docs: promote verified H4 L5 closure state

CURRENT_FOCUS.md

@@ -1,8 +1,10 @@
 # Current Focus
 
-**Updated:** 2026-08-14
+**Updated:** 2026-08-15 (H4 completed and promoted; B22 is next)
 **Review due:** 2026-09-06 (end of the bootstrap allocation window)
-**Full roadmap:** `ROADMAP.md` · **Research authority:** `NEXT_STEP.md`
+**Full roadmap:** `ROADMAP.md` · **Research authority:** `NEXT_STEP.md`, **except where this
+file supersedes it** — `NEXT_STEP.md` predates H4's completion and its L=5 targets are
+superseded (see its banner).
 
 ---
 
@@ -35,46 +37,62 @@ task produced.
 
 ## Currently active
 
-**No implementation task is currently authorized and in progress.**
+# `B22` — Stage-B constraint census
 
-This means no formally recorded implementation task is active. It does not mean the
-project is inactive: governance, documentation, and infrastructure work has merged
-since TASK-0001 under direct owner instruction rather than under a recorded successor
-implementation task. No successor implementation task has been selected or authorized.
+**Next research work, as preregistered.** B22 asks whether the bounded L=5 Stage-B test
+set admits useful normalization and deduplication, over the (K,pos)-window / 30-variable
+obstruction structure. See `OPEN_RESEARCH_QUESTIONS.md` B22 and the preregistration in the
+untracked intake; **the B22 preregistration is unchanged by H4 and must stay unchanged**
+unless the owner explicitly decides otherwise.
 
-## Long-running computation in flight — descriptive, not an authorized epic
+B22 is a census of the constraint structure. **It is not a rerun of the 3,316,540,933,500
+concrete codings.**
 
-**This is not an authorized epic and not a task.** It is recorded here so that a reader
-of this file alone learns that a large computation is occupying this machine, and does
-not start a conflicting or duplicate campaign.
+Nothing else is authorized. Governance, documentation, and infrastructure work continues
+under direct owner instruction rather than under a recorded implementation task.
+
+## H4 — completed, promoted, no longer active work
 
 ```text
 Name        L=5 Route-C full-family closure campaign ("H4")
-Location    C:\MSVC\L5_FULL_CLOSURE_CAMPAIGN_1\results_h4\   (outside this repository)
-State       IN FLIGHT — not complete
-Progress    read it from the campaign artifacts; deliberately not copied here
-Authority   results_h4\FINAL_AUDIT.txt — the only completion signal
-Claim       NONE, until that audit exists and independently passes
+Raw corpus  C:\MSVC\L5_FULL_CLOSURE_CAMPAIGN_1\results_h4\   (outside this repository)
+State       COMPLETE — audited, independently reviewed, merged
+Evidence    docs/evidence/h4-l5-full-closure-2026-08/        (in this repository)
+Claim       MATH_CLAIMS.md row 111 — COMPUTED (Level 1), bounded
+Result      2219/2219 shards, 0 missing, 0 duplicate, 0 survivors
 ```
 
-**No progress figure is recorded here on purpose.** A shard count is stale the moment it
-is written, and a stale count in a canonical file invites someone to reason from it. The
-campaign's own artifacts are the only current source; completion is signalled by the
-audit file, not by a count.
+**What the result is.** Every uniform coding `g : Σ₆ → Σ₃⁵` — all `243⁶` of them — is
+eliminated by at least one of two preregistered finite tests: Stage A's block-aligned
+Parikh obstruction from `h₆⁹(a)` (`m ∈ [2,120]`, `K = 5m ∈ [10,600]`, first 98,415 coded
+symbols), or Stage B's exact check (`K ∈ [6,40]`, first 3,645 coded symbols of
+`g(h₆⁶(a))`). Stage A retained 5,153,928 profiles = 3,316,540,933,500 codings; Stage B
+processed all of them and found 0 survivors.
+
+**What it is not.** It is a *finite*, Route-C, L=5 result. It does **not** establish
+Mäkelä's conjecture, and does not cover non-uniform morphisms, `L ≥ 6`, all ternary
+constructions, or source morphisms other than `h₆`. The combined window is **98,415**
+coded symbols, not 3,645 — Stage A's window is 27× larger (`EPISTEMIC_DISCIPLINE.md` §10).
+
+**Accepted provenance caveats — still open operational facts, not scientific failures:**
 
-**Partial progress is not a result.** A campaign that has not finished has decided nothing
-about the L=5 family: the shards not yet processed could contain a survivor, which would
-be a positive finding rather than an error. No `MATH_CLAIMS.md` row may cite this campaign
-until the campaign audit exists and independently passes. Row 49 is unaffected and must
-not be reworded on the strength of partial progress.
+```text
+per-shard executable identity   not recorded; same-binary identity across the
+                                campaign's resumes is not reconstructible per-shard
+real-survivor validation path   never naturally exercised (0 survivors); only a
+                                synthetic alert self-test exercised that path
+Stage-A artifact durability     no second physical backup yet (single physical disk)
+```
 
-**Naming hazard.** A file named `FINAL_AUDIT.txt` already exists elsewhere in the tree
-(`imported-artifacts\L5_FINAL_VERIFICATION_4\six_lane_results\`). That one is a
-**six-shard rehearsal** of the tooling and establishes nothing about the L=5 family. Only
-the campaign path above is authoritative.
+**Naming hazard — still live.** A file named `FINAL_AUDIT.txt` also exists at
+`imported-artifacts\L5_FINAL_VERIFICATION_4\six_lane_results\`. That one is a **six-shard
+rehearsal** of the tooling and establishes nothing about the L=5 family. Only the campaign
+path above, and the merged evidence package, are authoritative.
 
-**This block needs a revisit when the campaign ends** — at that point the audit result,
-not a progress figure, is what belongs here.
+**Not promoted, and not to be promoted without new work:** the idea that shallow L=5 kills
+would imply anything about L=6 is an **exploratory hypothesis only**. A kill-depth
+measurement remains a possible *separate* future exploration over aggregate distributions
+on a stratified sample — it is **not** part of B22.
 
 ---
 

KNOWLEDGE_STATE.md

@@ -97,6 +97,7 @@ exhausted. They stay closed.
 | The ternary abelian-square-free language ends at length 7 | exhaustive search | 1 |
 | **No** k-uniform ternary morphism, k ≤ 6, produces a Mäkelä fixed point | exhaustive sweep | 36 |
 | **Route (c):** no uniform map Σ₆ → Σ₃^L, L ≤ 5, produces a Mäkelä word from h₆'s fixed point | exhaustive sweep | 49 |
+| **Route (c) at L=5, full coding family:** all 243⁶ = 205,891,132,094,649 uniform codings eliminated by the two H4 preregistered finite tests; 0 survivors | two-stage exhaustive campaign (Parikh obstruction + exact prefix check) | 111 |
 | **Route (c) at L=6:** all 200,106 canonical codings avoiding K ∈ [2,5] also violate K ∈ [6,100] | locality-CSP + exhaustive escalation | 78, 79 |
 | **No coding at all** at L = 1, 2, 3 avoids the large periods, small window ignored (S_large(L) = 0) | exhaustive sweep | 79 |
 | No infinite [2,5]-free word exists over a two-letter sub-alphabet | cycle search in the container graph | 51 |