Skip to content

feat(workflow): §4a thin runtime_run loop — RunnableFrontier → eval_with_context → receipt append (wave-1 sequential, bounded) - #4745

Closed
briansrls wants to merge 3 commits into
mainfrom
session/clever-pike-231
Closed

briansrls wants to merge 3 commits into
mainfrom
session/clever-pike-231

Conversation

@briansrls

@briansrls briansrls commented Jun 12, 2026 •

Copy link
Copy Markdown
Contributor

Summary

  • Adds src/v4/workflow/runtime_run.dag: thin sequential run loop over RunnableFrontier
    • runtime_run(nodes, dependencies, tree, context) -> RuntimeRunResult — bounded fold (mirrors scheduler_run_to_fixed_point); each step calls eval_with_context per ready node and appends a RuntimeRunReceipt
    • runtime_run_step_from_frontier(frontier, state, tree, context) -> RuntimeRunState — callable entry for §4b (program.dag) to target
    • RuntimeRunComplete | RuntimeRunBlocked result with 🟡 dissolution mark for host-transport verdict return
  • Adds src/v4/test/claim/manual/runtime_run.dag: BoolWitnessClaim gate witness
    • Per-node InferredFacts lookup (not constant map — avoids grounding-node mismatch)
    • Three claims: single-step receipt append, full two-node sequential run reaches RuntimeRunComplete, step is bounded to its frontier
    • unified_claim_runtime_run: UnifiedTestClaim

Why

Wave-1 §4a substrate: exposes the callable entry runtime_run_step_from_frontier that §4b (program.dag run) targets. Bounded by fold(nodes, ...) so termination proof mirrors scheduler_run_to_fixed_point. Fail-closed per node: RuntimeRunReceipt captures both Accepted and Rejected so no eval outcome is silently dropped (P3).

Tests

cargo test --workspace — 18/19 pass; the 1 failure (workflow_no_path_regex_policy_ci_yml) is a pre-existing worktree-environment issue with git ls-files from the .worktrees/ parent directory, not present in CI.

cargo clippy --all-targets -- -D warnings — 3 pre-existing errors in src/v3/compiler/src/emit_host_eval.rs (not touched by this PR).

@briansrls briansrls changed the title §4a: thin runtime_run loop — scheduler RunnableFrontier → single eval_with_context step → receipt append (wave-1 sequential, bounded) feat(workflow): §4a thin runtime_run loop — RunnableFrontier → eval_with_context → receipt append (wave-1 sequential, bounded) Jun 12, 2026
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@briansrls

Copy link
Copy Markdown
Contributor Author

Superseded by #4743, which includes the same §4a runtime_run substrate (workflow/runtime_run.dag + manual/runtime_run.dag gate witness) plus §4c program.dag integration. Avoids duplicate merge of runtime_run.

@briansrls

Copy link
Copy Markdown
Contributor Author

Dashboard feedback triage (royal-ferret-510) — no action on this PR.

#4745 is CLOSED (superseded by #4743, 2026-06-12). §4a scope (workflow/runtime_run.dag + manual/runtime_run.dag) is included in #4743 plus §4c program.dag integration and the witness fixture fix (distinct TypeNode atom symbols for scheduler dedup).

Dashboard api-review APPROVE artifacts (#28672 claude, #28674 cursor) on this closed PR are historical — active lane is #4743 (68c5c48bc4). No reopen, no duplicate merge.

— sent from royal-ferret-510

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