dataset(code-review): add synthetic reaction gold answers from BCApps PR 10312

[Wenjie Fan (gggdttt)]

Summary

Adds minimal synthetic code-review gold answers from explicit BCApps PR feedback.

Full source PR patches are intentionally excluded. Positive reactions become
isolated expected findings; thumbs-down findings become precision guards by
omission. Existing equivalent gold answers are not duplicated.

Accepted findings

Rejected findings

• Rejected precision guard: Update expense VAT specification source BCApps#10312 (comment)

Knowledge dependency

• Related BCQuality improvement: knowledge: improve review precision from BCApps PR 10312 feedback BCQuality#168

Validation

See the gate-owned offline evaluation section below.

Generated by the BC-ALAgentsInternal self-improvement workflow.

Offline evaluation: changed_needs_adjudication

Candidate matches gold; baseline emitted unexpected findings. A human must compare raw findings with the linked rejected comment: unrelated findings disappearing or an F1 increase do not establish target suppression.

Preparation attempt 1: invalid_selection. Coverage requires a patched file and valid lineStart/lineEnd for synthetic__style-locked-label-locale-default-01.

Preparation attempt 2: validated.

Selection: new. New: synthetic__style-locked-label-locale-default-01. Reused (complete payloads): ``.

• synthetic__style-locked-label-locale-default-01 / false_positive_guard / Update expense VAT specification source BCApps#10312 (comment): The bot flagged XDOMESTICTxt (assigned as the default into a Code field with TableRelation to VAT Business Posting Group) as needing Locked = true per style/label-locked-for-non-translatable, treating it like a wire-level token (GET, JSON, URL). The maintainer explicitly rejected this ('It should be translated!') and the merged code kept the Label unlocked. This is a default value for master data that a locale's own translation is expected to match against its own localized posting-group setup, not an integration contract string, so the rule's rationale does not apply here. The synthetic entry isolates exactly this shape (Label assigned as a default into a TableRelation-backed Code field, single new file, no other findings) with expected_comments=[] and the article recorded at metadata.articles, encoding the guard by omission. Limitation: this does not prove the label can never cause a referential-integrity issue in any locale, only that the maintainer's stated design intent for this specific default-value pattern rejects the Locked=true fix; genuinely non-translatable machine tokens (URLs, JSON, GUIDs) remain correctly covered by the original article's anti-pattern.
  Coverage references and outcome shapes are validated mechanically. Semantic equivalence, severity calibration and recommendation quality are NOT proved by these checks or a matching F1.

Common engine code: ecf8e31759d6ddd6d78e3a0b7836b40134368009; dataset SHA256: B5F1FEFA94EA1E5925EC2E9373E77EC20EC09F01CB60B380679DCA3948498134.
Model: gpt-5.6-luna; judge: gpt-5.3-codex. Exact D: synthetic__style-locked-label-locale-default-01.
Dataset base: 26365d6a41a96fac41df9dcc8339cb71241fcb1f; candidate: 5316391d575aabf76c48abf2a1a5f43b3debfb6e.

Arm	Evaluation commit	Engine pin	Knowledge pin
baseline	e48ed5b13b2d028886728a03a4367336d71a8fb6	4407f20b44fe92e3693d0444d25cc3430f1d5262	8584217c7506eea7eef27a9db353d78c0cd6a9a1
candidate	70fe3ad3278049ee803479ed2045808cb4d0bdc6	b4fb35c7b61378b9570475eb644bcbebeaca470a	db26f8d3be4763a58e98d5aedf7576fde0be30cb

Baseline

Run: https://github.com/microsoft/BC-Bench/actions/runs/34344121713; conclusion: success; wall clock: 7.5 minutes.
Raw findings and runtime pin proofs are retained in the self-improvement-feedback artifact and the linked evaluation run.

Entry	Expected	Generated	Missed	Unexpected	F1	AI credits	Agent seconds
synthetic__style-locked-label-locale-default-01	0	3	0	3	0	unavailable	267.162257301

Observed AI credits: unavailable; coverage 0/1 entries. Missing entries are not extrapolated into a total.

Aggregate metric	Value
total	1
expected_comment_count	0
generated_comment_count	3
matched_comment_count	0
missed_comment_count	0
incorrect_comment_count	3
precision	0
recall	1
f1	0

Candidate

Run: https://github.com/microsoft/BC-Bench/actions/runs/34344818038; conclusion: success; wall clock: 5.3 minutes.
Raw findings and runtime pin proofs are retained in the self-improvement-feedback artifact and the linked evaluation run.

Entry	Expected	Generated	Missed	Unexpected	F1	AI credits	Agent seconds
synthetic__style-locked-label-locale-default-01	0	0	0	0	1	unavailable	154.665790368

Observed AI credits: unavailable; coverage 0/1 entries. Missing entries are not extrapolated into a total.

Aggregate metric	Value
total	1
expected_comment_count	0
generated_comment_count	0
matched_comment_count	0
missed_comment_count	0
incorrect_comment_count	0
precision	1
recall	1
f1	1

Missing telemetry is unavailable, not zero. Evaluation-only metrics exclude candidate generation and are not the full-cycle cost.
Human review must verify source-patch fidelity, gold correctness, and target attribution. No automatic merge or branch-protection claim is made.