← back to Harlequin Sample Price Analysis

artifacts/JUNK-MFRSKU-CYCLE5.md

29 lines

# Adjacent finding — fleet-wide junk `mfr_sku` = 'TRUE' (read-only, cycle 5)

Discovered during TK-10870's collision-sizing; **out of TK-10870 scope** — characterized here to
right-size the "open a new ticket" recommendation (Action 4 of the gated memo). **No writes.**

## Blast radius (right-sized)
- **`mfr_sku = 'TRUE'` on 3,675 rows across 10 vendors.** A boolean leaked into the text `mfr_sku`
  field — concentrated in specific scrape batches: **Fentucci 2,070 · Retro Walls 946 · Fentucci
  Fabrics 367 · Innovations USA 129 · Mind the Gap 110** (+ small tails).
- **Identity is NOT lost.** `has_real_sku` = row count for every vendor; **only 1 of 3,675** lacks a
  recoverable sku. Products remain identifiable/priceable by `sku`/`dw_sku` — only cross-referencing
  *by `mfr_sku`* is broken.
- **Customer-facing impact is small: 83 ACTIVE** (77 of them Innovations USA); the rest are draft/archived.
- Plus vendor-name leakage into `mfr_sku`: SANCAR 184 · Osborne 56 · York 55 · Innovations 52 (~347).

## Severity verdict
**LOW–MEDIUM, recoverable.** This is field-pollution cleanup, not a "products broken" emergency. The
scary "18,156 shared skus" fleet number (cycle 2) is dominated by exactly this recoverable junk +
legitimate pattern-level sharing — not real identity collisions like Cranes.

## Root-cause hypothesis (to confirm in the ticket)
A scraper/import wrote a boolean flag (e.g. `has_mfr_sku` / `in_stock` / an availability boolean)
into the `mfr_sku` column for the Fentucci / Retro Walls / Innovations batches. Fix = re-derive
`mfr_sku` from the source feed for those batches (they still have the real `sku`/`dw_sku` to re-key on).

## Recommendation
Open a LOW–MEDIUM ticket: "clean up junk `mfr_sku` field pollution (3,675 `'TRUE'` + ~347 vendor-name
leaks, 83 active, 99.97% recoverable)." Gated (Action 4) — Steve's call. NOT done here.