The Symbion non-Directo cohort (2,929 products) is the active publish target. Scored against two bars: MVP (can be sold — ID, name, pricing, categorisation) and Data-Rich (MVP plus descriptions, images, fuller competitive pricing). The cohort sits firmly in the Usable tier for MVP; the gap to Data-Rich is meaningful but not large.
| Severity | Title | Rows | Blocks MVP? |
|---|---|---|---|
| Medium | F11 / #56 — SaH classifier missing No code path populates support_at_home_category. Only true MVP blocker in the audit. |
All | Yes |
| High | F3a — Portal exporter not built No code produces the portal CSV; structural blocker before any publish. |
All 2,929 | Yes |
| Critical | Issue C — Symbion source categories unreliable ~30% of categories wrong due to bad source data. Risky for MVP; reduces quality of three category columns. |
~900 est.5 | Quality risk |
| Medium | F4 / #41 — Product name cleaner skips populated rows Raw Symbion strings still in product_name. Lifts name overall from 43 → 80+ when run. |
2,800 | Quality risk |
| High | F10 — IA/BrightSky matching coverage gap 94% of cohort prices have no market validation. Risky for MVP credibility. |
2,774 | Quality risk |
| Medium | Issue B — Out-of-scope products publishing Dishwasher tablets etc. have out_of_scope = false but category_1 = 'Out of scope'. |
29 | Quality risk |
| Medium | F6 — BrightSky category mapper bug BS-cohort only; doesn't affect Symbion publish. |
BS cohort | No (BS) |
| Low | Issue A — "Hygiene " trailing-space bug 148 rows table-wide. Quick fix, lifts category quality marginally. |
148 | Quality risk |
| Low | Matcher rethink proposal Discussion: pack equivalence, IA bulk-divide pricing. Lifts F10 architecture. |
— | No |
Indicative ranges. Assumes engineering capacity from Hazel and Faisal — validate before committing externally.
Cohort goes live without descriptions or images. Categories accurate enough for browsing, SaH classification working, names cleaned. Acceptable if customers tolerate plain listings.
Full enrichment. Descriptions and images on every product, categories reviewed via AI override, broader competitor coverage, polished listings.
Notes.
1 Product Name is populated on 95.6% of cohort rows but reflects raw supplier Description strings; the AI cleaner script was never run on these populated rows. Quality of 45 is an estimate; true score requires measurement of all 2,800 rows. Running the cleaner lifts overall from 43 to roughly 80+.
2 Size is shown as Data-Rich only because base_unit_qty already covers the UOM requirement for MVP. Size remains descriptive — useful for richer listings but not required to sell.
3 Categorisation completeness is high but accuracy is significantly compromised by Symbion source-data errors (Issue C). Quality score of 55 reflects a 30-row sample where roughly 30% of mappings were wrong. The full remediation (AI category override) is deferred to Data-Rich.
4 Pricing market-validation is the most material quality finding in this audit. The HCP engine sets every price using GREATEST(competitor_undercut, supplier × 1.22). When competitor data exists, the price reflects market reality. When it doesn't (94.7% of cohort), the formula sets the floor with no external reference. The weighted overall of 48 puts product_price_ex in the Risky tier.
5 The ~900 misclassification figure for Issue C is an extrapolation from a 30-row sample. True rate requires either AI classification across the cohort or manual spot-check at scale.
Scoring formulae: Overall = (Completeness × Quality) / 100, per column. MVP score = mean of overall across 20 MVP-required columns. Data-Rich score = mean of overall across all 26 columns.
Source: Supabase production products table queried 12 May 2026. Filter: symbion_price IS NOT NULL AND directo_sku IS NULL AND out_of_scope = false. Reference: audit_products_symbion_non_directo_v1_2.md.