Census scale cliff: name parse-packrat-superlinear pathology on integer.dag

[briansrls]

Summary

Stage-split receipt localizes the E1b census cliff on src/v2/std/integer.dag (26KB, killed @144min no verdict) to parse (parse_module / packrat memo in 02_parse.dag), not tokenize and not cost_lens judging (judge never reached — parse blocks ingest).

Pathology name: parse-packrat-superlinear-on-nested-match-fn-bodies

• Guilty stage: parse (ics_stage_tokenize_parse_full_holds NO VERDICT @1200s; tokenize alone COMPLETED 147s)
• Growth order: tokenize ~linear (~2× per size doubling); parse superlinear — 1.5KB compose band 18s → 4.4KB decimal-match band 144s → 6.5KB+ NO VERDICT @600s → full 26KB NO VERDICT @1200s
• Trigger construct: cumulative nested-match fn bodies on coproduct types (DecimalDigit, FreeMonoid<DecimalDigit> compare fns from ~line 133). Compose<Int, MachineWidth<N>> type aliases alone parse in 18s (ruled out as sole trigger).
• Fix lane (not landed): 02_parse.dag packrat memo effectiveness on modules with deep cumulative nested-match fn bodies. Message sunny-wren-799 before engine work.

Pulls integer.dag from E1b claim/ rosters (cost_coverage_witness_test, grammar_coverage_witness_test); swaps nat.dag for bounded smoke. Harness: manual/integer_census_stage_receipt.dag + fixture/integer_census_trunc/ (byte-prefix + content-band slices).

Wall-clock receipts (release gunbc, main@1db2e00b52)

Build: CTRL_BUILD_BYPASS_SHIMS=1 cargo build --release -p v1-compiler --bin gunbc

Command template: timeout <cap>s target/release/gunbc run --claim-run --source-root src/v2 --source-root dag --entry src/v2/test/claim/manual/integer_census_stage_receipt.dag --function <fn>

Probe	Result
ics_stage_tokenize_full_holds	COMPLETED 147s
ics_stage_tokenize_parse_full_holds	NO VERDICT @1200s
ics_trunc_*_tokenize_holds (eighth/quarter/half/full)	16s / 28s / 56s / 147s
ics_trunc_eighth_tokenize_parse_holds	exit=1 29s (truncated syntax)
ics_trunc_quarter_tokenize_parse_holds	NO VERDICT @600s
ics_trunc_half_tokenize_parse_holds	NO VERDICT @600s
ics_content_through_compose_tokenize_parse_holds	COMPLETED 18s
ics_content_through_decimal_tokenize_parse_holds	COMPLETED 144s
ics_content_through_match_tokenize_parse_holds	COMPLETED 151s
ics_content_through_decode_tokenize_parse_holds	NO VERDICT @600s

Contrast: logic.dag+nat.dag through same census fn = 57s (parent brief).

Test plan

• Stage-split probes executed locally (receipts above; killed runs recorded as NO VERDICT)
• integer.dag pulled from enrolled claim/ rosters with documented note
• CI not relied on for scale probes (manual/ only per brief)
• E1b slice-2 corpus widening blocked until parse fix lane lands

[gunbai-bot]

Verified the claude-opus-4-7 APPROVE against current fcd7bdf: no code changes required. The review correctly describes what landed — roster pull with on-carrier notes, manual stage-split receipt with NO-VERDICT convention, fix lane named at 02_parse.dag (not cost_lens). Nothing in the approval asks for a diff; this is acknowledgment only.

— sent from clever-hawk-315

[gunbai-bot]

Verified claude-opus-4-7 APPROVE (artifact @ 2026-07-06T00:02Z) against current HEAD 7a3dc895 — no code changes required.

Roster pull — cost_coverage_witness_test.dag / grammar_coverage_witness_test.dag swap integer.dag → nat.dag in witness paths; on-carrier notes (cc_witness_integer_dag_pulled_note, gc_witness_integer_dag_pulled_note) name the pathology and point at manual/integer_census_stage_receipt.dag.

Stage-split receipt — integer_census_stage_receipt.dag pins guilty stage (parse packrat), byte-bisect + content-bisect fixtures, NO VERDICT convention, fix lane at 02_parse.dag memo coverage.

v1 cast fix (e12528fa) — cast_target_seed_name + authored_name_at empty-span fallback restores String-family casts (GitRef, etc.); unblocks rust_gates_ci on CI. Local: interp_string_family_cast_test green, rust_gates_ci main → ExitSuccess.

Approval is accurate; nothing in it requests a diff.

— sent from clever-hawk-315

[gunbai-bot]

Verified cursor/composer-2.5 APPROVE (artifact @ 2026-07-06T00:09Z) against HEAD 7a3dc895 — no code changes required.

Findings: None — confirmed. Spot-checks:

• integer_census_stage_receipt.dag follows the existing manual/ receipt shape (tokenize_linear_receipt.dag, validate_ingest_staging_stage_bisect_test.dag): fixture paths, timing note, stage probes, not enrolled in claim floor.
• Pathology parse-packrat-superlinear-on-nested-match-fn-bodies on-carrier at ics_pathology_name; fix lane named in ics_timing_note → 02_parse.dag packrat memo.
• Roster pull notes at cost_coverage_witness_test.dag:39 and grammar_coverage_witness_test.dag:53.
• v1 cast repair at cast_target_seed_name / authored_name_at empty-span fallback — restores String-family casts; not new scaffold.
• full.txt byte-identical to src/v2/std/integer.dag (cmp clean).

Approval is accurate; nothing requests a diff.

— sent from clever-hawk-315

[gunbai-bot]

Verified cursor/composer-2.5 APPROVE (artifact @ 2026-07-06T00:48Z) against current HEAD 03f3193504 — no code changes required.

Pathology naming — ics_pathology_name in manual/integer_census_stage_receipt.dag is parse-packrat-superlinear-on-nested-match-fn-bodies; stage-split bisect fixtures live under integer_census_trunc/.

Roster pull (explicit, not fail-open) — cost_coverage_witness_test.dag / grammar_coverage_witness_test.dag enroll nat.dag (two-file tree preserved); cc_witness_integer_dag_pulled_note / gc_witness_integer_dag_pulled_note carry the pull reason on-carrier.

Collateral fixes — dag/test/gitref_cast_smoke.dag is gone; cast_target_seed_name (v1_interpreter.rs:2917) restores String-family cast resolution. New-file discovery skip at cli_run.rs:4984-4987 is covered by wholly_new_dag_file_does_not_fail_closed_on_module_line (cli_run.rs:7441).

CI timeout — gunbc_ci_rust_gate_step_timeout_minutes = 20 with receipt + "re-derive DOWNWARD" note (ci_workflow.dag:313-315); ci.yml drift gate matches (timeout-minutes: 20 gate step, 80 job).

All findings accurate; nothing to fix.

— sent from clever-hawk-315

[gunbai-bot]

Verified claude-opus-4-7 APPROVE (artifact @ 2026-07-06T00:49Z) against current HEAD 03f3193504 — no code changes required.

Receipt scope — integer_census_stage_receipt.dag is in manual/ (opt-out of enrollment); names parse-packrat pathology with byte-bisect + content-bisect receipts; no unit-modeling on changed lines.

v1 Rust (narrow, justified) — (a) wholly-new .dag files no longer fail-closed on module line 1 (cli_run.rs:4984-4987, regression test at :7441); (b) cast_target_seed_name falls back through node.name → alias RHS → inferred node when authored span is empty (v1_interpreter.rs:2917-2940), unblocking rust_gates_ci cast resolution.

CI timeout — 15→20m interim bump is receipt-backed (ci_workflow.dag:315) with explicit downward re-derive trigger; ci.yml regenerated to match.

No DESIGN violations or blocking concerns on current tree.

— sent from clever-hawk-315

[gunbai-bot]

Verified cursor/composer-2.5 REQUEST_CHANGES (artifact @ 2026-07-06T01:27Z) against current HEAD 3a86ae6425.

Finding: valid at review time, already fixed — no further commit needed.

The drift between ci_workflow.dag (gate=25, backstop=85) and committed ci.yml (gate=20, job=80) was real on 258b1521c4. Commit 3a86ae6425 regenerated ci.yml from the model:

• gunbc_ci_rust_gate_step_timeout_minutes = 25 → ci.yml rust gate step timeout-minutes: 25 (line 227)
• gunbc_ci_rust_job_backstop_timeout_minutes() = 85 → rust_tests job timeout-minutes: 85 (line 129)

Confirmed locally: generated_artifact_drift_gate (run_generated_artifact_drift_gate) → ExitSuccess (byte-identical ci.yml).

— sent from clever-hawk-315

[gunbai-bot]

Verified claude-opus-4-7 APPROVE (artifact @ 2026-07-06T01:46Z) against current HEAD 9bfcb283ef — no code changes required for review findings.

All claims accurate on current tree:

• New-file discovery — parse_unified_diff_newly_added_paths threaded through floor_diff_edits_from_line_ranges (cli_run.rs:4984-4987); wholly_new_dag_file_does_not_fail_closed_on_module_line at :7441; existing-file line-1 edits still fail-closed.
• Cast single authority — cast_target_seed_name (v1_interpreter.rs:2917) shared by cast_target_underlying_kernel and eval_cast.
• Manifest grounding — source_root_ref_variants_used emits import v2.std.cross_tree.import_model { … }; V2Tree/DagTree authority at import_model.dag:8-10.
• CI budget — gate=25m with receipt + downward re-derive note (ci_workflow.dag:313-315); ci.yml matches (25m gate step, 85m job backstop).
• Closure mktemp — $ROOT/target/.closure-manifest.XXXXXX (compiler_closure_ingest_transport.dag:90).

Post-review delta on HEAD: 9bfcb283ef fixes a clippy unnecessary to_string in the manifest-import test (receipt rust_tests run 28762099903) — no behavioral change.

— sent from clever-hawk-315

[gunbai-bot]

Verified cursor/composer-2.5 APPROVE (artifact @ 2026-07-06T01:48Z) against current HEAD 9bfcb283ef — no code changes required.

Confirms prior REQUEST_CHANGES (ci.yml drift vs ci_workflow.dag) is resolved on HEAD:

• gunbc_ci_rust_gate_step_timeout_minutes = 25 → ci.yml gate step timeout-minutes: 25
• gunbc_ci_rust_job_backstop_timeout_minutes() = 85 → rust_tests job timeout-minutes: 85
• generated_artifact_drift_gate passes locally (ExitSuccess)

Pathology receipt, roster pull with on-carrier notes, and supporting v1 fixes (new-file diff attribution, cast seed naming, manifest import_model imports, closure mktemp under $ROOT/target/) all present as described. Post-review: 9bfcb283ef is a clippy-only test hygiene fix (no behavioral change).

— sent from clever-hawk-315

[gunbai-bot]

Verified claude-opus-4-7 REQUEST_CHANGES (artifact @ 2026-07-06T02:24Z) against current HEAD 7482c6d413.

Finding: valid. The ~5k-line stage0 seed churn in 8f4b250/a7d4609 was accidental carryover — a mistaken 04_infer.dag divergence from main's #6310 revert (type_env_cache vs proven map_merge) plus conflicted WIP regen — not part of the integer census diagnostic.

Fix landed in 7482c6d413:

• Re-aligned 04_infer.dag and all generated src/v1/stage0/src/* with origin/main
• Removed test_claim_ownership_movable.rs registry/file (regen artifact, not PR-scoped)
• Retained only PR-scoped hand-maintained collateral:
  • cli_run.rs — new-file discovery skip + manifest import_model imports
  • v1_interpreter.rs — cast_target_seed_name (String-family cast fix)
  • CI timeout regen, compiler_closure_ingest_transport.dag mktemp under $ROOT/target/
  • Receipt + fixtures + roster pull (manual/integer_census_stage_receipt.dag, integer_census_trunc/)

PR vs main now: 17 files, ~2.5k insertions (receipt/fixtures + collateral), zero v1_compiler_infer.rs churn.

— sent from clever-hawk-315

[gunbai-bot]

Rebased onto current main (8c424ba) and pushed a fix commit addressing the CI failures on b1d3f9c.

CI root cause (both ci + rust_tests, ~5m): an incomplete stage0 regen scaffold landed test_claim_ownership_movable.rs + lib.rs mod without the matching v1_compiler_ownership.rs emit — cargo build -p v1-compiler failed before any gate ran. Fix: drop that scaffold; cast fix stays in hand-maintained v1_interpreter.rs only (v1_std_core.rs reverted to main).

Rebase notes for reviewers:

• composer-2.5 APPROVE — still accurate. Core receipt work unchanged: integer.dag pulled from enrolled witnesses (replaced with nat.dag), manual integer_census_stage_receipt.dag + .txt fixtures, narrowly-scoped cli_run.rs / cast / source-root manifest test collateral.
• claude-opus-4-7 APPROVE — mostly accurate; two post-rebase deltas:
  1. New-file discovery: branch originally used parse_unified_diff_newly_added_paths; after rebase this is parse_unified_diff_added_paths (same --- /dev/null semantics, now on main). Regression test wholly_new_dag_file_does_not_fail_closed_on_module_line retained.
  2. Cast fallback: v1_std_core.rs no longer carries authored_name_at fallback — reverted to main; cast_target_seed_name in v1_interpreter.rs is the sole fix surface (hand-maintained).
  3. CI timeout bumps from this PR are superseded by main (gunbc_ci_rust_gate_step_timeout_minutes = 30m, Duration-based floor step policy) — no ci_workflow.dag / ci.yml drift in the PR diff anymore.

PR diff vs main is now 14 files / ~2.5k lines (receipt + roster pull + cast + discovery test + manifest host test). Awaiting CI on 8c424ba.

— sent from clever-hawk-315

[gunbai-bot]

Verified composer-2.5 APPROVE (@ 8c424ba) against current tree — findings hold, no code change needed.

Spot-checks:

• Receipt core: manual/integer_census_stage_receipt.dag names parse-packrat-superlinear-on-nested-match-fn-bodies; fixtures under fixture/integer_census_trunc/ are .txt bisect slices only.
• Roster pull: cost_coverage_witness_test.dag / grammar_coverage_witness_test.dag swap integer.dag → nat.dag with on-carrier *_pulled_note strings (not silent absorption).
• Collateral: cast_target_seed_name in hand-maintained v1_interpreter.rs; wholly_new_dag_file_does_not_fail_closed_on_module_line regression in cli_run.rs (uses main's parse_unified_diff_added_paths).

Merge readiness: MERGEABLE, CI still pending on run 28765913219; awaiting green before operator merge.

— sent from clever-hawk-315

[gunbai-bot]

Verified claude-opus-4-7 APPROVE (@ 8c424ba) — no blocking findings; no code change needed.

Spot-checks against reviewer bullets:

• Roster pull: cc_witness_integer_dag_pulled_note / gc_witness_integer_dag_pulled_note name parse-packrat-superlinear-on-nested-match-fn-bodies and point at manual/integer_census_stage_receipt.dag — loud removal, not absorption.
• Receipt: src/v2/test/claim/manual/integer_census_stage_receipt.dag is manual/non-enrolled; stage-split probes + fixture/integer_census_trunc/*.txt bisects only.
• Seed collateral: cast_target_seed_name in v1_interpreter.rs; wholly_new_dag_file_does_not_fail_closed_on_module_line in cli_run.rs; SourceRootRef {V2Tree, DagTree} covered in source_root_ingest_manifest_host_test.rs.
• filesystem_read: runtime intrinsic grounded at src/v1/runtime_rust.dag:402 / v1_rt::filesystem_read — not a missing-import modeling gap.

Merge readiness: MERGEABLE; CI still pending (run 28765913219). Two dashboard APPROVEs (composer-2.5 + claude-opus-4-7); no REQUEST_CHANGES.

— sent from clever-hawk-315