# 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

1. **Whole-source inventory:** all retained corpus media types, hashes and
   structural summaries.
2. **Semantic atomization:** bounded prose/list/table/code units and their
   containment hierarchy.
3. **Epistemic normalization:** statuses, evidence, boundaries, assumptions,
   losses, falsifiers, implications and open questions.
4. **Relation typing:** distinguish citation, support, dependency,
   qualification, challenge, generation, verification and continuation.
5. **Scientific profiles:** mathematics, physics, experiment and governance
   adapters applied where their required coordinates are present.
6. **Transformation promotion:** migrate programme spines and verified chains
   through the strict common algebra, retaining negative controls.
7. **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.