Phase 0 · Foundation Assessment — BEA-12 WW ASIN Tracking
The permanent base layer (FAD) for all downstream work. Establishes what success means, who actually decides, what cannot move, and what we are assuming on faith. Written to be consistent with — and to anchor — Phases 1–5, which converged on Architecture B (two-tier, canonical-anchor assurance layer, built sweep-first) as the conditional-go path, with A as its staged Phase-1 and C killed.
1. Success criteria (measurable, time-bound)
- [ ] Coverage truth, not coverage theater: A single canonical view reports listing state for ≥ 90% of Beats' entitled marketplaces (target set is 22; the honest denominator is whatever SP-API/Vendor Central report entitlement actually exists per Phase 3.5 A8 / Phase 5 Priya). Measured at pilot exit (≤ 16 weeks from build start for the sweep-first core); a market is "covered" only if it logs a verified state at least once per its configured cadence. Markets not entitled are shown as explicit gaps, never implied as covered.
- [ ] Detection latency meets a stated SLA, not an assumed one: For each market, a Vendor Central listing change is reflected in the canonical view within the freshness target agreed with the owner — baseline ≤ 6–10h for poll/sweep-derived markets (accounting for SP-API report-generation latency, Phase 5 Priya) and ≤ 60 min for any market where push notifications are validated to exist. Measured by seeded-change spot-checks (Phase 3.5 A1 detection method) over a rolling 30-day window post-pilot, with ≥ 95% of seeded changes detected within target.
- [ ] Silent-failure elimination (the actual unmet need): Zero instances over a rolling 90-day post-launch window of a real listing discrepancy in an entitled market that the system failed to surface (i.e., no "wrong/missing ASIN in a country nobody checked" found by a human before the tool flagged it). Measured against an independent seeded-discrepancy harness plus logged human-found incidents.
- [ ] Labor displacement is realized, not just promised: Manual cross-marketplace reconciliation effort drops by ≥ 1.0 FTE-equivalent of attention (toward the ~1.3 FTE modeled in Phase 4 for B) within 6 months of launch, measured by tracked bypass/manual-check frequency (Phase 3.5 A5/B6 detection) trending to near-zero.
- [ ] Decision-gate pre-build gates closed before full commit: All three Phase-5 structural gates are answered with real data — (i) per-market notification coverage measured against actual Beats vendor entitlement, (ii) a named, sponsor-committed second consumer of the canonical stream (or option value is zeroed and B re-scored), (iii) an anchor-feasibility spike on the real Beats catalog modeling
intentionally-absentandregional-variantstates — each with a written yes/no before any funding of the fast-loop or tier features (target: within 4 weeks of decision-gate go).
2. Decision-maker profile (Three Ledgers)
- Public ledger (what we pitch): "Auto-log ASIN data across 22 countries/marketplaces, triggered by Vendor Central changes" — a worldwide, real-time-feeling catalog tracking tool that gives Beats one trustworthy view of every product listing everywhere it sells. Framed as production automation that replaces manual spreadsheet reconciliation.
- Shadow ledger (what the decision-maker actually optimizes for): Not being blamed for a wrong or missing listing in a market nobody was watching. The owner's true incentive (Phase 2.5 incentive-alignment) is converting silent, unattributable catalog failures into loud, owned, early alerts — defensibility and "we caught it first," more than raw automation. Secondarily: shipping something credible that survives a decision gate without over-committing headcount, and avoiding a tool that exposes regional teams' failures so harshly that it triggers political backlash (Phase 3.5 B8 / Phase 5 Sofia).
- True ledger (what will actually get built): The sweep-first "A-plus-anchor" core of Architecture B — a scheduled full/checksum reconciliation across entitled markets, normalized to a canonical schema, fingerprint-deduplicated, written to an append-only event log keyed on a curated canonical product anchor, with per-market coverage health scores and explicit verified/stale/missing tiers. The real-time push fast loop and any tier-driven workflow get added only per validated market and only if a named consumer/responder exists. Autonomous write-back and storefront crawling (Architecture C) will not be built. The word "triggered" will be reframed in writing as poll/notification-derived detection (Phase 5 Jordan/A4).
- Owner / sponsor: Jaclyn Kreidstein (WW catalog operations). Accountable for the freshness-SLA sign-off, naming the sustaining-ops owner + response SLA, and either producing a committed second-consumer team or accepting the descoped sweep-first floor.
3. Constraint inventory
| Constraint | Type (hard / soft / assumed) | Notes |
|---|---|---|
| Amazon Vendor Central exposes no true CDC / transaction-log feed to vendors | hard | The single binding reality across all phases. Every change signal is derived: SP-API report poll, SP-API notification, or storefront crawl — each rate-limited and ToS-governed. "Real-time trigger" in the one-liner is not literally achievable; detection is poll/notification-derived. |
| Amazon ToS prohibits/penalizes storefront crawling and autonomous write-back to live listings; aggressive polling is itself an account-health signal | hard | Kills Architecture C (Phase 3.5 C1/C2, Phase 5 Russell). Even "safe" A/B polling must be planned with Amazon vendor-relations proactively — throttling is a relationship risk, not just a freshness risk. |
| SP-API per-vendor-account rate limits are shared across all consumers, not per-marketplace; reports are generation-async (request-then-poll) | hard | Caps poll cadence and makes a naive nightly 22-market full sweep unbuildable at scale (Phase 3.5 A2/B5). Forces checksum-then-fetch-divergent reconciliation (Phase 5 Marcus) and bakes report-gen latency into the freshness SLA (Phase 5 Priya). |
| Apple/Beats change-control governance forbids unattended mutation of an external system of record (the live catalog) | hard | Independently blocks C's self-healing write-back even if Amazon allowed it (Phase 3.5 C2). Any future write-back must be human-gated. |
| SP-API report/credential entitlement may not exist for all 22 markets | assumed → must validate | If a viable majority isn't entitled, a "WW" label is false and the initiative is mislabeled (Phase 3.5 A8 kill criterion). Coverage denominator is entitlement, not the aspirational 22. |
| Headcount / build budget (synthetic): A ≈ 1–2 eng × 6–10 wk; B ≈ 2–3 eng × 12–18 wk (sweep-first staged); C ≈ 4–6 eng × 6–9 mo | soft | Phase 4 figures are AI-estimated synthetic, for relative comparison only. A 50% budget cut degrades B gracefully to its sweep-first core (Phase 4.5); the staged sequencing is the natural fit for a constrained budget. |
| Named sustaining-ops owner + runbook + paged staleness alerting must exist before any build | soft (org HARD GATE) | Phase 3.5 cross-architecture HARD GATE. "Automation" mistaken for "zero-maintenance" rots all three architectures within months (A7/C8). Non-negotiable precondition. |
| Data-localization / cross-jurisdiction privacy rules across 22 markets (esp. EU) | assumed | Low exposure for A/B (consume only vendor-entitled SP-API data, not scraped public data); high exposure for C's crawl (Phase 3.5 C11). Apply P13 federation only if a residency rule actually lands. |
| Canonical product↔ASIN mapping is messy (bundles, multipacks, regional-exclusive SKUs, frequent re-ASINing on relaunch) | assumed → must validate | The anchor is B's spine; if it churns or can't model intentional asymmetry it generates false "missing" alerts that destroy trust (Phase 3.5 B2/B11, Phase 5 Lena). Anchor-feasibility spike on real catalog is a pre-build gate. |
4. Scope boundary
- In scope: Detecting and logging Beats product-listing state changes across the entitled subset of the 22 Amazon country marketplaces, sourced from SP-API reports (and notifications where validated); normalizing to a canonical schema; append-only event-logged truth keyed on a curated canonical product anchor; per-market coverage health scores with verified/stale/missing degradation tiers; a checksum-then-fetch-divergent reconciliation sweep; operator-facing surfacing of discrepancies for human action. The deliverable is an observe-and-assure tool.
- Out of scope: Autonomous SP-API write-back / self-healing correction of live listings; independent storefront crawling and the zero-trust dual-source reconciler; the predictive-staleness ML model (all three are Architecture C, killed). Also out: pricing optimization, demand forecasting, any multi-sided "platform/network-effect" framing (legitimate value is an internal data-network effect only, and only once a second consumer is named). Non-Amazon marketplaces. Real-time push for markets where notifications are not validated to exist.
- Recursion budget acknowledged: max 3 per phase. Current
recursionCount = 0; Phase 5 explicitly recommended bounded refinement (gate-closing), not full re-architecture, leaving recursionCount unchanged.
5. Assumption log
| # | Assumption | Confidence (0–1) | Validated? |
|---|---|---|---|
| 1 | A canonical product anchor can be defined and curated for the real Beats catalog (modeling intentionally-absent and regional-variant states), stable enough that churn doesn't exceed its value. |
0.55 | No — pre-build anchor-feasibility spike required (Phase 5 Lena / B2/B11). |
| 2 | SP-API/Vendor Central notification (push) coverage exists for ≥ ~50% of entitled markets against Beats' actual vendor entitlement (not public docs). | 0.45 | No — must measure against real entitlement before costing the fast loop (Phase 5 Priya / B1). |
| 3 | SP-API report/credential entitlement exists for a viable majority of the 22 markets. | 0.65 | No — per-market entitlement audit pending (Phase 3.5 A8). |
| 4 | The business's true freshness need is satisfied by 6–10h sweep-derived detection for most markets (only a subset needs sub-hour). | 0.55 | No — owner must sign off freshness SLA per market (Phase 3.5 A5/B6, Phase 5 Priya). |
| 5 | A named, sponsor-committed second consumer (pricing / availability / regulatory-field audit) will subscribe to the canonical stream, realizing the ~$230K option value. | 0.40 | No — if unnamed, option value is zeroed and B re-scored as single-purpose (Phase 5 Devon / B's NPV swing). |
| 6 | A named sustaining-ops owner with a response SLA and executive air-cover for regional-score friction will exist before the tier feature is built. | 0.50 | No — Phase 3.5 HARD GATE, reaffirmed by Phase 5 Sofia (B3/B8). |
| 7 | Polling 22 markets within the shared per-vendor rate budget won't trip throttling or damage the Amazon vendor-account relationship. | 0.50 | No — capacity-plan + proactive vendor-relations engagement required (Phase 3.5 A2/B5, Phase 5 Russell). |
| 8 | Synthetic Phase-4 financials are directionally correct for relative A/B/C ranking (not for absolute budgeting). | 0.60 | Partial — labeled AI-estimated; decision rests on the three validation gates, not the dollars (Phase 5 Aimee). |
Gate met for 0-foundation: 5 time-bound measurable success criteria defined; decision-maker mapped across public/shadow/true ledgers with owner Jaclyn Kreidstein; 9-row constraint inventory spanning the no-CDC reality, Amazon ToS/rate limits, governance, entitlement, headcount, ops ownership, residency, and anchor-mapping; in/out scope set to the observe-and-assure tool (C explicitly out) with recursion budget acknowledged; 8-row assumption log with confidence and validation status. → set phaseGates.0-foundation = passed in manifest.json.