# Sources and transcription method

## Exact source copies used

**Mémoire 4.0(1).docx**

SHA-256: `cd7b9924479c1e976cbb4fd8c5cb2271897c658d8e5b35c967ed441b28c27949`

**Source copy 5.1 (2017–2018)**

SHA-256: `ab2c262e37aeefa264b70550992b12a4bfd0450f4c4154fdb7f67d3705ddbafc`

The hashes identify the exact source copies used; source-copy filenames are not required for identification. The newer archive’s inventory names a different whole-file hash for its 4.0 copy. The transcription binds the exact mounted originals above rather than copying that hash. All seven corresponding annex EMFs are byte-identical across the two copies actually inspected. Those seven hashes are recorded in `data/il.source.json` and verified from the delivered original EMFs.

## Body text

The IL synthesis is found in 4.0 paragraphs P129–P140 and 5.1 P142–P153. P133 / P146 describe the eight pillars and nonlinearity. P140 / P153 describe metacognition throughout the process and after its completion. Pxxx counts direct `w:p` children of `word/document.xml`’s `w:body`, including empty paragraphs. It is not a page reference.

The package includes 23 selected paragraphs: 13 from 4.0 and 10 from 5.1. The paragraph text is extracted from XML text runs. It is compared against the exact originals by `scripts/verify_originals.py`; the original wording is not regenerated from the preceding conversational analysis.

The graph’s historical-context view selects source-bound definitions rather than importing the later assistant’s interpretation of them. In particular, the original 5.1 treatment of constancy across independent STIs and truth value remains visible in its original French wording.

## Annexes

The source annexes are embedded Enhanced Metafiles (EMF), not ordinary editable DOCX tables. Text was read from `EMR_EXTTEXTOUTW` records with their coordinates; no OCR was used. The raw vector files and coordinate/text records are included. Original page images were rendered from 4.0 through LibreOffice and inspected.

| Content | 4.0 source | 5.1 source | Page of rendered 4.0 PDF |
|---|---|---|---:|
| Annex 1: comparison of source models | image1.emf | image2.emf | 45 |
| Annex 2: information need | image2.emf | image3.emf | 47 |
| Annex 2: finding/access | image3.emf | image4.emf | 48 |
| Annex 2: evaluation/organization | image4.emf | image5.emf | 49 |
| Annex 2: use/communication | image5.emf | image6.emf | 50 |
| Annex 2: ethics/metacognition | image6.emf | image7.emf | 51 |
| Annex 3: synthesis | image7.emf | image8.emf | 53 |

Annex 3 contains **15 competencies and 39 X marks**. NILIS’s column is hatched. A blank or hatched cell is transcribed as a document mark, not as a claim that the corresponding model has no relevant contribution.

Annex 2 contains **181 indicator occurrences** under the eight pillar headings: 19 need, 43 finding/access, 23 evaluation, 15 organization, 18 use, 17 communication, 19 ethics, 27 metacognition. Repeated indicators retain distinct occurrence IDs. For example, a source code appearing in both evaluation and organization is not deduplicated into a single occurrence or silently assigned to one of them.

Line-wrapped text runs in a table cell are joined with spaces for reading. Spelling and phrasing are retained. Exact coordinates and separate text runs remain available in `sources/*.text-runs.json`. Boldface, cell colors and hatching remain visible in the original rendered images; their graphical formatting is not turned into extra semantic assertions.

The graph records two kinds of annex connections without confusing them: an Annex 2 occurrence is attributed to its source model and classified under its original pillar; an Annex 3 X mark connects a source model with a synthesized competency. The annexes do not provide a complete fine-grained mapping between every indicator and every synthesized competency. No such mapping is invented.

## Explicit numbering inconsistency

The bodies give metacognition 7 and ethics/law 8. The annex’s vector text explicitly includes:

- Ethics row: x = 326, y = 476, text `7`.
- Metacognition row: x = 326, y = 522, text `8`.

These coordinates refer to 4.0 `word/media/image7.emf`, identical to 5.1 `image8.emf`. The rendered source image confirms the numerals. The claim that the annex has only visual ordering without explicit numbers is false for these sources.

The names remain stable graph IDs; the body and annex numbers remain separate fields. The annex additionally names responsibilities in its ethics heading; the shorter body heading is retained as its own wording.

## Editorial representation

3D geometry is an editorial reading aid, not a newly attributed historical figure. The first six pillars occupy one operational plane rather than a chain of mandatory transitions. Metacognition has source-bound “underlies” relations to the other pillars and a relation to reflection after the PRI. The ethics/law pillar frames the PRI through its own relation and a wire enclosure.

Two additional visual devices are **presentation-only guides**. A faint, undirected mesh among the six first pillars visualizes the body-text statement that one may navigate from one pillar to another at any point. A dashed directed guide from post-PRI reflection toward PRI visualizes learning across a later iteration. Neither guide is inserted into `data/il.ledger.json`; they exist only in `assets/layout.js`. This prevents a global prose statement from being misrepresented as fifteen separately asserted historical relations or as a literal same-episode loop.

Depth now separates presentation layers rather than alternating peer nodes for decoration. Operational pillars sit at z=0; synthesized competencies sit behind them; source-model provenance, quoted indicator occurrences and manuscript-context records use distinct depth bands. The crosswalk remains planar because its source is a table. Node size varies by record category, not by truth, influence or normative weight.

### Lexical fidelity

The model’s wording is not treated as interchangeable prose. In particular, the ethics/legal competency is preserved as: **“La personne infolettrée comprend et tient compte des enjeux éthiques et légaux qui entourent l'usage de l'information.”** This does not say that the person obeys law, complies with law, or treats legality as ethical authority. Replacing “comprend et tient compte” with stronger compliance language would change the model. The entry-point HTML contains a reviewer warning to prevent that drift in later revisions.

Canonical relations remain source-attributed navigational representations with no licensed inference. Presentation-only guides are visually distinct and are not listed as incident historical relations in the sidebar. Named semantic distinctions remain available through records even if an on-screen label is omitted to avoid overlap; the display reports its label count and the complete index is never truncated.

The English interface and translations are editorial. Quoted French source text is not silently translated in place, and quoted English source-model indicators remain in English in both interfaces.