Skip to content

Deepen Harness stream interpretation with semantic evidence - #641

Merged
JacobStephens2 merged 1 commit into
mainfrom
issue-640
Oct 10, 2026
Merged

JacobStephens2 merged 1 commit into
mainfrom
issue-640

Conversation

@JacobStephens2

@JacobStephens2 JacobStephens2 commented Oct 10, 2026 •

Copy link
Copy Markdown
Owner

Summary

Harness decoder
  parse one event → progress + semantic evidence
    Interpretation
      ordered observed Models
      optional Security Model progress and sticky refusals
      existing completion and retained-record reconciliation

Concentrate live protocol predicates in the existing dialect decoders for all six Harnesses. Keep Interpretation's caller interface and existing Session outcomes.

Closes #640

Evidence

  • Before: Interpretation sent each raw line to the decoder and separate Model/Security observers. The added assistant-attribution fixture passed against that implementation (cargo test --bin thirdshift harness::interpretation_tests: 13 passed), establishing behavior to preserve.
    After: the decoder returns an owned StreamUpdate; shared Model and Security policy contains no JSON parsing or Harness selection. cargo test --bin thirdshift harness::: 134 passed.
  • cargo check --all-targets, cargo fmt --check, and cargo clippy --all-targets -- -D warnings: passed.
  • Session recording regressions: 11 passed. The delegated-attribution Run notification, requested Codex Model across Resume, and refused Security reproduction regressions each passed by their exact test names.
  • SYSTEMD_UNIT_PATH=/usr/lib/systemd/system:/lib/systemd/system cargo test: 1713 passed, 1 existing test ignored, 0 failed across 41 test binaries.
  • Plain cargo test encountered a warning from this machine's unrelated /etc/systemd/system/career-dashboard-secrets.service in site_deploy::the_systemd_units_verify. The test and deployment units are unchanged from main; the exact test passed when systemd lookup was restricted to vendor units. The complete-suite command above uses that same environment isolation.

Test seams

  • Existing Interpretation interface: condense, observed_models, for_security, and finish, exercised through real Harness adapters.
  • Existing Session recording, Run notification, and Security-run regressions retain their lifecycle coverage.

Review

Independent Standards and Spec reviews against main at 86c3efc: 0 findings on each axis. Each reviewer read all nine changed files. Changed files left unread: none.

Unaddressed findings

Standards

None.

Spec

None.

Merge Danger

Door: two-way

Reversible internal refactor; no migration or new dependency.

Blast Radius: Sessions

All six Harnesses use this decoder interface for live progress and evidence.

Built with codex · gpt-6.1-sol · high

@JacobStephens2
JacobStephens2 merged commit 4fd3f71 into main Oct 10, 2026
17 checks passed
@JacobStephens2
JacobStephens2 deleted the issue-640 branch October 10, 2026 14:23
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.

Deepen Harness stream interpretation behind one evidence interface

1 participant