Screenshot flows — walkthrough with drop-off, artifact-first¶
The killer use-case: a persona walks through a real onboarding flow or prototype and shows exactly where it drops off. ~80% of that value is more reliable via reaction-to-artifacts than via live app control, so this path needs no browser anywhere — the flow is made of real screens the host actually looks at.
The flow¶
attach_asset(project, "welcome.png") … # the screenshots become evidence assets
define_flow(project, "Onboarding v2",
steps=[{asset_id, caption?}, …]) # the ORDERED flow, citable per step
brief_flow_walkthrough(persona, project, flow) # persona context + the ordered screens
↓ the HOST view_asset()s every screen (real pixels) and authors the timeline
record_usability_session(subject={kind:'flow', id}, fidelity='artifact', …)
↓ friction per step, dropoff_step + reason, predicted_behaviors with likelihoods
flow_funnel(project, flow) # the segment view
- Defining: steps reference the project's image/screenshot assets
(
attach_asset/attach_prototype_shotfirst); captions default to asset titles; a stablekeymakes re-definition an idempotent upsert. Document assets are rejected — personas react to pixels. - Walking: the brief carries the persona's loaded context plus per-step
view_asset(...)calls. The host looks at each screen before reacting, records one timeline step per screen, setsoutcome.dropoff_stepwhere this persona would genuinely bail (the reason lives in that step's verdict), and authorspredicted_behaviorson the canonical likelihood scale. - The funnel:
flow_funnelaggregates every session of the flow — per step entered / continued / dropped, the drop reasons, the dropping personas, the step captions, and abiggest_dropoffheadline: where the segment abandons, and why.
Downstream¶
Walkthrough predictions are scoreable: point record_prediction_outcome at a
session's prediction ({kind:'session', id, anchor:'pb1'}) when the real
funnel data arrives — the walkthrough feeds the calibration loop
(docs/calibration.md). Live actuation is the separate, complementary path
(the walkthrough/live modules); this one works from artifacts alone.
CLI: flow-define <json>, flows-list, flow-funnel.