feat(evals): add Muse Spark 1.3 measured results (2026-09-14)

[devin-ai-integration]

Summary

Adds Muse Spark 1.3 (native muse_cli client, 2026-09-14 campaign) as a third measured model in apps/evals, inline on the existing pages per the PR #89 conventions (counts displayed, graded-denominator sort key, unranked, Accuracy/Cost —).

Artifacts — byte-identical JSON subset of the supplied bundle under apps/evals/src/results/2026-09-14/muse-spark-1.3/ (comparison report, bundle manifest, provenance, campaign results/run plans/manifests, config, verification records). Excluded: baseline/ (identical to the 2026-09-11 files already checked in), genuine-replay.json, .jsonl trials, native traces, logs, source snapshot, PDF. Only the report's muse side is rendered; the sol side duplicates the existing GPT-5.6 Sol data.

Data model — measuredRun (a singleton hardcoded to the OMP run) is replaced by per-model campaign metadata:

interface MeasuredCampaign {
  id: "omp-2026-09-11" | "muse-2026-09-14";
  date; harness; repetitions; timeoutMs; sourceCommit;
  executableSourceCommit?; prUrl?; prLabel?; abortedRun?;   // OMP only
  limitations: readonly string[];                             // Muse: report.methodology
}
MeasuredModel.campaign: MeasuredCampaign
export const measuredCampaigns = [ompCampaign, museCampaign];

Muse normalisation (museResult/museCases): total = counts.dispatched, timeouts unscored, cases grouped from trials by caseId (verdict/checks/categories/duration/tokens; timeouts have null checks → {}); Spot cases reuse spotComparison.casePrompts for the shared 4 case ids. Numbers: Spot 9 / 12 (3 timeouts), Perps 42 / 54, Predictions 10 / 39 (1 timeout → sort key 10/38).

Pages — Model profile reads model.campaign.* (PR link / executable commit / aborted-run panel only when present; Limitations whenever the campaign has any). Leaderboard shows one indicator line per campaign and the row's own harness. Task explorer renders a per-model table for Perps/Predictions (was primary-model only); Muse's Spot row shows timeout cells. Methodology renders one block per campaign; the OMP block keeps the PR #85 link and limitations list.

Tests extended in apps/evals/__tests__/measured.test.ts. Typecheck, tests, fmt, lint, build, and check:public-boundary pass locally.

Link to Devin session: https://app.devin.ai/sessions/9924c1a614b94139a2fda346027389e0
Open in Devin Desktop: https://app.devin.ai/desktop/session/9924c1a614b94139a2fda346027389e0?variant=devin
Requested by: @ericjuta

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

---

Original prompt from Eric

make a pr from the patches @askgina/plugins in the zip

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
plugins-evals	Ready	Preview	Sep 14, 2026 3:35pm UTC

[devin-ai-integration]

Completed a recorded browser audit of PR #91, including the Methodology copy corrections in fbba9d8 (144b75f is formatting-only). No unresolved issues remain in the requested flows.

Muse campaign — verified UI

• Muse totals: 9 / 12 Spot, 42 / 54 Perps, 10 / 39 Predictions; measured rows unranked, Accuracy/Cost —.
• Graded-denominator sorting, stable search ranks, campaign-specific harness labels, latency-scatter inclusion / cost-scatter exclusion all behave as intended.
• Muse profile has its own 2026-09-14 provenance, correct dimensions, 4/18/13 cases, eleven limitations, and no OMP-only metadata (PR Fix native OMP eval catalog compatibility and tool evidence #85 link, executable commit, aborted run).
• Downloaded JSON has the expected filename and campaign object.

Muse profile	Breakdown at 700px

Tasks, methodology, regression, browser health

• Spot task matrices include Muse with timeout cells; Perps/Predictions show separate dated tables for Sol and Muse.
• Both Methodology campaign blocks render; two stale-copy issues found during testing were fixed and rechecked.
• Regression: GPT-5.6 Sol retains PR Fix native OMP eval catalog compatibility and tool evidence #85 link, executable commit, aborted-run panel, and 2026-09-11 footer.
• Pages fit at 700px; wide tables scroll internally. No console errors; all favicon assets return 200.