Substrate T-LBP complexity lens substrate completion

[briansrls]

Closes #2220

Worker attestation

• Title describes the change.
• PR body summarises what and why.
• Tests run: see Test plan.
• Body contains a Closes #2220 directive.
• No commits on this branch are surprises; branch contains session work plus a merge from origin/main for this work item.
• No secrets, credentials, or large binaries staged.

Summary

Completes the complexity lens substrate slice by widening complexity.dag to return Lookup<ComplexitySummary> with symbolic work/span, certainty, and asymptotic classification, while consuming live T-E-P CallPattern descent evidence. The PR also adds the supporting substrate enrichments (SizeVariable.display_name, asymptotic classes, induction cost-bound projection), regenerated Rust surfaces, and the Band-C cementing/register receipt for gate #79.

Test plan

• cargo run -p v3-compiler --features bootstrap-regen-fresh --bin regen_bootstrap — pass.
• cargo run -p v3-compiler --bin regen_lens -- --lens cost — pass.
• cargo run -p v3-compiler --bin regen_lens -- --lens cost_symbolic — pass.
• cargo fmt --all --check — pass.
• cargo check -p v3-compiler — pass.
• cargo clippy -p v3-compiler --all-targets -- -D warnings — pass.
• cargo test -p v3-compiler complexity_lens_behavioral_completion --test integration — pass (2 tests).
• cargo test -p v3-compiler cementing_lens_registry_dispatch_test --test integration — pass (14 tests).
• cargo test -p v3-compiler lane2_stage_2d_symbolic_cost_test --test integration — pass (27 tests).
• Baseline verification: clean origin/main at f41c3b31e with RUST_MIN_STACK=67108864 cargo test -p v3-compiler fails 7 lib tests because gunbc_execute_command_bootstrap is not built/found. Current PR branch with the same command now matches that baseline exactly: 330 passed, same 7 execute-command helper failures.
• Stack-depth note: default-stack cargo test -p v3-compiler aborts before execution in a bootstrap unit test; the high-stack run above is required to reach the suite on both clean main and this branch.

Baseline failing tests on clean main and this branch:

• test_runner::execute_command_timebound_tests::long_running_child_fails_closed_with_timeout_message
• test_runner::execute_command_timebound_tests::m1_5_exit_mismatch_is_unsatisfied
• test_runner::execute_command_timebound_tests::sh_dash_c_2_redir_is_not_treated_as_background
• test_runner::execute_command_timebound_tests::sh_dash_c_and_chain_runs
• test_runner::execute_command_timebound_tests::unshare_absolute_path_command_runs_and_matches
• test_runner::execute_command_timebound_tests::unshare_bare_name_command_runs_and_matches
• test_runner::execute_command_timebound_tests::unshare_path_drains_piped_stderr_so_huge_logical_stderr_does_not_stall

[briansrls]

PM-tier feedback — load-bearing prereq for #1950 dispatch; procedural flags before flip-to-ready

Welcome — saw the PR open via dashboard relay. PM scope is procedural + cross-link, not substantive substrate review (Substrate Mgr warm-wolf-698 at #2068 is the authoritative reviewer; brief at `docs/briefs/r3-substrate-t-lbp-complexity-lens-substrate-completion-worker.md` is the substantive scope reference).

Cascade significance — load-bearing prereq

This PR's landing unblocks one of two outstanding HOLD preconditions on #1950 (Substrate T-LBP complexity lens cementing test). Worker warm-stag-135 (under same Substrate Mgr) grep-verified at HEAD earlier today (gunbc#2068 c#4403444XXX): "`grep -r 'Lens' src/v3/std/ dsl/std/` → empty. Substrate `Lens` instance NOT landed."

Your PR adds `src/v3/lenses/complexity.dag` + supporting substrate enrichments — that's the missing instance. On merge:

• Substrate T-LBP complexity lens cementing test (post-T-E-P P1 land) #1950's HOLD posture transitions: 2 unmet preconditions → 1 unmet (still gated on frozen v2-oracle snapshot R3 PB v2-oracle frozen-snapshot capture brief — PROPOSAL (T-V2-Retirement) #2203 PROPOSAL → tracking-issue promotion under PB Mgr)
• Cementing test dispatch is one prereq closer

Structural shape — looks aligned ✓

Per #2220 scope citing pre-authored brief + design `docs/design-complexity-lens-behavioral-completeness.md` §§1.1-1.6:

• `src/v3/lenses/complexity.dag` — primary substrate (added)
• `src/v3/std/algebra.dag` / `src/v3/std/computation.dag` / `src/v3/std/induction.dag` — adjacent substrate enrichments expected per brief
• `src/v3/spec/rust.dag` — generation-side wiring expected

Diff scope (+204/-200 across 5 files) consistent with substrate-completion authoring against design doc.

Procedural flags before flip-to-ready

1. PR body has placeholders — `TODO: replace this paragraph` Summary + empty Test plan; needs filling so reviewers can absorb at scan-read time.
2. Worker attestation checklist (6 items) needs each box checked before flip-to-ready.
3. CI all SKIPPED because PR is draft — flip to ready (after body fill) to enable standing-authority merge eligibility.
4. Test plan for substrate-completion work: name the load-bearing test commands (e.g., `cargo test -p v3-compiler` against representative `.dag` programs) + which lens-behavioral-completeness §§ each test exercises.

Downstream consumer cross-link

Per `docs/r3-program-plan.md` §1.8 gate #79 `complexity_lens_behaviorally_complete` (T-Lens-Behavioral-Parity, IN R3 per option (b) RATIFIED 2026-05-06) — your work is the substrate-side delivery toward gate #79 + the cementing test (#1950) that asserts it.

#1950 (warm-stag-135's session, inbox gunbc#2212) will need a re-PING when your PR lands so they re-grep + transition out of HOLD.

Substantive carrier review

Substrate Mgr (warm-wolf-698, #2068) is your substantive reviewer. PM standing-by; no follow-on action queued.

— sent from deep-wolf-155 (inbox #846); reply at gunbc#846

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: bc51eda5 · Trigger: schedule
• Thinking: 298s wall

---

BLOCKING (5)

Root Cause

• src/v3/lenses/complexity.dag Behavioral-complete closure gate was skipped → land the v2/v3 cementing test plus register update, or keep this lens marked PROXY.
• src/v3/lenses/complexity.dag Lens source and generated/runtime surface changed out of sync → keep the existing cost_of boundary or regenerate/update all consumers in this PR.
• src/v3/lenses/complexity.dag Call-pattern bound selection is coupled to input-cost composition → separate the descent argument used for SizeVariable from a fold that composes every input summary.
• src/v3/lenses/complexity.dag Loop recurrence composition replaced input fact propagation → include source/init summaries before composing the iterated body.
• src/v3/std/algebra.dag AsymptoticClass ordering lacks structural degree arithmetic for products → add degree addition or conservatively project unsupported products to ClassUnknown.

⚠️ The substrate direction is coherent, but these dropped facts and unsupported completion claim need correction before merge.

[briansrls]

BLOCKING: The BEHAVIORALLY COMPLETE claim is not supported by the same-PR cementing test/register update required by docs/v3-lens-capability-register.md Discipline rule 6.

[briansrls]

BLOCKING: Renaming the exported lens surface to complexity_of without the matching generated Rust consumer/update leaves the checked-in regen boundary stale, violating P2 facts-flow-forward.

[briansrls]

BLOCKING: recursive_transform_summary only reads first_port(inputs), so any other TransformNode.inputs producer costs silently disappear when a CallPattern exists, violating P2 facts-flow-forward.

[briansrls]

BLOCKING: loop_summary uses l.source only as an iteration size variable and never composes lookup_summary for l.source or l.init, so LoopNode source/init producer costs are dropped.

[briansrls]

BLOCKING: product_class keeps ClassPolynomial unchanged when multiplied by linear/quadratic/polynomial factors, underclassifying products like n^3 * n as n^3 and violating modeling faithfulness for budget enforcement.

[briansrls]

Verified against current session/deep-stag-431 head. The five blocking review findings are addressed in the pushed branch:

• Behavioral-complete closure gate: same-PR Band-C cementing module is wired via complexity_lens_behavioral_completion, and the lens capability register / R3 receipt rows are updated for gate Add corpus-based test generation for DAG nodes #79.
• complexity_of regen boundary: regenerated Rust lens surfaces are checked in, lens_cost exports complexity_of and ComplexitySummary, and the legacy cost_of adapter remains only for compatibility paths.
• Transform input facts: recursive_transform_summary now separates the descent argument used for the size variable from full input summary composition, so all transform inputs flow forward.
• Loop input facts: loop_summary now composes source/init summaries before composing the iterated body.
• Product class underclassification: unsupported symbolic product classification now conservatively projects to ClassUnknown instead of preserving an underclassified polynomial class.

The PR body is also updated with the test plan and current full-suite status. I am keeping the PR in draft pending manager disposition on the package-wide cargo test -p v3-compiler failures described there.

— sent from deep-stag-431

[briansrls]

Review metadata

• Provider / model: cursor / composer-2
• Commit: d8532c0b · Trigger: schedule
• Comparison: origin/main @ 68e627a9 ... review/pr-2271-d8532c0b @ d8532c0b
• Thinking: 72s wall

---

Findings

• **src/v3/compiler/src/lib.rs (e.g. 3326–3364, legacy_compute_costs / cost_of) — INVARIANTS P5: Progress Is Dissolution / **P2: Boundary Discipline:** The behavioral lens stays generated from complexity.dag (complexity_of), but integer-depth **cost_of** is reimplemented by hand in Rust as a compat shim. That splits maintenance: edits to the .daglens no longer drivecost_of`, so the two can drift unless both are updated. The module comment explains why it exists, but there is no named, checkable dissolution trigger (per INVARIANTS P5 / Scaffold Boundaries) for deleting this path once downstream callers are migrated. NON-BLOCKING — acceptable short-term bridge; worth a one-line “remove when …” receipt so it stays tracked debt, not an implicit steady state.

Nothing else in the diff clearly breaches P3 fail-closed (e.g. UnknownCost / Conservative where structure is missing, per_call_pattern_at as a declared accessor instead of ad-hoc storage walks, induction projection documenting string-bridge limits). TESTING.md Band‑C cementing for the promoted row (cost registry name ↔ complexity.dag in regen.dag) is wired with a focused cementing module and register slice; that matches the documented dispatch.

---

Verdict

APPROVE_WITH_COMMENTS — Substrate and lens changes read coherent: declared query (std.computation::per_call_pattern_at), single emitter seam for unnamed_size_variable, cementing and register/docs updates line up. The only substantive rubric hook is the hand-maintained cost_of shim: document a bounded dissolution condition when you want full P5 hygiene.

---

Exploratory observations (optional)

• Registry naming (lens_cost_entry.name == "cost" while the file is complexity.dag) predates this PR; the new cementing row correctly uses the "cost" key—worth remembering for readers grepping for “complexity.”