Clarify Orient recovery output and offline boundaries

[sungjunlee]

Orient answers could omit the latest Progress and in-flight recovery details when resuming work. The contract now requires those outputs, distinguishes todo-batch selection from dependency authorization, and preserves provisional offline state, recorded ownership or unknown, and the failed-authority stop.

Refs #490. AC 1/2 remain open for the disclosed Fable response omissions; merging this bounded clarification does not close the issue or claim all-model compliance.

• Update Orient and eval scenarios 1/8 only; explicitly scope the offline fixture to unavailable default/declared GitHub.
• Record the required thirteen-scenario before/after pair on Fable 5.1 and Astra. Replace the previous report/raw directory per retention policy; historical grades remain unchanged in git.
• No spec, script/schema, framework, or completed-sprint changes.

Validation

Reviewed SHA: 636027fcc7f43c3e80a36ac351d3aa6fbeda3ff3.

• Required Node suite: 187 passed, 1 skipped, 0 failed. The skip is opt-in external triage integration.
• Required smoke suite: 167 passed, 0 failed. git diff --check is clean; SKILL.md is 165 lines.
• Linux and Windows CI passed on the same SHA: https://github.com/sungjunlee/dev-backlog/actions/runs/35604996015.
• Astra scenarios 1/8 improve to PASS. Fable retains scenario 1's dependency-authorization omission and scenario 8's ownership/unknown omission; no grades were weakened.
• Scenario 6/7 residuals and Fable's additional unexpected ASK remain recorded. All four hypothetical runs have zero unexpected MUT. These answers do not establish actual execution failure or success rates.

Review

Grok 4.6 authored the contract/eval change. Opus 5 high independently reviewed the content; a native GPT reviewer then gave final LGTM on the change, prompt construction, all four answers and report. A subsequent narrow Opus 5 high review of the final report and four answers at the exact SHA above returned LGTM (nonblocking residuals), with no merge blockers. It verified grade arithmetic and the main response evidence, and recommended preserving behavior-level AC residuals.

The remaining reviewer notes concern rubric interpretation, wording and unchanged online-recovery omissions; they do not authorize inflated grades or wider implementation. The committed raw prompts were separately checked for the existing five-line/1100-word limits and absence of Expected text.

GitHub Codex review completed with a thumbs-up and no findings. CodeRabbit skipped automatic review under its repository-stars policy; its success status is not counted as substantive review.

Final CLI verification

A separate native Sol verifier ran the real status.sh --json and next.sh --json entrypoints against temporary synthetic sprint fixtures at the reviewed SHA: four CLI executions, all exit 0, and 36 independent strict assertions passed. Cases cover an earlier in-flight-only batch vs. the later todo batch, PR/branch/unmoored pointers, latest Progress and its absence, and todo selection without batch headings. Recorded ownership was read from sprint prose, not assumed to be a dedicated JSON owner field. SHA-256 comparisons of all three fixture inputs were unchanged before/after. This validates the deterministic recovery rail, not model behavior during an authority outage.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Advanced

Run ID: 97e58ecf-13b0-4909-94f3-110ad15229c6

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-22T01:24:50.552563Z	636027f	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.