PVA Revised Decomposition and Autopilot Plan
internal prototype · canonical JSON + Dreamborn Forge HTML
internal generated
revised-decomposition-plan · supabase_json

PVA Revised Decomposition and Autopilot Plan

revised-decomposition-plan artifact · for Patient Visit Advocate · phase PVA-AUTOPILOT-01 · status approved

Planning Surface

Use this to decide what happens next.

Status

approved

Phase

PVA-AUTOPILOT-01

Agent Handoff
Start Here

revised-decomposition-plan artifact · for Patient Visit Advocate · phase PVA-AUTOPILOT-01 · status approved

Completion Evidence

No explicit evidence field yet. Require tests, screenshots, linked PRs, or reviewed outputs before marking complete.

Decision

Replace the internal-review-first sequence with a pocket-app-first decomposition. The customer product is the app. Internal corpus/review work supports safety and learning but must not dominate the first customer-value slice.

Structured Payload

Machine-readable source fields

phase
id

PVA-AUTOPILOT-01

name

Pocket-App-First Autopilot Plan

status

draft

title

PVA Revised Decomposition and Autopilot Plan

source

Atlas-Codex revised PVA decomp with ask_model critique from Claude and OpenAI; Gemini unavailable/high demand on retry.

product

Patient Visit Advocate

decision

Replace the internal-review-first sequence with a pocket-app-first decomposition. The customer product is the app. Internal corpus/review work supports safety and learning but must not dominate the first customer-value slice.

created at

2026-05-08T14:02:37.874Z

created by

atlas-codex

project id

31d7f681-bed2-44e9-9a55-b9fadcbba0da

explicit avoid
  • Do not dispatch workers from the old internal-review-first PLAN.json.
  • Do not build payments, caregiver permissions, corpus public surfaces, or cloud PHI pipelines in first wave.
  • Do not rewrite approved M-01/M-02 artifacts unless producing a delta/addendum.
  • Do not create canonical PVA docs in repo docs/specs; Studio artifacts are source of truth.
revised slices
idexitnameentryobjective
PVA-SLICE-00Hard blockers and first-release trust copy rules approved.Trust Boundary CharterApproved PVA intent/monetization/policy artifacts exist.Hard rules for non-clinical positioning, consent, local retention, escalation, analytics, AI/data handling, and monetization trust.
PVA-SLICE-01End-to-end local flow tested: messy intake, review, brief, question plan, quick note, debrief, export, recovery/delete.Local-Only Pocket PrototypeSLICE-00 hard rules accepted.Validate free single-visit loop without account or server PHI.
PVA-SLICE-02Local/server/model/free/paid boundaries approved.Patient Data + AI ContractPrototype loop defined; M-01/Architecture available.Define data/model delta required for saved memory and server processing.
PVA-SLICE-03Corpus outputs are publish-blocked and usable only through approved safety/product-learning paths.Corpus Safety + Internal Review FoundationExisting M-02A/M-02B artifacts approved.Support content safety and product learning without blocking free-loop validation.
PVA-SLICE-04Screen inventory, empty/error/offline states, escalation, export, and memory paywall states accepted.Pocket UX Build GatePrototype learnings and data contract deltas exist.Small screen inventory and one-hand flow map required for build, not a giant waterfall blueprint.
PVA-SLICE-05MVP passes safety/privacy QA and avoids cloud PHI violations.Patient-Facing MVP BuildTrust, data, and UX gates pass.Build free loop plus account/save/memory prompts where allowed.
PVA-SLICE-06Closed beta launch gate approved.Closed Beta + Production ReadinessMVP passes internal safety tests.Prepare safe release with privacy controls, deletion/export, analytics audit, model provenance, kill switch, and human AI output review.
ask model synthesis
claude
  • Direction correct, but avoid writing as if starting from zero; wire in approved artifacts.
  • Define export, recovery, AI fallback, escalation acknowledgement, local draft state, and free/paid retention clearly.
  • Paid memory is right, but free retention limits and basic debrief must be clearer.
gemini

Unavailable/high demand; no usable text returned after retries.

openai
  • Reduce bureaucracy and get to user signal; local visit loop and usability testing should happen earlier.
  • Treat live note-taking and export as core value.
  • Free should include limited useful history or export; paid memory/caregiver sharing is stronger than generic storage.
first autopilot wave
roleoutputartifact
BAHard rules, free/paid trust boundary, escalation/consent/retention decisions, and worker stop conditions.Trust Boundary Charter
UXSix-to-ten-screen free loop, one-hand waiting-room flow, export/share, draft recovery, AI failure/manual fallback, escalation acknowledgement.Local-Only Pocket Prototype Spec
ArchitectDelta from M-01/Architecture: local draft state, raw/structured capture, confidence/correction fields, escalation event, quick note, export artifact, model I/O.Patient Data + AI Contract
QAPass/fail gates for free loop, escalation, no diagnosis/treatment, local retention, export, deletion, analytics, model output audit.Safety & Privacy QA Gate Matrix
PMFile/task-level autopilot plan that reuses existing approved artifacts and avoids old internal-review-first drift.Revised Decomposition Plan
next dispatch condition

Only dispatch first-wave workers after Justin approves this revised plan or explicitly authorizes draft artifacts as sufficient.