Status: Proposal, written 2026-10-16 at the owner's direction after RDR-029 shipped (#708, live at 063e3a92). The owner lifted the Standard-Ebooks-only serving rule for the School only and asked how every one of the plan's 66 texts can become readable in RISE. This record holds the verified sourcing survey (one row per text, every link opened on 2026-10-16), the mechanism that admits them without touching the Archive canon, and the build order. Audience: the owner, and the engineers who build the slices.

Related: the plan · the pane's design · ARCHIVE-CANON-SPEC · RDR-029, RDR-030.

• The rule is one membership test. A work is served when its id is in CANON (src/content/archive/canon.js); nothing else counts. Lifting that for the School by widening CANON would make every School text a Library "Received" card, break the count pins at 15, give each a certification status of invalid (edition identity is hard-coded to Standard Ebooks), and put the Archive's own promise ("Standard Ebooks only") in a false position. Do not widen the canon. The Chapel already shows the right shape: a collection-scoped corpus with its own manifest, its own build loop, its own checksum gate and its own door, outside the canon machinery and never re-litigated by the August rule. The School gets the same: a School corpus (§3).
• The survey (§2) finds a free, proofed edition for 46 of the 62 unreadable texts today. Ten of those are a different edition from the one the plan names (a later printing or revision; recorded per row, and shown to the reader as the edition actually used). Nine texts exist only as scans or uncorrected OCR and need a proofing pass. Six have no free English at all and enter in their original language, as the plan's first selection principle allows; one of them is the Gate.
• The Gate has no English before 1931. The first complete English Oration is Forbes (1942); Wallis (1944/1949) is out of copyright under life+70 since 2015 but its US renewal is unverified; the Brown Pico Project's modern translation states no licence. The honest road is the one the plan already names: a RISE translation from the clean Latin (Latin Wikisource and Bibliotheca Augustana carry the Pico Project's numbered text), released CC BY as the school's first original work, made with a model and read line by line by the owner before it is served (§5).
• Seven of the nine withheld Archive payloads are not the way in. Two are usable now (Kandinsky, Zarathustra); the rest carry Wikisource boilerplate, prose-reflowed verse, a truncated extract, or OCR noise with a bibliography-named scheme. Each is re-sourced into the School corpus from a clean edition rather than repaired in place; the Archive's withheld copies stay as they are.
• The whole thing is six pull requests (§6). The first makes the mechanism and brings the ten Standard Ebooks texts in; after it, admitting a text is one recipe in one script, verified by the same gates every time.

1. The Archive's rule stands for the Archive. The School's texts never enter CANON, LIBRARY_TEXTS, the Library's Received shelf, the decision catalogue, or the release-readiness count. Nothing the Archive promises becomes less true.
2. Every School text is served with a named edition, a source URL, the source's bytes hashed at ingest, a rights basis for the US and a stated life+70 status. What the reader sees under Edition is what was ingested, not what the plan hoped for.
3. Structure is declared, never inferred. Each work's sections carry verbatim names and a per-section verse flag, read from the edition's own markup where it has any and from a hand-written recipe where it has none. Words in equal words out, or the ingest refuses to write (the Standard Ebooks importer's gate, applied to every source).
4. No distributor matter is served: no Gutenberg header, licence or "produced by", no Wikisource export banner, no CCEL notice, no library card pocket.
5. A text with no free English enters in its original language and says so. A RISE translation is an original work, labelled as RISE's, licensed CC BY, and never shown as if it were a period edition.

Legend. Tier A: proofed text, the plan's edition or a reprint of it. Tier B: proofed text, but a different edition than planned; the row says which, and the record will name it. Tier C: scan or uncorrected OCR only; needs a proofing pass against the scan before it can be served. Tier D: no free English; served in the original language, or translated by RISE. SE = Standard Ebooks, PG = Project Gutenberg (ebook number), WS = Wikisource, IA = Internet Archive. "Life+70" is the status in the EU and UK: PD, or in copyright until the year given; "?" where a translator's death year could not be found.

Totals. Tier A 36, Tier B 10 (46 proofed), Tier C 9 (scan or OCR: Kapp, Jung, Faust II, Haeckel, Goethe's Metamorphosis, Humboldt 2–5, Eckhart Evans, Ibn Khaldun, Kepler), Tier D 6 (no free English: Pico, Kapp, Marinetti, Ibn Khaldun, Vico, Kepler; four overlap C). Four already served. 66 in all.

Rights to show. US-only (life+70 still running): Forster and Čapek (2041), Spengler, Einstein (2031), Kandinsky (2028). Unknown translator dates: Wadsworth, Paul and Palmer, Bessaraboff, Page, Parker, Williams, Cox. The record's rights field grows a life70 value (pd | until:YYYY | unknown) and the Lectern says it in one line.

Copied from the Chapel, which has served a Gutenberg Bible under its own rule since July without touching the canon.

src/content/school/works/<record-id>.js   generated: SECTIONS [{name, content, verse?}] + META
src/content/school/corpus.js              SCHOOL_WORKS: [{ id, constName, checksum, edition, source, rights }]
scripts/school-ingest.mjs                 recipes per work; adapters per source; the gates
scripts/lib/content-plane.mjs             a school loop like the chapel loop → public/content/school/<sha>.json, manifest.school[]
src/core/content-store.js                 getSections('school', id) with the same hash check
src/content/school/index.js               readable if manifest.school has it, or (as today) the Archive serves it
src/app/school-reading.js                 loads a corpus work through the store; divisions via declaredScheme (verse propagates)

• META carries edition: { statement, translator, year }, source: { kind, url, retrieved, sha256 } (kind ∈ standard-ebooks · gutenberg · wikisource · ccel · sacred-texts · eebo-tcp · archive-org · rise-original), rights: { us: 'pd', life70 }, structure: 'declared', payloadChecksum. The content plane refuses a checksum mismatch, as it does for scripture.
• Adapters. Standard Ebooks: the existing spine-and-markup reader (src/content/archive/standard-ebooks.js) unchanged. Gutenberg: HTML where it exists (headings and verse are markup there), else txt; cut at the START/END markers; strip distributor matter; reconcile words exactly. Wikisource: the REST API's page HTML (never the ws-export epub). CCEL: txt. sacred-texts: fetched by hand into .ingest-cache/ because of the bot challenge; the recipe records the sha. EEBO-TCP: TEI. Scans: never ingested directly; a proofing pass produces a text file that the recipe then ingests with its sha (§4.3).
• Recipes, not regexes: each work declares its artifacts, where its body starts and ends, its units by name (or a heading vocabulary when the edition's own markup gives one), which units are verse, and which parts to extract when the text lives inside a bundle (James ch. IX, Heraclitus's chapter, Butler's essay, Novalis in Rampolli, Shelley's Defence, Schiller's Letters, Forster's story, Erewhon's three chapters).
• Gates, every work, every time: words reconciled against the cut artifact; no DISTRIBUTOR_MATTER anywhere; unit names unique and not bare numerals; verse units keep their line count; minimum size; the recipe's expected unit count. A gate failure refuses to write.
• What the School pane changes: schoolStatus reads manifest.school first; the Lectern's Edition line shows META's statement and the life+70 line; work: { corpus: true } on a record marks a corpus work. Nothing else in the pane moves.
• What stays untouched: canon.js, CANON, LIBRARY_TEXTS, the Library's cards and counts, certification, release readiness, the decision catalogue. The Archive's tests stay green by construction.

A recipe names the unit boundaries inside the artifact (by heading text, verified by the gate). The School record's edition states "from Principles of Psychology vol. 1, ch. IX".

Boole, Maxwell, Einstein: RISE presents words. Where a formula is in-line it stays as plain text as the edition set it; Boole's LaTeX is converted with a one-off converter to text with the equations in their TeX form in brackets; Maxwell's garbled equations are re-set by hand from the scan (the paper is nine pages). The Lectern notes "equations shown as text".

A proofing pass is a short lane of its own per text: OCR from the IA djvu text, corrected against the page images, with a model doing the first pass and a person reading the diff; the result is a plain text file committed under scripts/school-sources/<id>.txt with its provenance, then ingested like any artifact. Order by value: Faust Part Two, Eckhart (Evans vol. 1), Goethe's Metamorphosis (one article), Jung, Haeckel's supplement, Humboldt 2–5, Kapp, Ibn Khaldun (a selection), Kepler Book V.

Haeckel's Art Forms is 100 plates. It becomes a School work whose units are the plates: each unit's text is the plate's title and its organisms (from the supplement), and the reading's visual is the plate itself through the Gallery field (pinned works, as the science and Rijksmuseum imagery already are). This is the one text that needs a pane change beyond the Lectern: the record carries imagery: 'haeckel' and the launch sets the Gallery collection. Last in the order.

Kapp (German), Ibn Khaldun (French), Vico (French), Kepler (Latin), Marinetti (French), Pico (Latin) are served in their language with rights: 'orig-only' kept, the Lectern's state "Original language" and the language named. RISE's voice already speaks the browser's voices; a reading in French or Latin uses the language's voice where the device has one. The plan admits them this way until a free translation exists.

• Source: the Latin in the Pico Project's numbered paragraphs (Latin Wikisource / Augustana), ~7,500 words.
• Method: a model drafts the translation paragraph by paragraph with the Latin beside it; the owner reads it against the Latin and against Wallis (for sense only, nothing copied); every paragraph the owner has read is marked; the work is served only when all are.
• Licence: CC BY 4.0, "RISE, 2026, from the Latin of 1486". The record's edition says so; the Lectern says "RISE's own translation" and shows the Latin as a second unit so a reader can check.
• This is the school's first original work, as the plan proposed. It is also the pattern for Kapp, Vico, Ibn Khaldun and Kepler if the owner ever wants English for them.

PR 1 is the one that matters: after it the rule, the gates and the door exist, and every later text is a recipe.

1. Edition substitutions (Tier B). The ten rows marked B serve a later or different edition than the plan names. Accept them as listed, with the Lectern naming what was used?
2. Picture work. Haeckel through the Gallery field as described, or defer the work until a text edition of the supplement exists?
3. The translation. Proceed with the RISE Oration by the method in §5, with you as its reader?
4. Sizes. The Enneads, Thrice-Greatest Hermes vol. 2, the Golden Bough abridgement and Cosmos are large; each work is one lazy JSON, so first load is unchanged, but the content plane grows by roughly 8 MB. Fine, or cap the School at a chosen selection per large work?

Published from docs/product/discussions/2026-10-16-mystery-school-complete-library.md. Edits made here are overwritten on the next push to main.