Read as much out of an artefact as it will bear — without crossing into speculation. art-ops serves that aim by describing analysis as operations: single, nameable moves on material, each with a declared profile — plannable, combinable across traditions, checkable item by item.
Analytical traditions contradict each other in their justifications — but many of their moves are identical or combinable. art-ops works at the level of the move.
Every operation is a profile: what it needs from the material, what it demands of the analyst, what it produces, what it forbids. Analysis sequences are planned mechanically from these profiles — including the conflicts between operations.
Honesty is a construction principle: the register distinguishes empirically tested statements from reading-derived hypotheses, and every generated plan declares this line by line.
Distilled from 21 analysis-prompt files of a curated collection. Stage is a precondition ordering, not an importance ranking — stage 0 operations ask nothing of what came before.
Most of the register above is still hypothesis. A conflict is the register's term for a rule saying two operations can't run in the same analytical pass — most such rules are still untested reading. Four material findings stand. The most notable one is not a success story — a declared conflict was put to a test and did not survive it. The underlying test documentation is internal and not published with this repo — the ledger cites it; it cannot yet be independently inspected.
| Status | Finding | Support |
|---|---|---|
| falsified | O8 × O2 does not collide within the same pass — the first conflict of the register ever put to a test does not hold. | o8-knot-v3pre-registered test, blinded judges — control held clean 3/3 · judges passed their own calibration check 6/6 · judges agreed completely (AC1 1.0) |
| practice-proven | The authorship rule (both clauses — a source's own valuation may be reproduced, but only marked). | o8-knot v1–v3built out of two failed runs, held in the third (0/3 → 1/3 → 3/3) |
| provisionally relieved | O12 across two material grades — the a-priori "high" exposure rank stays formally listed, but the defect did not appear in practice (a null result at n=3 per cell, deliberately not stamped practice-proven). | o19-o12-authorshipthe defect never appeared, on either material grade, 3 trials each |
| real-but-probabilistic | O19's authorship defect is reachable — material-dependent, not knot-dependent. | gnomic vs. easily-attributableharder material: defect in 1 of 3 trials · easier material: never |
| reading | Everything else: 10 other O8 conflicts, ~35 cross-operation conflicts, 38 interop edges, 8 remaining exposed operations. | untested hypothesis — no trial run yet |
The register only claims an ordering or data edge between two operations where one is actually declared. Where none exists, the honest plan says so — instead of inventing a sequence.
The inclusive "we/us" folds speaker and hearer into one obligation-community — and quietly excludes the enslaved, whose status the war concerned.
The "personal" address is industrial reproduction — the same pointing finger, printed for millions, addresses no one in particular.
A bot fails on schema-exclusion; a human agent repairs by abandoning it. The channel switch is itself an address switch.
Mostly hypothesis. Profiles and conflicts are derived from source wording, not tested against material — see the evidence ledger above.
Plans; does not execute. A run mode is deliberately deferred until the register is more broadly validated.
Have no hard falsifier. Their outputs are justified readings with a mandatory counter-reading, not measurements.
Sound, moving image, numerical data and code are not covered. Image citation rests on an untested region convention (RCC).