LLM-Synthesized Essay

Curated Presentation

The Patina Catalog

Patina is an LLM skill to establish a warrant to communicate. Patina affiliates---words, phrases, habits of construction, analogies, allusions---so that a reader places the author inside a parlance and receives the subject in the light the author intends. Every text sounds like it comes from somewhere, and patina makes that a choice: the writing reads as coming from a community of people rather than from a machine.

(c) 2026 George Georgalis <george@iuxta.com> Unlimited use with attribution.

An introductory essay demonstrates the skill in three degrees natural 1317 words, brief 678 words, and an exaggerated vivid style of 1785 words. The three carry the same argument from the same parlance; what differs is the calibration, so reading any two together shows what a setting does.

The ontology is the foundational document used to develop the skill which is defined in a single file and leverages library files containing markers [words and phrases] to affiliate or displace, within a given parlance. It also carries the decision record, including the reversals, so a later revision can see what was tried and why it was set aside.

The library files published here are seeds; a working installation adds one per parlance, built by the harvest and discover workflows below. The printed settings are likewise a starting point. The effective configuration is synthesized per use and confirmed by fine tuning: render, read the calibration report, move one key, render again.

Quick Reference

Invocation. A purpose and a parlance are a complete invocation; everything else perturbs. The purpose may be omitted where the material shows it, and the parlance where the subject implies one.

/patina essay parlance=[kitchen]
/patina letter parlance=[nautical] cast=homely steep=antique
/patina memo parlance=[unix-admin] markers=[lexical,figurative] truss=0.30
/patina fiction parlance=[kitchen] neutral=saturated emblematic=0.90 poetic=free

Values read by position: 35, 3.5, 0.35, and 7/20 are the same setting. A region name may stand in place of a numeral, and a polar key may take a band, a:b.

Dispositions, derived from the request and never declared: compose with no draft, revoice with a draft, discover with an attached corpus. Harvest is a request, not a disposition.

key0.01.0regions
neutralsaturated with the parlanceno markers addedsaturated marked tinted neutral
emblematicoperative, noticed by membersemblematic, recognized by anyoneoperative mixed emblematic
steepnew in printantiquenew longstanding antique
glossleft for the reader to decodedecoded in contextundecoded contextual glossed
citedthe author speaks the parlancethe parlance quoted at a distanceinhabited mixed cited
trussno bridges, or minimalheavy use of bridgesminimal measured frequent abundant
poetictight ground, attested allusioninvented vehicles, loose groundliteral grounded loose free

Denoted settings: purpose, cast, parlance, markers, displace, span. The cast is the light the subject is received in, and it is fixed before any figure is chosen; no vehicle out of cast is admitted at any setting. poetic tracks truss unless stated. Every invocation emits a calibration report in the response body, never in the artifact, so a revision names a key rather than describing a symptom.

Harvest Workflow

The skill is built to be fed. It reads whatever library it is given and never writes to one, so every entry arrives by the author's decision; harvest is the mechanism that makes those decisions cheap.

  1. Render, then read the report. The invented bridges are listed with their locations, vehicles, tenors, and casts; library entries used are not listed.
  2. Ask for the harvest. The response returns library entries for every marker and bridge placed, each located, in the schema of _schema.yml. Forms already held come back with their current values, for amendment or exclusion.
  3. Amend by hand. Correct salience and the steep attestation, add cast tags and a gloss, set admit or forbid where a purpose should license or exclude the form, and delete what proved operative rather than emblematic.
  4. Research the evidence. Replace intuition-tier entries with an evidence block from the English print record; Ngram development arrives as YAML, never as a fabricated series.
  5. File it as one file per parlance, the file name matching parlance.name. An attached file supplements a directory file of the same name and replaces it entry by entry, so a session may carry amendments without disturbing what is installed.
  6. Re-render the same material against the grown library. The report's evidence line shows the intuition count falling as the library takes over.

For a parlance with a body of writing behind it, the discover disposition inverts the order: attach a corpus, and the skill measures form and construction frequency against a baseline and emits candidate entries, syntactic habits included, for the same amendment pass.