Skip to content

feat(analysis): add question-shaped run queries - #1339

Merged
andreasronge merged 5 commits into
mainfrom
codex/issue-1337-run-analysis
Aug 13, 2026
Merged

feat(analysis): add question-shaped run queries#1339
andreasronge merged 5 commits into
mainfrom
codex/issue-1337-run-analysis

Conversation

@andreasronge

@andreasronge andreasronge commented Aug 12, 2026

Copy link
Copy Markdown
Owner

Closes #1337

Summary

  • replace the split log/inspection vocabularies with six question-shaped analysis/* operations
  • share one bounded semantic read model across PTC-Lisp, the Viewer, Elixir embedders, and ptc transcript
  • preserve separate public/private authority while adding snapshot correlation, complete internal pagination, aggregate result bounds, and typed transcript diagnostics
  • remove raw Viewer inspection rendering and duplicate repo-analyst history aliases

Simplification and ease of use

  • four prompt namespaces and 23 exports become one namespace with six operations
  • repo-analyst run facade drops from 346 lines/13 join helpers to 52 lines/six question wrappers
  • production read/command surface is 619 lines smaller against current origin/main (2,364 deleted; 1,745 added)
  • conversation reconstruction is one semantic call; the flagship human path is one ptc transcript RUN_ID ... command with no record-shape knowledge
  • the same immutable evidence semantics now drive PTC-Lisp, Viewer rendering, embedding callbacks, CLI transcripts, public-only runs, private post-mortems, incomplete captures, and trace-only runs

Rebase and CI fixes

  • rebased onto current origin/main at be16ef00 (perf(cli): accelerate mix ptc startup (#1338))
  • updated repo-analyst live coverage to use the named mission contract
  • refreshed the replay prompt hash and supplied the required component-override environment
  • declared the named default mission in the GitHub MCP integration fixture
  • aligned Elixir-backed stdio fixture mission limits with installed discovery ceilings and constrained helper VMs to one scheduler under suite load
  • accepted both truthful bounded launcher finish reasons when a clean exit races the termination boundary after final stdout
  • aligned live E2E assertions with configured turn budgets, normalized provider narration, valid retry counts, and semantically equivalent extracted project names
  • lengthened an unrelated caller-cancellation test deadline so full-suite scheduler pressure cannot let its worker time out before the cancellation assertion
  • removed the completed disposable implementation plan while retaining durable contracts in docs/trace-log-contract.md

Review record

Verification

  • mix precommit: 6,212 root checks/tests plus Viewer, package, launcher, release, specs, conformance, duplication, and generated-artifact gates
  • MIX_ENV=dev mix docs --warnings-as-errors
  • pre-push hook: 6,212 root tests, Viewer 44 tests, Credo, generated artifacts, upstream audit, and Dialyzer (0 errors)
  • affected live Integration tests reproduced locally and passed after the fixes
  • caller-cancellation regression repeated 20 times at the failing seed
  • focused source-policy and immutable-capture regression suite: 85 tests

Replace split trace and inspection query vocabularies with one six-operation read model, semantic Viewer route, and transcript delivery. Migrate repo-analyst callers and delete raw join paths.\n\nVerified with mix precommit, the full root and Viewer suites, and the documentation warnings gate.
Reconcile conversation completeness with canonical events, preserve exact Viewer snapshot authority, and retain source-change classifications. Update CI fixtures and remove the completed implementation plan.\n\nVerified with mix precommit and MIX_ENV=dev mix docs --warnings-as-errors.
@andreasronge
andreasronge force-pushed the codex/issue-1337-run-analysis branch from 917cd3f to a40f6e4 Compare August 13, 2026 06:39
Match mission evaluation limits to the installed MCP discovery ceilings in the UTF-8 fixture and named-mission example, and keep their nested Elixir VMs to one scheduler under parallel suite load. Runtime defaults remain unchanged.\n\nVerified with eleven focused repetitions using the failing pre-push seed and mix precommit.
Align live-agent expectations with configured turn budgets and normalized provider responses. Keep assertions focused on mission routing and extracted semantics while accepting valid model retries and naming variants.\n\nVerified with the affected live E2E tests and mix precommit.
Keep the bounded worker alive long enough for the caller-kill assertion under full-suite scheduler pressure. The test still requires the linked worker to terminate with :killed.\n\nVerified with 20 focused repetitions at the failing seed and mix precommit.
@andreasronge
andreasronge marked this pull request as ready for review August 13, 2026 08:52
@andreasronge
andreasronge merged commit 363c555 into main Aug 13, 2026
18 checks passed
@andreasronge
andreasronge deleted the codex/issue-1337-run-analysis branch August 13, 2026 08:52
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.

Refactor the trace/inspection read side: one query surface, question-shaped vocabulary, restore conversation retrieval (breaking changes welcome)

1 participant