Group review, feedback, evaluation, observability, and entrypoint code into packages. Keep thin top-level compatibility shims for existing scripts and imports, and mirror the structure in the tests.
Remove the obsolete dashboard now that Langfuse is the analytics surface.\nIntroduce focused transport, model, and configuration modules while preserving the ai_review facade, and document the current runtime architecture.
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.
Adds the evaluation layer on top of the review traces: five deterministic
scores describing how the reviewer behaved, a bridge that turns human reactions
into ground truth, and a dataset seeded from the reviews already run.
The two are kept apart on purpose. feedback.db has recorded 113 reviews and
zero reactions, resolutions or replies — nobody has ever responded to a bot
comment — so an accuracy metric cannot be built yet. The scorers therefore
measure behaviour, which is computable from data in hand, and feedback_scores
turns verdicts into scores the moment any arrive.
eval_scores.py emits finding_rate, severity_info_ratio, severity_max,
dropped_findings and cost_per_finding into the same ingestion batch as the
trace. Undefined values are omitted rather than reported as zero: an info ratio
over a silent review is undefined, and charting it as 0 would read as perfect
calibration.
dropped_findings needed a parser change. Both parsers silently discard findings
with an unusable path/line, which made a model emitting garbage locations
indistinguishable from one that found nothing. last_parse_dropped() exposes the
delta, read at parse time — after apply_repo_config the drops are the config
working as intended, not the model misbehaving.
feedback_scores.py scores the session ("{repo}#{pr}"), because feedback arrives
days later against a PR and nothing records which re-run produced which
comment. review_acceptance is absent rather than 0 when nothing was engaged.
eval_bootstrap.py registers the score configs, seeds the pragent-reviews
dataset, and can backfill scores onto traces that predate the scorers.
expectedOutput is the reviewer's own prior output, flagged
labelled_by_human: false — a regression baseline, not verified truth.
Also fixes a silent telemetry failure: the ingestion endpoint answers 207 when
only some events succeed, so a batch with every event rejected still looked
like success. Score events were missing the required per-event timestamp and
ingested nothing while reporting 207. _warn_on_rejected_events now logs the
per-event errors under LANGFUSE_DEBUG.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Ship token spend, latency and equivalent cost for every review to the
self-hosted Langfuse so per-model behaviour is queryable as a trend rather
than one PR comment at a time.
langfuse_trace.py is stdlib-only and emits via the public ingestion API.
Traces split into `ollama` and `claude` environments keyed off the bare model
name, not the provider: both paths go through the same headroom proxy, so the
provider prefix says nothing about which spend story a review belongs to. The
pilot's own path bills $0, so the reported cost is the equivalent price from
cost_model.PRICES.
ai_review.py calls _emit_langfuse on both token-spending exit paths (the
normal post and the salvage path). Import and emission are wrapped in a
blanket except: with no LANGFUSE_HOST or key pair the whole thing is a silent
no-op, and a telemetry failure must never fail a review.
These files were previously deployed only by way of the image build's
`COPY . /app`, so a clean checkout would have silently dropped tracing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude
masi