Fix SG-0 census classification taxonomy

[briansrls]

Follow-up to #2653 / review at sha 40cead4.

Summary:

• Removes the ungenerated C5 audit class by teaching scripts/classify-sg0-census.py to classify the temporary ComplexitySummary / nested SymbolicCost expected-value blocker as existing Class C.
• Regenerates docs/audit/sg0-census-classification-2026-05-09.md from the current SG-0 census so the audit remains generated from A-G/STRUCTURAL only.

Test plan:

• python3 scripts/classify-sg0-census.py
• git diff --check
• cargo fmt --check

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: eb6bd45b · Trigger: schedule
• Thinking: 174s wall

---

Non-blocking — Strengths

• scripts/classify-sg0-census.py The generator keeps the SG-0 classification taxonomy derived and folds the ad-hoc C5 case into existing Class C without adding a parallel authority.

✅ No blocking concerns; the changed audit output and classifier rule are consistent with the Pure Bootstrap SG-0 census discipline.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: eb6bd45b · Trigger: manual
• Comparison: main @ cf28d320 ... session/silent-swift-300 @ eb6bd45b
• Conversation: View conversation

---

1. Story of the diff

This PR fixes the SG-0 census classifier so a temporary Rust receipt blocked by typed expected-value carrier gaps is no longer classified as an accidental C5 bucket, but folds into the real Class C taxonomy. The script change adds that rule in the classifier data table at scripts/classify-sg0-census.py:48, and the regenerated audit reflects it by reclassifying complexity_lens_behavioral_completion.rs as C | typed expected-value carrier gap at docs/audit/sg0-census-classification-2026-05-09.md:90. The audit snapshot also updates the live census shape: lens_testgen.rs moves out of the non-test file list, lens_testgen_body.txt is counted as a hand-authored fragment at docs/audit/sg0-census-classification-2026-05-09.md:193, new hand-authored tests are included in the test bucket, and the summary totals are recalculated at docs/audit/sg0-census-classification-2026-05-09.md:200-208.

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

N/A — the diff is an audit/classifier change only; it does not touch Dag-resident substrate types, dag.rs, cross-pass modeled facts, or new substrate variants.

1. INVARIANTS.md + modeling-discipline.md.

Compliant — Boundary Discipline / single authority is preserved by putting the taxonomy rule in the classifier table (scripts/classify-sg0-census.py:48) and regenerating the audit output from that rule (docs/audit/sg0-census-classification-2026-05-09.md:90), rather than leaving C5 as a parallel ad hoc class.

1. CODING.md.

Compliant — the implementation follows the data-table style already used by the script: a new regex row is added to the classification table at scripts/classify-sg0-census.py:48, with no new hidden state, object method surface, or branchy special-case helper.

1. TESTING.md.

Compliant — this is a generator/taxonomy fix, and the same PR includes the regenerated behavioral receipt: the old C5 row becomes Class C at docs/audit/sg0-census-classification-2026-05-09.md:90, and the summary totals are updated at docs/audit/sg0-census-classification-2026-05-09.md:200-208. I do not see a need for a separate compiler behavior test because no compiler runtime behavior changes.

1. LOCKED DESIGN DECISIONS.

Compliant — the diff does not weaken the Pure Bootstrap zero-floor target; it makes the census more faithful by separating EXPECTED_HAND_AUTHORED_FRAGMENTS at docs/audit/sg0-census-classification-2026-05-09.md:188 and increasing the visible total/untagged counts instead of hiding them at docs/audit/sg0-census-classification-2026-05-09.md:207-208.

1. TRACKED vs UNTRACKED DEBT.

Compliant — this PR does not introduce a new scaffold as implementation; it records census debt. Where debt is not fully tracked, the audit keeps that visible, for example lens_testgen_body.txt remains — UNTAGGED at docs/audit/sg0-census-classification-2026-05-09.md:193, and the summary reports Untagged explicitly at docs/audit/sg0-census-classification-2026-05-09.md:208.

2.5. Top-down PM intent review

Compliant — the highest-level intent here is to keep SG-0 census classification honest while driving the hand-authored surface toward zero, not to make residual Rust more acceptable. The diff supports that: it fixes an invalid/accidental taxonomy bucket through the classifier rule at scripts/classify-sg0-census.py:48, preserves the fragment bucket explicitly at docs/audit/sg0-census-classification-2026-05-09.md:188-193, and reports the increased total and untagged counts at docs/audit/sg0-census-classification-2026-05-09.md:207-208 rather than semantically diluting the zero-floor plan.

3. Verdict

APPROVE

The PR is narrow and consistent: classifier authority and generated audit output move together, the old C5 taxonomy leak is dissolved into Class C, and the census remains honest about fragment and untagged debt. I found no blocking invariant or PM-intent mismatch in the changed lines.