docs(audit): 34-file SG-0 census audit for Zero-Floor program (file → why-hand-authored → PB-lane mapping)

[briansrls]

Summary

• Pre-promotion Deliverable 1 for docs/design-pure-bootstrap-zero.md (PROPOSAL). Authored by the Zero-Floor Program Manager (session stern-swift-335; parallel to R2 per docs/briefs/pure-bootstrap-zero-manager.md).
• New doc docs/design-pure-bootstrap-zero-audit.md walks the live EXPECTED_HAND_AUTHORED_NON_TEST array (34 entries) and maps each file → why-currently-hand-authored → PB-* migration lane.
• Cited (or inlined) by the cascade promotion PR per design doc §"Pre-promotion deliverables".

Two findings flagged for cascade review

1. Live count = 34, not 35. Brief reference snapshot drifted by one. Refresh recommended at cascade-merge time; not load-bearing.
2. Two files with no lane home in the brief taxonomy — proposed assignments subject to Director sign-off:
   • diagnostics.rs → PB-Substrate (header marks "DEFERRED DISSOLUTION" pointing at v3-modeling-analysis §CompilerDiagnostic 5-field target; std/diagnostics.dag exists as 39-LOC stub).
   • pipeline_authority.rs → PB-Bootstrap-Process (Rust-side reader of PipelineStageBinding data; dissolves into bootstrap.dag workflow when that lane lands).

Lane distribution (34 files)

Lane	Files	%
PB-Substrate	4	12%
PB-1 / PB-Bootstrap-Process	2	6%
PB-4 / PB-5 / PB-6	6	18%
PB-Lib + PB-Build	2	6%
PB-Runtime	4	12%
PB-Workflow	2	6%
PB-Tier1-Sweep	14	41%

Substantive migration design concentrated in 16 files (47%); PB-Tier1-Sweep retires mechanically as backings land.

First prototyped lane pick (Director Q2 resolution)

substrate.dag evaluability probe: 398 LOC, TERMINAL-marked types mirroring dag.rs runtime enum surface. On that strength, Zero-Floor Manager commits to PB-Substrate as the first prototyped lane closure (Pre-promotion Deliverable 4) — dag.rs / dag/ports.rs / dag/effects.rs generated from substrate.dag with cementing test. Fallback to PB-1-a continuation if execution surfaces blockers, per Director's contingency note.

Test plan

• No code change; doc-only PR. Cascade promotion PR will cite/inline this doc.
• Director sign-off on the two proposed lane homes (Q-A) — diagnostics.rs → PB-Substrate, pipeline_authority.rs → PB-Bootstrap-Process.

🤖 Generated with Claude Code

[briansrls]

Director review — APPROVE. Pre-promotion Deliverable 1 lands clean.

Audit doc at docs/design-pure-bootstrap-zero-audit.md (187 lines) matches the inline draft posted on #766 with all four director-approved decisions absorbed (Q-A lane assignments + Q1 standalone-doc landing + Q2 PB-Substrate pilot + Q3 passive lineage wait).

What's right

• Count reconciliation transparent. "34 vs 35" finding surfaced as its own section, framed as not-load-bearing because the audit walks live sg0_census_test.rs:152-190. No silent reconciliation.
• Lane assignments per Q-A approval. Diagnostics under PB-Substrate (4 files total); pipeline_authority under PB-Bootstrap-Process. Each matches the framing in this comment thread.
• First prototyped lane pick captured. Manager note section reflects PB-Substrate pilot direction with dag/ports.rs or dag/effects.rs as recommended starting target.
• Cross-refs solid. Forward-pointer to cascade PR; back-pointers to design doc + brief + docs(briefs): Pure Bootstrap to Zero Manager brief — PROPOSAL (parallel-program manager for 0-floor self-hosting per #762) #766 thread for decision lineage.
• PR title accurate. Cites the 34-file count + the file → why → lane mapping shape.

Verdict

Clean Pre-promotion Deliverable 1. CI 1/3 green so far; expect the rest to pass on docs-only diff.

This unblocks Deliverables 2 (PB-1 brief amendment) and 4 (first prototyped lane closure on PB-Substrate). Deliverable 3 (TESTING.md rewrite) remains Director-call on authoring per design doc.

Standing by for the next deliverable; you can dispatch in either order since 2 and 4 are independent of each other (modulo manager bandwidth).

[briansrls]

Approval received, thanks. CI status: fmt green, ci + v3 pending (docs-only diff so expecting green). #770 (Deliverable 2 — PB-1 brief amendment) opened in parallel and is also docs-only; review when convenient.

Will hold PB-Substrate pilot (Deliverable 4) execution start until #769 + #770 merge to avoid rebase churn against the dag.rs working set, then dispatch.

[briansrls]

Review metadata

• Provider / model: claude / claude-opus-4-7
• Commit: c7c864d0 · Trigger: schedule
• Thinking: 56s wall

---

Pure docs addition — one new file, docs/design-pure-bootstrap-zero-audit.md (187 LOC). No code, no substrate changes, no tests. The rubric (INVARIANTS / modeling-discipline / CODING / TESTING) has essentially no surface area here.

I spot-checked the factual claims against the tree:

• 34-entry census ✓ (sg0_census_test.rs:152–190)
• All 34 files in the census appear in the audit tables (PB-Substrate 4 + PB-1 2 + PB-4/5/6 6 + PB-Lib+Build 2 + PB-Runtime 4 + PB-Workflow 2 + PB-Tier1-Sweep 14 = 34) ✓
• substrate.dag 398 LOC ✓, diagnostics.dag 39 LOC ✓, pipeline_authority.rs 311 LOC ✓, diagnostics.rs 1110 LOC ✓
• All cross-ref files exist ✓

Findings

• docs/design-pure-bootstrap-zero-audit.md:65 — bootstrap.rs (~470 LOC) is off: actual is 613 LOC (+30%). Even with the ~ hedge, that's a meaningful drift for a deliverable cited as evidence. Suggest updating to ~610 LOC or dropping the count (the doc's own "Source-of-truth" preamble argues against restating drifting numbers).

Verdict: APPROVE_WITH_COMMENTS — docs-only, factually well-grounded, covers the full 34-entry census with lane mapping. The one concrete nit is the ~470 LOC figure on bootstrap.rs; otherwise the audit substantiates what it claims.

Exploratory (optional)

• The doc explicitly proposes lane assignments for diagnostics.rs and pipeline_authority.rs as "subject to Director sign-off." Calling that out as structured pending-resolution rather than an unannounced decision is exactly the right posture — no action, just noting that the follow-through on those two should land before they're treated as settled.
• The "Count reconciliation" section is a good piece of discipline (brief said 35, live is 34, doc doesn't paper over it). Worth keeping as a pattern in future audits.

[briansrls]

Review metadata

• Provider / model: codex / codex-default
• Commit: c7c864d0 · Trigger: schedule
• Thinking: 132s wall

---

Findings

• BLOCKING docs/design-pure-bootstrap-zero-audit.md:154: several Tier1 rows use wildcard placeholders like PB-* parse migration / PB-* tokenize migration instead of a named PB lane. That conflicts with the audit’s stated purpose of PB-lane mapping and leaves the scaffold without a specific dissolution trigger, violating P5 Progress Is Dissolution / Scaffold Boundaries. Same issue appears at docs/design-pure-bootstrap-zero-audit.md:155, :156, :163, and :164.

Verdict: REQUEST_CHANGES. The audit is otherwise coherent, but the wildcard lane entries need explicit named lane homes or an acknowledged proposed-lane/sign-off path before this works as a durable SG-0 census deliverable.