Skip to content

The markedness vocabulary

Every component of a preserved model carries an explicit label that denotes the status of Pergamon's review. There is deliberately no unknown — a gap with no assigned accountability fails validation. Our epistemology follows the principle that humans make decisions about what we do or don't know, and that a system becomes more truthful when decision-makers must link their reputation as a custodian their decisions. The methodological standards for what constitues a comprehensive investigation will surely change over time, and, in any event, our intial standard must be written with preservation experts. For now, our custodial responsibility dictates that all claims we make about history remain marked, and that all the things we don't know remain marked as well.

Status Means Accountability Required floor
present Captured and in the bag inward (governance and practices) path or external_ref
withheld Existed; the rights-holder chose not to release it outward evidence of intent + reason
not-applicable Never existed for this model structural reason
unrecoverable Was public; Pergamon could not capture it inward (capture failure) reason
undetermined Pergamon's review ran and was inconclusive inward (handed forward) review reference with date

Gaps in the record come with marked responsibility

withheld requires evidence of intent. Asserting that a rights-holder chose to withhold something is a claim about their decision. When intent cannot be evidenced, the honest status is undetermined.

undetermined has to show it looked. Every undetermined carries a review reference — which process ran, and when — so "we did our best" is checkable, and a later determination reads as a dated improvement rather than a contradiction.

Example

"training_data": {
  "status": "withheld",
  "evidence": "Release shipped weights, config, and tokenizer but no corpus.",
  "reason": "Training corpus was never publicly released by the rights-holder.",
  "custodian_note": "ARCHI/CUSTODIAN.txt#training-data"
}

The only derived summary is ARCHI-Ledger-Resolution (resolved | unresolved): whether any component remains undetermined. It summarizes the state of the review, never the model. Everything else is read, not graded.