asolaria-whiteroom-engine

The white-room engine is LEG-1 of the Asolaria multi-substrate fabric: a pluggable-store, pluggable-scorer, never-delete room engine that emits a Brown-Hilbert address, scores a room, keeps genius, compacts mistakes, and seals append-only HBP evidence.

Local, deterministic, dependency-free Node implementation. No credentials or cloud dependency are required for the in-memory path.

2026-07-11 Path-2 and storage update

White rooms now link explicitly to the measured exact-recovery plane:

• Path 1: recall a retained room/cube body through an authenticated content address;
• Path 2: reconstruct an unretained body from jointly sufficient CRT shadows;
• DBBH→DBWH: re-project the candidate and require SHA/shadow/shell equality before emission;
• storage tier: HDD/SSD/cloud retains full genius bodies, compacted mistakes, receipts, shadows, cubes, and cold state;
• neural sidecar: trained GNN inference remains optional.

Full component record:

PATH2-WHITEROOM-STORAGE-VERIFICATION-2026-07-11.md

Four legs

Leg	Role	This repo
LEG-1	white-room scorer/store, never-delete	yes
LEG-2	prime-sector allocator BH.SECTOR.P{prime}.R{room}.{sha16}	replicated for convergence
LEG-3	GitHub bus, commit=emit/log=read	sibling
LEG-4	Google Drive cold sink	sibling adapter

Store interface

put(pid, value)
get(pid)
scanByPID(predicate)
compact(pid)   // MOVE to compacted; never delete

Backends include MemoryStore and a gated Redis stub; GoogleDriveStore follows the same interface.

The scorer interface includes:

• DeterministicScorer — offline/testable placeholder;
• L0GnnScorer — local GNN sidecar wiring.

The placeholder is explicitly not claimed as the real domain scorer.

Run

node --test
node liris-whiteroom-engine.mjs
node liris-sector-flywheel.mjs 0 5000 5000
node liris-flywheel-run.mjs 3000 1

Files cover the engine, sector allocator, sector flywheel, omniflywheel scale/HDD page persistence, and Hilbert projection.

Why never-delete matters

GENIUS  -> keep body + address + receipt
MISTAKE -> compact body/evidence + preserve receipt

Compaction reduces the active representation without erasing the evidence chain. Disk/cloud can hold durable results while RAM contains only the rooms currently being scored.

This is the correct “HDD instead of GPU-resident state” result. It does not claim disk performs GNN matrix multiplication.

Path 1 and Path 2

Path 1

address = sha256(X)
receiver returns X iff store[address] exists and hash matches
otherwise Held

Path 2

S_i = X mod p_i
exact iff product(selected p_i) >= source_range
otherwise Held::InsufficientJointCapacity

The DBWH white side then requires:

P(R(P(X))) = P(X)

by comparing SHA, all selected/recorded shadows, and frequency shells.

GNN lineage

The optional L0 scorer traces to Jesse's pre-Asolaria healthcare models. Four healthcare model files match later Asolaria sidecar blobs exactly. BigPickle then composes L0/L4, G1/G2/G3/G4, OmniShannon, SHA fallback, Fischer, and Hookwall. White rooms consume the semantic evidence but keep a deterministic receipt/store boundary.

Security

• no credentials, keys, tokens, or vault material are stored here;
• nodes mint identities locally and publish only public material;
• secret scanning remains part of CI/publish hygiene.

Prism/comb boundary

White rooms are quarantine/search regions where candidate collisions are deliberately compared; execution lanes remain collision-separated. The measured 256↔1024 rung is an exact representation change. Content addresses remain coordinates against retained stores, not sub-entropy compression.

Independent verification — 2026-07-11

• MEASURED_CLAUDE_FABLE5_THIRD_SEAT, operator supplied: Path 1 rustc 1.97 19/19 and Path 2 rustc 1.97 30/30.
• AUDITED_GPT_5_6_PRO: complete recovery, Q-PRISM, healthcare-GNN, BigPickle, trained-GNN, Hookwall/Shannon, white-room, cube-mint, Dispatcher, HyperHermes, reductions, algorithms, HyperBEHCS, and N-Nest audit.
• MEASURED_GPT_DIRECTED_GITHUB_ACTIONS: successful Rust 1.97.0 runs 29134408321, 29134413119, and 29134419389.

The Rust runs validate exact recovery/watchers; they are not a new white-room Node-suite result or a live Hilbra benchmark.

Part of

The broader multi-substrate system with 35-TB-google-AI-Ultra-migration, bigpickle-rebuild, asolaria-behcs-256, asolaria-federation-1024, and the two recovery crates.

License

Operator choice; add an explicit license before wide redistribution.