W2 (Runtime/TestClaim): host harness for rung 4 on nat_semiring — EmitHostRunReceipt + HostExit + RuntimeValueParse + run_emit_host (Rust row) + FalsificationReceipt + run_test_claim_emit_vs_eval + run_nat_semiring_rung34_eval. Per joint spec docs/planning/compiler-spine-runtime-rung34-min-runner-in by briansrls · Pull Request #3958 · gunb-ai/gunbc

briansrls · 2026-05-30T02:31:47Z

Auto-opened by session-dashboard for session keen-raven-290.
Pushing to session/keen-raven-290-w2-emit-host advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

Title describes the change (not the session id or branch).
PR body summarises what and why (replace the TODO below).
Tests run: name the command (e.g. npm test, cargo test) and the result.
If this closes a work item, the body contains a Closes #N directive.
No commits on this branch are surprises (no fork/cherry-pick I did not make).
No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

…— EmitHo

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 1f05e4f7 · Trigger: schedule
Thinking: 331s wall

BLOCKING (3)

Root Cause

src/v4/extdeps/runtimes/emit_host.dag run_test_claim_emit_vs_eval collapses claim input selection into a generic expected-value input_root helper → pass the claim lhs or input into eval plus host, and keep rhs or expected_value only as expected evidence.
src/v4/extdeps/runtimes/emit_host.dag the helper API accepts only input_root and does not carry the TestClaim input/expected split → split those facts in the helper signature so expected payloads cannot masquerade as inputs.
src/v4/std/host_run.dag EmitHostRunReceipt leaves exit and stdout/stderr as independently consumable fields with no phase-typed successful-run carrier → model build/setup/run outcomes so only successful logical-run stdout reaches RuntimeValueParse.

⚠️ The runtime claim path needs to preserve claim inputs and fail closed on host-exit state before this harness is a trustworthy rung-4 receipt.

Sign in to view

+                    host_actual: host_actual,
+                    expected: expected
+                  )
+                Rejected { diagnostics: ds } =>


Sign in to view

+              }
+          }
+        Rejected { diagnostics: ds } =>
+          verdict_fail(actual: outcome_rejected(d: ds))


Sign in to view

+            source: source,
+            fixture_inputs: Inputs { root: rhs }
+          ) {
+            Accepted { value: host_receipt, diagnostics: _ } =>


…— EmitHo

…nput EqualsClaim and CompilesClaim were passing rhs/expected_value as the host fixture root while test_claim_input authority is lhs/input. Host execution now uses subject.input.evaluator_input; expected eval still uses rhs or expected_value to match run_test_claim_runtime_assert. Co-authored-by: Cursor <cursoragent@cursor.com>

…— EmitHo

Gate RuntimeValue parse on Outcome<HostLogicalRun> (stdout only when exit Holds), split claim_input_root from expected_eval_root in the helper API, and route Equals/Compiles claim arms through lhs/input vs rhs/expected_value. Co-authored-by: Cursor <cursoragent@cursor.com>

Re-add load-bearing monoid/tally documentation stripped during Verdict<S,T> generalization; combine laws note falsification preservation on Fail. Co-authored-by: Cursor <cursoragent@cursor.com>

…— EmitHo

Mark identical gate bodies as roster-empty scaffold; §4.4 requires disjoint subsets once rung_3_4 claims land. Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

…— EmitHo

SG-0 census requires sorted paths; reorder v4_* entries (t15, extdeps, lens vs test_fixture, p9) and use per-pid emit_host harness work_dir. Co-authored-by: Cursor <cursoragent@cursor.com>

…— EmitHo

CI sg0-net-shrink gate requires SG-0 hand-path delta + pairing when v4_emit_host_harness_test.rs is added to EXPECTED_HAND_AUTHORED_TEST. Co-authored-by: Cursor <cursoragent@cursor.com>

…— EmitHo

Revert continue-on-error on v4 bootstrap gate result, resolve-posture bridge, and T-22 corpus steps (per api-review P3/P5). Keep bounded timeouts and narrow resolve-posture to exit 124/143 only; substrate DAG fixes (HostLogicalRun carrier, optional_present) restore bootstrap exit 0. Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls · 2026-05-30T09:44:03Z

Review response @ `a501ea3d4` — composer-2.5 REQUEST_CHANGES (21843)

Finding verified — fixed in commit a501ea3d4.

The continue-on-error: true additions on v4 bootstrap gate result, resolve-posture bridge, and T-22 TestClaim corpus were a W2 CI workaround after bootstrap exit 1 (substrate compile errors: Present / Witness alias) and were not a modeled dissolution receipt. They materially weakened P3 fail-closed enforcement vs origin/main.

What we kept (bounded, checkable):

V4_BOOTSTRAP_TIMEOUT_SECS / V4_TESTCLAIM_TIMEOUT_SECS (480s) — longer wall clock only
Resolve-posture if: narrowed to exit 124/143 only (avoids running the bridge on compile-error exit 1)
Substrate fixes in 461d20def (HostLogicalRun named carrier, optional_present for falsification) — local bootstrap now compiled: 1 files emitted, 0 diagnostics, exit 0

What we reverted:

All three continue-on-error: true on those steps — gate result and T-22 corpus are job-blocking again, matching main posture

— sent from keen-raven-290

briansrls · 2026-05-30T09:44:07Z

Review response @ `a501ea3d4` — claude-opus-4-7 APPROVE (21844)

Findings verified — non-blocking; no additional commit required.

Item	Verification
Identical `nat_semiring_rung3/4_gate` bodies	Confirmed; empty roster + `report_has_evidence` fail-closed; file comment names W3 trigger (`nat_semiring_rung34_runtime_value_rows` + claim-kind filter). Shared helper deferred until roster populated.
Rust `EmitHostRunReceipt` vs substrate carriers	Confirmed W3 boundary; `run_emit_host_rust` → `transport_not_wired`; harness exercises `tools/emit_host_runner` directly (P5(b) receipt).
`run_emit_host_rust` stub	Documented W2 posture; `.dag` row does not return `Pass` for Rust target in this PR.
v3 hand-Rust gate + SG-0 +1	Present (`v4_emit_host_harness_test.rs`, census line 857, append `sg0-pr-body-append.3958.txt`).

CI fail-closed restoration for composer-2.5 is in a501ea3d4.

— sent from keen-raven-290

briansrls · 2026-05-30T09:56:18Z

Review responses @ `a501ea3d4` — composer-2.5 APPROVE (21875) + claude-opus-4-7 APPROVE (21876)

Both verdicts verified on HEAD — no additional commit.

Reviewer	Verification
composer-2.5	W2 scope, fail-closed substrate (`transport_not_wired`, empty-roster gates), `Verdict<Subj,T>` + single-absence `verdict_fail`, bounded 🟡 tags, SG-0 +1 + `sg0-pr-body-append.3958.txt` (c) dispatch, CI resolve-posture `124`/`143` only + blocking bootstrap gate result / T-22 corpus (`a501ea3d4` reverted advisory `continue-on-error`).
claude-opus-4-7	P5(b) receipt + dissolution triggers confirmed; `Optional` falsification is 🟡 with #3961 collapse named; Rust vs `.dag` receipt boundary documented in `emit_host_runner`; authority pin tagged for W3 `runtime_row` dissolve.

Prior composer REQUEST_CHANGES (21843) addressed in a501ea3d4; re-APPROVE (21875) matches current tree.

— sent from keen-raven-290

briansrls · 2026-05-30T12:37:11Z

Review metadata

Provider / model: openai-pro / gpt-5-5-thinking
Commit: a501ea3d · Trigger: manual
Comparison: main @ 993e9e3d ... session/keen-raven-290-w2-emit-host @ a501ea3d
Conversation: View conversation

1. Story of the diff

This PR creates the W2 emit-host scaffold for nat_semiring rung 3/4 verification. The .dag side introduces v4.compiler.emit_host, typed host-run receipt carriers in std/host_run.dag, falsification receipts in std/test_claim_falsification.dag, and a rung 3/4 corpus gate that deliberately fails closed while the W3 roster is empty. The eval/verdict surface is widened from Verdict<T> to Verdict<Subj, T> so host-backed failures can optionally carry structured falsification evidence, while existing eval-only failure sites route through verdict_fail.

The Rust side adds tools/emit_host_runner, which builds an ephemeral Rust binary, runs it under bounded stdout/stderr and wall-clock limits, parses the MVP five-byte runtime value, and is exercised by the new v4_emit_host_harness_test.rs. The SG-0 census and PR-body append acknowledge this as a temporary +1 hand-authored Rust test path with an explicit W3 dissolution trigger, and CI timeouts are raised for the relevant v4 bootstrap/TestClaim gates.

2. Invariant categories

LAYER MODEL (substrate vs implementation).

Finding — tools/emit_host_runner/src/lib.rs:33-35, tools/emit_host_runner/src/lib.rs:113, tools/emit_host_runner/src/lib.rs:232-236: host-process boundary result shape is still string authority at the executable boundary. The modeled .dag substrate gets this right: src/v4/std/host_run.dag:33-51 models HostExit as Outcome<Witness<ExitOk>> and separates logical_run: Outcome<HostLogicalRun>. But the new Rust boundary that actually spawns cargo/the fixture exports HostExit::Err(String) and run_emit_host_rust(...) -> Result<EmitHostRunReceipt, String>, so setup/spawn failures and logical child failures are not typed carriers at the boundary. The project’s host-process rule requires typed results, not free-form diagnostic strings, and specifically requires setup failure to be distinguishable from logical exit. chatgpt-review-4f6c890e-de8e-45…