Skip to content

feat(fusion): v1 completion — CLI subcommand, evaluator model, confidence-aware difficulty gate - #392

Merged
Delqhi merged 1 commit into
mainfrom
feat/fusion-v1-completion
Jun 18, 2026
Merged

feat(fusion): v1 completion — CLI subcommand, evaluator model, confidence-aware difficulty gate#392
Delqhi merged 1 commit into
mainfrom
feat/fusion-v1-completion

Conversation

@Delqhi

@Delqhi Delqhi commented Jun 18, 2026

Copy link
Copy Markdown
Collaborator

Completes SIN Fusion v1 (issue #290, v3.22.0).

1. sin-code fusion CLI subcommand

  • fusion status: enabled/disabled, oracle mode, difficulty gate, provider count
  • fusion config: full config keys + env var overrides
  • fusion providers: lists Fireworks pool (model, base URL, max tokens)

2. Oracle judge prefers SIN_EVALUATOR_MODEL

  • WireFusion uses firstNonEmpty(SIN_EVALUATOR_MODEL, cfg.Model) for judge
  • SIN_EVALUATOR_BASE_URL creates a separate client (anti-bias)

3. Confidence-aware difficulty gate

  • TournamentRunner interface extended with ShouldRunWithConfidence
  • Loop passes verifyFailCount as attempt count
  • fusionAdapter uses ShouldTournamentWithConfidence when signals available
  • classifyVerifyError maps verify report to ErrorType

Tests

fusion ✅, loopbuilder ✅, orchestrator ✅, agentloop ✅ (all -race -count=1)

…nce, confidence-aware difficulty gate

1. sin-code fusion CLI subcommand (issue #290):
   - fusion status: enabled/disabled, oracle mode, difficulty gate, provider count
   - fusion config: full config keys + env var overrides (SIN_EVALUATOR_MODEL, etc.)
   - fusion providers: lists Fireworks pool (model, base URL, max tokens)
   - Read-only, no side effects. Registered in main.go

2. Oracle judge prefers SIN_EVALUATOR_MODEL (anti-bias):
   - WireFusion now uses firstNonEmpty(SIN_EVALUATOR_MODEL, cfg.Model) for judge
   - SIN_EVALUATOR_BASE_URL creates a separate client for the judge
   - Mirrors the stop-gate pattern at line 334

3. Confidence-aware difficulty gate fully wired:
   - TournamentRunner interface extended with ShouldRunWithConfidence
   - Loop passes verifyFailCount as attemptCount
   - fusionAdapter uses ShouldTournamentWithConfidence when signals available
   - classifyVerifyError maps verify.Result.Report to ErrorType
   - Falls back to text-only ShouldTournament when confidence=0 && attemptCount=0
   - All mock implementations updated (wiring_test, e2e_test)

Tests: fusion ✅, loopbuilder ✅, orchestrator ✅, agentloop ✅ (all -race -count=1)
@vercel

vercel Bot commented Jun 18, 2026

Copy link
Copy Markdown
Contributor

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

Project Deployment Actions Updated (UTC)
sin-code Ready Ready Preview, Comment, Open in v0 Jun 18, 2026 11:00am

@Delqhi
Delqhi merged commit 1d054c4 into main Jun 18, 2026
4 of 11 checks passed
@Delqhi
Delqhi deleted the feat/fusion-v1-completion branch June 18, 2026 11:00
@github-actions

Copy link
Copy Markdown

🏆 CEO Audit — A+ (100.0/100)

Metric Value
Grade A+
Score 100.0/100
Critical findings 0
High findings 0
Profile QUICK
Min grade gate B

📥 Download full report (Markdown)
📊 Download SARIF (for Code Scanning)

Run ~/.config/opencode/skills/ceo-audit/scripts/audit.sh . --profile=QUICK locally to reproduce.

@github-actions

Copy link
Copy Markdown

🏆 CEO Audit — A+ (100.0/100)

Metric Value
Grade A+
Score 100.0/100
Critical findings 0
High findings 0
Medium findings 0
Profile QUICK
Min grade gate B

📥 Download full report (Markdown)

Run ID: 27754904709 · Commit: ${github.sha}

Run ~/.config/opencode/skills/ceo-audit/scripts/audit.sh . --profile=QUICK locally to reproduce.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant