Series E · Provenance · The Method Turned On Itself

The Veracity Ledger

Doped-Toroid Inference Engine vs. The Industry · June 2026

The session's own discipline applied to the session's own claim. Every assertion about "is this in industry" is logged with a status and a dated source. VERIFIED = a source attests it. CONTESTED = sources disagree (disclosed, not resolved). SYNTHESIS = your construction, no product found — flagged honestly, neither inflated nor denied. The exterior witness is the cited source; where there is none, the ⊘ stands.

8 verified

2 contested

3 synthesis

§1 Verified · sourced, in industry now

C1Verified

Neuromorphic compute has crossed from academic prototype to commercial product.

596 patents filed through early 2026; patent activity surged 401% in 2025 alone (239 patents = 40% of the dataset).

src: PatSnap · 2026-04-01

C2Verified

Recurrent / event-driven inference breaks the von Neumann bottleneck — the toroid sheet's core mechanism.

Co-locates memory and processing in spiking neural nets; fires only on input events rather than clocking through instructions. (Structurally: relax-to-attractor, not fetch-execute.)

src: PatSnap · 2026-04-01

C3Verified

100–1000× energy efficiency vs GPU — but only on event-driven workloads.

Efficiency gain is real and large; scope-limited to sparse/event data, not a universal speedup. The caveat is part of the claim.

src: PatSnap · 2026-04-01

C4Verified

Named commercial players exist across four architectures.

Digital (BrainChip, Intel Loihi, SpiNNcloud); Analog/Mixed (Innatera); Superconducting Optoelectronic (Great Sky); Photonic (Q.ANT, Lightmatter).

src: ResearchAndMarkets · 2026-04-28

C5Verified

Large neuromorphic hardware has physically shipped.

Intel Hala Point: 1.15 billion neurons, 128 billion synapses, 2,600 W max draw — a real, powered, existing system.

src: DataM / openPR · 2026-05-03

C6Verified

TERNARY quantized weights — your trits — are in funded industrial R&D.

Fraunhofer IIS "Lo3-ML": data-flow architectures + distributed non-volatile weight memory + strongly (ternary) quantized weights. Horizon Europe, 2024–2026.

src: Fraunhofer IIS · 2026-03-27

C7Verified

DIODE-based in-memory compute — "the diode is the gate" — is real silicon.

AlScN ferroelectric diode memory demonstrated at >100 Mbit/mm² storage density; selector-free diode crossbar arrays.

src: Nano Lett. / Cell Device · 2025

C10Verified

Capital is flowing heavily into non-GPU AI hardware.

Over $5 billion in VC, M&A, and government funding into non-GPU AI computing architectures, 2024 – March 2026.

src: ResearchAndMarkets · 2026-04-28

§2 Contested · sources disagree, disclosed not resolved

C8Contested

The market is already large (~$8 B in 2025).

One report: $8.16 B (2025) → $34.48 B by 2033 at 19.6% CAGR. Counts broad "brain-inspired" scope.

src: DataM / openPR · 2026-05-03 · ⚠ conflicts with C9

C9Contested

Confirmed hardware revenue is tiny (~$5–50 M).

Stricter report: confirmed vendor revenue $5–10 M, upper bound $50 M. Notes published estimates span $28.5 M – $6.1 B for the same base year — a 200× variance across ten reports.

src: ResearchAndMarkets · 2026-04-28 · ⚠ the variance IS the finding

C8 vs C9 is not an error to fix — it is the honest state. The field is real and fast-growing AND mostly pre-revenue; "billion-dollar industry" and "$10 M rounding error" are both being published now, about the same thing, depending on scope. The ledger discloses the conflict rather than picking a side.

§3 Synthesis · yours, no product found

S1Synthesis ⊘

The DOPED-TOROID geometry as an inference loop.

Real builds are crossbar / planar SNN fabric, not toroidal. No product uses the torus geometry. (Consistent with the sheet's own conclusion: you cut the loop into a plane to scale — so industry lives on the plane.)

src: not found in search · honestly flagged

S2Synthesis ⊘

ABSTAIN-through-the-0 as a designed hardware output.

Industrial ternary hardware uses the third state for weight efficiency, not as a designed "insufficient evidence / I don't know" output. The ⊘-as-a-feature framing did not surface as a product.

src: not found in search · honestly flagged

S3Synthesis ⊘

The INTEGRATED engine — doped-toroid + ternary + abstain — as one object.

Every component is verified (C6 ternary weights, C7 diode compute, C2 recurrent inference). The integration + geometry + abstain-framing as a single thing is not a shelf product. Adjacent to the field, not behind it; assembled by you, not yet by anyone else.

src: not found in search · honestly flagged

The Codified Split

COMPONENTS → verified in industry now (ternary weights, diode in-memory compute, recurrent inference, the energy argument, named players, shipped hardware, heavy capital).
INTEGRATION + GEOMETRY + ABSTAIN-FRAMING → your synthesis, no product found. The bricks are in the catalog; the building is yours; the ⊘-as-output is the part nobody productized — because nobody else required the engine to be able to say "I don't know."

VERIFIED = A DATED SOURCE ATTESTS IT · CONTESTED = SOURCES DISAGREE, DISCLOSED · SYNTHESIS = NO PRODUCT FOUND, FLAGGED ⊘
THE COMPONENTS ARE REAL AND SHIPPING · THE INTEGRATED ABSTAINING TOROID IS NOT A SHELF PRODUCT
THE EXTERIOR WITNESS IS THE CITED SOURCE · WHERE THERE IS NONE, THE OBELUS STANDS
THE VERACITY LEDGER · SERIES E · JUNE 2026 · ALL SOURCES DATED FEB–MAY 2026