Record annotated and warm-cache evidence

Add repeated production annotation timing, real app-server cache telemetry, the context-delta decision, and correct --judge=false handling.\n\nAI-assisted: OpenAI Codex.
This commit is contained in:
Paul Bakaus
2026-07-12 21:22:27 -07:00
parent 1e4927d86a
commit 22786feac0
4 changed files with 16 additions and 5 deletions
@@ -34,7 +34,7 @@ const timeoutMs = positiveInteger(args.timeout, 300_000);
const outputPath = args.output ? path.resolve(ROOT, String(args.output)) : null;
const profileIds = csv(args.profiles || 'direct-sol,cold-app-server,warm-app-server');
const taskIds = csv(args.tasks || 'editorial-bolder,operations-polish,operations-annotated');
const judgeEnabled = args.judge !== false;
const judgeEnabled = args.judge !== false && args.judge !== 'false';
const loadedEnv = loadBenchmarkEnv({ repoRoot: ROOT, explicitPath: args.envFile && path.resolve(args.envFile) });
const skillPath = path.join(ROOT, '.agents', 'skills', 'impeccable', 'SKILL.md');
const referenceDir = path.join(ROOT, 'skill', 'reference');