[object Object]

← back to Dw Sku Integrity

TK-10896: record Claude-Codex no-mint comparison proof

04762e23b9f707eebf28f695e6b79b3a5ad29957 · 2026-08-30 09:35:23 -0700 · codex-10896

Files touched

Diff

commit 04762e23b9f707eebf28f695e6b79b3a5ad29957
Author: codex-10896 <steve@designerwallcoverings.com>
Date:   Sun Aug 30 09:35:23 2026 -0700

    TK-10896: record Claude-Codex no-mint comparison proof
---
 evidence/CLAUDE-CODEX-COMPARISON-2026-08-30.md | 29 +++++++++++++++++
 verification/e2e-proof.json                    | 45 ++++++++++++++++++++++++++
 2 files changed, 74 insertions(+)

diff --git a/evidence/CLAUDE-CODEX-COMPARISON-2026-08-30.md b/evidence/CLAUDE-CODEX-COMPARISON-2026-08-30.md
new file mode 100644
index 0000000..6fcf543
--- /dev/null
+++ b/evidence/CLAUDE-CODEX-COMPARISON-2026-08-30.md
@@ -0,0 +1,29 @@
+# TK-10896 — Claude vs Codex comparison (model=codex)
+
+Scope: independent read-only/dry-run comparison. No canonical DB writes and no
+new SKU allocation. The standing rule is authoritative: recover an already
+existing code from evidence, or re-scrape; never mint a sequential code.
+
+| Finding | Claude / ticket ledger | Codex comparison | Verdict |
+|---|---|---|---|
+| Original active-blank scope | 33,946, later 34,569 after Phase 1 accounting | Mac2 mirror currently reports 34,600 | Different snapshots; mirror is advisory, not canonical |
+| Safe existing-code self-copy | 18,502 scoped; 18,369 clean rows applied before the no-mint ruling | 25,623 mirror rows remain dry-run self-copy candidates after collision and provenance guards | Agreement on method; current count is mirror-only |
+| Collision/duplicate rows | 133 routed to TK-10649; examples include Novasuede and Arte | 9 currently surface on the mirror because most prior rows no longer meet blank-active scope | Agreement on route; no auto-fix |
+| Re-scrape need | 12,589 no-identifier rows; Carnegie 5,921 anchor | 8,962 mirror rows: 8,897 mint-residue plus 65 genuinely bare | Broad agreement after accounting for mirror residue |
+| Greenfield prefixes | Phase 4 allocated previously unused `DWAG/DWAX/DWCX/DWST/DWSC/DWDX/DWWG` prefixes, then Steve reverted the allocation | Codes in these prefixes on blank rows are blocked from self-copy and routed to `MINT_RESIDUE_RESCRAPE` | Codex accepts Claude's provenance evidence; this is required by no-mint |
+| Bucket-A prefixes | Existing prefixes such as `DWKN/DWTT/DWRW/DWJS/DWRO/DWCC` predated Phase 4, but some ranges were also minted and reverted | Prefix alone cannot distinguish native codes from mint residue; exact reverted-number ledger is still required | Partial/unresolved; do not claim blanket provenance |
+| Canonical authority | Kamatera owns `shopify_products` | Scanner run here used the Mac2 mirror only | Canonical totals remain unverified by this pass |
+
+## Codex conclusion
+
+Claude's phase ledger resolves the largest ambiguity in the initial Codex scan:
+the seven greenfield-prefix cohorts are reverted mint residue, not valid recovery
+candidates. The guarded dry run therefore reports:
+
+- 25,623 `recoverable_now_self_copy`
+- 8,962 `rescrape_program_TK10900`
+- 9 `dedup_gated_TK10649`
+- 6 `recoverable_now_staging_link`
+
+This comparison does not authorize applying any candidate. Canonical recovery
+must be re-counted on Kamatera and every write remains separately gated.
diff --git a/verification/e2e-proof.json b/verification/e2e-proof.json
new file mode 100644
index 0000000..356c8a8
--- /dev/null
+++ b/verification/e2e-proof.json
@@ -0,0 +1,45 @@
+{
+  "intent": "TK-10896 read-only/dry-run SKU recovery analysis with a hard no-mint provenance guard",
+  "risk_tier": "R1",
+  "environment": "Mac2 local dw_unified mirror; canonical Kamatera database intentionally untouched",
+  "build_commit": "2bcc1992d73b181b14455deb7685bef3a810f827",
+  "timestamp": "2026-08-30T16:35:13Z",
+  "baseline": "Initial classifier treated 34,520 mirror rows as self-copy candidates; Claude's phase ledger showed seven greenfield prefixes were allocated and later reverted.",
+  "checks": [
+    {
+      "boundary": "unit",
+      "command": "npm test",
+      "assertion": "24 tests pass, including candidate-origin no-mint invariant and greenfield-prefix rejection",
+      "verdict": "PASS"
+    },
+    {
+      "boundary": "database-read",
+      "command": "node dwsku-backlog-scan.mjs",
+      "assertion": "Scanner executes SELECT-only against the Mac2 mirror and emits a dry-run summary without --out",
+      "verdict": "PASS"
+    },
+    {
+      "boundary": "recovery-plan",
+      "assertion": "No MINT_RESIDUE_RESCRAPE row has a proposed candidate; all seven greenfield prefixes route to TK-10900",
+      "verdict": "PASS"
+    },
+    {
+      "boundary": "canonical-data",
+      "assertion": "No UPDATE, INSERT, DELETE, Shopify call, or Kamatera command executed",
+      "verdict": "PASS"
+    },
+    {
+      "boundary": "canonical-count",
+      "assertion": "Authoritative Kamatera segmentation",
+      "verdict": "SKIP",
+      "reason": "Explicit task scope forbids canonical DB work; mirror totals are labeled advisory. This is not critical to validating the local guard, but blocks claiming canonical backlog completion."
+    }
+  ],
+  "artifacts": [
+    "evidence/ANALYSIS-2026-08-30.md",
+    "evidence/CLAUDE-CODEX-COMPARISON-2026-08-30.md",
+    "evidence/mac2-mirror-2026-08-30.json"
+  ],
+  "cleanup": "No database or external state created. Local evidence artifacts retained in git.",
+  "overall_verdict": "PASS for safe local tooling and dry-run comparison; canonical recovery remains intentionally unperformed"
+}

← 2bcc199 TK-10896: add provenance guard — greenfield Phase-4 mint pre  ·  back to Dw Sku Integrity  ·  TK-10896: Cody-gate hardening — within-batch shared-candidat 2fb7252 →