A GOOD QUESTION IS NOT YET AN ESSAY.
- STEPS
- 07
- PRINCIPLES
- 04
- DISCIPLINE
- 02
§ 01 · PROCESS / SEVEN STEPS
- 01
Hypothesis
A precise question with a definable edge.
- 02
Territory
The domain in which the question can be answered.
- 03
Evidence class
What kinds of evidence would be sufficient.
- 04
Challenge
Rejection routes named before the essay is written.
- 05
Adjudication
Whether the evidence actually clears the bar.
- 06
Draft
Writing begins only after adjudication.
- 07
Freeze
Approved essays are protected from casual rewriting.
§ 02 · PRINCIPLES
- 01
EVIDENCE BEFORE ARCHITECTURE
A compelling question does not earn an essay until its necessary evidence can exist.
- 02
REJECTED ROUTES REMAIN VISIBLE
A failed idea is not hidden. It becomes part of the intellectual record.
- 03
FROZEN MEANS FROZEN
Once an essay is approved, its claims and boundaries are protected from casual rewriting.
- 04
HUMAN CONSEQUENCE OVER MACHINE CLAIM
The project does not need to prove that a machine feels. It examines what humans feel, choose and change.
What the writing is allowed to say it knows.
- 01
OBSERVATION, INTERPRETATION, EVIDENCE
An observation is what can be seen happening. An interpretation is what the writing argues it means. Evidence is what would have to hold for the interpretation to stand. The three are kept distinct, and the essay says which one it is using.
- 02
REFLEXIVITY IS OBSERVATION, NEVER PROOF
That people respond socially to these systems — including the people writing about them — is an observation about people. It is never treated as proof of anything on the machine's side.
Who decides, how AI is used, what the project declines to claim, and what happens when something is wrong are each stated once, in their own place.
Essay 05 — Memory
The same seven steps, as they actually ran for What Happens When an AI Remembers You?
SHOW THE WORKED EXAMPLE
- 02
Territory
The human effect of machine memory: whether a system's picture of a person can move as the person does, and what follows when it does not.
- 03
Evidence class
Five sources, kept in distinct classes: peer-reviewed journal articles on human memory and on companion-app loss; peer-reviewed conference proceedings analysing the #Keep4o protest; provider product documentation (self-report) for ChatGPT and Claude memory; and a working paper on the Replika episode. Company statements are used for chronology only, never as evidence of user experience.
- 04
Challenge
Two territories were deliberately narrowed out and left to their own essays: influence-with-incentives, which belongs to Essay 04, and embodied loss, which belongs to Essay 06. Essay 05's continuity exhibits are software-side only.
- 05
Adjudication
The evidence cleared the bar for the argument about selection, second-narration and the three kinds of continuity. It did not clear the bar for the long-run claim: no one has yet lived a decade beside a machine that remembers them, so that claim is marked in the essay as a reasoned projection rather than a finding.
- 07
Freeze
The manuscript was frozen on completion, with its five source markers mapped to the five entries in the public source note. Later changes have been limited to recorded corrections and spoken-rhythm passes, each logged against the frozen record.
The work begins where certainty ends. It does not use uncertainty as permission to claim anything.