[dnth]

Summary

Profiles declare their provider's peak-price windows once; the resolver boosts that candidate's rank everywhere outside them and leaves ranking untouched inside them.

 { "harness": "omp", "model": "opencode-go/deepseek-v4.1-flash", "provider": "opencode-go",
+  "peak_pricing": {
+    "multiplier": 2,
+    "windows": [
+      { "weekdays": ["mon","tue","wed","thu","fri"], "start": "01:00", "end": "04:00" },
+      { "weekdays": ["mon","tue","wed","thu","fri"], "start": "06:00", "end": "10:00" } ] } }

rank(candidate)
  no peak_pricing, or inside a declared window  -> exp(spendPriority)
  outside every declared window (incl. weekend) -> exp(spendPriority) * multiplier   candidate line: offpeak=xN

The ranking weight is exp(spendPriority).
quota-axi's scalar is signed, so a literal multiplication by a boost factor would deepen a negative deficit instead of favoring the candidate; exp is order-preserving, so peak ranking is identical to a run with no declaration.
Eligibility, quota floors, exhausted-runway refusals, approvals, and the genuine-tie rule read the same unadjusted evidence.
The UTC clock is read once per run; FM_DISPATCH_CLOCK_UTC pins it for tests; a malformed declaration exits 2 in the resolver and bootstrap reports the same configuration error.

Evidence

• Before: the new resolver cases failed on the unfixed code, for example not ok - off-peak boosts a below-leader weight past the leader (missing: profile '--harness omp --model opencode-go/deepseek-v4.1-flash'); the 0.3 candidate stayed behind the 0.4 candidate.
  After: bash tests/fm-dispatch-resolve.test.sh exits 0 with every case green, including off-peak and weekend boosts, peak output identical to no declaration, eligibility and adjusted-tie handling, malformed declarations, single midnight-crossing clock sampling, and trailing-newline rejection.

• Substitute for the cancelled macOS check (firstmate-approved; the captain decides the merge): the Stock macOS Bash snapshot compatibility job was cancelled by the provider twice with no verdict.
  Its own workflow commands were run under the exact required interpreter, GNU bash 3.2.57:

  • bash --version -> GNU bash, version 3.2.57(1)-release (x86_64-pc-linux-musl)
  • parse sweep (bin/fm-lint.sh --list-files, then bash -n per file) -> 443 files, PASS
  • bash tests/fm-fleet-snapshot-view.test.sh -> 17 ok (CI expects 17)
  • bash tests/fm-bearings-snapshot.test.sh -> 47 ok (CI expects 47)
  • Date portability: bin/fm-dispatch-resolve.sh:365-367 tries GNU date -u -d and falls back to BSD date -u -j -f '%Y-%m-%dT%H:%M:%SZ'; the live path at :369 runs date -u '+%u %H:%M' with no date flags and POSIX/BSD strftime formats.

• bin/fm-lint.sh (pinned ShellCheck 0.11.0): exit 0; bin/fm-doc-audience-check.sh: ok, surfaces=98, local_links=530.

Merge Danger

Door: two-way.
The field is optional, absent-field behavior is byte-identical, and reverting the commits restores the previous ranking.

Blast Radius: dispatch ranking only.
No behavior changes until peak_pricing is added to the private config/crew-dispatch.json after merge.

Pipeline

Updates from git push no-mistakes

✅ **Intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

✅ **Test** - passed

✅ No issues found.

• Live validation: ✅ go - 7 of 7 scenarios driven live against the product

Scenario	Result	Live	Evidence
Local resolver with synthetic quota inputs boosts the lower-weight candidate off-peak and displays offpeak=x2	✅ pass	live	Product CLI transcripts: weekday-offpeak, negative-signed-priorities
Local resolver applies the weekend boost even inside declared window clock times	✅ pass	live	Product CLI transcripts: saturday-inside-clock-window, sunday-second-window
Local resolver preserves undeclared ranking during peak hours and respects half-open window boundaries	✅ pass	live	Product CLI transcripts: first-window-start, first-window-interior, first-window-end, second-window-start, second-window-end and undeclared comparisons
Local resolver refuses ineligible candidates and preserves captain approval despite off-peak pricing	✅ pass	live	Product CLI transcripts: floor-refusal, exhausted-runway, approval-required and undeclared comparisons
Local resolver escalates a genuine adjusted-weight tie without choosing a profile	✅ pass	live	Product CLI transcripts: adjusted-weight-tie
Resolver rejects malformed rule and default pricing declarations; detect-only bootstrap reports the same configuration errors	✅ pass	live	Product CLI transcripts: malformed multiplier, time, weekday, newline and corresponding bootstrap cases
Local resolver samples weekday and time together once and rejects an invalid clock override	✅ pass	live	Structured CLI outputs: date_calls; Product CLI transcripts: system-clock, malformed-clock-override

• bash tests/fm-dispatch-resolve.test.sh — exit 0; supplemental fixture-based regression coverage.
• python3 .tmp-pricing-live/drive.py — exit 0; 47 real resolver/bootstrap CLI executions with isolated synthetic inputs and localhost HTTPS.
• Inspected resolver-header and configuration-owner documentation; did not run prohibited lint or static-analysis commands.
• Retained CLI evidence, stopped the HTTPS emulator, removed disposable workspace fixtures, and confirmed a clean worktree.

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.