changelog-bot：从 PR 中提取「为什么」而非让 AI 编造

## 为什么

The previous implementation could leave stale release data after an update.

### 已修复
- Fix cache invalidation

  Why: The previous implementation could leave stale release data after an update.

## 描述

Extract shared runtime configuration into ProviderBase.

Update OpenAI, Anthropic, and Gemini adapters to use the shared implementation.

This reduces duplicated setup logic and makes adding new providers easier.

This reduces duplicated setup logic and makes adding new providers easier.

Why was this change made?

const STRONG_CANONICAL_SECTION_NAMES = new Set([
  "why"，
  "reason"，
  "because"，
  "motivation"，
  "context"，
  "background",
  "problem",
  "rationale",
]);

const RATIONALE_MARKER_RE =
  /\b(why|because|so that|in order to|reason|rationale|motivation|to avoid|to prevent|context|problem)\b|,\s*so\b/i;

const candidates = buildWhyCandidateSnippets(
  extracted.sections,
  body,
  options.maxCharsPerPr,
  !extracted.hasTargetSection,
);

[
  "Extract shared runtime configuration into ProviderBase.",
  "This reduces duplicated setup logic and makes adding new providers easier.",
];

问题： "目标候选是否明确说明了为什么做出这个 changelog 变更？";

We changed the cache invalidation logic.

The README was updated because the previous explanation was confusing.

问题： "目标候选是否陈述了一个适用于所识别 changelog 变更的理由？";

{
  probability: explicitWhyProbability,
  relevanceProbability: changeRelevanceProbability,
}

Is this a reason?

AND

Is it the reason for this change?

const bestCandidate = candidateProbabilities.reduce((best, candidate) =>
  combinedWhyCandidateProbability(candidate) >
  combinedWhyCandidateProbability(best)
    ? candidate
    : best,
);

const accepted =
  selectedCandidate !== undefined &&
  bestCandidate.probability >= JEV_MIN_EXPLICIT_WHY_PROBABILITY &&
  bestCandidate.relevanceProbability >= JEV_MIN_CHANGE_RELEVANCE_PROBABILITY;

if(!accepted) continue;

There is no reliable WHY in this PR.

items.push({
  prNumber: item.prNumber,
  why: selectedCandidate,
  confidence: mappedConfidence,
});

why: selectedCandidate;

PR description
      ↓
candidate extraction
      ↓
Jev evaluates candidates
      ↓
select candidate
      ↓
original source text
      ↓
CHANGELOG

source changes
      ↓
ReleaseChange[]
      ↓
deterministic classification
      ↓
complete ReleaseDraft
      ↓
optional editorial enrichment
      +
optional WHY enrichment
      ↓
ReleaseResult
      ↓
deterministic renderer
      ↓
CHANGELOG

ReleaseDraft / ReleaseChange[]
      ↓
find eligible WHY targets
      ↓
collect PR rationale candidates
      ↓
LLM or Jev
      ↓
trust / confidence validation
      ↓
attach structured WHY notes
      ↓
ReleaseResult

{
  prNumber,
  why: selectedCandidate,
  confidence,
}

- Fix cache invalidation
- Why: The previous implementation could leave stale release data after an update.

WHY engine
→ find the evidence

Renderer
→ format the evidence

0.50?
0.60?
0.80?

{
  candidates: [
    "Separate changelog execution into input-resolution and output-finalization phases.",
    "This simplifies the main orchestration flow and runs independent GitHub lookups concurrently.",
  ],
  expectedSelectedCandidateIndex: 1,
}

{
  candidates: [
    "Extract shared provider runtime configuration into ProviderBase.",
    "Refactor OpenAI, Anthropic, and Gemini adapters to inherit it.",
  ],
  expectedSelectedCandidateIndex: null,
}

Explicit Rationale
Implementation-Only
Template & Noise
Multilingual PRs

[0.4， 0.45， 0.5， 0.55， 0.6， 0.65， 0.7， 0.75， 0.8， 0.85， 0.9];

Pull Request
     ↓
Normalize PR body
     ↓
Extract WHY candidates
     ↓
Remove noise
     ↓
Local trust check
     ↓
Jev:
  explicit rationale?
  relevant to this change?
     ↓
Apply threshold
     ↓
Candidate or none
     ↓
Original PR text
     ↓
CHANGELOG

How can AI write a good reason?

How can we find evidence that the reason already exists?

Use deterministic code to find evidence.

Use Jev to evaluate the evidence.

Keep the original evidence as the output.