Current position and next steps

Self-serve ABI and OpenAI account connections

The website now includes a browser-local Digital DNA design-brief workshop and a local, read-only OpenAI Advertiser API connector. These are separate journeys: an Ads account key does not upload a persona or grant access to private ChatGPT conversations. Hosted generation, multi-client authentication, an open-source SDK and production account onboarding remain proposed.

See Self-serve ABI and account connections for the product model, permissions and delivery sequence.

Current status and supporting material

Guinness: illustrative voice prototype with user listening feedback and a separate character review candidate. Pre: local voice enabled for review, ten athlete dossiers, ten selected sporting events, five campaigns and optional mental skills; new delivery awaits listening. Marriott Hotels: source-backed research draft, voice gated. None is an authorised client deployment. Learning pipeline, production approval service and autonomous Gauntlet remain planned.

See framework-stress-test.md for the 19 structural gaps; starter-brand-research.md for the new research scope; guinness-research.md for Guinness; live-voice.md for transport. Structural gaps must be addressed before production scale. This whitepaper must stay honest about those limits.

Official voice reference checked 3 October 2026: https://developers.openai.com/api/docs/guides/voice-webrtc

Athlete evidence roadmap

1. Discovery: ask the brand what athlete, shoe-use and event records exist, who owns them, and what can be used publicly. Inspect a small authorised sample before estimating cost or promising coverage.
2. Pilot: build a reviewed set of exact-model appearances with independently checked results. Agree field definitions, consent, publication rights and access boundaries; evaluate retrieval and the Gauntlet on a held-out set.
3. Integration: connect the approved client source, show provenance when requested, support corrections and revocation, and publish only after human review. Begin with one shoe category; expand according to verified coverage.

This work sits within the existing research/recommendation and client-integration roadmap. Dates depend on a partner, data access and review; no launch date is committed.

Next working session

1. Listen to Pre and Guinness; record specific character and connection failures. Turn confirmed failures into review cases.
2. Build the first bounded behavioural Gauntlet runner and human review report, with an explicit provider-use budget.
3. Scope one sports-data pilot: choose a competition, compare provider coverage and rights, and design a replay-based watch-together demonstration before commissioning live access.

The website whitepaper is generated from this document. Supporting research stays separately linked. Update the working edition and present implementation status honestly when the product changes.