill(args.db, environment=args.environment, dry_run=args.dry_run)
+    print(json.dumps(summary, indent=2))
+    return 0 if summary.get("posted") or args.dry_run else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

pilot/langfuse_trace.py

@@ -0,0 +1,468 @@
+#!/usr/bin/env python3
+"""pragent pilot — Langfuse trace emission.
+
+Ships one trace per PR review to a self-hosted Langfuse (v3) so the reviewer's
+token spend, latency and per-model behaviour are queryable outside the review
+body. The review body already renders a usage table; that table is per-PR and
+disappears into Gitea. This is the same numbers, aggregated.
+
+Why hand-rolled instead of the `langfuse` SDK: the pilot image is stdlib-only
+(see pilot/Dockerfile — no requirements.txt anywhere in the repo), and the
+ingestion API is a single authenticated POST of a JSON batch. Pulling an SDK
+plus its otel dependency tree into a fail-open telemetry side-path is a bad
+trade.
+
+Provider split
+--------------
+`environment` on every trace is either `ollama` or `claude`, derived from the
+resolved display model (`resolve_environment`). That is what keeps the two
+spend stories separate in Langfuse: every dashboard, filter and cost breakdown
+takes an environment selector, so "what did the local/self-hosted path cost"
+and "what did the Claude path cost" are two views of one project rather than
+two projects with two key pairs to rotate. Tags carry the finer split
+(`provider:headroom`, `model:...`, `engine:opencode`).
+
+Cost
+----
+The pilot's own path bills $0 (headroom proxy, no per-token charge), so the
+`cost` reported to Langfuse is the *equivalent* cost from `cost_model` — what
+the same tokens would bill on the comparison model. That is the number worth
+trending; a chart of $0.00 is not.
+
+A model is "free" when `cost_model.PRICES` has no entry for it (MiniMax-M2.7,
+glm-5.2:cloud) or when its entry is all zeros (the self-hosted vLLM qwen). In
+both cases the reported cost is priced against the comparison target instead —
+same precedence the review body uses: `.pr-review.json:cost_target` >
+`PRAGENT_PRICE_TARGET` > `claude-sonnet-5`. A paid model is priced as itself.
+
+Because a hypothetical and a real charge must never be read as the same
+number, every trace is tagged `cost:actual` or `cost:equivalent:<target>`, and
+the generation's metadata carries `cost_basis`.
+
+Fail-open: every entry point swallows its own exceptions. Telemetry must never
+cost a review.
+
+Env:
+  LANGFUSE_HOST         e.g. http://langfuse-web.langfuse.svc.cluster.local:3000
+  LANGFUSE_PUBLIC_KEY   pk-lf-...
+  LANGFUSE_SECRET_KEY   sk-lf-...
+  LANGFUSE_TIMEOUT      seconds, default 5
+  LANGFUSE_DEBUG        1 to log ingestion failures to stderr
+Disabled (silently) when host or either key is unset.
+"""
+
+from __future__ import annotations
+
+import base64
+import json
+import os
+import sys
+import time
+import urllib.error
+import urllib.request
+import uuid
+from datetime import datetime, timezone
+
+INGESTION_PATH = "/api/public/ingestion"
+
+# Model-key prefixes that mean "this review ran against Anthropic-shaped
+# billing". Everything else (glm, MiniMax, qwen, local vLLM) is the ollama /
+# self-hosted side of the split.
+_CLAUDE_PREFIXES = ("claude-", "anthropic/")
+
+
+def _now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
+
+
+def _enabled() -> tuple[str, str, str] | None:
+    host = (os.environ.get("LANGFUSE_HOST") or "").strip().rstrip("/")
+    pk = (os.environ.get("LANGFUSE_PUBLIC_KEY") or "").strip()
+    sk = (os.environ.get("LANGFUSE_SECRET_KEY") or "").strip()
+    if not host or not pk or not sk:
+        return None
+    return host, pk, sk
+
+
+def _debug(msg: str) -> None:
+    if os.environ.get("LANGFUSE_DEBUG"):
+        print(f"pragent/langfuse: {msg}", file=sys.stderr, flush=True)
+
+
+def strip_provider(model: str) -> str:
+    """`headroom/claude-sonnet-5` -> `claude-sonnet-5`. Bare names pass through."""
+    return model.split("/", 1)[1] if "/" in model else model
+
+
+def provider_of(model: str) -> str:
+    """The opencode provider block a display model routes through."""
+    return model.split("/", 1)[0] if "/" in model else "headroom"
+
+
+def resolve_environment(model: str) -> str:
+    """Which spend story this review belongs to: `claude` or `ollama`.
+
+    Keyed off the bare model name, not the provider, because both paths route
+    through the same `headroom` proxy — `headroom/claude-sonnet-5` is Claude
+    spend, `headroom/glm-5.2:cloud` is not.
+    """
+    bare = strip_provider(model).lower()
+    return "claude" if bare.startswith(_CLAUDE_PREFIXES) else "ollama"
+
+
+def _usage_details(usage: dict) -> dict:
+    """opencode's usage dict -> Langfuse `usageDetails`.
+
+    Langfuse sums every key except the ones it knows are derived, so `input`
+    here is the *uncached* portion: reporting both `input` (which opencode
+    reports as the full input, cache included) and `cache_read_input_tokens`
+    would double-count.
+    """
+    inp = int(usage.get("input") or 0)
+    cache_read = int(usage.get("cache_read") or 0)
+    cache_write = int(usage.get("cache_write") or 0)
+    details = {
+        "input": max(0, inp - cache_read),
+        "output": int(usage.get("output") or 0),
+    }
+    if cache_read:
+        details["cache_read_input_tokens"] = cache_read
+    if cache_write:
+        details["cache_write_input_tokens"] = cache_write
+    reasoning = int(usage.get("reasoning") or 0)
+    if reasoning:
+        details["reasoning"] = reasoning
+    return details
+
+
+DEFAULT_PRICE_TARGET = "claude-sonnet-5"
+
+
+def resolve_price_target(price_target: str | None = None) -> str:
+    """The model to price free/unknown runs against.
+
+    Mirrors `ai_review._resolve_price_target`: an explicit target (which the
+    caller reads from `.pr-review.json:cost_target`) wins, then
+    `PRAGENT_PRICE_TARGET`, then Sonnet.
+    """
+    if price_target and price_target.strip():
+        return price_target.strip()
+    env = os.environ.get("PRAGENT_PRICE_TARGET", "").strip()
+    return env or DEFAULT_PRICE_TARGET
+
+
+def _is_free(price) -> bool:
+    """A price entry that charges nothing — self-hosted or proxied at no cost."""
+    return price.input == 0 and price.output == 0
+
+
+def _cost_details(usage: dict, model: str, price_target: str | None = None) -> tuple[dict, str]:
+    """USD for this usage plus the basis it was computed on.
+
+    Returns `({"total": …}, basis)` where basis is `actual` for a model that
+    genuinely bills, or `equivalent:<target>` for one that does not. `({}, "")`
+    when nothing can be priced at all — better no number than a wrong one.
+
+    Local import + broad except: `cost_model` is only present on the opencode
+    path, and an unknown model key must not break telemetry.
+    """
+    try:
+        from cost_model import PRICES, Usage, cost
+
+        bare = strip_provider(model)
+        price = PRICES.get(bare)
+        basis = "actual"
+        if price is None or _is_free(price):
+            # MiniMax / glm / self-hosted qwen: $0 through the proxy, so the
+            # useful number is what these tokens would have billed elsewhere.
+            target = resolve_price_target(price_target)
+            price = PRICES.get(target)
+            if price is None:
+                _debug(f"comparison target {target!r} not in PRICES")
+                return {}, ""
+            basis = f"equivalent:{target}"
+
+        u = Usage(
+            uncached_input=max(0, int(usage.get("input") or 0) - int(usage.get("cache_read") or 0)),
+            cached_input=int(usage.get("cache_read") or 0),
+            cache_writes=int(usage.get("cache_write") or 0),
+            output=int(usage.get("output") or 0),
+        )
+        return {"total": round(cost(u, price), 6)}, basis
+    except Exception as e:  # pragma: no cover - defensive
+        _debug(f"cost lookup failed for {model!r}: {e}")
+        return {}, ""
+
+
+def _severity_counts(findings: list[dict] | None) -> dict:
+    counts: dict[str, int] = {}
+    for f in findings or []:
+        sev = str(f.get("severity") or "unknown").lower()
+        counts[sev] = counts.get(sev, 0) + 1
+    return counts
+
+
+def build_batch(
+    *,
+    repo: str,
+    index: str,
+    sha: str,
+    title: str,
+    model: str,
+    usage: dict | None,
+    findings: list[dict] | None = None,
+    summary: str = "",
+    engine: str = "opencode",
+    tier: str = "",
+    lenses: list[str] | None = None,
+    trace_id: str | None = None,
+    release: str = "",
+    price_target: str | None = None,
+    dropped_count: float | None = None,
+) -> list[dict]:
+    """The ingestion batch for one review: a trace, a generation, and scores.
+
+    Split out from `emit_review_trace` so the shape is testable without a
+    Langfuse to POST to.
+
+    `dropped_count` is how many findings the parser rejected for an unusable
+    `path`/`line`, measured where the model output was parsed. Passing it turns
+    on the `dropped_findings` score; leaving it `None` omits that score rather
+    than reporting a zero the caller never measured.
+    """
+    usage = usage or {}
+    tid = trace_id or str(uuid.uuid4())
+    ts = _now_iso()
+    env = resolve_environment(model)
+    duration = float(usage.get("duration_s") or 0.0)
+    started = datetime.fromtimestamp(
+        time.time() - duration, tz=timezone.utc
+    ).isoformat().replace("+00:00", "Z")
+
+    tags = [
+        f"provider:{provider_of(model)}",
+        f"model:{strip_provider(model)}",
+        f"engine:{engine}",
+        f"repo:{repo}",
+    ]
+    if tier:
+        tags.append(f"tier:{tier}")
+    for lens in lenses or []:
+        tags.append(f"lens:{lens}")
+
+    costs, cost_basis = _cost_details(usage, model, price_target) if usage else ({}, "")
+    if cost_basis:
+        # Filterable in Langfuse, so an equivalent-cost chart can never be
+        # mistaken for money actually spent.
+        tags.append(f"cost:{cost_basis}")
+
+    metadata = {
+        "repo": repo,
+        "pr": index,
+        "sha": sha,
+        "engine": engine,
+        "steps": usage.get("steps"),
+        "duration_s": duration or None,
+        "findings": len(findings or []),
+        "severities": _severity_counts(findings),
+        "provider_cost_usd": usage.get("cost"),
+        "cost_basis": cost_basis or None,
+    }
+    if lenses:
+        metadata["lenses"] = lenses
+    if tier:
+        metadata["tier"] = tier
+    metadata = {k: v for k, v in metadata.items() if v not in (None, {}, [])}
+
+    trace_body = {
+        "id": tid,
+        "name": "pr-review",
+        "timestamp": ts,
+        "environment": env,
+        "sessionId": f"{repo}#{index}",
+        "input": _review_input(repo, index, sha, title),
+        "output": _review_output(summary, findings),
+        "metadata": metadata,
+        "tags": tags,
+    }
+    if release:
+        trace_body["release"] = release
+
+    events = [
+        {
+            "id": str(uuid.uuid4()),
+            "type": "trace-create",
+            "timestamp": ts,
+            "body": trace_body,
+        }
+    ]
+
+    if usage:
+        gen_body = {
+            "id": str(uuid.uuid4()),
+            "traceId": tid,
+            "type": "GENERATION",
+            "name": f"{engine}-review",
+            "environment": env,
+            "startTime": started,
+            "endTime": ts,
+            "model": strip_provider(model),
+            "usageDetails": _usage_details(usage),
+            "metadata": metadata,
+            "level": "DEFAULT",
+            # Repeated from the trace on purpose: an evaluator's variable
+            # mapping reads the *observation's* input/output, so a generation
+            # left blank cannot be judged at all.
+            "input": _review_input(repo, index, sha, title),
+            "output": _review_output(summary, findings),
+        }
+        if costs:
+            gen_body["costDetails"] = costs
+        events.append(
+            {
+                "id": str(uuid.uuid4()),
+                "type": "generation-create",
+                "timestamp": ts,
+                "body": gen_body,
+            }
+        )
+
+    events.extend(
+        _score_events(
+            trace_id=tid,
+            findings=findings,
+            environment=env,
+            cost_usd=costs.get("total"),
+            dropped_count=dropped_count,
+            timestamp=ts,
+            cost_basis=cost_basis,
+        )
+    )
+
+    return events
+
+
+MAX_JUDGED_FINDINGS = 25
+_FIELD_CAP = 600
+
+
+def _review_input(repo: str, index, sha: str, title: str) -> dict:
+    return {"repo": repo, "pr": index, "sha": sha, "title": title}
+
+
+def _review_output(summary: str, findings) -> dict:
+    """What the reviewer actually said, in a shape an evaluator can read.
+
+    The findings themselves are included, not just their count. A judge given
+    only `{"summary": ..., "findings": 3}` can say nothing about whether those
+    three findings are specific, actionable, or consistent with the summary —
+    which is the whole question worth asking of a reviewer that has no ground
+    truth to check against.
+
+    Capped rather than complete: this rides in every ingestion batch, and a
+    review with 80 findings would push the payload past what is reasonable to
+    store per trace. `finding_count` stays exact so nothing reading the count
+    is misled by the cap.
+    """
+    items = list(findings or [])
+    return {
+        "summary": summary[:2000],
+        "finding_count": len(items),
+        "findings_truncated": len(items) > MAX_JUDGED_FINDINGS,
+        "findings": [
+            {
+                "path": f.get("path"),
+                "line": f.get("line"),
+                "severity": f.get("severity"),
+                "problem": str(f.get("problem") or "")[:_FIELD_CAP],
+                "fix": str(f.get("fix") or "")[:_FIELD_CAP],
+            }
+            for f in items[:MAX_JUDGED_FINDINGS]
+        ],
+    }
+
+
+def _score_events(*, cost_basis: str, **kwargs) -> list[dict]:
+    """Deterministic scores for this review, or [] if the scorer is missing.
+
+    Local import + blanket except for the same reason the rest of this module
+    swallows: `eval_scores` is optional, and a scoring bug must not cost the
+    trace it was supposed to annotate.
+    """
+    try:
+        import eval_scores
+
+        # The cost score is only meaningful next to its basis — a $/finding
+        # figure computed from an equivalent price is not money that was spent.
+        comment = f"cost basis: {cost_basis}" if cost_basis else ""
+        return eval_scores.build_scores(comment=comment, **kwargs)
+    except Exception as e:  # pragma: no cover - defensive
+        _debug(f"scoring failed: {e}")
+        return []
+
+
+def _post(host: str, pk: str, sk: str, batch: list[dict], timeout: float) -> int:
+    payload = json.dumps({"batch": batch}).encode("utf-8")
+    auth = base64.b64encode(f"{pk}:{sk}".encode("utf-8")).decode("ascii")
+    req = urllib.request.Request(
+        host + INGESTION_PATH,
+        data=payload,
+        headers={
+            "Content-Type": "application/json",
+            "Authorization": f"Basic {auth}",
+            "User-Agent": "pragent-pilot/1.0",
+        },
+        method="POST",
+    )
+    with urllib.request.urlopen(req, timeout=timeout) as resp:
+        _warn_on_rejected_events(resp.read())
+        return resp.status
+
+
+def _warn_on_rejected_events(raw: bytes) -> None:
+    """Surface per-event rejections hiding inside a 207.
+
+    The ingestion endpoint answers 207 Multi-Status when *some* events failed,
+    so a caller that only checks the status code reads a batch where every
+    single event was rejected as a success. That failure mode is invisible
+    exactly when it matters — the traces simply never appear.
+    """
+    try:
+        body = json.loads(raw or b"{}")
+        errors = body.get("errors") or []
+        if errors:
+            first = errors[0]
+            _debug(
+                f"{len(errors)} event(s) rejected by ingestion; "
+                f"first: status={first.get('status')} {first.get('error')}"
+            )
+    except Exception:  # pragma: no cover - never let logging break emission
+        pass
+
+
+def emit_review_trace(**kwargs) -> bool:
+    """Ship one review's trace. Returns True if Langfuse accepted it.
+
+    No-op (False) when Langfuse is unconfigured. Never raises — a telemetry
+    outage must not turn into a failed review.
+    """
+    conf = _enabled()
+    if conf is None:
+        return False
+    host, pk, sk = conf
+    try:
+        timeout = float(os.environ.get("LANGFUSE_TIMEOUT", "5"))
+    except ValueError:
+        timeout = 5.0
+    try:
+        batch = build_batch(**kwargs)
+        status = _post(host, pk, sk, batch, timeout)
+        if status not in (200, 201, 207):
+            _debug(f"ingestion returned HTTP {status}")
+            return False
+        return True
+    except urllib.error.HTTPError as e:
+        _debug(f"ingestion HTTP {e.code}: {e.read()[:300]!r}")
+    except Exception as e:
+        _debug(f"ingestion failed: {e}")
+    return False

pilot/opencode_review.py

@@ -397,7 +397,7 @@ def install_config(src: str, dst: str) -> bool:
     private-network addresses. Real values are supplied at runtime and patched
     in here.
 
-    Env var convention (case-sensitive provider name — `headroom`, `local`):
+    Env var convention (case-sensitive provider name — `headroom`, `vllm-qwen38`):
 
       PRAGENT_<NAME>_BASE_URL   — per-provider endpoint override
       PRAGENT_<NAME>_API_KEY    — per-provider API key override
@@ -406,9 +406,9 @@ def install_config(src: str, dst: str) -> bool:
       PRAGENT_MODEL_API_KEY     — legacy catchall (same)
 
     Per-provider wins over the catchall. The first 2 win when the operator
-    needs a different endpoint per upstream (e.g. headroom → MiniMax, local →
-    ai-workstation). The catchall keeps the single-provider deploys from
-    needing any env config.
+    needs a different endpoint per upstream (e.g. headroom → MiniMax,
+    vllm-qwen38 → ai-workstation). The catchall keeps the single-provider
+    deploys from needing any env config.
 
     This is done in Python rather than with opencode's own `{env:VAR}` config
     templating because the reviewer subprocess runs with an allow-listed

tests/pilot/test_ai_review.py

@@ -462,11 +462,12 @@ def test_resolve_display_model_precedence(monkeypatch):
         ai_review._resolve_display_model("MiniMax-M2.7", {"model": "claude-sonnet-5"})
         == "headroom/claude-sonnet-5"
     )
-    # Self-hosted models carry provider="local" → routes to the `local`
-    # provider block in opencode.json (AI workstation on 192.168.1.79:18020).
+    # Self-hosted models carry provider="vllm-qwen38" → routes to the
+    # matching provider block in opencode.json (AI workstation on
+    # 192.168.1.79:18020).
     assert (
         ai_review._resolve_display_model("MiniMax-M2.7", {"model": "qwen3.8-27b"})
-        == "local/qwen3.8-27b"
+        == "vllm-qwen38/qwen3.8-27b"
     )
     # 3. Env wins over config
     monkeypatch.setenv("OPENCODE_MODEL", "headroom/MiniMax-M2.7")

tests/pilot/test_eval_bootstrap.py

@@ -0,0 +1,158 @@
+"""Tests for the eval bootstrap's dataset-item construction."""
+import os
+import sqlite3
+import sys
+import urllib.parse
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "pilot"))
+
+import eval_bootstrap as eb  # noqa: E402
+
+
+# --- item_id --------------------------------------------------------------
+
+def test_item_id_has_no_path_separator():
+    """A `/` would split the UI's item route into extra path segments."""
+    assert "/" not in eb.item_id("netcracker/interview", 29)
+
+
+def test_item_id_has_no_fragment_marker():
+    """Everything after a `#` is a fragment the browser never sends."""
+    assert "#" not in eb.item_id("netcracker/interview", 29)
+
+
+def test_item_id_survives_a_url_round_trip():
+    """The id must appear verbatim in a path, needing no percent-encoding."""
+    ident = eb.item_id("netcracker/interview", 29)
+    assert urllib.parse.quote(ident, safe="") == ident
+
+
+def test_item_id_keeps_repo_and_pr_readable():
+    assert eb.item_id("netcracker/interview", 29) == "netcracker__interview__pr29"
+
+
+def test_item_id_is_unique_per_pr():
+    assert eb.item_id("o/r", 1) != eb.item_id("o/r", 2)
+
+
+def test_item_id_is_unique_per_repo():
+    assert eb.item_id("o/one", 1) != eb.item_id("o/two", 1)
+
+
+def test_item_id_accepts_a_string_pr():
+    assert eb.item_id("o/r", "29") == eb.item_id("o/r", 29)
+
+
+# --- read_review_items ----------------------------------------------------
+
+def _db(tmp_path, rows, findings=()):
+    path = str(tmp_path / "feedback.db")
+    conn = sqlite3.connect(path)
+    conn.execute(
+        "CREATE TABLE review (repo TEXT, pr INTEGER, posted_at INTEGER, head_sha TEXT)"
+    )
+    conn.execute(
+        "CREATE TABLE inline_finding (repo TEXT, pr INTEGER, path TEXT, line INTEGER,"
+        " severity TEXT, problem TEXT, fix TEXT)"
+    )
+    conn.executemany("INSERT INTO review VALUES (?,?,?,?)", rows)
+    conn.executemany("INSERT INTO inline_finding VALUES (?,?,?,?,?,?,?)", findings)
+    conn.commit()
+    conn.close()
+    return path
+
+
+def test_items_use_url_safe_ids(tmp_path):
+    path = _db(tmp_path, [("netcracker/interview", 29, 100, "abc")])
+    items = eb.read_review_items(path)
+    assert [i["id"] for i in items] == ["netcracker__interview__pr29"]
+
+
+def test_item_input_keeps_the_real_repo_name(tmp_path):
+    """The id is mangled for the URL; the payload must stay faithful."""
+    path = _db(tmp_path, [("netcracker/interview", 29, 100, "abc")])
+    item = eb.read_review_items(path)[0]
+    assert item["input"]["repo"] == "netcracker/interview"
+    assert item["input"]["pr"] == 29
+
+
+def test_one_item_per_pr_not_per_review(tmp_path):
+    path = _db(
+        tmp_path,
+        [
+            ("o/r", 1, 100, "a"),
+            ("o/r", 1, 200, "b"),
+            ("o/r", 2, 300, "c"),
+        ],
+    )
+    items = eb.read_review_items(path)
+    assert [i["id"] for i in items] == ["o__r__pr1", "o__r__pr2"]
+    assert items[0]["metadata"]["reviews_run"] == 2
+
+
+def test_items_are_not_flagged_as_human_labelled(tmp_path):
+    path = _db(tmp_path, [("o/r", 1, 100, "a")])
+    assert eb.read_review_items(path)[0]["metadata"]["labelled_by_human"] is False
+
+
+# --- metadata facets ------------------------------------------------------
+
+def _md(findings=(), repo="netcracker/interview", pr=29):
+    return eb._item_metadata(
+        repo=repo, pr=pr, head_sha="abc", reviews_run=2, last_seen=1788189422,
+        findings=[{"severity": s} for s in findings],
+    )
+
+
+def test_metadata_carries_the_repo_for_filtering():
+    assert _md()["repo"] == "netcracker/interview"
+
+
+def test_metadata_splits_owner_from_repo_name():
+    """A filter on the joined repo can match one repo; owner matches an org."""
+    md = _md()
+    assert md["owner"] == "netcracker"
+    assert md["repo_name"] == "interview"
+
+
+def test_owner_falls_back_when_the_repo_is_unqualified():
+    md = _md(repo="standalone")
+    assert md["owner"] == "standalone"
+    assert md["repo_name"] == "standalone"
+
+
+def test_metadata_values_are_filterable_primitives():
+    """Nested objects and lists are not reachable from the filter bar."""
+    for key, value in _md(["high"]).items():
+        assert isinstance(value, (str, int, float, bool)), key
+
+
+def test_max_severity_is_the_worst_finding():
+    assert _md(["low", "critical", "medium"])["max_severity"] == "critical"
+
+
+def test_max_severity_is_none_not_absent_for_a_silent_review():
+    md = _md([])
+    assert md["max_severity"] == "none"
+    assert md["has_findings"] is False
+
+
+def test_unknown_severity_does_not_win_the_max():
+    assert _md(["banana", "low"])["max_severity"] == "low"
+
+
+def test_severity_comparison_ignores_case():
+    assert _md(["HIGH"])["max_severity"] == "high"
+
+
+def test_finding_count_matches_the_findings():
+    md = _md(["low", "low"])
+    assert md["finding_count"] == 2
+    assert md["has_findings"] is True
+
+
+def test_last_reviewed_is_exposed_both_ways():
+    """The epoch sorts; the ISO string is what a human reads in a filter."""
+    md = _md()
+    assert md["last_reviewed_at"] == 1788189422
+    assert md["last_reviewed_iso"].startswith("2026-08-31T")

tests/pilot/test_eval_experiment.py

@@ -0,0 +1,159 @@
+"""Tests for linking existing review traces into dataset runs."""
+import os
+import sys
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "pilot"))
+
+import eval_experiment as ex  # noqa: E402
+
+
+def trace(tid, repo="o/r", pr=1, model="M2", ts="2026-08-01T00:00:00Z", **md):
+    meta = {"repo": repo, "pr": pr}
+    meta.update(md)
+    return {
+        "id": tid,
+        "timestamp": ts,
+        "tags": [f"model:{model}", "engine:opencode"],
+        "metadata": meta,
+    }
+
+
+# --- trace_model ----------------------------------------------------------
+
+def test_model_read_from_tag():
+    assert ex.trace_model(trace("t1", model="MiniMax-M2.7")) == "MiniMax-M2.7"
+
+
+def test_model_falls_back_when_untagged():
+    assert ex.trace_model({"tags": ["engine:opencode"]}) == "unknown"
+
+
+def test_model_falls_back_when_tags_absent():
+    assert ex.trace_model({}) == "unknown"
+
+
+# --- trace_item_id --------------------------------------------------------
+
+def test_item_id_matches_the_bootstrap_scheme():
+    assert ex.trace_item_id(trace("t1", repo="netcracker/interview", pr=29)) == \
+        "netcracker__interview__pr29"
+
+
+def test_trace_without_repo_is_not_an_item():
+    assert ex.trace_item_id({"metadata": {"pr": 1}}) is None
+
+
+def test_trace_without_pr_is_not_an_item():
+    assert ex.trace_item_id({"metadata": {"repo": "o/r"}}) is None
+
+
+def test_trace_without_metadata_is_not_an_item():
+    assert ex.trace_item_id({}) is None
+
+
+# --- plan_runs ------------------------------------------------------------
+
+ITEMS = {"o__r__pr1", "o__r__pr2"}
+
+
+def test_traces_group_by_model():
+    plan = ex.plan_runs(
+        [trace("a", pr=1, model="x"), trace("b", pr=2, model="y")], ITEMS
+    )
+    assert set(plan["runs"]) == {"x", "y"}
+
+
+def test_group_by_none_collapses_to_one_run():
+    plan = ex.plan_runs(
+        [trace("a", pr=1, model="x"), trace("b", pr=2, model="y")],
+        ITEMS,
+        group_by="none",
+    )
+    assert list(plan["runs"]) == ["all-traces"]
+
+
+def test_only_the_newest_trace_per_item_is_kept():
+    """A re-reviewed PR has many traces; a run takes one output per input."""
+    plan = ex.plan_runs(
+        [
+            trace("old", pr=1, ts="2026-08-01T00:00:00Z"),
+            trace("new", pr=1, ts="2026-08-09T00:00:00Z"),
+        ],
+        ITEMS,
+    )
+    assert plan["runs"]["M2"]["o__r__pr1"]["id"] == "new"
+
+
+def test_newest_wins_regardless_of_input_order():
+    older = trace("old", pr=1, ts="2026-08-01T00:00:00Z")
+    newer = trace("new", pr=1, ts="2026-08-09T00:00:00Z")
+    for order in ([older, newer], [newer, older]):
+        plan = ex.plan_runs(order, ITEMS)
+        assert plan["runs"]["M2"]["o__r__pr1"]["id"] == "new"
+
+
+def test_trace_for_a_pr_outside_the_dataset_is_skipped():
+    plan = ex.plan_runs([trace("a", pr=99)], ITEMS)
+    assert plan["runs"] == {}
+    assert plan["skipped_not_in_dataset"] == 1
+
+
+def test_non_review_trace_is_counted_separately():
+    plan = ex.plan_runs([{"id": "x", "metadata": {}}], ITEMS)
+    assert plan["skipped_not_a_review"] == 1
+    assert plan["skipped_not_in_dataset"] == 0
+
+
+def test_same_pr_different_models_lands_in_both_runs():
+    plan = ex.plan_runs([trace("a", pr=1, model="x"), trace("b", pr=1, model="y")], ITEMS)
+    assert plan["runs"]["x"]["o__r__pr1"]["id"] == "a"
+    assert plan["runs"]["y"]["o__r__pr1"]["id"] == "b"
+
+
+# --- run_name -------------------------------------------------------------
+
+def test_run_name_prefixed():
+    assert ex.run_name("baseline", "MiniMax-M2.7") == "baseline-MiniMax-M2.7"
+
+
+def test_empty_prefix_leaves_the_key_bare():
+    assert ex.run_name("", "MiniMax-M2.7") == "MiniMax-M2.7"
+
+
+# --- create_run -----------------------------------------------------------
+
+def test_create_run_posts_one_item_per_pair(monkeypatch):
+    calls = []
+
+    def fake_call(method, path, body=None, timeout=20.0):
+        calls.append((method, path, body))
+        return 201, {}
+
+    monkeypatch.setattr(ex.eb, "_call", fake_call)
+    res = ex.create_run("run-1", {"o__r__pr1": trace("t1"), "o__r__pr2": trace("t2", pr=2)})
+    assert res["items_linked"] == 2
+    assert res["failed"] == []
+    assert {c[1] for c in calls} == {"/api/public/dataset-run-items"}
+    assert {c[2]["runName"] for c in calls} == {"run-1"}
+
+
+def test_create_run_links_the_trace_to_the_item(monkeypatch):
+    seen = {}
+
+    def fake_call(method, path, body=None, timeout=20.0):
+        seen.update(body)
+        return 201, {}
+
+    monkeypatch.setattr(ex.eb, "_call", fake_call)
+    ex.create_run("run-1", {"o__r__pr1": trace("t1")})
+    assert seen["datasetItemId"] == "o__r__pr1"
+    assert seen["traceId"] == "t1"
+    assert seen["metadata"]["model"] == "M2"
+
+
+def test_create_run_reports_rejected_items(monkeypatch):
+    monkeypatch.setattr(ex.eb, "_call", lambda *a, **k: (400, "nope"))
+    res = ex.create_run("run-1", {"o__r__pr1": trace("t1")})
+    assert res["items_linked"] == 0
+    assert res["failed"][0]["item"] == "o__r__pr1"
+    assert res["failed"][0]["status"] == 400

tests/pilot/test_eval_judges.py

@@ -0,0 +1,126 @@
+
+
+"""Tests for the LLM-as-judge evaluator bootstrap."""
+import os
+import sys
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "pilot"))
+
+import eval_judges as ej  # noqa: E402
+
+
+# --- rule_body ------------------------------------------------------------
+
+def test_rule_body_targets_observations():
+    """Trace-level rules wouldn't see observation input/output."""
+    body = ej.rule_body("rule-x", "finding_actionability", 1.0)
+    assert body["target"] == "observation"
+    assert body["enabled"] is True
+
+
+def test_rule_body_filters_on_trace_name():
+    """`name` isn't a stringOptions column; only `traceName` is."""
+    body = ej.rule_body("rule-x", "finding_actionability", 1.0)
+    f = body["filter"][0]
+    assert f["column"] == "traceName"
+    assert f["operator"] == "any of"
+    assert f["type"] == "stringOptions"
+    assert "pr-review" in f["value"]
+
+
+def test_rule_body_references_evaluator_by_name():
+    """Ids are version-specific; rules must name the evaluator across versions."""
+    body = ej.rule_body("rule-x", "finding_actionability", 1.0)
+    assert body["evaluator"]["name"] == "finding_actionability"
+    assert body["evaluator"]["scope"] == "project"
+
+
+def test_rule_body_maps_input_and_output():
+    """Both judges read the observation's own input/output."""
+    body = ej.rule_body("rule-x", "any", 1.0)
+    sources = {m["source"] for m in body["mapping"]}
+    assert sources == {"input", "output"}
+
+
+def test_rule_body_carries_mapping_at_both_levels():
+    """The server validates `mapping` at the rule root and echoes it on the evaluator."""
+    body = ej.rule_body("rule-x", "any", 1.0)
+    assert body["mapping"]
+    assert body["evaluator"]["variableMapping"] == body["mapping"]
+
+
+def test_rule_body_passes_sampling_through():
+    assert ej.rule_body("r", "any", 0.25)["sampling"] == 0.25
+
+
+# --- ensure_evaluators idempotency ---------------------------------------
+
+def test_ensure_evaluators_skips_existing(monkeypatch):
+    seen = []
+
+    def fake_call(method, path, body=None, timeout=20.0):
+        seen.append(path)
+        return 200, {}
+
+    monkeypatch.setattr(ej.eb, "_call", fake_call)
+    monkeypatch.setattr(ej, "existing_evaluators",
+                        lambda: {"finding_actionability": "id-1", "review_self_consistency": "id-2"})
+    res = ej.ensure_evaluators()
+    assert res["created"] == {}
+    assert sorted(res["skipped"]) == ["finding_actionability", "review_self_consistency"]
+    assert res["failed"] == []
+    assert seen == []
+
+
+def test_ensure_evaluators_records_failures(monkeypatch):
+    def fake_call(method, path, body=None, timeout=20.0):
+        return 422, "boom"
+
+    monkeypatch.setattr(ej.eb, "_call", fake_call)
+    monkeypatch.setattr(ej, "existing_evaluators", lambda: {})
+    res = ej.ensure_evaluators()
+    assert res["created"] == {}
+    assert res["failed"][0]["status"] == 422
+
+
+# --- ensure_rules idempotency --------------------------------------------
+
+def test_ensure_rules_skips_existing(monkeypatch):
+    calls = []
+    monkeypatch.setattr(ej.eb, "_call",
+                        lambda *a, **k: calls.append(a) or (200, {}))
+    monkeypatch.setattr(ej, "existing_evaluators",
+                        lambda: {"finding_actionability": "id-1",
+                                 "review_self_consistency": "id-2"})
+    monkeypatch.setattr(ej, "existing_rule_names",
+                        lambda: {"finding_actionability-on-reviews",
+                                 "review_self_consistency-on-reviews"})
+    res = ej.ensure_rules({"finding_actionability": "id-1",
+                           "review_self_consistency": "id-2"}, 1.0)
+    assert res["created"] == []
+    assert sorted(res["skipped"]) == ["finding_actionability", "review_self_consistency"]
+    assert calls == []
+
+
+def test_ensure_rules_creates_when_missing(monkeypatch):
+    calls = []
+    monkeypatch.setattr(ej.eb, "_call",
+                        lambda *a, **k: calls.append(a) or (201, {}))
+    monkeypatch.setattr(ej, "existing_rule_names", lambda: set())
+    res = ej.ensure_rules({"finding_actionability": "id-1"}, 1.0)
+    assert res["created"] == ["finding_actionability"]
+    assert calls[0][0] == "POST"
+    assert calls[0][1] == "/api/public/unstable/evaluation-rules"
+
+
+# --- judge shape ----------------------------------------------------------
+
+def test_judges_have_required_keys():
+    for j in ej.JUDGES:
+        assert j["prompt"]
+        assert j["outputDefinition"]["dataType"] in ("NUMERIC", "BOOLEAN", "CATEGORICAL")
+
+
+def test_default_base_url_points_at_the_thinking_patch_proxy():
+    """`8802` is the judge-proxy that adds a `signature` to thinking blocks."""
+    assert "8802" in ej.JUDGE_BASE_URL

tests/pilot/test_eval_scores.py

@@ -0,0 +1,200 @@
+"""Tests for the deterministic review scorers."""
+import os
+import sys
+
+import pytest
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "pilot"))
+
+import eval_scores as es  # noqa: E402
+
+
+def f(sev, path="a.py", line=1):
+    return {"severity": sev, "path": path, "line": line, "problem": "p", "fix": ""}
+
+
+# --- finding_rate ---------------------------------------------------------
+
+def test_finding_rate_counts_findings():
+    assert es.finding_rate([f("high"), f("low")]) == 2.0
+
+
+def test_finding_rate_zero_for_silent_review():
+    assert es.finding_rate([]) == 0.0
+    assert es.finding_rate(None) == 0.0
+
+
+# --- severity_info_ratio --------------------------------------------------
+
+def test_info_ratio_all_advisory():
+    assert es.severity_info_ratio([f("info"), f("trivial")]) == 1.0
+
+
+def test_info_ratio_mixed():
+    assert es.severity_info_ratio([f("info"), f("high")]) == 0.5
+
+
+def test_info_ratio_none_when_no_findings():
+    # Undefined, not zero — zero would read as perfectly calibrated.
+    assert es.severity_info_ratio([]) is None
+
+
+def test_info_ratio_unknown_severity_treated_as_medium():
+    # Matches _normalize_finding's fallback, so an odd severity is not
+    # silently counted as advisory.
+    assert es.severity_info_ratio([f("bogus")]) == 0.0
+
+
+# --- severity_max ---------------------------------------------------------
+
+def test_severity_max_picks_highest():
+    assert es.severity_max([f("info"), f("critical"), f("low")]) == "critical"
+
+
+def test_severity_max_none_when_silent():
+    assert es.severity_max([]) == "none"
+
+
+def test_severity_max_case_insensitive():
+    assert es.severity_max([f("HIGH")]) == "high"
+
+
+# --- dropped_findings -----------------------------------------------------
+
+def test_dropped_findings_delta():
+    assert es.dropped_findings(5, 2) == 3.0
+
+
+def test_dropped_findings_never_negative():
+    assert es.dropped_findings(1, 3) == 0.0
+
+
+def test_dropped_findings_none_when_unknown():
+    assert es.dropped_findings(None, 2) is None
+
+
+# --- cost_per_finding -----------------------------------------------------
+
+def test_cost_per_finding_divides():
+    assert es.cost_per_finding(1.0, [f("high"), f("low")]) == 0.5
+
+
+def test_cost_per_finding_silent_review_divides_by_one():
+    # The run still cost money; attributing all of it to "found nothing" is
+    # the honest reading, and it avoids a division by zero.
+    assert es.cost_per_finding(0.25, []) == 0.25
+
+
+def test_cost_per_finding_none_when_unpriced():
+    assert es.cost_per_finding(None, [f("high")]) is None
+
+
+def test_cost_per_finding_none_on_garbage():
+    assert es.cost_per_finding("abc", [f("high")]) is None
+
+
+# --- build_scores ---------------------------------------------------------
+
+def _by_name(events):
+    return {e["body"]["name"]: e["body"] for e in events}
+
+
+def test_build_scores_emits_expected_set():
+    events = es.build_scores(
+        trace_id="t1", findings=[f("high"), f("info")], environment="claude",
+        cost_usd=0.5, dropped_count=2, timestamp="2026-01-01T00:00:00Z",
+    )
+    names = _by_name(events)
+    assert set(names) == {
+        es.FINDING_RATE, es.SEVERITY_INFO_RATIO, es.SEVERITY_MAX,
+        es.DROPPED_FINDINGS, es.COST_PER_FINDING,
+    }
+    assert names[es.FINDING_RATE]["value"] == 2.0
+    assert names[es.SEVERITY_MAX]["value"] == "high"
+    assert names[es.DROPPED_FINDINGS]["value"] == 2.0
+    assert names[es.COST_PER_FINDING]["value"] == 0.25
+
+
+def test_build_scores_all_events_are_score_create_on_the_trace():
+    events = es.build_scores(
+        trace_id="t9", findings=[f("low")], environment="ollama", cost_usd=1.0,
+    )
+    assert all(e["type"] == "score-create" for e in events)
+    assert all(e["body"]["traceId"] == "t9" for e in events)
+    assert all(e["body"]["environment"] == "ollama" for e in events)
+
+
+def test_build_scores_omits_undefined_scores():
+    # No cost and no drop count measured -> those scores are absent, not zero.
+    events = es.build_scores(trace_id="t2", findings=[], environment="ollama")
+    names = set(_by_name(events))
+    assert es.COST_PER_FINDING not in names
+    assert es.DROPPED_FINDINGS not in names
+    assert es.SEVERITY_INFO_RATIO not in names
+    assert names == {es.FINDING_RATE, es.SEVERITY_MAX}
+
+
+def test_build_scores_categorical_value_is_string():
+    events = es.build_scores(trace_id="t3", findings=[f("high")], environment="claude")
+    sev = _by_name(events)[es.SEVERITY_MAX]
+    assert sev["dataType"] == "CATEGORICAL"
+    assert isinstance(sev["value"], str)
+
+
+def test_build_scores_numeric_values_are_floats():
+    events = es.build_scores(
+        trace_id="t4", findings=[f("high")], environment="claude", cost_usd=1,
+    )
+    for name, body in _by_name(events).items():
+        if body["dataType"] == "NUMERIC":
+            assert isinstance(body["value"], float), name
+
+
+def test_build_scores_comment_propagates():
+    events = es.build_scores(
+        trace_id="t5", findings=[f("high")], environment="claude",
+        cost_usd=1.0, comment="cost basis: equivalent:claude-sonnet-5",
+    )
+    assert all("equivalent" in e["body"]["comment"] for e in events)
+
+
+# --- score configs --------------------------------------------------------
+
+def test_every_emitted_score_has_a_config():
+    configured = {c["name"] for c in es.SCORE_CONFIGS}
+    events = es.build_scores(
+        trace_id="t6", findings=[f("high")], environment="claude",
+        cost_usd=1.0, dropped_count=0,
+    )
+    assert set(_by_name(events)) <= configured
+
+
+def test_severity_max_config_covers_every_severity_it_can_emit():
+    labels = {c["label"] for c in
+              next(c for c in es.SCORE_CONFIGS if c["name"] == es.SEVERITY_MAX)["categories"]}
+    assert set(es.SEVERITY_RANK) | {"none"} == labels
+
+
+# --- ingestion envelope ---------------------------------------------------
+
+def test_every_event_carries_a_timestamp():
+    # Ingestion rejects events without one, and reports the rejection as a
+    # per-event 400 inside an HTTP 207 that reads as success.
+    events = es.build_scores(
+        trace_id="t7", findings=[f("high")], environment="claude", cost_usd=1.0,
+    )
+    assert events
+    assert all(e.get("timestamp") for e in events)
+
+
+def test_timestamp_defaults_when_caller_omits_it():
+    events = es.build_scores(trace_id="t8", findings=[f("low")], environment="claude")
+    assert all(isinstance(e["timestamp"], str) and e["timestamp"].endswith("Z") for e in events)
+
+
+def test_explicit_timestamp_is_used():
+    events = es.build_scores(
+        trace_id="t9", findings=[f("low")], environment="claude",
+        timestamp="2026-01-02T03:04:05Z",
+    )
+    assert all(e["timestamp"] == "2026-01-02T03:04:05Z" for e in events)

tests/pilot/test_feedback_scores.py

@@ -0,0 +1,207 @@
+"""Tests for the feedback.db -> Langfuse score bridge."""
+import os
+import sys
+
+import pytest
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "pilot"))
+
+import feedback  # noqa: E402
+import feedback_scores as fs  # noqa: E402
+
+
+@pytest.fixture
+def db(tmp_path):
+    conn = feedback.init(str(tmp_path / "fb.db"))
+    yield conn
+    conn.close()
+
+
+def _seed_finding(conn, repo="o/r", pr=1, comment_id=100, path="a.py", line=1):
+    cur = conn.execute(
+        "INSERT INTO review (repo, pr, head_sha, posted_at) VALUES (?,?,?,?)",
+        (repo, pr, "deadbeef", 1000),
+    )
+    review_id = cur.lastrowid
+    cur = conn.execute(
+        """INSERT INTO inline_finding
+           (review_id, repo, pr, path, line, severity, problem, comment_id, posthash, posted_at)
+           VALUES (?,?,?,?,?,?,?,?,?,?)""",
+        (review_id, repo, pr, path, line, "HIGH", "problem", comment_id, f"h{comment_id}", 1000),
+    )
+    conn.commit()
+    return cur.lastrowid
+
+
+# --- score_pr maths -------------------------------------------------------
+
+def test_engagement_zero_when_nobody_responded():
+    v = fs.score_pr({"total": 4, "engaged": 0, "positive": 0, "negative": 0})
+    assert v[fs.REVIEW_ENGAGEMENT] == 0.0
+
+
+def test_acceptance_absent_when_nobody_engaged():
+    # Not 0.0 — zero would claim humans judged it neutral.
+    v = fs.score_pr({"total": 4, "engaged": 0, "positive": 0, "negative": 0})
+    assert v[fs.REVIEW_ACCEPTANCE] is None
+
+
+def test_engagement_is_a_share_of_findings():
+    v = fs.score_pr({"total": 4, "engaged": 1, "positive": 1, "negative": 0})
+    assert v[fs.REVIEW_ENGAGEMENT] == 0.25
+
+
+def test_acceptance_all_positive():
+    v = fs.score_pr({"total": 2, "engaged": 2, "positive": 3, "negative": 0})
+    assert v[fs.REVIEW_ACCEPTANCE] == 1.0
+
+
+def test_acceptance_all_negative():
+    v = fs.score_pr({"total": 2, "engaged": 2, "positive": 0, "negative": 2})
+    assert v[fs.REVIEW_ACCEPTANCE] == -1.0
+
+
+def test_acceptance_mixed_is_normalised():
+    v = fs.score_pr({"total": 4, "engaged": 4, "positive": 3, "negative": 1})
+    assert v[fs.REVIEW_ACCEPTANCE] == 0.5
+
+
+def test_engagement_absent_when_no_findings_at_all():
+    v = fs.score_pr({"total": 0, "engaged": 0, "positive": 0, "negative": 0})
+    assert v[fs.REVIEW_ENGAGEMENT] is None
+
+
+# --- collect_pr_feedback over a real sqlite ------------------------------
+
+def test_collect_counts_nothing_on_untouched_findings(db):
+    _seed_finding(db)
+    tally = fs.collect_pr_feedback(db, "o/r", 1)
+    assert tally == {"total": 1, "engaged": 0, "positive": 0, "negative": 0}
+
+
+def test_collect_counts_positive_reaction(db):
+    _seed_finding(db, comment_id=101)
+    db.execute(
+        "INSERT INTO reaction (comment_id, user, content, created_at) VALUES (?,?,?,?)",
+        (101, "alice", "+1", 1),
+    )
+    db.commit()
+    tally = fs.collect_pr_feedback(db, "o/r", 1)
+    assert tally["positive"] == 1 and tally["engaged"] == 1
+
+
+def test_collect_counts_negative_reaction(db):
+    _seed_finding(db, comment_id=102)
+    db.execute(
+        "INSERT INTO reaction (comment_id, user, content, created_at) VALUES (?,?,?,?)",
+        (102, "bob", "-1", 1),
+    )
+    db.commit()
+    tally = fs.collect_pr_feedback(db, "o/r", 1)
+    assert tally["negative"] == 1 and tally["engaged"] == 1
+
+
+def test_resolved_thread_counts_positive(db):
+    fid = _seed_finding(db, comment_id=103)
+    db.execute(
+        "INSERT INTO thread_state (finding_id, resolved, checked_at) VALUES (?,?,?)",
+        (fid, 1, 1),
+    )
+    db.commit()
+    tally = fs.collect_pr_feedback(db, "o/r", 1)
+    assert tally["positive"] == 1 and tally["engaged"] == 1
+
+
+def test_unresolved_thread_is_not_a_vote(db):
+    fid = _seed_finding(db, comment_id=104)
+    db.execute(
+        "INSERT INTO thread_state (finding_id, resolved, checked_at) VALUES (?,?,?)",
+        (fid, 0, 1),
+    )
+    db.commit()
+    tally = fs.collect_pr_feedback(db, "o/r", 1)
+    assert tally == {"total": 1, "engaged": 0, "positive": 0, "negative": 0}
+
+
+def test_negation_reply_counts_negative(db):
+    fid = _seed_finding(db, comment_id=105)
+    db.execute(
+        "INSERT INTO reply (finding_id, author, body, created_at) VALUES (?,?,?,?)",
+        (fid, "carol", "this is a false positive", 1),
+    )
+    db.commit()
+    tally = fs.collect_pr_feedback(db, "o/r", 1)
+    assert tally["negative"] == 1 and tally["engaged"] == 1
+
+
+def test_neutral_reply_is_engagement_but_not_a_vote(db):
+    fid = _seed_finding(db, comment_id=106)
+    db.execute(
+        "INSERT INTO reply (finding_id, author, body, created_at) VALUES (?,?,?,?)",
+        (fid, "dave", "done", 1),
+    )
+    db.commit()
+    tally = fs.collect_pr_feedback(db, "o/r", 1)
+    assert tally["engaged"] == 1
+    assert tally["positive"] == 0 and tally["negative"] == 0
+
+
+# --- event shape ----------------------------------------------------------
+
+def test_build_score_events_shape():
+    events = fs.build_score_events("o/r", 7, {fs.REVIEW_ENGAGEMENT: 0.5}, "claude")
+    assert len(events) == 1
+    body = events[0]["body"]
+    assert events[0]["type"] == "score-create"
+    assert body["sessionId"] == "o/r#7"
+    assert body["value"] == 0.5
+    assert body["environment"] == "claude"
+
+
+def test_build_score_events_skips_none():
+    events = fs.build_score_events("o/r", 7, {fs.REVIEW_ACCEPTANCE: None})
+    assert events == []
+
+
+def test_score_ids_are_stable_across_runs():
+    # A backfill re-run must update, not duplicate.
+    a = fs.build_score_events("o/r", 7, {fs.REVIEW_ENGAGEMENT: 0.5})[0]["body"]["id"]
+    b = fs.build_score_events("o/r", 7, {fs.REVIEW_ENGAGEMENT: 0.9})[0]["body"]["id"]
+    assert a == b
+
+
+def test_score_ids_differ_per_pr_and_name():
+    e1 = fs.build_score_events("o/r", 7, {fs.REVIEW_ENGAGEMENT: 1})[0]["body"]["id"]
+    e2 = fs.build_score_events("o/r", 8, {fs.REVIEW_ENGAGEMENT: 1})[0]["body"]["id"]
+    e3 = fs.build_score_events("o/r", 7, {fs.REVIEW_ACCEPTANCE: 1})[0]["body"]["id"]
+    assert len({e1, e2, e3}) == 3
+
+
+def test_backfill_dry_run_reports_without_posting(db, tmp_path):
+    _seed_finding(db, comment_id=107)
+    db.commit()
+    path = db.execute("PRAGMA database_list").fetchone()[2]
+    summary = fs.backfill(path, dry_run=True)
+    assert summary["prs_scanned"] == 1
+    assert summary["prs_with_engagement"] == 0
+    assert summary["posted"] is False
+
+
+def test_every_emitted_score_has_a_config():
+    configured = {c["name"] for c in fs.SCORE_CONFIGS}
+    assert {fs.REVIEW_ENGAGEMENT, fs.REVIEW_ACCEPTANCE} == configured
+
+
+def test_every_event_carries_a_timestamp():
+    # Without one the ingestion endpoint 400s the event inside a 207 that the
+    # caller reads as success.
+    events = fs.build_score_events("o/r", 1, {fs.REVIEW_ENGAGEMENT: 0.0})
+    assert events
+    assert all(e.get("timestamp") for e in events)
+
+
+def test_explicit_timestamp_is_used():
+    events = fs.build_score_events(
+        "o/r", 1, {fs.REVIEW_ENGAGEMENT: 0.0}, timestamp="2026-01-02T03:04:05Z"
+    )
+    assert events[0]["timestamp"] == "2026-01-02T03:04:05Z"

tests/pilot/test_langfuse_trace.py

@@ -0,0 +1,326 @@
+"""Unit tests for Langfuse trace emission. No network.
+
+`_post` is monkeypatched everywhere a POST would happen; a test that reaches
+the real network is a bug in the test, not a slow test.
+"""
+import json
+import os
+import sys
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+ROOT = os.path.abspath(os.path.join(HERE, "..", ".."))
+sys.path.insert(0, os.path.join(ROOT, "pilot"))
+
+import langfuse_trace as lt  # noqa: E402
+
+
+USAGE = {
+    "input": 2_000_000,
+    "output": 17_000,
+    "reasoning": 500,
+    "cache_read": 400_000,
+    "cache_write": 50_000,
+    "total": 2_017_000,
+    "cost": 0.0,
+    "steps": 28,
+    "duration_s": 348.3,
+}
+
+BASE = dict(
+    repo="techspark/pragent",
+    index="42",
+    sha="2613b3e1122334455",
+    title="Harden the review path",
+    usage=USAGE,
+    findings=[
+        {"severity": "critical", "path": "a.py"},
+        {"severity": "minor", "path": "b.py"},
+        {"severity": "minor", "path": "c.py"},
+    ],
+    summary="Three findings.",
+)
+
+
+# ---------------------------------------------------------------------------
+# model -> environment split (the whole point of the integration)
+# ---------------------------------------------------------------------------
+
+
+def test_claude_models_land_in_the_claude_environment():
+    assert lt.resolve_environment("headroom/claude-sonnet-5") == "claude"
+    assert lt.resolve_environment("claude-opus-5") == "claude"
+
+
+def test_everything_else_lands_in_the_ollama_environment():
+    for m in (
+        "headroom/glm-5.2:cloud",
+        "headroom/MiniMax-M2.7",
+        "vllm-qwen38/qwen3.8-27b",
+        "gpt-5",
+    ):
+        assert lt.resolve_environment(m) == "ollama", m
+
+
+def test_provider_and_bare_model_are_split_on_the_first_slash_only():
+    assert lt.provider_of("vllm-qwen38/qwen3.8-27b") == "vllm-qwen38"
+    assert lt.strip_provider("headroom/glm-5.2:cloud") == "glm-5.2:cloud"
+    # A bare name has no provider prefix; default to the pilot's proxy.
+    assert lt.provider_of("glm-5.2:cloud") == "headroom"
+    assert lt.strip_provider("glm-5.2:cloud") == "glm-5.2:cloud"
+
+
+# ---------------------------------------------------------------------------
+# usage accounting
+# ---------------------------------------------------------------------------
+
+
+def test_cache_reads_are_subtracted_from_input_not_added():
+    # Langfuse sums usageDetails keys; opencode reports cache_read *inside*
+    # input, so reporting both raw would bill the prefix twice.
+    d = lt._usage_details(USAGE)
+    assert d["input"] == 2_000_000 - 400_000
+    assert d["cache_read_input_tokens"] == 400_000
+    assert d["cache_write_input_tokens"] == 50_000
+    assert d["output"] == 17_000
+    assert d["reasoning"] == 500
+
+
+def test_zero_cache_fields_are_omitted_rather_than_sent_as_zero():
+    d = lt._usage_details({"input": 100, "output": 10})
+    assert d == {"input": 100, "output": 10}
+
+
+def test_a_paid_model_is_priced_as_itself():
+    costs, basis = lt._cost_details(USAGE, "headroom/claude-sonnet-5")
+    assert costs["total"] > 0
+    assert basis == "actual"
+
+
+def test_minimax_is_priced_against_the_comparison_target_not_zero():
+    # MiniMax-M2.7 is the model the webhook actually runs and it is absent from
+    # PRICES; charting it at $0 would make the whole dashboard a flat line.
+    costs, basis = lt._cost_details(USAGE, "headroom/MiniMax-M2.7")
+    assert costs["total"] > 0
+    assert basis == "equivalent:claude-sonnet-5"
+
+
+def test_glm_is_priced_against_the_comparison_target():
+    costs, basis = lt._cost_details(USAGE, "headroom/glm-5.2:cloud")
+    assert costs["total"] > 0
+    assert basis.startswith("equivalent:")
+
+
+def test_an_all_zero_price_entry_counts_as_free_not_as_priced():
+    # The self-hosted vLLM qwen IS in PRICES, at 0.00 across the board.
+    costs, basis = lt._cost_details(USAGE, "vllm-qwen38/qwen3.8-27b")
+    assert costs["total"] > 0
+    assert basis.startswith("equivalent:")
+
+
+def test_explicit_price_target_wins_over_the_default():
+    costs, basis = lt._cost_details(USAGE, "headroom/MiniMax-M2.7", "claude-opus-5")
+    assert basis == "equivalent:claude-opus-5"
+    sonnet, _ = lt._cost_details(USAGE, "headroom/MiniMax-M2.7", "claude-sonnet-5")
+    assert costs["total"] > sonnet["total"]
+
+
+def test_env_overrides_the_default_target(monkeypatch):
+    monkeypatch.setenv("PRAGENT_PRICE_TARGET", "claude-haiku-4-5")
+    assert lt.resolve_price_target() == "claude-haiku-4-5"
+    # An explicit argument still beats the env.
+    assert lt.resolve_price_target("gpt-5") == "gpt-5"
+
+
+def test_unknown_comparison_target_yields_no_cost_block_rather_than_a_wrong_one():
+    costs, basis = lt._cost_details(USAGE, "headroom/MiniMax-M2.7", "not-a-real-model")
+    assert costs == {}
+    assert basis == ""
+
+
+# ---------------------------------------------------------------------------
+# batch shape
+# ---------------------------------------------------------------------------
+
+
+def test_batch_has_a_trace_and_a_generation_linked_by_trace_id():
+    batch = lt.build_batch(model="headroom/claude-sonnet-5", **BASE)
+    types = [e["type"] for e in batch]
+    # Scores ride in the same batch; the trace and generation lead it.
+    assert types[:2] == ["trace-create", "generation-create"]
+    trace, gen = batch[0], batch[1]
+    assert gen["body"]["traceId"] == trace["body"]["id"]
+    assert trace["body"]["environment"] == gen["body"]["environment"] == "claude"
+
+
+def test_batch_without_usage_has_no_generation():
+    batch = lt.build_batch(model="headroom/glm-5.2:cloud", **{**BASE, "usage": None})
+    types = [e["type"] for e in batch]
+    assert "generation-create" not in types
+    assert types[0] == "trace-create"
+
+
+def test_trace_carries_repo_pr_session_and_severity_counts():
+    batch = lt.build_batch(model="headroom/glm-5.2:cloud", **BASE)
+    body = batch[0]["body"]
+    assert body["sessionId"] == "techspark/pragent#42"
+    assert body["metadata"]["severities"] == {"critical": 1, "minor": 2}
+    assert body["metadata"]["findings"] == 3
+    assert "provider:headroom" in body["tags"]
+    assert "model:glm-5.2:cloud" in body["tags"]
+
+
+def test_lens_names_become_tags():
+    batch = lt.build_batch(
+        model="headroom/glm-5.2:cloud", lenses=["security", "tests"], **BASE
+    )
+    assert "lens:security" in batch[0]["body"]["tags"]
+    assert "lens:tests" in batch[0]["body"]["tags"]
+
+
+def test_cost_basis_is_tagged_so_equivalent_is_never_read_as_spend():
+    batch = lt.build_batch(model="headroom/MiniMax-M2.7", **BASE)
+    trace = batch[0]["body"]
+    assert "cost:equivalent:claude-sonnet-5" in trace["tags"]
+    assert trace["metadata"]["cost_basis"] == "equivalent:claude-sonnet-5"
+
+    paid = lt.build_batch(model="headroom/claude-sonnet-5", **BASE)
+    assert "cost:actual" in paid[0]["body"]["tags"]
+
+
+def test_minimax_generation_carries_a_nonzero_cost():
+    batch = lt.build_batch(model="headroom/MiniMax-M2.7", **BASE)
+    assert batch[1]["body"]["costDetails"]["total"] > 0
+
+
+def test_batch_is_json_serializable():
+    batch = lt.build_batch(model="headroom/claude-sonnet-5", **BASE)
+    json.dumps({"batch": batch})
+
+
+# ---------------------------------------------------------------------------
+# emit_review_trace — config gate and fail-open
+# ---------------------------------------------------------------------------
+
+
+def _configure(monkeypatch):
+    monkeypatch.setenv("LANGFUSE_HOST", "http://langfuse.test:3000/")
+    monkeypatch.setenv("LANGFUSE_PUBLIC_KEY", "pk-lf-test")
+    monkeypatch.setenv("LANGFUSE_SECRET_KEY", "sk-lf-test")
+
+
+def test_no_config_means_no_post_and_no_error(monkeypatch):
+    for k in ("LANGFUSE_HOST", "LANGFUSE_PUBLIC_KEY", "LANGFUSE_SECRET_KEY"):
+        monkeypatch.delenv(k, raising=False)
+    calls = []
+    monkeypatch.setattr(lt, "_post", lambda *a, **k: calls.append(a) or 200)
+    assert lt.emit_review_trace(model="headroom/glm-5.2:cloud", **BASE) is False
+    assert calls == []
+
+
+def test_configured_emit_posts_to_the_ingestion_endpoint(monkeypatch):
+    _configure(monkeypatch)
+    seen = {}
+
+    def fake_post(host, pk, sk, batch, timeout):
+        seen.update(host=host, pk=pk, sk=sk, batch=batch, timeout=timeout)
+        return 207
+
+    monkeypatch.setattr(lt, "_post", fake_post)
+    assert lt.emit_review_trace(model="headroom/claude-sonnet-5", **BASE) is True
+    # Trailing slash stripped so the path is not doubled.
+    assert seen["host"] == "http://langfuse.test:3000"
+    kinds = [e["type"] for e in seen["batch"]]
+    assert kinds[:2] == ["trace-create", "generation-create"]
+    assert "score-create" in kinds
+
+
+def test_transport_failure_is_swallowed(monkeypatch):
+    _configure(monkeypatch)
+
+    def boom(*a, **k):
+        raise OSError("connection refused")
+
+    monkeypatch.setattr(lt, "_post", boom)
+    assert lt.emit_review_trace(model="headroom/glm-5.2:cloud", **BASE) is False
+
+
+def test_non_success_status_reports_failure_without_raising(monkeypatch):
+    _configure(monkeypatch)
+    monkeypatch.setattr(lt, "_post", lambda *a, **k: 401)
+    assert lt.emit_review_trace(model="headroom/glm-5.2:cloud", **BASE) is False
+
+
+# ---------------------------------------------------------------------------
+# Scores folded into the review batch (added with eval_scores)
+# ---------------------------------------------------------------------------
+
+def _scores(events):
+    return {e["body"]["name"]: e["body"] for e in events if e["type"] == "score-create"}
+
+
+def test_build_batch_appends_scores():
+    events = lt.build_batch(
+        repo="o/r", index="1", sha="abc", title="t",
+        model="headroom/claude-sonnet-5",
+        usage={"input": 100, "output": 10},
+        findings=[{"severity": "high", "path": "a.py", "line": 1}],
+    )
+    names = set(_scores(events))
+    assert "finding_rate" in names
+    assert "severity_max" in names
+
+
+def test_scores_attach_to_the_same_trace():
+    events = lt.build_batch(
+        repo="o/r", index="1", sha="abc", title="t", model="m",
+        usage={"input": 1, "output": 1}, findings=[], trace_id="fixed-id",
+    )
+    for body in _scores(events).values():
+        assert body["traceId"] == "fixed-id"
+
+
+def test_scores_inherit_the_trace_environment():
+    events = lt.build_batch(
+        repo="o/r", index="1", sha="abc", title="t",
+        model="headroom/glm-5.2:cloud",
+        usage={"input": 1, "output": 1}, findings=[],
+    )
+    for body in _scores(events).values():
+        assert body["environment"] == "ollama"
+
+
+def test_dropped_findings_scored_when_provided():
+    events = lt.build_batch(
+        repo="o/r", index="1", sha="abc", title="t", model="m",
+        usage={"input": 1, "output": 1}, findings=[], dropped_count=3,
+    )
+    assert _scores(events)["dropped_findings"]["value"] == 3.0
+
+
+def test_dropped_findings_absent_when_not_measured():
+    events = lt.build_batch(
+        repo="o/r", index="1", sha="abc", title="t", model="m",
+        usage={"input": 1, "output": 1}, findings=[],
+    )
+    assert "dropped_findings" not in _scores(events)
+
+
+def test_cost_score_carries_its_basis_in_the_comment():
+    # An equivalent-cost $/finding must never be read as money spent.
+    events = lt.build_batch(
+        repo="o/r", index="1", sha="abc", title="t",
+        model="headroom/glm-5.2:cloud",
+        usage={"input": 1000, "output": 100}, findings=[{"severity": "low", "path": "a", "line": 1}],
+    )
+    cpf = _scores(events).get("cost_per_finding")
+    if cpf is not None:  # only when cost_model could price the comparison target
+        assert "equivalent" in cpf["comment"]
+
+
+def test_batch_without_usage_still_scores_findings():
+    # A run with no usage report still produced findings worth scoring.
+    events = lt.build_batch(
+        repo="o/r", index="1", sha="abc", title="t", model="m",
+        usage=None, findings=[{"severity": "critical", "path": "a", "line": 2}],
+    )
+    assert _scores(events)["severity_max"]["value"] == "critical"

tests/pilot/test_parse_dropped.py

@@ -0,0 +1,90 @@
+"""The parse-time drop counter feeding the `dropped_findings` score.
+
+A model that emits findings at unusable locations produces an empty findings
+list, exactly like a model that found nothing. These tests pin the signal that
+tells the two apart.
+"""
+import json
+import os
+import sys
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, os.path.abspath(os.path.join(HERE, "..", "..", "pilot")))
+
+import ai_review  # noqa: E402
+
+
+def _payload(findings):
+    return "```json\n" + json.dumps({"summary": "s", "findings": findings}) + "\n```"
+
+
+GOOD = {"severity": "high", "path": "a.py", "line": 3, "problem": "p", "fix": "f"}
+NO_PATH = {"severity": "high", "line": 3, "problem": "p"}
+NO_LINE = {"severity": "high", "path": "a.py", "problem": "p"}
+BAD_LINE = {"severity": "high", "path": "a.py", "line": 0, "problem": "p"}
+
+
+def test_no_drops_on_clean_output():
+    _, findings, *_ = ai_review.parse_review_output(_payload([GOOD, GOOD]))
+    assert len(findings) == 2
+    assert ai_review.last_parse_dropped() == 0
+
+
+def test_counts_findings_missing_path():
+    _, findings, *_ = ai_review.parse_review_output(_payload([GOOD, NO_PATH]))
+    assert len(findings) == 1
+    assert ai_review.last_parse_dropped() == 1
+
+
+def test_counts_findings_missing_line():
+    _, findings, *_ = ai_review.parse_review_output(_payload([NO_LINE, NO_LINE]))
+    assert findings == []
+    assert ai_review.last_parse_dropped() == 2
+
+
+def test_counts_findings_with_unusable_line():
+    _, findings, *_ = ai_review.parse_review_output(_payload([BAD_LINE]))
+    assert findings == []
+    assert ai_review.last_parse_dropped() == 1
+
+
+def test_all_dropped_is_distinguishable_from_found_nothing():
+    ai_review.parse_review_output(_payload([NO_PATH, NO_PATH, NO_PATH]))
+    all_dropped = ai_review.last_parse_dropped()
+    ai_review.parse_review_output(_payload([]))
+    found_nothing = ai_review.last_parse_dropped()
+    assert all_dropped == 3 and found_nothing == 0
+
+
+def test_counter_resets_on_unparseable_output():
+    # Otherwise a salvage-path review inherits the previous review's count.
+    ai_review.parse_review_output(_payload([NO_PATH, NO_PATH]))
+    assert ai_review.last_parse_dropped() == 2
+    ai_review.parse_review_output("no json here at all")
+    assert ai_review.last_parse_dropped() == 0
+
+
+def test_counter_resets_on_malformed_json():
+    ai_review.parse_review_output(_payload([NO_PATH]))
+    ai_review.parse_review_output("```json\n{not valid json,,,}\n```")
+    assert ai_review.last_parse_dropped() == 0
+
+
+def test_parse_findings_tracks_drops_too():
+    # The non-opencode path must be scored on the same basis.
+    findings = ai_review.parse_findings(json.dumps({"findings": [GOOD, NO_PATH]}))
+    assert len(findings) == 1
+    assert ai_review.last_parse_dropped() == 1
+
+
+def test_parse_findings_resets_on_garbage():
+    ai_review.parse_findings(json.dumps({"findings": [NO_PATH]}))
+    assert ai_review.last_parse_dropped() == 1
+    ai_review.parse_findings("not json")
+    assert ai_review.last_parse_dropped() == 0
+
+
+def test_bare_array_output_is_counted():
+    _, findings, *_ = ai_review.parse_review_output("```json\n" + json.dumps([GOOD, NO_PATH]) + "\n```")
+    assert len(findings) == 1
+    assert ai_review.last_parse_dropped() == 1