Summary

• Adds a fast-check property suite for the forecast engine's pure decision layer — recommendForecastAccount, summarizeForecast, and buildForecastExplanation in lib/forecast.ts. These drive the codex-multi-auth forecast recommendation and its --explain output; the suite pins the selection policy over the full availability/risk/wait/flag space instead of hand-picked pools.

What Changed

New test/property/forecast.property.test.ts (7 properties over generated ForecastAccountResult pools, 0–8 accounts):

1. Candidate soundness — a non-null recommendation always points at a recommendable result (not disabled, not hard-failed, not exhausted, not unavailable); null exactly when no such candidate exists.
2. Ready dominance — if any ready candidate exists, the pick is ready with the minimal risk score among ready candidates, and the reason says so.
3. Delayed fallback — with only delayed candidates, the shortest wait wins.
4. Blocker-class guidance — an empty candidate pool yields "blocked or exhausted" guidance iff some account is blocked/exhausted rather than disabled/hard-failed. This is the regression guard for the exhausted-flag fix the code comments document (an all-exhausted pool must not produce a bogus "shortest wait" pick).
5. Order invariance — the recommendation doesn't depend on input order (the comparator's tie-breaks are total over unique indexes).
6. Summary partition — ready + delayed + unavailable === total exactly, with high-risk counted by risk level.
7. Explanation fidelity — considered mirrors the inputs in order, and selected is true on exactly the recommended index.

No SUT bugs found.

Validation

• npm test -- test/property/forecast.property.test.ts test/forecast.test.ts — 42/42 (new 7 + existing 35 untouched)
• npm run typecheck (also via pre-commit hook)
• npx eslint test/property/forecast.property.test.ts --max-warnings=0

Docs and Governance Checklist

• Test-only; no behavior or docs surface changed

Risk and Rollback

• Risk level: minimal — additive test file; conventions match the existing test/property/ suites.
• Rollback plan: revert the single commit.

https://claude.ai/code/session_01XNtnkLbBiXZxfQQYLMpucB

Generated by Claude Code

note: greptile review for oc-chatgpt-multi-auth. cite files like lib/foo.ts:123. confirm regression tests + windows concurrency/token redaction coverage.

Greptile Summary

adds a 7-property fast-check suite for the forecast engine's pure decision layer (recommendForecastAccount, summarizeForecast, buildForecastExplanation), covering soundness, ready-dominance, delayed-fallback, blocker guidance, order-invariance, summary partition, and explanation fidelity. addresses both items from the previous review cycle: the exhausted → availability = "delayed" domain constraint is now enforced in arbResults, and order-invariance now uses fc.shuffledSubarray for arbitrary permutations rather than reversal alone.

• the arbResults generator derives availability from exhausted to restrict the test space to the reachable domain, preventing false positives if a future fast-path relies on exhausted ⟹ availability !== "ready".
• the summary partition test asserts ready + delayed + unavailable === total and pins summary.ready, but does not individually assert summary.delayed or summary.unavailable, leaving a transposition of those two counts undetected.

Confidence Score: 5/5

additive test-only file with no behavior or api surface changes; safe to merge.

the change is a single new test file exercising pure, side-effect-free functions. no production code is modified, no new dependencies are introduced, and the suite's 7 properties are internally consistent with the sut. the one gap (missing individual delayed/unavailable assertions in the summary test) is a coverage nit that doesn't risk a regression.

test/property/forecast.property.test.ts — the summary partition test could be tightened with individual delayed and unavailable count assertions.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[results pool] --> B{filter candidates}
    B -->|not disabled, not hardFailure, not exhausted, not unavailable| C[candidates]
    B -->|empty| D{hasBlockedOrExhausted?}
    D -->|yes| E[null + blocked-or-exhausted reason]
    D -->|no| F[null + no-healthy-accounts reason]
    C --> G[sort via compareForecastResults]
    G --> H{best availability}
    H -->|ready| I[pick best index - Lowest risk ready account]
    H -->|delayed| J[pick best index - pick shortest wait]

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
test/property/forecast.property.test.ts:156-172
**summary partition misses individual `delayed`/`unavailable` pins** — the `ready + delayed + unavailable === total` check combined with the single `summary.ready` assertion leaves `delayed` and `unavailable` unconstrained against each other. a bug that transposes those two counts satisfies every current assertion. add `expect(summary.delayed).toBe(results.filter(r => r.availability === "delayed").length)` and the equivalent for `unavailable` to close the gap.

_{Reviews (2): Last reviewed commit: "test: keep forecast generator in the rea..." | Re-trigger Greptile}