Skip to content

Latest commit

 

History

History
74 lines (57 loc) · 6.62 KB

prd.md

File metadata and controls

74 lines (57 loc) · 6.62 KB

PRD — Explorable Body Record

Version: 1.1 Status: Accepted scope; implementation verification pending Date: 2026-09-12

Problem and evidence

A consultation combines measurements, spoken explanation and pointing. Afterward, the client may struggle to reconnect a report with what the coach explained. The founder described this experience; one gym owner described assembling reports with Claude and sending them through WhatsApp/email. These motivate the product but do not establish market demand or clinical benefit.

Sources: user research, latest team discussion.

Users and proposition

  • Primary user: client revisiting their assessment and adding observations.
  • Contributor: coach/physio recording and reviewing session explanations.
  • Buyer hypothesis: a practice or gym includes the service in its offering.
  • Proposition: keep explanations and evidence attached to the body regions they concern.

Core journey

Open prepared client record → coach starts recording → speaks about findings, optionally selecting regions → Finish → transcription/extraction → coach reviews and confirms proposals → client sees updates, explores sources and saves an observation.

UI specifications

Detailed screen/JTBD contract: experience PRD. Expanded view and orientation contract: viewer PRD. Build on Vite/React and the current Node backend per ADR-006.

Requirements

ID Requirement Acceptance evidence
PR-01 Body is primary navigation Select and focus at least two supported broad regions on the demo device
PR-02 One app has session and client modes Same record is visible in both; switching modes is explicitly demo-only
PR-03 Capture is intentional Start requires capture agreement; recording indicator, timer, Finish and Discard are visible
PR-04 Short live voice note Real recorded audio becomes visible transcript after Finish
PR-05 Preserve pointing context Timestamped selections are retained; no invented speech alignment
PR-06 Source-backed extraction Each proposed annotation resolves to report/transcript evidence and an allowed region, or is marked unplaced
PR-07 Coach controls publication Edit/remove/confirm actions; no draft appears as a confirmed client update
PR-08 Revision-aware confirmation Changes invalidate confirmation; only the reviewed revision is committed
PR-09 Client discovers updates New marker with author/date; Show update focuses the region and opens evidence
PR-10 Distinguish evidence types Measurement, coach statement, generated explanation and client observation are labelled
PR-11 Client contributes Exact text can be previewed, saved, edited and deleted at the selected region
PR-12 Persistence and replay safety Reload preserves confirmed records; replay duplicates nothing; later distinct sessions remain allowed
PR-13 Recovery from failure Failed recording/model/storage operations retain useful draft state and permit retry or labelled fallback
PR-14 Repeatable demo Explicit reset restores the fictional fixture without affecting unrelated browser data
PR-15 Expanded body Fill app viewport; preserve pose/selection; exit and Escape work
PR-16 Predictable camera views Front/Back/Left/Right and two isometric presets; smooth cancellable movement
PR-17 Useful framing Fit whole body and Focus selected; manual actions require no model
PR-18 Body-first screen states Overview, region detail, capture, review and confirmed record meet the experience PRD
PR-19 (P1) Recorded history Earlier snapshot with explicit date and return-to-latest; no interpolation

Interface priorities

Use Kingsley's calm green/cream visual treatment as inspiration. Give the body most of the workspace. Adjacent details show evidence, session review or selected-region notes. Start session records a consultation; it must not resemble the prototype's workout-start action.

Agent behaviour

Use active assessment, allowed region map, transcript and selection context to propose annotations. Use a validated action bridge to expose semantic context and deliberate viewer actions. CopilotKit is conditional on a short integration check; the real model-to-record workflow is mandatory. Do not require a chatbox to navigate ordinary controls.

The model proposes mappings; deterministic code validates IDs/sources and computes numerical comparisons. Coach confirmation grants permission to publish the current draft. Save client wording exactly; no model rewrite is necessary to persist a note.

Evidence rules

A joint movement test does not establish a particular injured muscle. Missing side remains unspecified. Selected anatomy alone is not evidence of a clinical finding. The reference body is not a client scan. The supplied report's monitoring thresholds are fixture-specific.

Demo data

Fictional client with de-identified report values and optionally one labelled fictional historical entry. Use the script in scope. The live spoken observation is fictional too; no actual client data is required.

P0 exclusions

Exercise planning/logging, golf technique analysis, body or recovery scores, tissue-healing simulation, trend analytics, hardware/WhatsApp integration, separate coach dashboard, multi-user access control, automatic prescriptions and diagnosis.

Recovery and performance are contextual stories, not additional implemented flows.

These exclusions apply to delivery of the consultation P0. Existing prototype recovery/routine features, the assessment story, optional WhatsApp integration and golf mockup source are retained in the combined repository; their presence does not satisfy or expand PR-01 through PR-18. See combined implementation status.

Definition of done

PR-01 through PR-18 pass on the selected demo device. Demonstrate one real recording-to-confirmed-update cycle and one client observation surviving reload. Run twice, record actual latency and disclose all seeded elements. A transcript fallback is useful but does not count as proving PR-04.

Open product checks

Can a client find the relevant explanation without instruction? Are source labels understandable? Is review fast enough for a coach? These are questions for rehearsal, not established benefits.

Change control

Only expand P0 through an explicit decision with a time/scope tradeoff. ADRs record architectural changes; tickets record execution.