Phase 0 · Foundation Assessment — BEA-04 Competitive Analysis Tool
The permanent base layer (FAD) for all BEA-04 work. Written to be consistent with the downstream phases already completed (1 → 5): three pillars (launches / social sentiment / reviews), a deliberately thin V1 aggregator, the A→B→hold-C sequencing the model and validators converged on, and the three binding pre-build conditions surfaced in Phase 5 (reliability/corroboration gate, decisions-changed instrumentation, legal-sourcing + funded-owner gates).
1. Success criteria (measurable, time-bound)
These are deliberately written as adoption- and decision-grounded, not "items surfaced" — the Phase 2.5 incentive-alignment warning and the Phase 5 Brandt/Venkatesh objection both insist the metric is decisions changed, not dashboard impressions.
- [ ] Decision relevance (the real metric): Within 12 weeks of pilot launch, ≥ 5 distinct competitive/positioning decisions are recorded in which a Beats PMM cites a BEA-04 surfaced signal as a material input — captured via a one-click "this changed/informed a decision" log instrumented from pilot day one. A surfaced-item count is explicitly not a success criterion.
- [ ] Sustained adoption (anti-theater floor): Weekly active users ≥ 5 of the target PMM/competitive-strategy cohort sustained across weeks 4–12 of pilot, with median session depth beyond the first screen. Falling below WAU < 5 by month 3 triggers the Phase 3.5 REVERT-to-A-scope kill criterion.
- [ ] Hours saved (the modeled IV anchor): Pilot users self-report (and a sampled time-and-motion check confirms) ≥ 3 analyst-hours/user/week saved on manual cross-tab triage by week 8, validating the Phase 4 Immediate-Value base (6 users × ~3 hr/wk × $150/hr).
- [ ] Honest data hygiene: By GA, ≥ 2 of 3 pillars run on a legally-sourced / licensed path with documented provenance, and 0 single-source unverified items are presented without an "unconfirmed" label. (Directly enforces the Phase 3.5 / Phase 5 legal-sourcing gate.)
- [ ] Calibration honesty (gates B's alerting): If/when scoring (Architecture B) ships, scores launch in observe-only mode with uncertainty bands and alerting is enabled only after week-over-week score variance with no real underlying change stays within a defined guardrail across a ≥ 6-week live window (Okafor's internal-consistency calibration, Phase 5).
2. Decision-maker profile (Three Ledgers)
- Public ledger (what we pitch): "A multi-pillar competitive-intelligence tool that gives Beats one normalized view of competitors across product launches, social sentiment, and product reviews — replacing eight manual browser tabs with a single reconciled, freshness-aware substrate, built internally because Apple data policy makes external CI SaaS (Klue/Crayon/AlphaSense) a non-starter."
- Shadow ledger (what the decision-maker actually optimizes for): Patrick O'Brien (competitive-strategy / PMM org) is measured on defensible positioning decisions and not being blindsided by a competitor launch — not on tool usage. His true optimization is credibility in the room: a narrative he can defend to leadership without later being burned by a fake signal (Brandt's exact objection). He also quietly optimizes for avoiding a Legal/PR incident over an Apple-branded tool scraping third-party data, and for not owning a maintenance liability that rots after launch.
- True ledger (what will actually get built): Realistically a thin Architecture A substrate first (bronze/silver/gold per-pillar feed + freshness chips, one or two licensed pillars), then Architecture B scoring layered behind a calibration gate with a coarse reliability/corroboration check borrowed from C — if a named owner and licensed sourcing are secured. Architecture C (the I&W graph) is held behind two hard pre-build kill-tests (indicator lead-time > 0; funded curation owner) and most likely is not built in this cycle. The honest base case is "A ships, B is a funded follow-on, C stays a documented option."
- Owner / sponsor: Patrick O'Brien (per manifest
owner). Sponsor attention span (~2 quarters, Phase 3.5 scenario 31) is itself a risk the build sequence must respect by delivering incremental value every 6–8 weeks.
3. Constraint inventory
| Constraint | Type (hard / soft / assumed) | Notes |
|---|---|---|
| Apple Legal/Privacy will not permit an Apple-branded internal tool to scrape Amazon reviews / X content at scale; ≥ 2 of 3 pillars must use licensed/first-party sources | hard | The single most under-mitigated dependency across all phases (3.5 scenarios 7/35; Venkatesh in Phase 5). No engineering mitigation overrides it — it is a decision-gate KILL condition, not a risk. |
| Apple AI-model / data-handling policy governs any LLM used for auto-narratives (provider, data egress, on-device vs. hosted) | hard | B's auto-narrative is LLM-assisted with provider/model TBD; Apple policy plus the Phase 4 cost/latency kill-gate constrain it. Templated (non-LLM) narratives are the safe V1 default; see Assumption #6. |
| No external data egress of Apple-adjacent competitive data to a third-party SaaS | hard | This is the actual buy-vs-build justification (data governance, per Costa/Venkatesh) — not feature parity, which Beats would lose to Klue/Crayon. |
| Single-tenant internal tool — no revenue line, no external users | hard | Value is monetized only as analyst-hours-saved + decision-quality (Phase 4 modeling convention). Any revenue or network-effect claim is theater. |
| Named, funded maintenance/curation owner + ongoing ops budget must exist before launch | hard (elevated from soft in Phase 5) | Connectors are a service, not a project (Ramaswamy). Unowned scrapers rot silently (3.5 scenarios 6/22/27). Absence of an owner is a launch gate, not a footnote. |
| Headcount: ~3.5 FTE for A, ~4.5 FTE (incl. 1 data scientist) for B, ~5 FTE program for C | soft | Negotiable across the A→B→C sequence; but B's value lives entirely in the data scientist / calibration loop (Phase 4.5 — B fails on budget cut). |
| Built on the internal Vibe Elements component stack for the dashboard layer | soft | Convenient and consistent with the host repo, but the stack is evolving; a breaking change can strand the dashboard (3.5 scenario 9). Keep the UI layer thin and swappable. |
| Beats competitive peer set (Bose, Sony, JBL, Apple/AirPods, Skullcandy) is analyst-declared and relatively stable | assumed | Plausible for consumer audio, but a static list misses emerging entrants (Nothing, Anker/Soundcore, Baseus) — 3.5 scenario 4. Needs a quarterly review ritual. |
| Source APIs/scrapers stay accessible and stable enough for cadence-tiered polling | assumed | Fragile — ToS/rate-limit changes break pillars silently (3.5 scenario 1). The most operationally challengeable assumption. |
| Competitive events decompose into observable indicator sequences that fire before the public launch | assumed (C-only, kill-tested) | The make-or-break bet for Architecture C; cheaply falsifiable via a 2-week lead-time backtest (Reyes, Phase 5). FCC filings for ANC/Bluetooth radios genuinely lead; $40 accessories may have zero lead time. |
4. Scope boundary
- In scope: A multi-pillar competitive-intel substrate for Beats, covering exactly three pillars — product launches, social sentiment, product reviews — over a defined, analyst-declared peer set of consumer-audio competitors. Per-pillar normalized ingestion (bronze/silver/gold), per-pillar freshness indicators, and a reconciled single-view dashboard (Architecture A). Layered scoring + velocity-delta alerting behind a calibration gate and a coarse source-reliability/corroboration check (Architecture B, conditional). Decisions-changed instrumentation from pilot day one. Licensed/first-party data sourcing for ≥ 2 pillars.
- Out of scope: External monetization or any multi-tenant/SaaS product; competitor sets outside Beats' consumer-audio peer group; the full Architecture C order-of-battle graph + I&W rule engine + sustained curation program (held behind hard kill-tests, not in this cycle); the Phase 2.5 "impossible but worth exploring" moonshots (deception-budget index, self-Brier scoring, red-team-mirror, synthetic-indicator pre-registration) — carried at $0 value and explicitly off the V1 critical path; any scraping of sources Apple Legal has not cleared; "order-of-battle / tripwire / intelligence" military framing in the product surface (drop for plain "competitive-research / early-warning" language per Costa).
- Recursion budget acknowledged: max 3 per phase. (Phase 5 explicitly recommended no full recursion — proceed to the Decision Gate with three binding pre-build conditions carried forward.)
5. Assumption log
| # | Assumption | Confidence (0–1) | Validated? |
|---|---|---|---|
| 1 | A legally-sourced/licensed path exists for ≥ 2 of 3 pillars and can be confirmed before any connector is written | 0.55 | No — this is the hardest external gate; Venkatesh (Phase 5) warns Apple Legal blocks at-scale scraping. Must be confirmed at the Decision Gate; KILL if it fails. |
| 2 | Beats can name and fund a maintenance/curation owner with recurring time before launch | 0.45 | No — every internal CI tool Ramaswamy/Venkatesh saw die, died from unowned connector rot. Treated as a hard launch gate, currently unconfirmed. |
| 3 | Target PMM cohort will change ≥ 5 defensible decisions on BEA-04 signals within 12 weeks (i.e., it's not "glorified RSS reader" theater) | 0.50 | No — Brandt's core objection; only provable via day-one decisions-changed instrumentation in the pilot. |
| 4 | The analyst-declared peer set (Bose/Sony/JBL/AirPods/Skullcandy) is stable enough that a static V1 list won't miss the actual #2 challenger | 0.65 | Partially — plausible for audio, but emerging entrants (Nothing, Soundcore, Baseus) need a quarterly review ritual; flagged in 3.5 scenario 4. |
| 5 | Per-pillar 0–100 scores (Architecture B) can be calibrated into something internally consistent and stable week-over-week, even absent ground-truth "real move" labels | 0.50 | No — Okafor's central caveat; the dependent variable doesn't exist, so V1 must calibrate on internal consistency and ship observe-only until accept/dismiss labels accrue. Bimodal outcome (Phase 4). |
| 6 | LLM auto-narrative cost/latency stays within the value threshold under Apple AI policy, or templated narratives suffice for V1 | 0.55 | No — provider/model TBD; "TBD cost on a per-event feature" is how it becomes uneconomic at scale (Ramaswamy). Phase 4 carries narrative prose at $0 value and as a kill-gate; templates are the safe default. |
| 7 | For Architecture C only — competitor launch indicator sequences (FCC filing → job postings → retailer listing → embargo lift) reliably precede the public announcement with median lead-time > 0 | 0.40 | No — the make-or-break C bet; cheaply falsifiable via Reyes's 2-week historical backtest on the last 8–10 audio launches before any graph is built. |
| 8 | A coarse source-reliability + ≥2-channel corroboration gate can be added to the B baseline without C's full graph cost | 0.70 | Partially — Phase 5 (Klein + Reyes) elevated this from a C-only mechanism to a mandatory B precondition; structurally feasible as a lightweight tag, not yet implemented. |
| 9 | Recency ("last polled") can be surfaced in the UI without users misreading it as accuracy | 0.60 | No — Pae (Phase 5) trusted a green "12 min ago" chip as "this is true"; the chip must separate "last polled" from "last changed" from "coverage %" or every new coordinator repeats the error. |
Gate: success criteria defined, constraints mapped, decision-maker identified → set phaseGates.0-foundation = passed in manifest.json.