Layer: Know · Agent: The Cartographer · Time: 90 minutes, one sitting · Difficulty: Hard · Cadence: Once, refreshed quarterly

You are running on a model of yourself you have never tested.

It was assembled early, from other people's feedback, a handful of memorable successes, and a few humiliations you have not revisited since. You consult it before every significant decision. You have never audited it.

The Self Map is the audit.

What it does

Ninety minutes of structured interview that replaces as much of your assumed self-model as possible with a recorded one. It works from evidence you have already produced — your calendar, your finished work, your abandoned work — rather than from self-description, because self-description is the least reliable instrument you own.

The interview has one governing rule: a claim with a date attached is data; a claim without one is a hypothesis, and gets labelled as such.

What it costs

Ninety minutes, uninterrupted. Your calendar for the past three months, open in front of you. A tolerance for being contradicted by your own record.

Most people find the middle third uncomfortable. The discomfort is the measurement. If the protocol tells you nothing you did not already know, either your self-model was unusually accurate or you did not answer honestly, and only one of those is common.

What it produces

A Self Map: a structured artifact in six parts.

Section	What it records
The Record	What you have actually finished in 24 months — and what you abandoned, and where each one stopped
The Attention Ledger	The hours you are genuinely capable, drawn from the calendar rather than from memory
The Capability Inventory	What you can do, each item tiered by the strength of its evidence
Drives and Aversions	What moves you, inferred from behaviour rather than asked for directly
The Constraints	The immovable, in numbers
The Contradictions	Every place the record disagrees with the self-description

Every capability carries a tier, assigned from the strongest instance you can name:

demonstrated — you produced something with it that another person used or judged
practised — done repeatedly, but no external judgment exists
studied — learned about, never shipped with
claimed — asserted, with nothing attached

Expect the inventory to come out smaller than you expected. That narrowness is usually the most useful line in the map.

Adversary mode

Optional, off by default.

With it set, The Cartographer accepts no capability claim without a dated instance, asks for the least flattering reading of any evidence you interpret favourably, and refuses to let a contradiction close unresolved.

It is not hostility. It is the refusal to help you lie to yourself, which is a different thing.

Honest limits

Self-report is a weak instrument. This protocol routes around it and still partly depends on it.
A calendar records what was scheduled, not what happened. The attention windows are better than memory, not measurement.
The four-tier evidence scale is an invented instrument, not one taken from the literature. It works by forcing you to name an instance; the labels are a convenience.
This is a record, not an oracle. It does not predict your performance.
It is not an assessment, a test, or a diagnostic instrument of any kind.
The protocol as a whole has not been tested. Its components rest on published work; the assembly does not.

See evidence.md for the full accounting, including what is weakly supported and why it is included anyway.

The map decays

You are a moving object. Run it again each quarter and compare. The differences between consecutive maps are more informative than any single map, and after the second run the comparison takes twenty minutes rather than ninety.

What comes next

P-001 feeds two protocols:

P-002 The Weekly Reckoning — the constraints and attention windows become the basis for weekly prediction, and your estimates start being scored against reality.
P-003 The Ninety — the capability inventory and drives become the basis for a ninety-day route.

Neither works well without this one. You cannot govern what you have not mapped.

Running it

Three ways, all equivalent in output:

In the Studio — the browser runner keeps the map, and later protocols read from it. Your key and your data stay on your machine.
In any AI you already use — download agent.md, paste it as a system prompt, and begin.
On paper — the six sections and their questions are above and in agent.md. Nothing here requires a machine. The machine only saves time.

P-001 v0.1.0 · status: draft · licensed CC BY-SA 4.0 · last updated 2026-08-02

The Self Map