feat(examples): P1 graph-vs-loop parity harness — offline-proven, live-ready wiring - #710
Merged
Conversation
…ell (#694) Replay one identical CellSpec { task, coderProfile, reviewerProfile, shots, budget } through both loop forms and return one comparable ParityRecord per arm: - arms.ts — runLoopArm (agent-eval multishot loop: reviewer = driver leg, coder = agent leg, maxTurns = shot budget) and runGraphArm (runGraph over the two-node reviewer->coder topology, shot budget on the delegates edge, completion oracle on the deliverable). The record documents the real asymmetries instead of hiding them: only the graph arm has an edge ledger (the loop's is always undefined, never synthesized), the loop cannot stop early (no deliverable gate), and the conserved budget reaches only the graph. - offline.ts — scripted seams generated from one per-shot script for both arms (reuses examples/graphs/shared.ts leafSeam/scriptedBrain); zero network, zero env, $0; captures each arm's seam-level inputs for equivalence proofs. - run-parity.ts — CLI: --backend offline|cli-bridge --cells N --shots S. offline is the CI-safe proof path; cli-bridge is the one-command live entry wiring VB_CLI_BRIDGE_URL / VB_CLI_BRIDGE_BEARER / VB_PARITY_MODEL (not executed by any gate). - parity.test.ts — asserts input equivalence on CAPTURED inputs (task text, profiles, and shot budget reach both execution seams), well-formed records with the measured asymmetries (graph settles at the passing shot, loop burns the full budget), and the honest cap-refusal non-convergence path. tsconfig.examples.json gains the missing ./durable path mapping — the pre-existing typecheck:examples failure in examples/chat-handler (present on origin/main) resolved by mapping the subpath the example already imports.
tangletools
approved these changes
Aug 3, 2026
tangletools
left a comment
tangletools
left a comment
Contributor
There was a problem hiding this comment.
✅ Auto-approved drewstone PR — d1d2ff69
This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.
tangletools · auto-approval · reason: drewstone_author · 2026-08-03T06:28:26Z
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The measurement harness for #694 P1: replay the SAME cell through agent-eval's real
runMultishotloop (the legacy arm — audited as the genuine machinery, not a reimplementation) and throughrunGraph's 2-node shot topology, behind one identicalCellSpec, returning comparableParityRecords with every asymmetry documented rather than papered over (the loop has no ledger, no early-stop, no enforceable budget — that asymmetry IS the migration's argument).Adversarially audited SHIP: input-equivalence killed by 3 targeted mutations (task text, profile corruption, shot budget), offline run proven with zero inet connects under strace + scrubbed env, suite 2346 green, zero
src/changes.The
--backend cli-bridgelive entry is wiring only and NOT yet valid to run — three audited gaps, to be fixed before any live number is quoted:parity/unspecified);Offline measurement already demonstrates the qualitative delta on one scripted cell: graph settles at shot 2/3 via the deliverable gate; the loop burns all 3 (no gate to stop it).