---
title: "Three-Lane Strategy: sequence, spine, and stop conditions"
status: "Proposed strategy"
version: "0.1"
date: "2026-08-23"
applicability: "Strategic record. It commits this project to a build order and to stated stop conditions. It creates no product, cohort, pricing, partnership, institutional relationship, clinical validation, or authority to process PHI."
---

# Three-Lane Strategy

## 1. The operating rule

> **Interview and observe all three archetypes continuously. Build exactly one lane per 90-day cycle.**

The three-group structure is sound as a long-term product map and dangerous as a quarterly plan. Each archetype asks for a different system — the Learner wants study, explanation, memory, and progress; the Manager wants planning, policy, governance, communication, and measurement; the Builder wants projects, content, collaboration, distribution, and intellectual property. Building toward all three at once produces one more broad platform before any single job is proven, which is precisely the failure this project is closest to.

Discovery is cheap and parallel. Construction is expensive and serial. The rule separates them.

## 2. The sequence, and why it is this one

```text
Cycle 1: Manager  →  Cycle 2: Certification Learner  →  Cycle 3: Builder–Organizer
```

| Cycle | Lane | What it is chosen to prove | Why it comes here |
|---|---|---|---|
| **1** | Manager | That governance improves consequential professional work, survives a reviewer, and someone will pay for a bounded outcome | Closest to the differentiated asset; least contested by consumer AI; the only lane with institutional budget within reach |
| **2** | Certification learner | Frequency, retention, personalization, longitudinal value, and direct individual payment | Tests what Cycle 1 cannot: whether memory and return behavior create value over months rather than in one artifact |
| **3** | Builder–Organizer | Extensibility, contribution, reusability, rights-aware distribution, network effects | Requires a proven capability worth packaging; packaging nothing is how marketplaces die |

**Why the Manager is first.** The Manager combines clear professional pain, a valuable bounded artifact, weak competition from general-purpose assistants, institutional visibility, the ability to sponsor adoption, plausible budget access, direct connection to governance, and no need for PHI or clinical execution to be useful. No other lane has more than four of those.

**Why not the Learner first**, despite it being the easier build: individual learning tools compete head-on with general-purpose chat, where the honest differentiator is governance discipline that a solo learner does not yet feel the absence of. The Learner lane is a retention experiment, and retention experiments need something to retain around.

**Why not the Builder first**, despite it being the most exciting: generic tools already support business planning, content, and project management well. This lane wins only through nursing context, governance, community accountability, rights-aware contribution, traceable evidence, reusable packaging, and nurse-network distribution — and most of those are downstream of a workflow that has already proven worth reusing.

## 3. The spine: build once, instantiate three times

The reason one-lane-per-cycle does not mean three disconnected products is that roughly seventy percent of each lane is the same machinery. The spine is built in Cycle 1 as general infrastructure with a single consumer, and each later lane is a schema plus a workflow on top of it.

| Spine component | What it does | Built in | Reused as |
|---|---|---|---|
| **Intake contract** | Collects a bounded input set, refuses out-of-boundary content at the door, records what was declined | Cycle 1 | Learner goal/baseline/sources intake; Builder initiative intake |
| **Refusal set** | Named, testable refusals with user-visible reasons — PHI, identifiable staff performance, employer-confidential, rights-encumbered source, out-of-scope clinical | Cycle 1 | Extended per lane, never weakened per lane |
| **Source and rights ledger** | Every source recorded with origin, permission basis, and whether it may appear in a distributable artifact | Cycle 1 | The dominant constraint in Cycle 2; the licensing substrate in Cycle 3 |
| **Section engine** | Renders a schema of required sections, tracks which are unanswered, marks what is missing rather than inventing it | Cycle 1 | Learning plan sections; initiative charter sections |
| **Uncertainty and provenance display** | Shows what is sourced, what is inferred, what is unknown | Cycle 1 | Unchanged across lanes |
| **Human correction and override record** | Captures every edit the human makes, as product telemetry and as accountability evidence | Cycle 1 | Unchanged across lanes |
| **Session record** | A reviewable trace of what was asked, retrieved, drafted, refused, and approved | Cycle 1 | Unchanged across lanes |
| **Evaluation harness** | Runs stored cases against the current model and flags regressions | Cycle 1 | Grows into the durable asset |

What is *not* spine, and must be rebuilt per lane: the artifact schema, the guided workflow, the refusal extensions, the evidence gate, and the pricing test.

This is the difference between "we will get to the Learner in Cycle 2" and "the Learner lane is a schema, a workflow, and three refusals away."

## 4. The commercial ladder

One ladder. Not four simultaneous offers.

```text
Free governed workflow
        ↓
Paid professional workspace or cohort
        ↓
Nurse-Led AI Governance Readiness Sprint
        ↓
Institutional pilot
        ↓
Governance, evaluation, and assurance platform
```

The first revenue question is not what the eventual platform is worth. It is:

> **Will one nurse leader, educator, or institution pay for one governed outcome now?**

The Sprint rung matters more than it looks. It is the only rung that converts a founder's scarce time into institutional evidence while the software is still thin, and it is where the implementation patterns that later become capability packs are first observed under real conditions.

## 5. The prelicensure lane is research-only

Prelicensure students are studied now and built for only with an educator or program partner. There is no direct-to-student build in this plan.

Permitted direction when a partner exists: synthetic clinical-judgment practice, Socratic questioning, simulation prebrief and debrief, study-load reflection, academic-integrity guidance, program-policy interpretation from authenticated sources, and learning-portfolio development.

Hard non-goals, which are refusals rather than roadmap items:

- Completing graded assignments
- Generating exam-bank substitutes
- Ingesting or reasoning over clinical-rotation patient information
- Claiming competence
- Replacing faculty supervision
- Implying completion of accredited clinical or simulation hours

## 6. Governance ceiling for all three lanes

Every lane in this plan operates at or below the public ceiling: **D0/D1 data, Green/Yellow risk, Observe/Draft/Recommend action, with Recommend beginning at Yellow.** The system proposes; a named human decides and performs every consequential action.

| Lane | Typical data class | Typical risk | Action ceiling | Why |
|---|---|---|---|---|
| Manager | D1 | **Yellow** | Recommend | Institution-specific content names real units, staffing conditions, and local policy even with no PHI |
| Certification learner | D0–D1 | Green, Yellow for employer-sponsored | Recommend | Personal learning content is low risk; sponsorship introduces an employment-adjacent boundary |
| Builder–Organizer | D0–D1 | Yellow | Recommend | Community and rights exposure, and anything distributable |

Three prohibitions hold across all lanes without exception: no PHI, no employment or competency determination about a named person, and no clinical decision support. A lane that needs one of these to be useful is not ready and does not proceed by being useful anyway.

## 7. What each lane teaches, beyond revenue

Sequencing is also a learning order, and the ordering is chosen so each cycle produces an input the next cycle needs.

**Cycle 1 teaches** whether governance improves real professional work, whether output survives executive or committee review, whether rework falls, whether managers return with a second project, which institutional workflows repeat, what an organization will pay for, what must remain human, where authenticated local policy retrieval becomes unavoidable, and how personal use expands into an institutional conversation.

**Cycle 2 teaches** whether personalization compounds, whether memory creates value a user can feel, what actually causes return behavior, the line between appropriate reliance and answer dependence, how to scaffold thinking without deskilling, and what an individual nurse will pay with their own money.

**Cycle 3 teaches** whether users can build rather than only consume, which capabilities are genuinely reusable, how contribution and attribution should work, whether packs create network value, how community knowledge enters without appropriation, how licensing must work, and whether builders recruit other users.

Together they compound in one direction:

```text
Managers generate implementation evidence
        ↓
Learners generate longitudinal and reliance evidence
        ↓
Builders convert validated patterns into reusable capabilities
        ↓
The community reviews and improves them
        ↓
The evaluation corpus grows — the one asset that is hard to copy
```

The compounding asset is not the Constitution. It is the accumulated evidence about whether real workflows conform to it.

## 8. Stop conditions

These are stated in advance so that continuing becomes a decision rather than a default.

**Stop the sequence and re-plan if:**

- Cycle 1 fails its evidence gate ([EVIDENCE §3](EVIDENCE.md#3-the-cycle-gates)) and the failure is demand, not execution — no reviewer accepted a packet, or no manager returned with a second problem.
- Two consecutive cycles produce enthusiasm and no payment.
- The refusal set has to be weakened for any lane to be useful.

**Stop a lane immediately if:**

- Any workflow requires PHI, identifiable staff performance data, or rights-encumbered material to produce its artifact.
- Users routinely bypass the governance steps rather than merely finding them tedious. Bypass means the governance is theater; tedium means it needs design work. The two are distinguished by asking what they did next, not by asking how they felt.
- Output is accepted by reviewers because it looks authoritative rather than because it is complete. This is the most dangerous success signal in the plan and is tested for deliberately.

**Do not start Cycle 3 at all if** contribution cannot be made consented, attributed, versioned, evaluated, and revocable. A marketplace built without those is a liability with a landing page.

## 9. What this cycle does not build

Deferred, deliberately, with no apology: more ecosystem names, more generalized agents, agent swarms, broad EHR integration, PHI workflows, a native model, a marketplace, certification claims, universal clinical terminology integration, another architecture report, and institutional platform features without a named design partner.

Each of those may belong later. None resolves the proof gap, and every one of them consumes the only cycle available.