Programme Algebra Conversion Register
Date: 26 October 2026
Status: L0–L6 conversion complete with explicit boundaries; parallel lane remains non-replacing
Completion criterion
Conversion is complete when every retained corpus item has the deepest honest algebraic representation supported by its source. Completeness does not mean calling every relation a physical transformation. It means that every item is either admitted at a declared level or carries an explicit boundary explaining why deeper promotion is unavailable.
Conversion levels
| Level | Object | Completion test |
|---|---|---|
| L0 | retained file | exact path, media type, byte hash and source authority |
| L1 | addressable content | headings, bounded prose units and machine-record boundaries have stable identities |
| L2 | epistemic object | kind, status, provenance, support and authority are explicit or explicitly unresolved |
| L3 | typed relation | source, target, relation kind and observational authority are declared |
| L4 | admissible transformation | endpoint profiles, changed coordinates, dimensions, assumptions, losses, evidence and authority pass validation |
| L5 | composed route | accepted and rejected paths retain ALE valuations and bounded-search receipts |
| L6 | empirical return | preparation, intervention, detector, nuisance model, record and decision rule are joined without authority inflation |
An item stops at the highest level its evidence licenses. A recorded L2 boundary is complete conversion for material that does not assert a transformation. It is not a scientific failure or an excuse to infer missing structure.
Material classes
| Material class | Required conversion |
|---|---|
| Markdown research sources | L0--L2 for every source and semantic unit; L3--L6 where explicitly warranted |
| JSON registers and ledgers | L0 plus structural record summary; schema-specific objects and relations where an adapter is declared |
| Images and diagrams | L0 plus media geometry; interpretive claims remain attached to their authored captions or surrounding source |
| Verification code and evidence referenced by the corpus | referenced-artifact identity and provenance first; executable claims require a declared verifier adapter |
| External literature | exact citation/reference object; scientific authority remains with the source and the corpus's explicit use of it |
| Generated algebra and HTML | disposable projections with source hashes; never independent authority |
Tranches
- Whole-source inventory: all retained corpus media types, hashes and structural summaries.
- Semantic atomization: bounded prose/list/table/code units and their containment hierarchy.
- Epistemic normalization: statuses, evidence, boundaries, assumptions, losses, falsifiers, implications and open questions.
- Relation typing: distinguish citation, support, dependency, qualification, challenge, generation, verification and continuation.
- Scientific profiles: mathematics, physics, experiment and governance adapters applied where their required coordinates are present.
- Transformation promotion: migrate programme spines and verified chains through the strict common algebra, retaining negative controls.
- Reading projection and comparison: algebraic journeys, held-out retrieval tests and replacement/rollback decision.
The generated coverage report is the controlling progress measure. Counts may rise as atomization becomes finer; status is determined by closed gates rather than by a shrinking file count alone.
The compiler's conversionAudit closes this conversion criterion only when all six levels pass. It deliberately does not close the replacement gate: held-out retrieval/reasoning comparison, human reading quality and an explicit migration decision remain separate work.