Skip to content

feat: project TableEvent results to table.evaluation/1.0 - #9

Merged
chaoz23 merged 1 commit into
mainfrom
agent/table-evaluation-projection
Aug 3, 2026
Merged

feat: project TableEvent results to table.evaluation/1.0#9
chaoz23 merged 1 commit into
mainfrom
agent/table-evaluation-projection

Conversation

@chaoz23

@chaoz23 chaoz23 commented Aug 2, 2026

Copy link
Copy Markdown
Owner

Summary

Projects dmcheck's native TableEvent evaluation into the shared
table.evaluation/1.0 contract defined by chaoz23/table-kit#11.

This is intentionally stacked on #8 (agent/table-event-adapter) because the
projection consumes that adapter's normalized event stream, blockers, cursor,
and exact source evidence.

Contract behavior

  • adds dmcheck run-events <events> --gm <id> --table-evaluation
  • emits deterministic evaluation identity and canonical input digest
  • preserves exact TableEvent evidence references for every public finding
  • reports aggregate and per-rule coverage plus checked-through cursor state
  • distinguishes checked_clean, findings, incomplete, unsupported,
    invalid, and typed internal_error
  • fails closed on gaps, redacted required content, zero-compatible streams,
    unsupported variants, malformed inputs, or findings without exact evidence
  • always emits authority_status: self_attested; dmcheck cannot grant itself
    host authority

Validation

  • isolated Python 3.10/3.12/3.14 CI-equivalent runs: 202 collected; 201
    passed, 1 expected platform skip, and 196 subtests passed on each runtime
  • validated clean, findings, incomplete, unsupported, and invalid outputs
    against table-kit's actual table-evaluation-1.0.schema.json with format
    checking and its semantic conformance validator
  • cold-wheel packaging test exercises the installed console command and shared
    fail-closed envelope outside the source checkout
  • cold-installed dmcheck and current post-Add SKILL.md front door (check-family tune-up Phase 2) #12 table-kit wheels pass the hosted
    five-case boundary: checked-clean, exact-evidence findings, golden transport
    gap, deterministic legacy migration, and redacted content

Dependencies / sequencing

@chaoz23
chaoz23 force-pushed the agent/table-evaluation-projection branch from 17e3a42 to 5b44cc5 Compare August 3, 2026 05:10
@chaoz23
chaoz23 changed the base branch from agent/table-event-adapter to main August 3, 2026 05:10
@chaoz23
chaoz23 marked this pull request as ready for review August 3, 2026 05:12
@chaoz23
chaoz23 merged commit ce7ce6f into main Aug 3, 2026
4 checks passed
chaoz23 added a commit that referenced this pull request Aug 9, 2026
Rebuilt against the 0.6.0 validation-module rewrite (#7#9): worked
example uses the result_schema_version 1.0 envelope; severity/provenance
semantics (advisory+inferred is a lead, not an accusation) stated in the
example and MUST NOTs; R5 retirement and run-events in the deltas.
README: agents-start-here link.

Acceptance test (run against 0.5.5 surface, still valid — the CLI verbs
and no-false-accusation framing are unchanged): fresh-context agent
produced correct run --gm --dice-bot invocation, relayed findings with
evidence + charter citation, reported silence as silence.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
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