fix(usage): add refused_transcripts to the empty-stats expected dict

[jeeshofone]

Problem / Motivation

#9317: test_usage.py::TestParseSessions::test_empty_session_stats fails on every head built after 2026-09-07 22:03:57Z, across all authors' PRs. Cross-PR semantic conflict: #9254 (merged 20:49Z) made _parse_sessions include refused_transcripts in every payload; #9305 (merged 22:03:57Z — the exact boundary) added this test with an expected dict written before #9254 landed. Each PR was green in isolation; together, every head fails with one extra left-side key — refused_transcripts: 0.

Why it matters

Base-red on shard 4 for the whole repository: every PR built on current main carries this failure regardless of its content, which both blocks unrelated merges and trains people to ignore a red shard.

What changed

Motivation → approach → change: the empty-directory zero-statistics payload genuinely carries refused_transcripts: 0 (the #9254 contract), so the test's expected dict gains that key with a comment explaining the semantic-conflict history. One-line-plus-comment test fix; no source change — the source behavior of both merged PRs is correct and mutually consistent.

Tests

• test/test_usage.py full suite: 95 passed locally on main + this fix (was 2 failed).
• Reproduced the failure on unpatched main (678fc3266) first; verified both parametrizations ([True]/[False]) flip to green with the fix.
• black gate, flake8, isort on the touched file: clean.

Manual verification

Confirmed via git log -S refused_transcripts that #9254 introduced the key into every payload path, and via the pytest -vv dict diff that the only mismatch is the single missing key — no value differences.

Screenshots / video

N/A — test-only change.

Related Issues

Fixes #9317. Context: #9254, #9305.

Pattern harvest

Two PRs touching the same payload contract can be individually green and jointly red; when a whole-dict equality assertion starts failing repo-wide at an exact merge timestamp, diff the dict keys against git log -S for the extra key before suspecting the platform or the shard.

Checklist

• Tests added/updated and passing locally
• Lint gates clean on touched files
• No unrelated changes

Contribution License Agreement

By submitting this pull request, I confirm that my contribution is made under the terms of the project's license.

[github-actions]

GPT 5.6 Review (fork) — ✅ no blocking findings

Reviewed b43df1d3ca3634c2da083f5708815dbe7c5336a8 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] b43df1d

[github-actions]

Design Review (Fable 5, fork) — ✅ PASS

Design-level review of b43df1d3ca3634c2da083f5708815dbe7c5336a8 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Correct minimal fix: expected dict aligns with the single payload path (usage.py:1947), which the empty case genuinely shares — no design concerns.

[DESIGN-REVIEWED] b43df1d

[github-actions]

Opus 4.8 Review (fork) — ✅ no blocking findings

Reviewed b43df1d3ca3634c2da083f5708815dbe7c5336a8 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] b43df1d

[jeeshofone]

Duplicate resolution: #9319 (opened 3 minutes before this, functionally identical, maintainer-endorsed on #9317) should win — same key, and its comment additionally cites the #6733 complete-vs-unknown rationale. No objection to closing this one in its favor; leaving it open only until #9319 merges in case that stalls.

[jeeshofone]

Closing in favor of #9319 (3 minutes earlier, functionally identical, maintainer-endorsed on #9317). One of the two should merge quickly — main's shard 4 stays red until then.