[RFC] Phase 0: failure cases, adapter checklist, and public evaluation path

[wcqxgjy6d8-pixel]

Problem

VALP now has solid repository proof: schemas, unit tests, example audits, release notes, and CI. The next risk is public comprehension. New contributors need to understand what false-done failures VALP catches, how to evaluate runtime adapters, and which changes should go through RFC review.

Proposal

Use v0.2.0 as a stable evaluation baseline and focus Phase 0 community work on three public artifacts:

1. Failure cases: collect small examples where a runtime or agent says done but VALP evidence catches the gap.
2. Runtime adapter checklist: use a common checklist before claiming Full Mode, Remote Mode, or Manual Mode compatibility.
3. RFC process: discuss protocol, schema, evidence, adapter, and governance changes before turning them into issues or PRs.

Relevant entry points:

• Release: https://github.com/wcqxgjy6d8-pixel/Visible-Agent-Loop-Protocol/releases/tag/v0.2.0
• Failure gallery: https://github.com/wcqxgjy6d8-pixel/Visible-Agent-Loop-Protocol/blob/main/docs/failure-gallery.md
• Adapter checklist: https://github.com/wcqxgjy6d8-pixel/Visible-Agent-Loop-Protocol/blob/main/docs/adapter-checklist.md
• Project status: https://github.com/wcqxgjy6d8-pixel/Visible-Agent-Loop-Protocol/blob/main/docs/project-status.md

Evidence Impact

This RFC does not weaken any gates. It should make the existing gates easier to evaluate:

• dispatch receipts must still prove submitted vs inserted;
• expected evidence must still exist before completion;
• selected-agent recommendations must still be resolved with scope control;
• adapter compatibility must still be tied to exported evidence, not runtime marketing.

Compatibility

Backwards-compatible. This is a public-process and onboarding RFC. It does not change existing schemas, examples, or CLI behavior.

Smallest Testable Artifact

A useful response to this RFC can be one of:

• one concrete false-done failure case;
• one adapter checklist result for a real runtime;
• one proposed wording change that avoids overclaiming;
• one small fixture or docs patch that makes audit behavior clearer.

The main question: what should VALP require before a runtime, workflow, or example is allowed to claim Full Mode compatibility?

[wcqxgjy6d8-pixel]

Update after the cold-start cleanup:

• Public explainer: https://wcqxgjy6d8-pixel.github.io/Visible-Agent-Loop-Protocol/when-agent-done-is-not-done
• Failure gallery: https://github.com/wcqxgjy6d8-pixel/Visible-Agent-Loop-Protocol/blob/main/docs/failure-gallery.md
• Adapter checklist: https://github.com/wcqxgjy6d8-pixel/Visible-Agent-Loop-Protocol/blob/main/docs/adapter-checklist.md
• Adapter feedback thread: [Runtime adapter] Request for adapter checklist feedback #9

Good first issues now exist for concrete feedback:

• Run the adapter checklist against one runtime: Run the adapter checklist against one runtime #10
• Add one false-done case to the failure gallery: Add one false-done case to the failure gallery #11
• Improve the Pages demo: Improve the Pages demo for Agent done is not done #12