feat(eval): LLM-as-judge evaluators for finding actionability and review self-consistency

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.

pilot/README-eval.md

@@ -127,6 +127,51 @@ 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