Claim status is tied to identified evidence.

SFT B18 R1 deterministic context comparison

B18 R1 deterministic context construction passed all required-fact, missing-evidence refusal, solver-parity, and revision-traceability gates, but failed its predeclared median serialized-context efficiency threshold.

• The seven cases are authored fixtures and do not establish production-scale retrieval, geometry, interoperability, robotics, or digital-twin performance.
• LEXICAL_TOP_K and GRAPH_NEIGHBORHOOD are transparent local baselines, not production Vector RAG or GraphRAG systems.
• The predeclared median context-size target failed: SFT used 78.5827% of full-scene bytes versus a maximum target of 60%.
• The pinned-model holdout was frozen but not run because the deterministic efficiency gate failed.

Fractapp Fractal Memory Machine

External Base-4 integer-addressed reference machine implementing exact 160-level byte addresses, paged storage, direct seek, range reads, streaming reconstruction, SHA-256 verification, path-as-data, portable packs, and executable program bytes in the same recursive geometry.

• The current evidence was produced inside the Fractapp project and has not yet completed the independent B19 review protocol.
• The published benchmark uses an in-memory backend and does not represent disk, object-storage, network, or multi-device latency.
• The 64-bit logical offset field describes address capacity, not available physical browser storage.
• Path-as-data is exact representation and must not be described as arbitrary compression.
• The Fractal VM is a deterministic reference instruction set, not a production CPU or WebAssembly replacement.
• The canonical repository is access controlled; outside source review requires owner authorization.

SchemaStack Signal Gate live audit

390 completed local Ollama rows across three models, five modes, and three small benchmark families.

• Small sample.
• XSTest and OR-Bench inputs are seed subsets, not full official datasets.
• No general safety conclusion is supported.

GlyphDrive/FOVEA proof-deck independent verification

Reproduced 12/12 shipped self-tests and responsive/browser checks; independent audit also found open proof and security defects.

• Green fold verification was self-hashed rather than checked against a pinned receipt.
• Same-frame restore was not actually proven.
• Long address arithmetic and unescaped input were release blockers in the audited deck.

MicroScope Resolution Frontier Phase 5 checkpoint

Checkpoint records baseline, ablation, restoration, neighbor control, benchmark-family, metric, UI, and export implementation.

• CRS ordering and compressibility proxy remain speculative.
• No unique-circuit or closed-provider causal claim is supported.

EL-JEFFE 4B specification and Phase 0 control package

241-file specification/build-control package with schemas, gates, synthetic tests, and Phase 0 verification.

• No trained checkpoint.
• No model-quality gate has passed.
• No production mobile runtime or comparative result is claimed.

WorkPage OS v1.1 pressure-tested build pack

Specification, pressure test, nineteen-phase runbook, and build-first instructions.

• The source pack is build-ready documentation, not a current completed desktop-application receipt.

Source bytes remain behind the publication boundary; the reviewed hash and scoped interpretation are public.

GlyphCAS design report

Source report supporting the typed identity, deterministic canonicalization, reuse, and proof architecture.

• The public paper does not establish universal canonicalization, universal cache reuse, or production latency gains.
• Source bytes remain behind the paper's public-disclosure boundary.

Source bytes remain behind the publication boundary; the reviewed hash and scoped interpretation are public.

GlyphDrive/FOVEA/OOC technical assessment

Technical assessment and v0.2 research plan supporting narrow mechanism claims and explicit future gates.

• Founder-local Rust conformance, full codec baselines, and cognition experiments remain unrun.

Source bytes remain behind the publication boundary; the reviewed hash and scoped interpretation are public.