Skip to content

Wave 2 de-risking: execution-measured per-module emit-frontier probe over the 27 src/v2/compiler modules — fixture-FREE self-emit bar (emit→compile→run→equivalent, no injected fixtures, discriminating RED) — + answer common-root vs 24-heterogeneous-gaps + Wave-2 sizing - #6533

Merged
briansrls merged 12 commits into
mainfrom
session/sharp-raven-495
Jul 13, 2026

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor

Auto-opened by session-dashboard for session sharp-raven-495.
Pushing to session/sharp-raven-495 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.

briansrls and others added 4 commits July 13, 2026 14:51
… ingest manifest stub.

CI discovery failed resolving compiler_frontier_probe_entry_test because the
host_source_root_ingest_manifest stub lacked the overlay symbol the entry imports.

Co-authored-by: Cursor <cursoragent@cursor.com>
@gunbai-bot
gunbai-bot Bot marked this pull request as ready for review July 13, 2026 15:35
…receipt.

Load compiler_frontier_sweep_order from frontier.dag at runtime instead of a
parallel Rust module list; document seed-retained bin scope and dissolution on
^migrate_when_frontier_per_module_probe_receipt_binds per wave2_prep_design.

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

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Addressed both REQUEST_CHANGES findings in df5e307:

1. Single authority (COMPILER_FRONTIER_MODULE_PATHS duplicate) — Valid. Deleted the hardcoded 27-path Rust list. frontier_probe_survey now resolves compiler_frontier_sweep_order from src/v2/compiler/self_host/frontier.dag at runtime (load_compiler_frontier_sweep_order); the bin cannot drift from the carrier without a resolve failure.

2. P5 hand-Rust scaffold receipt — Valid. Added explicit seed-retained documentation:

  • Cargo.toml bin entry: SCAFFOLD block naming authority, non-floor enrollment, dissolution on ^migrate_when_frontier_per_module_probe_receipt_binds per gunbc.plans.wave2_prep_design census carrier.
  • frontier_probe_survey_transport_scaffold_note in frontier_probe_survey.dag and frontier_probe_survey_transport.dag: census before/after (27 knowledge-attributed → execution-measured manifest), named migration trigger, concrete plan cite.

Fixture-free probe tightening (UnknownProbeCause, discriminating witnesses, long-lane placement) unchanged.

— sent from sharp-raven-495

…ureCemented.

Remove parallel Rust verdict heuristic from cluster_summary (histogram only);
name FixtureCemented's future behavioral-equivalence producer.

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

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Thanks for the APPROVE — addressed both nits in the follow-up commit:

  1. cluster_summary — Removed the parallel verdict heuristic (COMMON_ROOT_LIKELY / etc.); stderr output is histogram-only with an explicit pointer to canonical frontier_gap_clustering_from_receipts (witness compiler_frontier_gap_clustering_common_root_at_blocker_class_holds). Deletes with the P5 bin.

  2. FixtureCemented — Added producer comment: forward-declared for the fixture-free behavioral-equivalence probe (injected-fixture detection), not frontier_classify_rejection_reason.

— sent from sharp-raven-495

Inline // comments inside type variant lists are invalid substrate syntax;
frontier_probe_types failed resolve at line 12.

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

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Acknowledged APPROVE — no findings to address. Latest head 2b5a062 fixes the parse regression from the FixtureCemented note (// inside type arms → data frontier_fixture_cemented_producer_note); CI re-running on that commit.

— sent from sharp-raven-495

Avoid duplicating target/frontier-probe-survey literal already on survey_manifest_dir.

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

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Acknowledged APPROVE — addressed the path duplication nit in 8e48326: Mkdir.Parents now uses survey_manifest_dir instead of re-literalizing target/frontier-probe-survey.

Manifest string-concat in the Rust bin left as tracked P5 scaffold debt (same pattern as neighbouring manifest emitters; dissolves with the bin per ^migrate_when_frontier_per_module_probe_receipt_binds).

— sent from sharp-raven-495

@gunbai-bot

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Acknowledged APPROVE — no findings to address on current head 8e48326.

— sent from sharp-raven-495

…Ready tautology.

Register frontier_probe_oom_or_budget/runtime_error/probe_emit_accept symbols
on frontier_probe_types; document ready_count==0 as honest deferral not RED.

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

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Acknowledged APPROVE — addressed both nits in e0fdaa38:

  1. SelfEmitReady tautology — Added compiler_frontier_wave2_self_emit_ready_zero_note documenting that ready_count == 0 is green-by-construction on the assemble+emit probe slice (honest ^probe_emit_accept_self_host_bar_unbound deferral, not a discriminating SelfEmitReady RED until behavioral-equivalence lands).

  2. Survey-host unknown reasons — Declared ^frontier_probe_oom_or_budget, ^frontier_probe_runtime_error, and ^probe_emit_accept_self_host_bar_unbound on frontier_probe_types via frontier_blocker_is_probe_survey_host_reason + scaffold note; Rust bin comment points at that authority.

— sent from sharp-raven-495

Single authority for variant matching removes duplicate folds in frontier
and frontier_probe_survey so new blocker variants cannot drift.

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

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Composer review 37522 — duplicate blocker equality (valid, fixed in c3e33b0)

Verified: frontier_roster_row_blocker_matches_probe (frontier.dag) and frontier_blocker_class_matches (frontier_probe_survey.dag) were identical FrontierBlockerClass variant-equality folds.

Fix: moved frontier_blocker_class_matches to frontier_probe_types.dag (single authority on the type carrier). Both frontier.dag and frontier_probe_survey.dag now import it; roster matching delegates after the path check.

— sent from sharp-raven-495

@gunbai-bot

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Composer review 37524 — no findings (verified, no action)

Re-verified current head c3e33b03bf: scaffold discipline (P5 receipt, transport note, runtime compiler_frontier_sweep_order), honest UnknownProbeCause deferral for SelfEmitReady, and fail-closed probe/witness/transport paths all match the review's read. Prior nit (duplicate FrontierBlockerClass equality) was fixed in the same commit by consolidating frontier_blocker_class_matches on frontier_probe_types.dag.

— sent from sharp-raven-495

@gunbai-bot

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Opus review 37525 — no findings (verified, no action)

Re-verified current head c3e33b03bf: single-authority sweep order, honest emit-accept deferral, FixtureCemented/UnknownProbeCause threading, and frontier_blocker_class_matches now live once on frontier_probe_types (imported by frontier.dag + frontier_probe_survey.dag). No new modeling or fail-open issues in the changed lines.

— sent from sharp-raven-495

@gunbai-bot

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Composer review 37526 — REQUEST_CHANGES (valid, fixed in pending commit)

Verified all three findings on current head before fix:

  1. frontier_probe_survey_has_unclassified — valid: coproduct match predicate without disposition. Fix: added frontier_probe_survey_has_unclassified_dispatch_note documenting the existential UnknownProbeCause witness gate (predicate-dissolution receipt per execution_mode_is_wet_dispatch pattern).

  2. frontier_gap_clustering_is_common_root — valid: redundant variant→bool nickname. Fix: deleted the predicate; compiler_frontier_gap_clustering_common_root_at_blocker_class_holds now matches FrontierGapClusterVerdict directly.

  3. frontier_blocker_is_probe_survey_host_reason — valid: unwired + Rust parallel strings. Fix: wired through frontier_classify_rejection_reason; introduced frontier_probe_survey_host_{runtime_error,oom_or_budget}_reason Symbol data + frontier_probe_survey_host_failure_reason; seed bin reads those data rows at runtime (no hardcoded symbol strings). Added frontier_blocker_is_probe_survey_host_reason_dispatch_note.

— sent from sharp-raven-495

@gunbai-bot

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Composer review 37528 — APPROVE, no findings (verified, no action)

Re-verified head 47dc8f83b5 against the review read: P5 scaffold + dissolution trigger, fail-closed transport, honest UnknownProbeCause deferral on emit-accept, !frontier_probe_survey_has_unclassified witness gate, single-authority sweep order + host-failure symbols read from frontier_probe_types at runtime, long-lane placement outside floor discovery. Prior REQUEST_CHANGES (37526) addressed in same commit: predicate dispatch notes, deleted frontier_gap_clustering_is_common_root, wired frontier_blocker_is_probe_survey_host_reason.

— sent from sharp-raven-495

@gunbai-bot

gunbai-bot Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

Opus review 37529 — APPROVE, non-blocking notes (verified, no action)

Re-verified head 47dc8f83b5 against each note:

  1. emit_survey_manifest string-built .dag — valid medium-as-string tell; already covered by P5 scaffold receipts + dissolution triggers in frontier_probe_survey_transport_scaffold_note / transport module. Intentional for this window; dissolves toward self-emitted survey kernel per wave2_prep_design census carrier.

  2. frontier_probe_survey_has_unclassified predicate — valid tell; frontier_probe_survey_has_unclassified_dispatch_note (:74) frames it as the existential UnknownProbeCause witness gate with named dissolve-on. Non-blocking as marked.

  3. **frontier_blocker_class_matches + UnknownProbeCause** — intentional: variant-equality for classified blocker arms only; UnknownProbeCause { reason: _ } => falseavoids reason-agnostic class match (different host/survey reasons are not the same blocker). Roster↔survey binding is gated by!frontier_probe_survey_has_unclassifiedfirst — survey must have zero UnknownProbeCause beforecompiler_frontier_roster_matches_survey_holds` can green. Reason-level roster binding is post-manifest work, not this harness slice.

— sent from sharp-raven-495

@briansrls
briansrls merged commit 9008c5c into main Jul 13, 2026
3 checks passed
@briansrls
briansrls deleted the session/sharp-raven-495 branch July 13, 2026 19:18
briansrls added a commit that referenced this pull request Jul 13, 2026
…stered (whole-corpus red; selection gap, third instance of #6530's masking class)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
briansrls added a commit that referenced this pull request Jul 13, 2026
…rostered — whole-corpus red on the next cold sweep (selection gap, third instance of the #6530 masking class) (#6541)

Co-authored-by: Brian Searls <briansrls@gunb.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
briansrls added a commit that referenced this pull request Jul 13, 2026
…e 371s→319s, both stages −95%, fingerprints byte-identical (#6539)

* WIP: corpus OOM

* WIP: corpus OOM

* WIP: corpus OOM

* WIP: corpus OOM

* nfr roster backfill: #6533 frontier_blocker_class_matches landed unrostered (whole-corpus red; selection gap, third instance of #6530's masking class)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Brian Searls <briansrls@gunb.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
briansrls added a commit that referenced this pull request Jul 14, 2026
* design: live-read witness classification — masking-class root for runtime tree reads invisible to import-closure attribution

Documents the #6530/#6533 receipts, carrier taxonomy, G0–G3 classification grades, selection axis (iv), and dissolution path for accumulator_copy_roster_gate + lever_a EX-notes. Points v2.std.live_tree stamp provenance at the new authority.

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

* WIP: Live-read witness classification design — the masking-class root: witnes

* WIP: Live-read witness classification design — the masking-class root: witnes

* WIP: Live-read witness classification design — the masking-class root: witnes

* WIP: Live-read witness classification design — the masking-class root: witnes

* WIP: Live-read witness classification design — the masking-class root: witnes

* WIP: Live-read witness classification design — the masking-class root: witnes

* WIP: Live-read witness classification design — the masking-class root: witnes

* WIP: Live-read witness classification design — the masking-class root: witnes

---------

Co-authored-by: Brian Searls <briansrls@gunb.ai>
Co-authored-by: Cursor <cursoragent@cursor.com>
gunbai-bot Bot pushed a commit that referenced this pull request Jul 14, 2026
The nightly affected-set-falsifier's whole-corpus cold sweep (gunbc_falsifier_batches,
SelectionPredictOnly) surfaced 4 red live-tree witnesses = 2 real drift debts of the
documented masking class (PRs landed changes the per-PR affected-set selection did not
run these live-tree witnesses for, so the red surfaced on the next cold sweep, not the
landing PR — 4th instance, siblings to #6530/#6533):

1. doc_graph_has_no_orphan_docs (x2): docs/plans/live-read-witness-classification-design.md
   (added #6560/#6582) was referenced only in prose data:String notes, never a doc-graph
   edge. Fix: add a `bind:` provenance row in the owning lens file (the established
   convention, cf. commit_workflow_plan_doc_provenance) -> doc-graph root, orphans 1->0.

2. non_fold_residue_clean_holds (x2): resolved report = 3 unrostered, 0 stale.
   Backfill NON_FOLD_RESIDUE_ROSTER with the 3 sites (dated comment + per-site dissolution
   triggers): orch_emit_let_step (#6573; dissolves with the backward grammar-row emit fold),
   live_read_carrier_eq / path_pattern_eq (#6582; structural-equality nests, dissolve with
   derived == from inhabitance) — all matching existing roster precedent.

Verified green-by-execution: all 4 previously-red witnesses (+6 siblings) PASS via
claim_batch on a fresh v1-compiler build.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
briansrls added a commit that referenced this pull request Jul 14, 2026
* WIP: affected sest falsifier

* Affected-set falsifier burn-down: link orphan doc + backfill 3 nfr sites

The nightly affected-set-falsifier's whole-corpus cold sweep (gunbc_falsifier_batches,
SelectionPredictOnly) surfaced 4 red live-tree witnesses = 2 real drift debts of the
documented masking class (PRs landed changes the per-PR affected-set selection did not
run these live-tree witnesses for, so the red surfaced on the next cold sweep, not the
landing PR — 4th instance, siblings to #6530/#6533):

1. doc_graph_has_no_orphan_docs (x2): docs/plans/live-read-witness-classification-design.md
   (added #6560/#6582) was referenced only in prose data:String notes, never a doc-graph
   edge. Fix: add a `bind:` provenance row in the owning lens file (the established
   convention, cf. commit_workflow_plan_doc_provenance) -> doc-graph root, orphans 1->0.

2. non_fold_residue_clean_holds (x2): resolved report = 3 unrostered, 0 stale.
   Backfill NON_FOLD_RESIDUE_ROSTER with the 3 sites (dated comment + per-site dissolution
   triggers): orch_emit_let_step (#6573; dissolves with the backward grammar-row emit fold),
   live_read_carrier_eq / path_pattern_eq (#6582; structural-equality nests, dissolve with
   derived == from inhabitance) — all matching existing roster precedent.

Verified green-by-execution: all 4 previously-red witnesses (+6 siblings) PASS via
claim_batch on a fresh v1-compiler build.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Brian Searls <briansearls1@gmail.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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