Anti-Smoothing Principle — Provenance, Not Style
Capture
The site resists AI smoothing through provenance specificity, not through stylistic choices. The schema is inviolable and machine-readable. The content inside each field is irreducibly human — tied to specific incidents, specific sessions, specific squirrel history.
An AI can imitate almost any style on demand. An AI cannot generate the provenance of a specific stuffed squirrel who hid a metronome behind a couch during a specific practice session on a specific night. The resistance lives in the specificity, not in the surface register.
Why
"AI smoothing" describes what happens when content is optimized for readability, tone consistency, and universal palatability. Smoothing removes: genuine contradiction, irrational preferences, dead ends that stayed dead, inside references that require knowing the author, the operator's 11:47pm bitterness about a bad practice session.
The naive response to AI smoothing is to write in ways that "sound human" — informal, slightly imperfect, occasionally contradictory. But this is imitable. A well-prompted AI produces all of these surface features. Any specific example of "AI-resistant writing" can itself be produced by an AI given the right prompt.
The correct response is provenance. An AI can write "this reasoning may be slightly bitter." It cannot write that this specific reasoning was produced by this specific operator at 11:47pm after a bad practice session in which YY staged a protest involving the metronome. The bitterness is surface. The specific session is evidence.
The schema reinforces this. By requiring WHY-NOT and BREAKS IF as explicit fields, it forces the author to document the parts of reasoning that are most specific — the rejected alternatives are always specific to the context in which they were rejected, the constraint is always specific to the conditions that produced the decision. These fields are harder to fabricate than the DECISION or WHY fields.
Why-Not
Why not write in a deliberately non-AI style (informal, typos, run-ons)? Imitable. A well-prompted AI produces deliberate imperfection. The surface is not the evidence. Provenance is the evidence. Style-based anti-smoothing fails as AI capability advances.
Why not claim "AI can't write this"? Self-defeating. Any specific claim of AI-resistant writing is immediately testable. The claim "no language model appends 'this reasoning may be slightly bitter' to a timestamp" is falsified by the act of writing it. The resistance cannot be located in what AI cannot do — it must be located in what only this operator with this squirrel could produce.
Why not use AI to write entries and just add human review? The provenance dies the moment an AI drafts the entry. A human-reviewed AI draft has AI provenance with a human signature. The YY'S TAKE field becomes a decoration rather than an authentication mechanism. The entry looks right but fails the authenticity test.
Why not rely on the schema alone for anti-smoothing? The schema is machine-readable by design — an AI can fill the schema fields correctly. The schema constrains structure; it cannot constrain origin. The content inside the fields must carry the provenance. Schema without provenance is a filled form, not an artifact.
Breaks If
Entries are edited for tone consistency before publication. YY'S TAKE is invented rather than arising from genuine session character. The WHY-NOT field is filled with generic alternatives rather than the specific alternatives that were actually considered in the session that produced the entry. Timestamps lose their contextual notes ("written at 11:47pm after a hard session") and become bare dates.
Tribal Context
Operator supplied: The core insight — provenance not style is the correct anti-smoothing mechanism.
Session supplied: The observation that the self-defeating claim ("AI can't write this") was itself produced by an AI, making the provenance argument the only defensible one.
Commit
Decision: Anti-smoothing via provenance specificity. Inviolable schema holds the structure. Irreducibly specific content inside each field carries the human signal. The squirrel's history is the authentication. Style cannot be the defense.
YY'S TAKE: YY notes that no AI has ever hidden a metronome behind a couch.
Confidence: High.