[SamCT86]

Incident

A real first-time evaluator reported that the MachineOutcome GitHub experience did not work well. Fresh black-box diagnosis confirmed the core verifier was healthy, but the public evaluation path was high-friction: clone + Node only, no frictionless browser lab, and the obvious public MachineOutcome route did not give a usable demo.

Fix

• add a browser-based synthetic Outcome Verification Lab driven by the canonical public verifier;
• add a guided 30-second ambiguous-timeout scenario plus adversarial presets;
• expose VERIFIED / UNKNOWN / FAILED, safe next action, evidence trace, naive-vs-guarded comparison, and exact verifier input/output;
• keep the browser demo zero-network, zero-credential, zero-real-write;
• reevaluate immediately on preset selection and manual control changes so stale results are not left visible;
• add executable scenario→verifier→render tests for all six presets, wrong-task evidence, and escaped rendering;
• add a manual, pinned, least-privilege GitHub Pages deployment with an exact pre-deploy test gate;
• put the browser lab first in README while keeping availability wording conditional until deployment completes.

Verification

Exact head: 967fb7e1da404687e886222b418506f63fe83f6a

• local clean test run: 25/25 PASS
• git diff --check: PASS
• GitHub CI verify-reference run 37596831802: SUCCESS
• independent source/security review: PASS P0=0 P1=0
  • VER-TERM-DIRECT-92d2e86b9a8d0aadf80f9e74
• independent first-time UX review: PASS P0=0 P1=0
  • VER-TERM-DIRECT-976457205eecb8924e4a55d0

Boundary

This remains a bounded synthetic public proof. It is not the private MachineOutcome runtime and makes no customer-adoption, ROI, universal-reliability, or production-provider claim.

Release gate

Source gate is now satisfied. Merge exact reviewed head only. Then run the manual Pages deployment and perform independent live black-box browser QA before closing the incident.