Phase 0 · Foundation Assessment — BEA-03 Deck & Creative Asset Generation Engine
Establish the foundation before any analysis. The permanent base layer (FAD) for all future work. Retroactively reconstructed: this project shipped to live status without ever passing through the methodology, so Phase 0 is being written against an already-deployed artifact.
1. Success criteria (measurable, time-bound)
- [ ] Output quality bar: ≥80% of generated slide scaffolds are accepted by a Beats marketing/creative reviewer with "light edits only" (no full rewrite) on the first generation, measured across a rolling sample of 50 briefs within the first 90 days post-launch. (This is the literal promise in the one-liner — "80%-quality" — and is currently asserted, not instrumented.)
- [ ] Cycle-time reduction: Median brief-to-first-reviewable-deck time drops from the current manual baseline (estimated 4–8 working hours per deck) to ≤30 minutes of human-attended time, validated on ≥20 real Beats briefs within 60 days of launch.
- [ ] Adoption / retention: ≥15 distinct Beats creators generate at least one deck, and ≥40% of them return for a second deck within their first 30 days of access, measured 90 days after rollout.
- [ ] Export integrity: ≥95% of PPTX and Markdown exports open without manual repair in the target tools (Keynote/PowerPoint for PPTX, standard Markdown renderers), measured over the first 200 exports.
- [ ] Regeneration usefulness: Median regenerations-per-accepted-deck ≤3 (i.e., the review→regenerate loop converges quickly rather than churning), measured over the first 90 days.
2. Decision-maker profile (Three Ledgers)
- Public ledger (what we pitch): "A brief-to-deck engine that gives Beats creators an 80%-quality first draft — slides, messaging, and asset direction — in minutes instead of hours, with human review, one-click regenerate, and clean PPTX/Markdown export." Framed as creative leverage and speed, not headcount reduction.
- Shadow ledger (what the decision-maker actually optimizes for): A defensible, ownable demonstration that Beats can ship a proprietary AI capability on Apple-enterprise rails — proving internal AI-generation competence to leadership and de-risking dependence on outside creative agencies. Speed-to-deck is the visible win; the real win is institutional credibility and a reusable internal generation pipeline that Buku/Tom can point to.
- True ledger (what will actually get built / has been built): A Vue SPA + service-backed generation pipeline that ingests a brief, produces a structured slide scaffold (slide breakdown, messaging copy, asset direction notes) at "good-enough-to-edit" quality, supports per-slide review and regenerate, and exports to PPTX and Markdown — shipped already as
live, ahead of the analytical backing that would have set its quality and adoption targets. - Owner / sponsor: Buku / Tom (per manifest
owner). Buku/Tom own both the build mandate and the internal political case; the True ledger is constrained by what they could ship fast on Beats-proprietary assets.
3. Constraint inventory
| Constraint | Type (hard / soft / assumed) | Notes |
|---|---|---|
| Beats-proprietary creative assets, brand, and tone must never leak to third-party model providers | Hard | README marks project proprietary: true and "not reusable beyond Beats." Forces on-prem / Apple-internal model rails or strict data-handling guarantees; rules out naive public-API generation. |
| Output must export cleanly to PPTX (Keynote/PowerPoint) and Markdown | Hard | Stated in one-liner; broken exports destroy the entire value prop since the deck must land in the creator's real toolchain. |
| Brand/tone fidelity to Beats voice | Soft | "80% quality" implies brand-acceptable but human-finishable; exact fidelity threshold is negotiable and not yet instrumented (see Assumption 1). |
| Human-in-the-loop review + regenerate is required, not optional | Hard | The product is explicitly a scaffold with review/regenerate, not autonomous final-deck generation. Removing the human gate changes the risk profile and the brand-safety story. |
| Built on Apple-enterprise / SkaFld monorepo rails (Vue SPA, design-token system, base/functional component library) | Soft | Phase 6 build plan reuses the vibe-elements design-token and component library; deviating is possible but costly. |
Shipped live before methodology backing existed |
Assumed | This retroactive Phase 0 cannot set go/no-go gates the way a pre-build Phase 0 would; success criteria here are reverse-engineered and must be treated as hypotheses to instrument, not pre-commitments that were met. |
| "80% quality" is a marketing figure, not a measured one | Assumed | No evidence in the dossier folder that the 80% bar was ever defined operationally or measured; the figure anchors expectations without a rubric. |
| Generation latency acceptable to creators | Assumed | A multi-slide generation + regenerate loop must feel faster than manual work; no latency budget is documented. |
4. Scope boundary
- In scope: Brief ingestion; structured slide-scaffold generation (slide breakdown + messaging copy + asset direction — textual creative guidance, not finished art); per-slide human review; regenerate loop; PPTX and Markdown export; Beats-internal use only.
- Out of scope: Final pixel-perfect art / image rendering (engine gives asset direction, not produced assets); autonomous deck delivery without human review; non-Beats / external client use; real-time collaborative multi-user editing; integration into Keynote/PowerPoint as a live plugin; localization/translation of generated copy.
- Recursion budget acknowledged: max 3 per phase.
5. Assumption log
| # | Assumption | Confidence (0–1) | Validated? |
|---|---|---|---|
| 1 | "80% quality" corresponds to a reviewer-acceptable first draft needing only light edits | 0.45 | No — never operationalized into a rubric or measured |
| 2 | Manual baseline is 4–8 hours per deck, so a 30-min engine is a large, felt win | 0.55 | No — baseline asserted, not timed against real Beats workflows |
| 3 | Beats creators will trust and adopt an AI-generated scaffold rather than starting from blank | 0.50 | No — adoption/retention not yet measured post-launch |
| 4 | Brand/tone fidelity is achievable at scaffold quality without per-brief fine-tuning | 0.50 | Partially — shipped output exists, but fidelity acceptance rate is uninstrumented |
| 5 | PPTX export survives round-trip into Keynote/PowerPoint without manual repair | 0.60 | Partially — export feature shipped; >95% integrity rate not yet measured |
| 6 | The proprietary-asset constraint is fully satisfied by the chosen model/data rails | 0.55 | No — data-handling guarantee assumed, not audited in this dossier |
| 7 | The review→regenerate loop converges (≤3 regenerations) rather than frustrating users | 0.40 | No — no convergence data captured |
Gate: success criteria defined, constraints mapped, decision-maker identified → set phaseGates.0-foundation = passed in manifest.json.