Record Hy3 reasoning dogfood comparison

[elanthus]

Summary

• archive the August 15 OpenRouter Tencent Hy3 dogfood run with reasoning disabled
• archive the same-corpus reasoning-enabled follow-up, including the original length-exhausted attempt
• preserve first and final briefings, raw token/cost manifests, the shared corpus, and configuration snapshot
• raise the API-adapter completion-token default from 8,192 to 100,000 and document the environment override

Recorded outcomes

• Reasoning disabled: final ERROR with 3 errors and 6 warnings; OpenRouter reported $0.014364728.
• Reasoning enabled: final WARN with 0 errors and 27 warnings; completed calls cost $0.0261915078, and the all-attempt estimate is approximately $0.0349072038 because the initial failed call's billed envelope was not persisted.

Review

Agentic-preflight reviewed all 13 delivered diff units and recorded no findings. Its deterministic risk verdict is low/pass, with no human-review path matched. The attestation and local stage results are attached to the commit through the pushed git note; repository CI is the authoritative remote check.

This review record proves what the gate reported, including complete unit coverage. It is an audit trail, not proof that the review judgment was correct, and it does not replace human review.

This repository requires manual merge for this PR. Auto-merge must remain disabled.

Summary by CodeRabbit

• Documentation

  • Added August 15, 2026 daily briefings covering politics, news, world events, AI, and development practices.
  • Included source links, excluded-topic rationale, corpus-health reports, validation findings, and reproducibility details.
  • Documented briefing configuration and model-generation usage, including token and cost reporting.

• Enhancements

  • Increased the evaluator’s default completion limit to 100,000 tokens.
  • Added guidance for overriding the limit through environment settings.

• Tests

  • Updated evaluator validation to reflect the increased token limit.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: fda15922-5300-4878-a4d2-be47d8aca2a9

📥 Commits

Reviewing files that changed from the base of the PR and between ba9da53 and 9f97c67.

📒 Files selected for processing (13)

• docs/dogfooding.md
• docs/runs/2026-08-15/briefing-config.json
• docs/runs/2026-08-15/briefing-first.md
• docs/runs/2026-08-15/briefing.md
• docs/runs/2026-08-15/corpus-2026-08-15.json
• docs/runs/2026-08-15/generation-usage.json
• docs/runs/2026-08-15/hy3-reasoning-enabled/briefing-first.md
• docs/runs/2026-08-15/hy3-reasoning-enabled/briefing.md
• docs/runs/2026-08-15/hy3-reasoning-enabled/generation-usage.json
• evaluator/.env.example
• evaluator/README.md
• evaluator/adapters.py
• evaluator/tests/test_evaluator.py

---

📝 Walkthrough

Walkthrough

The PR raises the evaluator default completion limit to 100,000 tokens and records two August 15, 2026 Tencent Hy3 briefing runs. It adds run configuration, generated briefings, validation and corpus-health reports, usage metrics, costs, and reproduction details.

Changes

Briefing runs and evaluator support

Layer / File(s)	Summary
Evaluator completion budget evaluator/.env.example, evaluator/adapters.py, evaluator/README.md, evaluator/tests/test_evaluator.py	The default API completion limit increases from 8,192 to 100,000 tokens. EVALUATOR_MAX_TOKENS remains an override. Documentation and the payload test reflect the new default.
Run configuration and usage records docs/runs/2026-08-15/briefing-config.json, docs/runs/2026-08-15/generation-usage.json, docs/runs/2026-08-15/hy3-reasoning-enabled/generation-usage.json, docs/dogfooding.md	The run configuration defines six briefing sections. Usage records and dogfooding logs capture model calls, retries, checker results, token counts, costs, and reproduction commands.
Generated briefing outputs docs/runs/2026-08-15/briefing*.md, docs/runs/2026-08-15/hy3-reasoning-enabled/briefing*.md	Two briefing variants add categorized stories, source links, excluded-topic logs, corpus-health reports, validation results, and source issue records.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 9f97c

This PR archives the comparison results and updates the API-adapter completion-token default with an environment override; no actionable merge-blocking risk remains, so it is merge-ready after normal checks and review.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: recording the Hy3 dogfooding comparison with reasoning disabled and enabled.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch dogfood/hy3-2026-08-15

---

_{Comment @coderabbitai help to get the list of available commands.}

[claude]

Code review

No high-confidence issues found. Checked correctness, edge cases, security, and repository-specific requirements.

Notes (non-blocking):

• The only functional code change is evaluator/adapters.py:235, raising the EVALUATOR_MAX_TOKENS default from 8192 to 100000, mirrored consistently in evaluator/.env.example, evaluator/README.md, and the test assertion in evaluator/tests/test_evaluator.py. No other call site reads this variable, so the change is self-contained and correct.
• The PR adds generated corpora, briefings, and evaluator run artifacts under docs/runs/2026-08-15/. This looks like it could conflict with CONTRIBUTING.md's "Never commit ... generated corpora, briefings, or evaluator run artifacts," but the same pattern already exists on main for prior dates (docs/runs/2026-08-10/, 2026-08-12/, 2026-08-13/), so this PR is following established repo convention rather than introducing a new violation.
• No credentials, API keys, or other secrets were found in the diff (including the added JSON manifests and .env.example update).