audit: Sangetsu spec/desc gap is source-data (not merge) — Lilycolor is the near-term publish path

d7dea0f13b33d73346670d289473c7371af78d13 · 2026-07-01 14:23:17 -0700 · Steve Abrams

Files touched

A onboarding/sangetsu-lilycolor/SANGETSU-SPEC-GAP-063026A.md

Diff

commit d7dea0f13b33d73346670d289473c7371af78d13
Author: Steve Abrams <steve@designerwallcoverings.com>
Date:   Wed Jul 1 14:23:17 2026 -0700

    audit: Sangetsu spec/desc gap is source-data (not merge) — Lilycolor is the near-term publish path
---
 .../SANGETSU-SPEC-GAP-063026A.md                   | 38 ++++++++++++++++++++++
 1 file changed, 38 insertions(+)

diff --git a/onboarding/sangetsu-lilycolor/SANGETSU-SPEC-GAP-063026A.md b/onboarding/sangetsu-lilycolor/SANGETSU-SPEC-GAP-063026A.md
new file mode 100644
index 00000000..5a48eeb5
--- /dev/null
+++ b/onboarding/sangetsu-lilycolor/SANGETSU-SPEC-GAP-063026A.md
@@ -0,0 +1,38 @@
+# Sangetsu spec/description gap — source analysis (2026-07-01)
+
+**Question:** Can Sangetsu's Activation-Gate gaps (0 descriptions, ~8.8% specs from
+the readiness audit) be closed *deterministically* from staged source data — no LLM,
+no Henry mount — while the Lilycolor enrichment runs?
+
+**Answer: No.** The gap is a **source-data gap, not a merge gap.** The data simply
+isn't in what was scraped.
+
+## Evidence (`staging/sangetsu-staging.jsonl`, 622 pattern rows → 10,532 colorways)
+
+| Field | Populated | Note |
+|---|---|---|
+| `overview` (description candidate) | **4 / 622** | Not a usable description source — 99.4% empty |
+| `spec` (dict present) | 622 / 622 | Dict exists but fields are sparse ↓ |
+| ↳ `spec.type` | 64 / 622 | Type II etc. |
+| ↳ `spec.weight` | 49 / 622 | |
+| ↳ `spec.backing` | 42 / 622 | |
+| ↳ `spec.width` | 28 / 622 | |
+| ↳ `is_metallic` / `is_grasscloth` (flags) | 104 / 37 | booleans, not full specs |
+
+Only ~9% of patterns carry real width/type/weight — consistent with the audit's 8.8%.
+
+## Conclusion / what Sangetsu actually needs
+- **Descriptions:** cannot be mapped from `overview`. Require either (a) LLM generation
+  from pattern name + spec + material (do this once the Lilycolor enrichment frees the
+  local ollama), or (b) copywriting.
+- **Specs:** ~91% of patterns genuinely lack width/type/weight in the scrape. Require a
+  **fresh sangetsu.com spec pass or a vendor spec sheet** — not recoverable from staging.
+
+## Priority implication (for Steve / purchasing)
+Sangetsu is **not near-term gate-ready regardless of cost** — it needs a source-data
+build (specs + descriptions) before publish is even possible. **Lilycolor** (2,543 SKUs,
+100% images/specs/list-price, enrichment recovering to ~full coverage now) is the
+near-term publish path. → **Confirm Lilycolor net cost first;** treat Sangetsu spec/desc
+sourcing as a separate work item, not a blocker on Lilycolor.
+
+_Read-only analysis. No writes to dw_unified/Shopify. $0 (local)._

← 324313fd recover: enrich Lily swatches from local copy (Henry USB unm · back to Designer Wallcoverings · cost-sheet: catalog-level Lilycolor net-cost skeleton (5 row fd27b7e1 →