Synthetic readers
The instrument that tests the pages, kept deliberately outside the pack because it feeds the pack rather than belonging to it. Two agents: one is a browser and nothing else, one is an archetype that receives pixels and nothing else and points at where to go next. Patience is set from outside the model, so abandonment is a measured event rather than a story. Findings flow one way — into the pack's change control.
The three archetypes
Property lists, not portraits — composed so that they map to no individual, with the sources recorded nowhere. The full definitions are in the archetypes document.
R1 — The shipping founder
Ships product with AI coding agents every day. A power user of agents who has never once looked at what they can reach. Turned the permission prompts off early because they interrupted.
- Grant position
- position 3 → 4 · vibe coding drifting into professional development
- Prior confidence
- “My repo, basically” — said with more confidence than has been tested
- Predicted failure
- Bounces off jargon before reaching the finding
R2 — The agent-security practitioner
Secures agent deployments for a living, and runs more agents than the people they advise. Arrives already holding opinions, and stops reading at the first loose word.
- Grant position
- position 4 · CLI agent, code-host token, several machines
- Prior confidence
- High and partly earned — has actually looked, though not recently and not everywhere
- Predicted failure
- Rejects the whole instrument over one inaccurate claim
R3 — The adoption executive
Pushing agent adoption down through a team, measured on productivity rather than on security. Expects a security argument aimed at slowing the thing they are championing.
- Grant position
- position 1–2 personally, responsible for a team spanning 1 to 5
- Prior confidence
- Has not thought about it as a question with an answer
- Predicted failure
- Reads it as an argument against the thing they are championing
Runs performed
| Run | Archetype | Type | Arrival | Outcome |
|---|---|---|---|---|
| Run 001 | The shipping founder | Type A — tabletop (blind, pixels only) | Pitched — the fixed elevator pitch, and nothing else | Budget spent — no abandonment |
Scheduled, and not yet run
Published as a schedule rather than as results, so that what has not been done is as visible as what has.
| Run | Archetype | Type | Arrival | Status |
|---|---|---|---|---|
| 002 | The shipping founder | Type B — informed analysis | — | Run — published beside run 001 |
| 003 | The agent-security practitioner | Type A — tabletop | Pitched | Scheduled |
| 004 | The agent-security practitioner | Type B — informed analysis | — | Scheduled |
| 005 | The adoption executive | Type A — tabletop | Pitched | Scheduled |
| 006 | The adoption executive | Type B — informed analysis | — | Scheduled |
| 007 | The shipping founder | Type A — tabletop | Cold — told nothing at all | Scheduled — keeps document 08’s rule 3 alive |
| 008 | Any | Type A — re-run after the MC5–MC8 fixes | Pitched | Blocked on the fixes shipping |
Findings so far
| # | Finding | From | Status |
|---|---|---|---|
| F1 | “and 5 more” is an affordance that does nothing, at exactly the point the reader wants more | Run 001 | → MC7 F1 |
| F2 | The entry point speaks the project's vocabulary rather than the reader's | Run 001 | → MC7 F2 · open (MC-D23) |
| F3 | The escalation sentence is the most interesting and the least understood claim on the page | Run 001 | → MC7 F3 |
| F4 | Loading an example drops the reader into the middle of the snapshot card | Run 001 | → MC7 F4 |
| F5 | A realistic budget ends before the controls section — the top card is carrying the page | Run 001 | Recorded — informs the v3 order |
| F6 | Confirmed: the chokepoint sentence is what turns the page into an action | Run 001 | Confirmed |
| F7 | Confirmed: offering only reachable capabilities at the intent step defuses reactance | Run 001 | Confirmed |
| F8 | Truncating a frightening-first list inverts P6 — the hidden rows are the mildest, so the page reads as more alarming than its data | Run 002 (informed) | → MC10 · the one recommended fix |
| F9 | The escalation problem is routing, not wording — and story V6's test verifies availability rather than discoverability | Run 002 (informed) | → MC10 |
| F10 | The self-scroll cause is a missing scroll-margin-top under a sticky nav | Run 002 (informed) | → MC10 |
| F11 | The flagship example omits a benign capability, inflating its own headline count from 8 to 9 | Run 002 (informed) | → MC10 |