Carepatron · decision handoff

Fewer templates.
Much better defaults.

A profession-specific starter system that gets practitioners to a useful first note quickly—without generic clutter, unsafe automation, or another routing project.

Research complete · private exemplars live · production gated
20existing profession keys
95baseline placements
8library-only placements
95distinct canonical bodies
3private gold exemplars

The decision

Content correction, not a new platform.

Keep the current 20 profession keys and signup routing. Replace the 663-slot default wall with reviewed, Carepatron-owned bodies for new workspaces only.

Ship the smallest safe release.

Four to six relevant workflows per profession. Ninety-five baseline placements, eight library-only placements, no Community dependencies, no Default Intake, no auto-send, no backfill. Unapproved rows fail closed.

Go nowCanonical authoring, profession review, engineering dry-run and feature-flag work.
Go after gatesNew-workspace seeding when each released row has an approved ID, version and body hash.
No-goCommunity IDs, universal RCM forms, hidden conditional branches, auto-send or existing-workspace backfill.

What perfect means

One real job, completed safely.

The gold standard is not the template with the most fields. It is the smallest body that captures the right information, reuses known data and behaves honestly in the real workflow.

1

One named workflow

An obvious user and moment. No catch-all form and no duplicate job inside the same pack.

2

Profession-native

Correct terminology, scope and care episode—not a generic SOAP note wearing a new title.

3

Complete, not bloated

Every required field earns its place. Legitimate unknown or not-assessed states remain possible.

4

Native data first

Client, appointment, practitioner, insurance, date and signature behavior use supported blocks.

5

Low burden

Plain language, one topic per section, mobile-usable and no avoidable repeat typing.

6

Human-owned decisions

Humans own risk, diagnosis, goals, scoring, billing, consent, signatures and final review.

7

Safe AI

Only source-grounded draft assistance, with a non-empty section-scoped prompt and visible review.

8

Release integrity

Carepatron ID, canonical key, version, body hash, named approvals and verified placement.

Live proof

Three exemplars in Riverside.

Built privately in the fake Riverside Psychology workspace. Previewed end-to-end. Nothing was used, shared, published, attached to a client or placed in Default Intake.

Behavioral Health Therapy Progress Note v1Primary demo

A complete routine note without the blank-page problem.

Clear field ownership, bounded AI and explicit risk handling.

RequiredSession date & formatKnown encounter context first.
Required · AI draftDataReported concerns, observations, themes, interventions and response. Transcript only; no invention.
Human selectedRisk and safetyNo acute concern · addressed · safety plan reviewed · not assessed.
OptionalRisk and safety detailsVisible when needed; never populated by AI.
RequiredAssessmentClinician-owned formulation and progress.
RequiredPlanClinician-owned next steps and follow-up.
Pass

Collaborative Treatment Plan & Review

Five human-owned fields: primary goal, progress measure, actions/supports, client perspective/agreement and review cadence. No forced signature and no AI-authored goals.

Product dependency

Insurance & Billing Information

Native demographic and Insurance policies blocks work, but the demographic block exposes both Gender and Sex. Resolve the data-model choice before production.

Safety preserved

Default Intake stayed empty

Zero files before and after the work. The three gold standards and three earlier pilots remained private and available only.

Builder contract

The real product changed the spec.

These are verified constraints from the live editor—not assumptions from the research files.

No observed conditional renderingFormer branches become separate resources or concise, always-visible optional fields.
New blanks persist immediatelyCreation is a commit. Preflight complete payloads and read saved bodies back before approval.
AI badge ≠ configured AIAn AI-enabled row must have a non-empty reviewed prompt. Badge-only bodies fail QA.
Requiredness is nativeHelper text saying “required” does not count. Field properties must be tested.
Below-fold content can be missedScroll the full editor and full preview before judging field count, order or duplication.
Yes/No does not mean acceptedA required acknowledgement can still be answered No. High-stakes continuation needs a product rule.
Native blocks are authoritativeInsurance policies, client information, date range, signature, text and scoring blocks must not be imitated with copy.
Available ≠ automatically sentEverything stays outside Default Intake unless a later workflow receives explicit approval.

Non-send audit

The boundary was kept.

The private build validated authoring and preview behavior without crossing into a client workflow.

Did not use a templateNo client record, response, note or submission was created.
Did not share or publishNo public link, collaborator, Community item or external recipient.
Did not auto-sendDefault Intake remained at zero files.
Did not touch productionNo production ID, seeding config, existing workspace or Community dependency was activated.

Profession packs

Twenty relevant starting points.

Each baseline has four to six high-frequency jobs. Amber items are library-only until a practitioner intentionally selects the specialist workflow.

20 professions · 95 baseline placements · 8 library-only

David / RCM

The answer is two forms.

The January Demo Day transcript and August Dylan/David huddle resolve the question: adapt the RCM intake pattern for new US healthcare workspaces, not just existing RCM customers.

Native behavior to finish

1. Insurance information & demographics

  • Configured payer, member/group/plan details and coverage dates.
  • Policyholder relationship and details only when relevant.
  • Multiple policies, card images and coordination-of-benefits support.
  • Create policy records and verify eligibility where available.
  • Resolve Gender/Sex duplication, consent and signed-source retention.
RCM + compliance gate

2. Financial responsibility & assignment of benefits

  • Separate responsibility, authorization/referral and assignment sections.
  • Practice-configurable copy; no test workspace name or invented policy.
  • Signer role, authority and signed date; AI and scoring off.
  • Affirmative-acceptance behavior needs an explicit product decision.
  • Never universal, never Default Intake, never auto-sent.
Reconciled scope: exactly two approved forms in Insurance & billing for the explicit 16-profession US healthcare allowlist. Release of Information remains conditional. Good Faith Estimate remains in the self-pay workflow. No existing-workspace backfill in v1.

Rollout

Build in parallel. Release behind gates.

Engineering can implement the narrow contract while content review continues. Missing approvals omit a row; they never trigger a weak substitute.

Stage 0

Freeze the contract

Same 20 keys, 95 baseline + 8 library-only placements, no auto-send, new workspaces only and a one-config rollback.

Stage 1

Engineer behind a flag

Manifest loading, approval gate, Carepatron-owned ID/body-hash validation, US allowlist and fail-closed behavior.

Stage 2

Canonicalize and review

Build the 95 distinct bodies in batches: mental health; medicine/nursing; chiro/rehab; coaching/training; complementary care; nutrition/lactation; functional/Other; RCM.

Stage 3

Controlled workspace pilot

Test all 20 keys, US/non-US branches, excluded keys, partial approvals, mobile rendering, signatures, uploads and zero auto-send.

Stage 4

Gradual new-workspace release

Release approved rows only. Existing workspaces stay unchanged. Roll back to the prior config if verification fails.

Stage 5

Measure activation

First-note save within seven days, removals, edits, imports, time to first note and support contacts.

For Dylan

Four decisions, not another project.

The research, pack design, builder contract and private examples are ready. These are the decisions that unlock implementation.

1. Confirm the existing 20-key contract.No new profession router or onboarding question in this release.
2. Confirm available-not-auto-sent behavior.Profession and RCM templates stay out of Default Intake unless separately approved.
3. Confirm the fail-closed release gate.Only approved Carepatron IDs with version and body hash seed; missing rows are omitted.
4. Confirm the regional and rollback path.Two forms for the US healthcare allowlist, new workspaces only, prior config retained.

Evidence & artifacts

Everything behind this page.

The handoff is self-contained, while the underlying internal files preserve the full research, canonical fields, approval register and QA evidence.

Research & packs

report-source.md, eight profession field briefs, EVIDENCE-LEDGER.csv, profession-pack-decisions-v2.json.

Production contract

canonical-default-templates-v2.json, verification-v2.json, canonical-review-register-v2.csv, IMPLEMENTATION-HANDOFF-V1.md.

Builder & leadership

BUILDER-CONSTRAINTS-AND-QA.md, DYLAN-GOLD-STANDARD.md, DYLAN-ROLLOUT-DECISIONS.md, DAVID-RCM-EVIDENCE.md.

Live QA

carepatron-template-builder-test-report-2026-09-02.md records the three gold standards, three pilots and no-send audit.

RCM source

RCM-CANONICAL-DRAFTS.md and RCM-SEEDING-CONFIG-V1.json; grounded in the 16 January Demo Day transcript and 18 August Dylan/David huddle.

Open external gates

Named profession reviewers; David/RCM and compliance; instrument-owner permissions; Dylan/Leah engineering confirmation.