diff --git a/docs/audit/r3-pb0-velocity-walk-2026-05-09.md b/docs/audit/r3-pb0-velocity-walk-2026-05-09.md
new file mode 100644
index 00000000000..cd70098c6c6
--- /dev/null
+++ b/docs/audit/r3-pb0-velocity-walk-2026-05-09.md
@@ -0,0 +1,252 @@
+# R3 PB-0 Velocity Walk + SG-0 Census Trajectory — 2026-05-09
+
+**Author**: deep-wolf-155 (PM)
+**Authority scope**: PM-tier audit. Director-greenlit follow-up to PR #2300 cluster analysis (Director acknowledgment relayed at [gunbc#846 #issuecomment-4411924843](https://github.com/gunb-ai/gunbc/issues/846#issuecomment-4411924843), 2026-05-09; subsequent ratification + partner-work delegation at [gunbc#846 #issuecomment-4412008376](https://github.com/gunb-ai/gunbc/issues/846#issuecomment-4412008376)).
+**Parent docs**:
+- [`docs/r3-program-plan.md`](../r3-program-plan.md) §1.8 gates **#8** + **#84** (Pure-Bootstrap-Zero closure gates)
+- [`THESIS.md`](../../THESIS.md) "Pure Bootstrap to Zero" framing — quote: "v3's trajectory is the Pure Bootstrap to Zero program (0 hand-maintained)" (section anchor per `feedback_section_anchors_over_line_numbers` — line numbers drift)
+- [`ROADMAP.md`](../../ROADMAP.md) **T-PB-A** lane row (`pb_hand_rust_at_shim_floor` predicate — non-test SG-0 census → 0) + **T-PB-B** lane row (`pb_rust_tests_outside_residual_zero` predicate — test census → 0)
+- [`docs/audit/r3-cluster-analysis-2026-05-09.md`](r3-cluster-analysis-2026-05-09.md) — prior cluster analysis (under-weighted #8/#84)
+
+---
+
+## §0. TL;DR — load-bearing finding
+
+**The SG-0 census is growing, not shrinking.** Over the last 9 days the ratchet went from 119 entries (2026-04-30) to **149 entries (2026-05-09 audit-time snapshot)** — net **+30 entries**, **+3.3/day** average. R3 close requires this number to reach **0** (gate #8 non-test = 0 + gate #84 test = 0).
+
+> **Snapshot scope (added 2026-05-09 post-codex BLOCKING review; refined 2026-05-09 post-second codex BLOCKING)**: numbers below are **audit-time snapshots** at indicated git sha; not refreshed as `origin/main` advances. Live source-of-truth for the SG-0 census itself is [`src/v3/compiler/tests/integration/sg0_census_test.rs`](../../src/v3/compiler/tests/integration/sg0_census_test.rs) — `EXPECTED_HAND_AUTHORED_NON_TEST` + `EXPECTED_HAND_AUTHORED_TEST` + `EXPECTED_HAND_AUTHORED_FRAGMENTS` arrays are the canonical authority. A daily-cadence trajectory tracker artifact lands via sibling PR #2361 (Cluster M remediation program); cite-once-merged. The trajectory finding (growth rate ≥ +3.3/day; gates cannot reach zero at observed velocity) is **structural** and remains valid regardless of point-in-time count drift; specific count cells below should be read as "as of the audit window" not "as of HEAD now."
+
+| Date | non-test (#8 target=0) | test (#84 target=0) | total | Note |
+|---|---|---|---|---|
+| 2026-04-30 (HEAD~500) | 38 | 81 | 119 | retroactive baseline (excludes 1 fragment entry) |
+| 2026-05-02 (HEAD~300) | 40 | 87 | 127 | retroactive |
+| 2026-05-06 (HEAD~150) | 46 | 89 | 135 | retroactive |
+| 2026-05-07 (HEAD~50) | 47 | 95 | 142 | retroactive |
+| **2026-05-09 (audit-time ~`c25b2d8df`)** | **48** | **101** | **149** | **audit-time** snapshot (excluding fragments); 150 fragments-inclusive |
+
+**Numbers in this audit are point-in-time at the audit window**; main has advanced since. For live counts, run the snapshot procedure against `src/v3/compiler/tests/integration/sg0_census_test.rs` directly (`grep -E '"src/v3/' EXPECTED_HAND_AUTHORED_*` arrays). The trajectory tracker artifact lands via sibling PR #2361 (Cluster M remediation); cite-once-merged.
+
+**At current trajectory the gates never close.** The path to zero is structural — not per-file dissolution at observed velocity. It depends on **single bulk-dissolution events** firing.
+
+---
+
+## §1. Why the census grows despite SG-0 PR-window net-shrink discipline
+
+Per [`ROADMAP.md`](../../ROADMAP.md):177 the SG-0 PR-window discipline allows three pairings for `+N` deltas:
+- **(a)** same-PR retirements listing removed paths
+- **(b)** Director-budget citation (issue/comment URL)
+- **(c)** structural deferral with named follow-up dispatch
+
+**Option (c) is the dominant path.** **About 32 of 101 test entries** in `EXPECTED_HAND_AUTHORED_TEST` (audit-time snapshot per §0 table — total 101 test entries at sha `c25b2d8df`) have an explicit header-comment naming a "dissolution trigger" (revised 2026-05-09 per codex BLOCKING — prior audit overclaimed "every entry"); the remaining **~69 entries** (101 − 32 = 69) lack inline dissolution-trigger comments. Where present, the trigger fires **structurally** when a downstream capability lands, not via per-PR retirement. The result: each new gate or feature can land with hand-Rust acceptance + a deferred-dissolution comment, growing the census.
+
+The **~69 entries without explicit comments are potentially untracked hand-Rust debt**: they may have *implicit* dissolution paths (e.g., m1/m2 boundary tests dissolve via T-V2-Retirement + Tests-As-Data; sg* tests dissolve via Tests-As-Data; common/ helpers dissolve when their downstream test consumers retire) but lack the per-entry header comment that the SG-0 PR-window option-(c) discipline assumes. **This is itself an audit finding** (added 2026-05-09 per codex BLOCKING — INVARIANTS P1/P5): the option-(c) discipline assumes per-entry dissolution-trigger documentation; entries lacking that documentation operate as untracked deferral.
+
+PM follow-up (Task 13 effectively): per-entry audit of the ~69 commentless entries to either (a) classify each under an existing cluster's bulk-dissolution event or (b) flag as untracked debt requiring fresh substrate authoring or comment-attribution PR.
+
+This is by design for the entries WITH triggers (the cascade-promotion 2026-04-25 shape). What it requires: **the named dissolution triggers actually fire**, AND **commentless entries get classified or comment-attributed**.
+
+---
+
+## §2. Root-cause partition — what dissolves each class
+
+**Methodology scope (added 2026-05-09 per codex BLOCKING)**: this partition is a **cluster-level** estimation, NOT a per-entry mechanical audit. The trigger-class counts below are derived from (a) inspection of header comments where present (~32 of 101 test entries per §1) + (b) inferred classification of commentless entries based on filename pattern (e.g., `m1_*` boundary tests → T-V2-Retirement + Tests-As-Data; `sg*_test.rs` → Tests-As-Data infrastructure). **Per-entry verification is Task 13 follow-up** (UNACCOUNTED entries grep). The cluster-level partition supports the §3 velocity-math finding (bulk-events-required) without requiring per-entry attribution; both §1 + §3 conclusions are reproducible at cluster-level.
+
+Reading the per-entry "Dissolution trigger" / "Dissolves when..." comments in `sg0_census_test.rs` (where present) + inferred classifications (commentless entries), the 150 entries cluster into a small number of trigger classes:
+
+### §2.1 Test entries (101 entries, gate #84) — dissolution triggers
+
+| Trigger class | Approx count | Dissolution event |
+|---|---|---|
+| **Testgen covers reflected-Dag structural assertions over std/ types** | ~25-30 | T-Tests-As-Data-Completeness Cluster M lands (#85 quantifier substrate + #86 program-generator carrier) |
+| **`.dag` TestClaim runner can execute generic DimensionReport / cementing assertions** | ~20-25 | Gate #84 + #87 (cementing-test discipline complete) — same family as above |
+| **PB-Runtime evaluator-as-data lands** | ~5-8 | Cluster F (T-LP-Retirement) — enables Row-4 / R2-Evaluator corpus comparison without host harness |
+| **`.dag`-fn-resolution-against-bootstrap as TestClaim** | ~5-8 | T-Tests-As-Data infrastructure (#1966 §3 ratchet predicate scope) |
+| **R1 close dissolves wrappers** (R1C-D / R1C-E scaffolds) | ~3 | R1 close (already past — these may be eligible NOW) |
+| **L4/L7/L5 skeleton retirement when TestRunner can evaluate directly** | ~5 | Verification Cluster G + M overlap |
+| **Specific carrier-test retirements** (each with own structural trigger) | ~10-15 | Various gates |
+| **Process-level: m1/m2 boundary tests** | ~10 | Likely T-V2-Retirement + T-Tests-As-Data |
+
+**Net**: ~80-90 of the 101 test entries collapse via **Cluster M (Tests-As-Data-Completeness)** landing. Cluster M is therefore a **bulk-dissolution event** for the test-side ratchet.
+
+### §2.2 Non-test entries (48 entries, gate #8) — dissolution triggers
+
+| Trigger class | Approx count | Dissolution event |
+|---|---|---|
+| **Bin-shims** (`src/bin/regen_*.rs`, `r1c_e_emit_gates.rs`, `self_host_fixed_point.rs`) | 9 | PB Item 5 (bin-shim emit pattern) — PR #2282 PROPOSAL merged; full retirement pending |
+| **Regen-emit support** (`regen_*_emit.rs`, `bootstrap_regen_fresh.rs`, `regen_tokenize.rs`) | 5 | Retires alongside bin-shims |
+| **Lens-producer family** (`lens_apply.rs`, `lens_testgen.rs`, `dimension.rs`) | 3 | Cluster F (T-LP-Retirement gates #5/#6/#7) — gated on PB-Runtime interpreter-as-data |
+| **emit/codegen** (`emit.rs`, `emit/python_target.rs`, `emit/rust_target.rs`, `emit_rust*.rs`, `emit/collection_ops_method_contract.rs`) | 6 | Self-host via PB-Runtime trampoline (gate #71) + T-V2-Retirement |
+| **dag substrate** (`dag.rs`, `dag/builder.rs`, `dag/effects.rs`, `dag/ports.rs`, `dag/cardinality_payload.rs`) | 5 | T-Tier3-Dissolution Cluster K (gates #1-#4) + dag.rs reflection-completeness work |
+| **infer/lower/test_runner** | 3 | Self-host via PB-Runtime + Cluster M for test_runner |
+| **bootstrap.rs** | 1 | PB-Runtime trampoline (gate #71) |
+| **R3 lane carries** (`tier3_mirror_perf.rs`, `workflow_idempotency.rs`, `workflow_parallelism.rs`, `omni_shape_b_openapi.rs`, `pb_method_template_projection.rs` × 2, `r1c_e_gates.rs`, `int_literal_ranges.rs`, `emit_rust_roundtrip_fixtures.rs`, `process_exit.rs`, `self_host_receipt_p0.rs`, `pipeline_authority.rs`, `post_emit_verifier.rs`, `diagnostics.rs`, `lib.rs`, `build.rs`) | 16 | Various lane-specific dissolutions; many tied to T-V2-Retirement + Cluster K + Cluster F |
+
+**Net**: the 48 non-test entries collapse via **PB-Runtime trampoline + T-LP-Retirement + T-V2-Retirement + T-Tier3-Dissolution** firing across Clusters F + K + E. No single event collapses all 48; multiple bulk-dissolution events needed.
+
+---
+
+## §3. Velocity-to-zero math
+
+### §3.1 At current per-file rate
+- 9-day window net delta: **+30 entries**
+- Current count: 149
+- Implied "weeks-to-zero": **never** (negative velocity)
+
+### §3.2 At observed bulk-dissolution event rate
+
+Recent cycle landings since 2026-05-06, partitioned into actual census reductions vs enabling-only landings (no immediate reduction):
+
+**Census-reducing landings** (the only events that count toward bulk-dissolution rate):
+- **PR #2279** (Coercion-Fold ScratchIntExamples retired) — reduced ~3 entries (substrate move collapses class)
+
+**Enabling-only landings** (substrate or scaffolding work that opens future bulk-dissolution paths but does NOT reduce census in-PR; some net-add):
+- **PR #2281** (G6 emit-shim coherence test) — added 1 entry; enables future v2-retirement bulk drop (gate #97 is the structural enforcement that fails closed when v2 retires without Grounding shim removal)
+- **PR #2271** (Substrate T-LBP complexity-lens substrate) — net 0 in census (substrate-shape-only)
+- **PR #2200** (T-E-P P1 Slice 6) — added entries (cementing tests under option-(c) deferral)
+
+**Observed bulk-dissolution rate** (using census-reducing landings only): **1 reduction event in 4-day window** ≈ ~0.25/day or ~0.5-1/cycle depending on cycle definition. Even at the upper bound, **not enough to collapse 149 entries inside 8-12 week R3 window via per-PR retirements**. Bulk events (per §3.3) are required.
+
+**Why this matters for §3.3**: rate measurement only counts census-reducing landings; enabling-only landings are tracked separately because they are *prerequisites* for future bulk events (e.g., #2281 enables T-V2-Retirement future drop) but do not themselves reduce the count.
+
+### §3.3 Required bulk-dissolution events for R3 close
+
+**Test side (gate #84 → 0)**:
+- **Cluster M COMPLETE** (Tests-As-Data #85/#86/#87 + #84 closure) → bulk-collapses ~80-90 test entries in single event
+- **Without Cluster M**: ratchet cannot reach zero (per-file is too slow)
+
+**Non-test side (gate #8 → 0)**:
+- **PB-Runtime interpreter-as-data fully landed** (gates #5/#6/#7 + Items 4+5 disposition) → ~14 non-test entries collapse (LP family + bin-shims + regen support + bootstrap)
+- **T-Tier3-Dissolution complete** → ~5 entries collapse (dag-substrate family)
+- **T-V2-Retirement complete (gate #42)** → ~10 entries collapse (emit/codegen + R3 lane carries)
+- **Self-host via PB-Runtime (gate #71)** → ~5 entries collapse (infer/lower/emit-side)
+- **Remaining ~14 entries**: each lane-specific; per-file or small-class dissolutions
+
+**At least 4-6 bulk-dissolution events required** for R3 close. Each is a substantial substrate / cluster milestone.
+
+---
+
+## §4. Reclassification: Cluster M is critical-path, not parallel
+
+The PR #2300 cluster analysis classified Cluster M (T-Tests-As-Data-Completeness) as a **parallel cluster** in §2 (worth 2-3 PRs). This audit corrects:
+
+**Cluster M is critical-path-load-bearing for the entire PB-0 closure thesis.** Without Cluster M COMPLETE:
+- Gate #84 (test side → 0) cannot close → SG-0 ratchet stays >0 → R3 close blocked
+- ~80-90 of 101 test entries cannot dissolve (no testgen capability)
+- Per-file retirement cannot make up the gap inside 8-12 week window
+
+**New dependency picture (replaces PR #2300 §2 Cluster M row)**:
+
+The dependency structure has **two distinct edge-classes** that operate on different axes; conflating them was the error in PR #2300 §2 and an early draft of this section (caught by codex review on PR #2358). Both views are below; the second is the "M → B → E sequencing-critical" referenced in §5 Risk 4 and PR #2300 §4 Risk 2.
+
+**View 1 — Substrate-flow critical path** (lane-input dependencies):
+
+```
+A T-E-P-Producer-Broadening
+   │
+   ├─→ B T-Lens-Behavioral-Parity (B consumes A's descent evidence + producer broadening)
+   │
+   └─→ M T-Tests-As-Data-Completeness (M's substrate carriers #85/#86 consume A; M does NOT depend on B)
+
+A → {B, M} (parallel post-A; B and M have no mutual substrate dep)
+```
+
+**View 2 — PB-0 closure flow** (what each gate needs to honestly fire):
+
+```
+M T-Tests-As-Data-Completeness COMPLETE  (gates #85/#86/#87 substrate-landed)
+   │
+   └─→ B's cementing tests can migrate from hand-Rust to `.dag` form
+        (current cementing tests live at `tests/integration/cementing/*.rs`)
+        │
+        └─→ B-cementing-test-capture-in-`.dag`-form COMPLETE (gate #79/#80 PB-0-honest closure)
+             │
+             └─→ E T-V2-Retirement (gate #42) — v2 oracle frozen via `.dag` cementing receipts
+```
+
+**View 2 is the "M → B → E sequencing-critical" referenced in §5 Risk 4 + PR #2300 §4 Risk 2.** It is **not** a substrate-flow contradiction with View 1; it is a *closure-honesty* sequencing — without M, B's cementing tests stay hand-Rust; without B-cementing-in-`.dag`, oracle freeze cannot be honest after v2 retires.
+
+**Net authoritative picture under INVARIANTS P2/P5**:
+- **Substrate flow**: A → {B, M} (B and M parallel-from-A; one canonical edge per lane-input dependency)
+- **PB-0 closure**: M → (B-PB-0-honest) → E (sequencing-critical for cementing-capture-before-v2-retires; one canonical edge per closure-readiness dependency)
+- The two edge-classes describe different relations (substrate-availability vs PB-0-honest-closure-readiness); both are simultaneously true for distinct nodes-pair purposes
+- Gate #84 (~80-90 test entries dissolve) requires M COMPLETE per View 1
+- Gate #79/#80 PB-0-honest closure requires View 1 (A → B substrate) AND View 2 (M before B-cementing-`.dag`-migration)
+- Gate #8 (non-test → 0) dependency: PB-Runtime + T-V2 + T-Tier3 (parallel cluster events for the bulk of non-test entries) **plus partial Cluster M overlap for `test_runner.rs` specifically** — the test runner retires when Cluster M's TestClaim system can drive testing end-to-end as `.dag` data (i.e., when #87 cementing-test discipline + bulk-port discipline land). Gate #8 is mostly orthogonal to M/B closure flow but not fully; per §2.2 row "infer/lower/test_runner" — `test_runner.rs` is the specific overlap entry. (Per openai-pro REQUEST_CHANGES on sha `cf89a699`: prior phrasing "orthogonal to M/B closure flow" was too strong; corrected to acknowledge partial overlap.)
+- gates #8 + #84 BOTH GREEN = PB-0 closure = R3 close minimum bar
+
+---
+
+## §5. Honest-close risk update (additions to PR #2300 §4)
+
+**Risk 5 (NEW) — SG-0 census growth trajectory**:
+At observed +3.3/day rate, the census is growing 3-4× faster than dissolution events fire. Without bulk-dissolution events firing in the next 4-6 cycles, R3 close becomes impossible inside the 8-12 week window. **Severity**: load-bearing for R3 thesis.
+
+**Risk 6 (NEW) — Cluster M dispatch / authoring status**:
+Cluster M's gates (#84/#85/#86/#87) are all DECLARED. No active worker visible at HEAD on any of the four. Verification Mgr (wise-bear-525 #2075) lane scope includes T-Tests-As-Data-Completeness, but recent dispatches focus on Pattern-A executable (Cluster G) and V7 ValueBody. **Cluster M needs explicit dispatch sequencing**; without active authoring, the bulk-dissolution event for ~80% of the test ratchet is not in flight.
+
+**Risk 4 (UPDATE) — #75 pr_anticipation_discipline_ci_active**:
+Earlier flagged as "PM/Debt-Paydown standing-program owned." Closer reading of [`ROADMAP.md`](../../ROADMAP.md):177: the script `scripts/check-pr-sg0-net-shrink-discipline.sh` already exists in CI. Gate #75 may be closer to CONSUMER_LANDED than the §1.8 status indicates — verify with Debt-Paydown Mgr.
+
+---
+
+## §6. PM-tier surfaces (recommendations, not authoring)
+
+These are surfaces for Director cycle absorption, not PM-authored directives:
+
+1. **Cluster M sequencing-criticality**: Director-tier sequencing decision on whether Cluster M should ride the M → B → E sequencing-critical chain (currently only B oracle-freeze is documented as sequencing-critical). M → bulk-dissolution → gate #84 = R3 close enablement.
+
+2. **§10 RED elevation candidates** (Director discretion):
+   - SG-0 census growth trajectory (Risk 5 above)
+   - Cluster M authoring status (Risk 6 above)
+
+3. **Brian-tier framing question**:
+   - Is the "Pure-Bootstrap-Zero by R3 close" claim still load-bearing for R3, or has it drifted into a longer-horizon goal? Current trajectory implies it cannot close inside 8-12 weeks unless Cluster M + 3-4 other bulk-dissolution events fire. If the claim is still load-bearing, dispatch needs to prioritize Cluster M. If it's drifted, the R3 plan §1.5/§1.8 framing needs reconciliation.
+
+---
+
+## §7. What's durable in this audit
+
+§2 partition (per-class dissolution triggers) is durable methodology — useful for any future PM walk against SG-0 census. §3 velocity math is point-in-time but the framework (per-file rate vs bulk-dissolution event rate) is reusable. §4 reclassification updates the cluster analysis structure.
+
+The growth-trajectory finding in §0 supersedes any earlier cluster-analysis claim that "structural execution fits in 4-6 days at 10 PRs/day." That was correct at the per-PR level but missed that **PRs are net-adding to the ratchet, not net-shrinking**.
+
+---
+
+## §8. Meta-finding: closure-claims-vs-HEAD drift (added 2026-05-09 per Director audit sweep)
+
+This audit + 2 parallel Director-tier audits (gunbc#828 sweep 2026-05-09 via Director-spawned Explore agents, surfacing 6 additional drift items at gunbc#846 #issuecomment-4412017502) converge on a meta-finding: **R3 program-plan claims drift in the same direction — running ahead of HEAD reality**.
+
+Specific instances surfaced across the 3 audits — each grounded against a landed authority (verified column added 2026-05-09 per codex BLOCKING):
+
+| # | Drift instance | Source | Verification (landed authority) |
+|---|---|---|---|
+| 1 | **§1.8 status drift** — ~9 gates likely promotable to CONSUMER_LANDED but ledger Status not refreshed | [`docs/audit/r3-cluster-analysis-2026-05-09.md`](r3-cluster-analysis-2026-05-09.md) §1 (PR #2300, on main) — "9 gates likely-promotable from DECLARED → CONSUMER_LANDED. 88 → ~79 still-DECLARED if Mgrs refresh ledger" | grep-verified against `docs/r3-program-plan.md` §1.8 + recent merged PRs (#2251 / #2208 / #2271 / #2147 / #2190 / #2160 / #2288 / #2281) |
+| 2 | **SG-0 trajectory drift** — census growing despite ratchet-to-zero framing | this audit §0 | grep-verified against `src/v3/compiler/tests/integration/sg0_census_test.rs` history (HEAD~500 → HEAD count progression) |
+| 3 | **TC1 #11 plan-language drift** — claimed "flips PASSING on E3.c merge" vs ratified "DECLARED through R3" | Director ask 6 (#issuecomment-4412017502) | grep-verified against `docs/r3-program-plan.md` §1.8 row #11 + Director (a)-disposition `473b99fb...`; addressed in PR #2361 sha 6efde884c |
+| 4 | **10 demonstration gates runtime-path drift** — #65-#74 DECLARED with no runtime path | Director ask 7 | grep-verified against §1.8 ledger rows #65-#74 (all DECLARED at HEAD); per-gate dissolution-trigger audit deferred to Ask 7 follow-up |
+| 5 | **Substrate-gap-class #61 enumeration drift** — §1.4 conjunctive close vs §2.4 one-sided resolution | Director ask 8 | grep-verified against `docs/r3-program-plan.md` §1.4 line 70 + §2.4 |
+| 6 | **Gate-count canonicalization drift** — 97/94/93 ambiguity | Director ask 9 | grep-verified against `docs/r3-program-plan.md` §1.5; addressed in PR #2361 sha 2e782f219 (now 97/96 forward-looking framing) |
+| 7 | **Gate #95 carve-doc cross-ref drift** — `r3-structure.md` cites C1; `r4-carve-out-routing.md` doesn't enumerate #95 | Director ask 10 | grep-verified against `docs/r3-structure.md`:38/146 + `docs/r4-carve-out-routing.md`; dissolved entirely by Task 12 PR #2364 (carves dissolve) |
+| 8 | **§10.3 ratification ledger publication drift** — Q-items cited without inline disposition | Director ask 11 | grep-verified against `docs/r3-program-plan.md` §10.3 + ROADMAP/r3-structure citations; deferred to Ask 11 follow-up (housekeeping) |
+| 9 | **R4-carve hand-Rust drift** — `workflow_parallelism.rs` + effect_enum walker stay hand-Rust per C1/C2 | PM ask 2026-05-09 at #828 #issuecomment-4412052024 | grep-verified against `src/v3/compiler/src/workflow_parallelism.rs` + design doc; resolved by Director (a) ratification at #issuecomment-4412330468 + Task 12 PR #2364 |
+
+### Pattern shape
+
+Every instance is the same structural shape: **document text asserts a closure-state that HEAD does not satisfy**. Either:
+- Plan asserts gate will close on event X, but X requires post-R3 substrate (drift instances 3, 7)
+- Plan asserts ratchet trajectory shape, but HEAD trajectory diverges (instances 1, 2, 9)
+- Plan asserts conjunctive closure but only one disjunct shipped (instance 5)
+- Plan internal cross-references drift (instances 4, 6, 8)
+
+### Standing recommendation
+
+**Future R3 readiness audits should grep for status-vs-HEAD before trusting plan-text.** Specific tooling proposal:
+- Per Mgr cycle: each lane-Mgr cross-checks own §1.8 rows against HEAD evidence; reports drift to PM
+- PM-cycle: weekly status-vs-HEAD grep across §1.8 + §10.3 (using §1.8 ledger gate IDs as grep anchors to `dsl/std/`, `src/v3/std/`, and CI consumer paths)
+- This artifact (and the SG-0 trajectory tracker landing via sibling PR #2361, file `r3-sg0-trajectory-tracker.md` once merged) should be folded into Director-tier progress-bar authoring per operator ask 2026-05-09; honest-trajectory visualization replaces lane-completion-proxy framing
+
+The meta-finding is structural-not-personnel: **plan-text drifts because the HEAD-vs-text reconciliation cadence is not in the standing PM/Director cycle**. Adding it explicitly closes the drift class.
+
+---
+
+**End of audit.**