JTBD Positioning Findings - AFPS Tracker

The framework-selection gate was approved (compiled YAML, approval_status: ready-for-agent-review), which is the Stage-1 scope approval for the whole selected set. The run manifest research/afps-tracker/_working/positioning-run.yaml was written, and this session ran the first pending framework, jtbd-positioning, inline at its research stage.

One-line job: hold trustworthy, resumable command of a multi-branch AI research portfolio without manual state reconstruction.

This is a progress the operator is trying to make — staying in command of a research portfolio as AI accelerates artifact production — not a feature (“a dashboard”). The hire event is reconstruction cost crossing a threshold; the fire target is the manual reconstruction bundle (editor + rg + reading several files) and, secondarily, asking an AI agent to re-read the repo.

• ICP core job (verbatim): “keep research branches, evidence, approval state, and next actions understandable while AI-assisted workflows produce many artifacts quickly” (icp.md Customer Profile).
• Journey-map synthesis names the same thing: “AFPS Tracker is hired when an AFPS operator can no longer cheaply reconstruct research-portfolio state from the source artifacts alone … a trust-preservation problem, not a generic productivity-dashboard problem.”
• Shared aha moment: both ICP and journey map describe a correct, source-linked branch + pipeline-stage view that matches repo state without manual data entry.

Secondary (related, non-anchoring) jobs

Switch / Moesta four-forces grounding for why and when the hire happens. Push and Pull move the operator toward AFPS Tracker; Anxiety and Habit hold them on native files.

Force imbalance → Push is real but Habit + Anxiety are jointly strong, so the hire happens only past an urgency threshold (multi-path / multi-artifact sprawl) and only if the product reads canonically and verifiably — read-first, source-linked, non-destructive. This directly shapes the job-outcome positioning below.

What the job accomplishes practically.

Functional success metric (customer’s own): time-to-confident-state-answer drops sharply versus opening 3+ files, with every claim verifiable to source — and no new file to maintain.

How doing the job affects how others perceive the operator.

Social risk if done poorly: presenting stale or mis-parsed state to a collaborator / client is worse than presenting nothing — it actively damages the “rigorous” identity the job is meant to reinforce. Reinforces read-first / verifiable design.

How doing the job makes the operator feel.

Emotional success signal: the shift from “I have to rebuild this again” to “I can see it’s right, and I can prove it” — confidence without a verification tax.

What “done well” looks like: the operator opens one source-linked view and, within seconds, knows which paths are active / parallel, what stage each is in, and what evidence and approval record backs each state — and can click through to the canonical file to confirm any claim. Nothing was manually entered; nothing canonical was mutated without consent; nothing is silently stale.

AFPS Tracker should be positioned as the passive, source-of-truth-preserving way to stay in command of a multi-branch AI research portfolio — it makes the real workflow state (paths, stages, evidence, approvals) instantly legible and verifiable without duplicating or mutating the canonical files the operator already trusts.

The job-winning frame is trust-preserving legibility, not productivity dashboarding. The product wins the job specifically because it (a) reads canonical artifacts faithfully, (b) links every claim back to source, and (c) stays subordinate and non-destructive — exactly what neutralizes the strong Habit + Anxiety forces that otherwise keep the operator on native files.

Winning dimensions

Positioning implication (for parent synthesis)

• Anchor on the primary job, framing the category around trust-preserving research-portfolio legibility, not generic PM / portfolio dashboards (which competitive + journey evidence explicitly reject).
• Lead with the read-first, source-linked, non-destructive attributes — they are the literal forces that win the hire, not secondary features.
• Keep write-back and agent-memory framings as latent expansion, not headline jobs (anxiety + scope evidence).
• Hands synthesis a job-anchored differentiator hypothesis to test against strategic-canvas (eliminate/reduce/raise/create) and moore-positioning (template + category), where the research-target wedge gets pressure-tested.

• No direct AFPS-user interviews or observed switching sessions; the job is synthesized from approved research + repo artifacts (journey-map gap).
• No measured reconstruction-time baseline to size the functional outcome.
• No prototype test of the source-linked read-through (the predicted aha).
• No direct WTP evidence for the job; adjacent budgets are directional only.
• Customer language (“context switching”, “keeping up with AI speed”, “knowing what’s being worked on with proper context”) is from interrogation + ICP, not live interviews — wording should be confirmed.

On approval, only the JTBD framework intermediate is written. No canonical research/afps-tracker/positioning.md synthesis is created from this framework — that is the later parent --synthesize step after strategic-canvas and moore-positioning also complete.