Where this sits
Integration takes the pages and the bridges built in Phases 4 to 6 and connects them to the systems they came from.
The ways in
Every way in opens the hub, and from there the 2D and 3D pages. tests/links.cjs checks each one.
| From | Where | Opens |
|---|---|---|
| The site catalog | pages.html, a card after Hyperbolic Oracles | The hub, the 2D page, the 3D page, the bridges explainer |
| Hyperbolic Oracles’ site bar | Elsewhere menu, on every page of that system | The hub |
| The Geometric Calculator’s site bar | Simulations menu, beside Hyperbolic Oracles | The hub |
| The Hyperbolic Oracles hub | A “Built on these cores” note under its oracles, which says what Engram takes from each core | The hub, the 2D page, the 3D page |
| Its system map | An Engram Mind’s Eye node under a new band, “Used by”, fed by the four cores Engram loads from there | The hub |
Its README.md | A “Used by” section listing every name Engram uses from each file, and the four commands to run after changing one | This folder’s README |
scripts/geometric-build/README.md | Names this system as the first user of orbital_core.js (since Phase 1) | — |
The system’s own pages carry its own menu set, engram-minds-eye, with three menus: the hub and the two pages, the ten explainers, and the systems it builds on. All three menu sets live in scripts/page-chrome-targets.json. node scripts/install-page-chrome.mjs writes the bar into every page, and --check confirms every page has it.
Owl Academy does not link back. That would mean changing the Owl-Academy repository, which this project only reads: it pins Owl Academy’s code and checks against it, and never edits it. A link from Owl Academy’s Engram Mind Eye page to this hub is a change for that repository’s owners to make.
What it stands on
The engram layer is new. Everything under it is loaded by relative path from the system that owns it, in Node and in the pages, and is never copied.
| File | Owned by | What Engram uses | Checked by |
|---|---|---|---|
| orbital_core.js | Geometric Calculator (scripts/geometric-build/, Phase 1) | Fraction, state, packTile, merkleChain, gaugeExact, readout, pumpOf, wheelSector, muxIndex, sha256 and its byte helpers | tests/orbital-core.cjs (18/18); rebuild-identical.sh for the four orbital pages |
| hyperbolic_core.js | Hyperbolic Oracles | The norm-5 tree P (its reduced words are the addresses), Rng, and the clock ON/OFF, which Engram checks against its own | tests/parity.cjs (169/169) |
| decoder3d.js | Hyperbolic Oracles | TreeEmbedding3D, the memory tree in H³ | tests/parity3d.cjs (28/28) |
| twin3d.js | Hyperbolic Oracles | qmul, qconj, qnormExact, qnormFloat, rotOfQuat: the coupler’s mirror law and the frames | tests/parity-twin.cjs (21/21) |
| fence.js | Hyperbolic Oracles | fence, guardCheck, defect, distanceMatrix, centrePoint, generate, corrupt, treeWords, LN2: the watch layer | tests/parity-fence.cjs (101/101) |
| ttmpt_core_v1.js | Owl Academy, copied verbatim and marked so | engram/1: the key, the grids, crystallization, the tokenizers | The golden corpus, E1 (613 cases in the parity check) |
| strix_library_v1.js, lattice_seed_v1.js, biochain_v1.js | Owl Academy, generated from its code | STRIX’s library and reply rule; LATTICE-SEED/1; the biomesh codec | Their pin scripts with --check |
The browser core checks one of these assumptions as it loads: if hyperbolic_core.js’s clock schedule ever differs from Engram’s, engram_core.js refuses to start and says so. The rest are checked by the parity lines, which compare exact values with the Python reference. A change in a core’s results shows up as failing lines, never as a page that quietly does something different.
Who stands on it
The Hyperbolic Lexicon loads two of this folder’s files unchanged and sends its shards back through importBiochain. scripts/cores.json records every name each consumer uses, and scripts/check-cores.mjs fails when a consumer uses a name the core lacks or the manifest does not list. tests/all.cjs runs the Lexicon’s two checks of this folder, so a change here that breaks the Lexicon fails here first.
| File | Used by | What it uses | Checked by |
|---|---|---|---|
| engram_core.js | Hyperbolic Lexicon: lexicon_core.js, its pages | chainQuaternion, qprod, qeq, qnorm, unfold, packetQuaternions, tokenizeBiomesh, hexLsb, crystallize, computeJ, L; the pages: importAny, detectImport, verifyBiochain, liveOf, Coupler; its shards: importBiochain | chain-lexicon.cjs (30 checks), bridges-lexicon.cjs (40) |
| biochain_v1.js | Hyperbolic Lexicon: lexicon_core.js, its pages | CHUNK, MAX_GROW_BYTES, SEGMENTS, STREAM_VERSION, crystalQuat, crystallizeBytes, encodeChunk, decodeChunk, frameStream, popcountEven, qmul, qn, quatHex | the same two; make_biomesh_compat_v1.cjs --check reports its bodies against Owl’s grower |
| reference/engram/ | Hyperbolic Lexicon’s Python reference | engram.address, engram.ttmpt, imported in place | the Lexicon’s pytest suite |
engram_core.js changes only by addition, and the _v1 pins are never edited: a new pin is a _v2 file beside them (decision D4 of the integration overhaul).
One command
tests/all.cjs runs every check of the system and of the cores it loads, the Lexicon’s checks of Engram, the site bar, the Python reference and the four pins. -j N runs them N at a time; the root tests/run-all.cjs runs this list with the other systems’.
The pins compare content, not commits. Each one rebuilds its _v1 files from the pinned source content, which it finds by SHA-256 in the Owl-Academy checkout or, after upstream commits, in its git history (biochain_v1.provenance.json names the commits), and the result must match byte for byte. Then it runs Owl Academy’s current code on the same inputs, and the records and verdicts must equal the pinned ones. An upstream commit that changes a file but not that behaviour passes, with NOTE lines naming what moved; a change in behaviour fails. The checkout is the one next to this repository, OWL_ACADEMY_DIR, or a path given as the argument. Without it, or when the pinned content is not in its history (a shallow clone), a pin is reported as skipped, never as passed. The Python reference’s 45 tests run in the same command.
tests/links.cjs checks every relative href and src in the folder’s pages and explainers, and every #anchor against the ids of the page it points to. It also checks every link on the system map, every file in every site-bar menu set, and each way in from the table above. It finds the folder’s pages itself, so a new page is covered without editing the test. Its three controls must come out right on every run: a missing file and a missing anchor must be caught, and a real anchor must pass.
When something changes
Six kinds of change, and what each one needs.
| Change | Do |
|---|---|
Code inside a segment of orbital_core.js | Rebuild with build.py. rebuild-identical.sh lists the orbital pages that moved; commit them if the move was intended. Then run tests/all.cjs. If an engram result changes, port the change to reference/engram/orbital.py too, regenerate the traces (run_traces.py) and commit both together. The parity check will not let the two drift apart. |
| One of Hyperbolic Oracles’ four cores | Run that core’s parity test, then tests/all.cjs. The names Engram depends on are listed in Hyperbolic Oracles’ README under “Used by”. Renaming one breaks this system. |
| Owl Academy changes its engram code | If the code the pins take behaves as before, they pass and print NOTE provenance or a NOTE body changed line: behaviour identical when a differential set (records with one field altered, the grow cap) gave the pinned code’s verdicts too, same results on the v1 inputs when only v1’s inputs were run. Nothing needs doing. If its behaviour changes, the live contract fails and names the first field or input that differs; so does a name defined differently in biomesh-core.js and biomesh.js, since the pages would run the copy the pins do not. Keep the _v1 files as they are: they define engram/1, LATTICE-SEED/1 and the version 1 session format that existing chains, seeds and exports rely on. Pin the new version beside them as _v2, with its own script. If a claim is to rest on it, register an amendment first, as with Amendments 1 and 2. |
| A shared core here or in Hyperbolic Oracles | node scripts/check-cores.mjs fails until scripts/cores.json is updated in the same change (--update) and the consumers’ suites pass; CI runs both on every pull request. A frozen core or a v1 pin changes only by the user’s decision: on a pull request --base fails unless cores.json records it on that core as "sanctioned": { "sha256": …, "reason": … }. |
| A menu or the site bar | Edit scripts/page-chrome-targets.json (four-space indent, final newline), run node scripts/install-page-chrome.mjs, then --check. The bar is part of the four built orbital pages, so rebuild-identical.sh will show them as moved: commit the reinstalled pages. |
| A new module or page here | The five things on the hub: the code, an explainer, a node on the map, a ledger row for anything it claims, and tests. Add its test to tests/all.cjs. tests/links.cjs picks up the new page itself. |
The record
| Phase | Delivered | Explainer |
|---|---|---|
| 0 · Scaffold | Hub, map, preregistration (E1–E9), golden corpus, four spikes | 1, 2 |
| 1 · Shared cores | orbital_core.js, with the four orbital pages byte-identical | 3 |
| 2 · Python reference | Twelve modules; E1–E7 and E9 supported, E8 inconclusive | 4 |
| 3 · Browser core | engram_core.js, line for line with the reference (849/849 since Phase 7) | 5 |
| 4 · The 2D page | STRIX on the clock; v1 exports byte-identical to Owl Academy | 6 |
| 5 · The 3D page | The memory in H³, the twin system, the fence and the guard | 7 |
| 6 · Bridges | Owl Academy in and out, each import verified by its own code | 8 |
| 7 · Follow-ups | E10 inconclusive (its exact parts supported), E11 supported | 9 |
| 8 · Integration | The ways in, the dependency record, one command for every check | 10, this page |
Eleven claims have been decided: nine supported, two inconclusive, none refuted. The two inconclusive claims are about the same thing. The guard names a corrupted stored distance correctly 96 times in 100, short of the 99 its rule asked for. E10 showed why. Each miss is a tie, or a corruption that leaves no quadruple above ln 2: the guard runs short of witnesses, not of blame. Naming the pair from fewer witnesses would need more than four-point counts, such as the stored distances’ own history. That would be a new claim, registered before any code for it.