MINDCHAIN

MindChain: The Substrate Identity — One Witness, Five Products, and the Graveyard That Paid for Them

Date: 2026-07-10 Research corpus codename: NOOSPHERE Provenance: Compiled entirely from on-disk artifacts of the MindChain program — noosphere/ (chapters 01–09, frozen), kt-verification-ladder/, nel-dispute-ladder/, zkvm-port/, nel-real-inference/, dream-lane/, frontier-lab/math/, and international-lab/ — with no claim in this paper that lacks a named artifact behind it.


Abstract

MindChain is built on one identity: every joule of committed tensor work can emit a useful thought, a security certificate, and a checkable proof at once — and, composed further, a learning signal and a priced counterfactual. This paper is the program's accomplishments record. It states the five axioms as labeled claims, gives the construction that cashes them (one committed span witness emitting five products; eleven organs consuming them), and publishes two tables side by side: the measured spine — Freivalds verification at 1.8669 cycles/MAC MEASURED, 2026-07-10, the 1/T dispute law 958,090/T + 11,602 cycles/chunk MEASURED, 2026-07-10, a 12.67× dispute-vs-recompute advantage at T=128 MEASURED, 2026-07-10, a real 288-byte Groth16 settlement receipt MEASURED, 2026-07-10, and a frozen integer inference profile with digest 1ce6c47f… reproduced bit-exactly across independent implementations MEASURED, 2026-07-10 — and the graveyard: a retired universal verification conjecture, a foresight market killed by its own preregistered gate at 21.7× the fee SIMULATED, 2026-07-10, eighteen of twenty invention seeds killed under a hostile gate, and nine complete prior L1 designs abandoned on the way here. The graveyard is not an appendix. It is the evidence that the spine numbers mean what they say: the same discipline that measured them is the discipline that killed everything that could not be measured. The pentagon itself remains THEORY, and the paper names the gates that will decide it.

Claims

The five axioms of noosphere/README.md, restated as claims with the labels their evidence has actually earned. Labels do not average; a proud sentence and its label travel together.

C1 — Substrate Identity. THEORY; measured edge anchors Work, security, and evidence are one dataflow. The model is the miner. The inference is the block. The proof is the exhaust. Mechanized as conjecture I-PENTAGON (noosphere/09-the-identity.md §1.3): for security, evidence, and learning, marginal cost from a shared committed witness is O(n²) per span against Ω(n³) standalone. The verifier law is measured — scaling exponent n^1.997 on the 32³→64³ two-point probe MEASURED, 2026-07-10, kt-verification-ladder/README.md; the composite claim stays THEORY until E-IDENT-01 runs all five products from literally one witness. No such end-to-end run exists yet; the gate opens when it does.

Same-build economic boundary SIMULATED 2026-07-10. E-DEMAND-WASH-01 kills one interpretation of “security”: useful thought or burned demand does not receive finality weight. The lab priced pure wash into a loss in 34,560/34,560 grid points, but the frozen absolute security inequality failed in 206,175/4,723,920 comparisons; profitable genuine use was the worst family, and the stake-only control also failed 66/729 payoff vectors. Accordingly, product 2 below means a verification/security certificate consumed by a shadow or non-consensus accounting path — not consensus influence or an issuance redirect — unless a future, separately preregistered security object passes its own gate.

C2 — Meaning is law; speed is a market. THEORY; substrate is design-complete [PAPER] Consensus semantics fit in the 12-opcode Grain calculus; every fast implementation is a jet admitted only with a machine-checkable equivalence certificate and slashed on divergence (noosphere/README.md axiom 2; noosphere/06-weft-language.md). The certificate market is THEORY until E-WEFT-01..08 pass; the package explicitly invites builders to be the ones who fail it.

C3 — The mind is a quotient, not a file. THEORY The world model has no canonical weights; it is a proof-connected equivalence class of implementations (noosphere/README.md axiom 3; noosphere/03-living-model.md). You cannot confiscate a quotient. Equivalence-class soundness is gated on E-SPECIES-01; the frozen integer profile (Results, row 8) is the first measured object of the required kind — one semantic identity, two independent implementations, zero divergence MEASURED, 2026-07-10.

C4 — Privacy is a fiber, not a feature. PAPER; component mechanisms THEORY Every owner carries a sealed causal world — encrypted state, proof-carrying concurrency, one realized head — wired into consensus from genesis, never a library on the side (noosphere/README.md axiom 4; noosphere/01-architecture.md Umbra). No Umbra code runs today, and this paper says so plainly; the corpus's cautionary example of the alternative is a privacy library whose only verifier returns true (noosphere/README.md, founding observation).

C5 — The chain must dream. DREAM; first production contract DISPROVED Idle cycles compute sealed counterfactual futures; foresight is a first-class, priced, falsifiable product (noosphere/README.md axiom 5). The axiom survives; its first market did not. The tested activation contract was killed by its own preregistered gate — manipulation profit +542 µT/event against a 25 µT/event forecast fee, 21.7× SIMULATED, 2026-07-10, dream-lane/README.md — and the lane ships disabled (dream_lane_enabled = false). A kill executed on schedule is this axiom's current proudest artifact.

Construction

The witness

The atomic object is the committed span witness of the Canonical INT8 Span Spec v0 (kt-verification-ladder/README.md; quoted in noosphere/09-the-identity.md §1.1): int8 operands A, B, the exact int32 accumulator C32, the requantized output C8, and a transcript root SHA256(A || B || C32 || params) committed before the challenge beacon. Two inherited properties make it load-bearing:

The pentagon: five products from one witness

Per noosphere/09-the-identity.md §1.2:

#ProductFunction of the witnessConsumed by
1Useful thoughtC8, delivered against escrowthe requester
2Security weightchallenge-dependent verifier execution under a post-commit beaconthe Braid — capped ≤ 0.10 Loom, ≤ 20% proofpower, both zero at genesis
3Settlement evidenceFreivalds accept bit + journaled bindings, escalating to the ZK leaf on disputeescrow settlement, slashing, Chorus testimony
4Learning signalforward roots bound into an adjoint-dual receipt; the backward GEMMs share the same leaf calculusthe reaction network
5Foresightsealed branch families rooted at the terminal state commitmentthe dream lane (disabled)
flowchart TD
    W["Committed span witness: A, B, C32, C8 + transcript root, committed pre-beacon"]
    W --> T["THOUGHT: C8 delivered against escrow"]
    W --> S["SECURITY: post-commit verifier execution as proofpower"]
    W --> P["PROOF: Freivalds accept bit, escalating to the 288 B Groth16 receipt on dispute"]
    W --> M["MEMORY: forward roots bound into the adjoint-dual training receipt"]
    W --> F["FORESIGHT: sealed branch families rooted at the terminal state commitment"]
    T --> T2["the requester"]
    S --> S2["the Braid - capped 0.10 Loom / 20% proofpower, both zero at genesis"]
    P --> P2["escrow settlement, slashing, Chorus testimony"]
    M --> M2["the reaction network"]
    F --> F2["the dream lane - ships disabled"]

Figure 1 — The pentagon: one committed span witness (Canonical INT8 Span Spec v0) emitting five products — thought, security, proof, memory, foresight — and the organ that consumes each. The composite five-from-one claim is conjecture I-PENTAGON; the gate opens when E-IDENT-01 runs all five from literally one witness. THEORY; measured edge anchors 2026-07-10.

The invoice, composing only cited rates (noosphere/09-the-identity.md §1.5): a serving-shaped job that emits thought + evidence + security + trust costs roughly 1.015× the thought MODELED, 2026-07-10, plus 51.06 B/token of availability MODELED, 2026-07-10 and a $0.38 contingent dispute liability MODELED, 2026-07-10; a training-shaped job that also emits its learning signal costs roughly 1.06× MODELED, 2026-07-10. Against five standalone products — each Ω(thought) — the four priced extras together cost about six percent. That six percent is the Substrate Identity, stated as an invoice.

Why nobody built this before: in floating point the composition really was five stacked research programs — without a canonical bit-string for "the right answer," the security edge cannot slash, the evidence edge needs tolerance bands that hide cheats, and the trust edge dies at the first cross-vendor mismatch (noosphere/09-the-identity.md §6). The exact integer witness discharges the prerequisite for all five edges at once. The five products were never five technologies; they were five customers of one discipline.

The organism

Eleven organs consume the five products (noosphere/README.md, organ table): the Braid (tensor-work lottery braided with a proofpower-weighted finality ring), Lumen (open UTXO realm), Umbra (sealed causal fibers), Grain (12-opcode total noun calculus), the Species (proof-connected model population), Chorus (phones testify, they don't mine), the Neural Lane (consensus-native token-by-token LLM execution), Weft (proof-carrying surface language), Hearth & Swarm (households as bonded executors), Loam (rights-metabolic memory), and ¤NOOS (duplex issuance). Base security starts boring and honest: the Work Loom holds zero consensus influence at genesis (work_loom_weight_cap = 0, protocol maximum 0.10); the useful-work lane must earn influence through published gates. The gate opens when the E-* ladder says so, not before.

Results

The measured spine

Every row verified against its artifact on 2026-07-10. Denominators travel with numerators.

#QuantityValueLabelArtifact
1Freivalds verifier cost, 32³ span1.8669 c/MAC (61,175 cycles / 32,768 MACs) at 2⁻⁶⁴ soundnessMEASURED, 2026-07-10kt-verification-ladder/README.md (METRIC zk_cycles_per_mac=1.8669)
2Direct in-VM recompute comparator2.7198 c/MAC; verifier 1.457× faster in-VMMEASURED, 2026-07-10kt-verification-ladder/README.md
3Verifier scaling exponent, 32³→64³n^1.997 (two-point probe)MEASURED, 2026-07-10kt-verification-ladder/README.md
4Per-span soundness2⁻⁶⁴ (two 32-bit challenge vectors, exact mod-2⁶⁴ Freivalds); union bound ~2⁻⁶⁰ per layer-chunkPAPER bound on MEASURED harnessnel-dispute-ladder/README.md
5The 1/T dispute law, (K,N)=(256,256)958,090/T + 11,602 cycles/chunk; measured 131,363 / 71,481 / 41,542 at T=8/16/32MEASURED, 2026-07-10nel-dispute-ladder/README.md
6Full-layer dispute vs monolithic recompute, T=12812.67× (440,897,006 vs 5,586,651,389 cycles/chunk; per-token dispute cost 3,444,507 cycles)MEASURED, 2026-07-10 cycle modelnel-dispute-ladder/README.md
7ZK settlement receipt288 B total: 256 B Groth16 seal + 32 B journal claim; 0.0352 B/span at S=8,192MEASURED, 2026-07-10zkvm-port/README.md (dispute_settlement_bytes = 288)
8Frozen integer profile digest1ce6c47fd57f3d338104a09f95b1610a75c6fd865b353a7c843426059e7b81d9 — 24-layer Qwen2.5-0.5B-Instruct q8_0 (model SHA-256 ca59ca7f…), digest-identical across two independently ordered implementationsMEASURED, 2026-07-10nel-real-inference/runs/*.json (second_impl.json: profile_lineage_ok: true)
9Pentagon sum (serving / training-shaped)1.015× / 1.06× the thoughtMODELED, 2026-07-10noosphere/09-the-identity.md §1.5
10Conditional foresight rate (research-only)29.1998522 mB/dream at the smallest pool clearing the ≥75 mB gateSIMULATED, 2026-07-10dream-lane/README.md

The honest boundary of the spine, stated in the same breath: rows 1–7 are deterministic RV32IM trace-cycle measurements, not wall-clock, prover-memory, gas, or GPU claims; row 8 is CPU-only — no GPU integer kernel exists yet, so cross-vendor bit-exactness on real silicon remains the program's Gate 1. The gate opens when certified NVIDIA and AMD kernels reproduce digest 1ce6c47f… bit for bit; until then the vendor-invariance claim is construction, not measurement (noosphere/research/nel-b1-determinism-memo.md).

flowchart LR
    A["Frozen integer profile 1ce6c47f..."] -->|"digest-identical across two implementations"| B["Layer chunk claims"]
    B -->|"958,090/T + 11,602 cycles/chunk"| C["1/T dispute law"]
    C -->|"12.67x vs monolithic recompute at T=128"| D["Freivalds leaf: 1.8669 c/MAC"]
    D -->|"2^-64 soundness per span"| E["288 B Groth16 settlement receipt"]
    E -->|"0.0352 B/span at S=8,192"| F["Escrow settlement and slashing"]

Figure 2 — The measured spine as one dataflow: from the frozen profile digest 1ce6c47f… through chunk claims, the 1/T dispute law (958,090/T + 11,602 cycles/chunk), the 12.67× dispute advantage at T=128, the 1.8669 c/MAC Freivalds leaf at 2⁻⁶⁴ per-span soundness, to the 288 B Groth16 receipt at 0.0352 B/span. Every edge number is a row of the spine table above. MEASURED 2026-07-10.

The graveyard

Every chain has a graveyard. We publish ours with the same precision as the spine, because the kills are what make the surviving numbers believable.

#CasualtyWhat diedThe decisive numberArtifact
1M-HDF universal conjecture — RETIREDThe claim that signed-Hadamard diffusion universally improves verification sampling. Killed by four counterexample classes: characteristic-two collapse (32,767 nonzero 4×4 residuals vanish after a mod-2 transform; a concrete witness accepted 16/16 transformed challenges vs 8/16 standard-basis), pointwise-dominance failure (the all-ones 2×2 error: standard-basis miss 0, post-diffusion miss 3/4; exhaustive enumeration 40 worsened / 8 improved / 32 tied), adaptive-timing spikes, and a quantization-floor attack (any spike with rounding quantum satisfying \δ\< Δ√(mn)/2 erased)Exhaustive enumeration, 10/10 checker gates MEASURED, 2026-07-10. Survivor: M-HDF-ENERGY — unbiased sampled-energy estimator with Var ≤ 8‖E‖⁴_F/s under precommitment, and nothing morenoosphere/02-mathematics.md §3; frontier-lab/math/checker.py
2Dream market v1 production contract — DISPROVEDThe proper-scoring sealed-counterfactual market as an activatable product. Proper scoring cut manipulation profit 3,863 → 542 µT/event, eliminated truthful regret (0.0 µT/event on the preregistered temperature grid), and neutralized wash trading (−55.7 µT/event) — and still failed its own gate: the outcome manipulator nets +542 µT/event against a 25 µT/event fee = 21.7×, while manipulation makes observed Brier look better (ΔBrier = −0.0540)268,000 simulated event instances, 100,000-event main comparison, seed 20260710, exit 0 SIMULATED, 2026-07-10. Lane ships disableddream-lane/README.md; dream-lane/results.json
318 of 20 invention seeds — KILLEDA hostile gate over 201 source artifacts named the closest prior art or hard limit for every seed: 4 straight prior-art kills, plus duplicate-composition, fair-exchange-gap, compiler-not-mechanism, no-comparative-claim, PKI-analogue, database-prior-art, rebrand, observability-impossibility, local/global-fallacy, economics, representation-only, algebra, soundness-gap, and physics/distributed-limit — one each. Zero N4 (category-defining) claims survivedkilled=18, survivors=2, kill_ratio=0.90, final_N4=0 MEASURED, 2026-07-10. Survivors bounded at [SURVIVOR-EXPERIMENT]: S01 genealogical quorum (12 clone aliases collapse to weight 1; 96 permutations identical) and S02 molecular quarantine (0 double payouts across 4,324,320 causal interleavings)international-lab/invention-final.md §5.1–5.2
49 complete prior L1 designs — DEADSee epitaph table below. All superseded before MindChain was designed; MindChain was deliberately not derived from any of them9 designs, ~450-report corpus consumed PAPER, 2026-07-09nockchain-octra-ai-research-handoff/L1-plans/README.md
xychart-beta
    title "The graveyard: kills by class"
    x-axis ["Retired conjecture", "Killed market", "Killed seeds", "Dead prior designs"]
    y-axis "Kills" 0 --> 20
    bar [1, 1, 18, 9]

Figure 3 — Kills by casualty class, from the graveyard table: 1 retired conjecture (M-HDF), 1 killed market (dream v1), 18 killed invention seeds, 9 dead prior L1 designs (the prior-design count carries PAPER in the table). MEASURED 2026-07-10.

The nine epitaphs, one line each:

DesignEpitaph
SOMASelf-Organizing Model Architecture: CapsuleDAG plus "Proof of Resonant Work" — the resonance metaphor never became a slashing rule. L1-plans/SOMA/
NOEMA"The chain that thinks": chain state as cognitive state, consensus as cognition — poetry the spec could not cash; MindChain's own ch. 03 disclaims the same sentence. L1-plans/NOEMA/
AXIOMPolyglot proof-carrying AI L1 betting that intelligence is compositional multi-agent construction — a Rust skeleton and a bet, no witness object. L1-plans/AXIOM/
NOESISProof-carrying sovereign-AI L1 from 172 international mechanism reports — explicitly superseded by ALETHEIA where they conflict. L1-plans/NOESIS/
ALETHEIAThe reasoning-verifying L1 that killed NOESIS's weaker ideas, then was archived by the same divergent-then-convergent process it pioneered. L1-plans/ALETHEIA/
QUIPUThe "living knowledge network" for federated compositional intelligence — revolutionary framing, no priced verification path. L1-plans/QUIPU/
OMNIAThe polycentric, self-evolving L1 composing 75 reports' best mechanisms — a composition without a conservation law. L1-plans/OMNIA/
HYPHA"The chain IS the model": coherence consensus over a distributed MoE — consensus-by-convergent-inference never got an adversarial bound. L1-plans/HYPHA/
TESSERALedger state as a Bitcoin-anchored routed library of expert deltas — its own red team killed the marquee "frontier-model-from-soup" claim; the honest remnant was a library, not a mind. L1-plans/TESSERA/

Read the two tables together. The spine is short because the graveyard is long: eighteen seed kills, nine dead designs, one retired conjecture, and one market executed by its own preregistered gate are the price of ten rows of numbers that survive hostile reading. The program fails only if a favorite theory becomes too important to kill (noosphere/04-experiments-and-falsifiers.md, closing rule).

Falsifiers & kill thresholds

The standing universal kill rules, quoted verbatim from noosphere/04-experiments-and-falsifiers.md §2.1 — these override every mechanism-specific threshold:

Regardless of a mechanism-specific threshold, any of the following kills its consensus influence immediately: - a false proof acceptance in a consensus-reachable path; - a consensus split between conforming independent clients; - unauthorized value creation, spend, decryption, capability use, or counterfactual activation; - a safety failure during a partition within the declared BFT/adversary bound; - inability to disable the mechanism without stopping base Braid/Lumen operation; - evidence that the experiment measured a weaker property than the published claim; - an undisclosed benchmark exclusion, self-dealing demand source, verifier dependency, or shared implementation that invalidates independence. "Kill" means set the relevant influence to zero, reject new uses, preserve historical verification, and execute the named fallback. It does not mean delete evidence or quietly retune parameters.

Standing falsifiers specific to this paper's claims:

Reproduction

Every spine and graveyard number above regenerates from these commands. All Python labs are stdlib-only; fixed seeds throughout.

#Result reproducedcwdCommandExpected
1Rows 1–4: 1.8669 c/MAC, comparator, n^1.997kt-verification-ladderbash autoresearch.sh (or python harness/pouw_analysis.py for the PoUW tables)exit 0; METRIC zk_cycles_per_mac=1.8669
2Rows 5–6: 1/T law, 12.67× at T=128nel-dispute-ladderbash autoresearch.sh (or python harness/bench.py)exit 0, gate_failures=0, ~70 s
3Row 7: 288 B Groth16 settlement. (repo root)bash zkvm-port/autoresearch.sh (requires WSL Ubuntu + rzup)dispute_settlement_bytes = 288
4Row 8: profile digest, dual-implementation equalitynel-real-inferencepython second_impl_gate.pyexit 0; profile_lineage_ok: true in runs/second_impl.json
5Row 10 + graveyard row 2: dream rate and the 21.7× killdream-lanepython dream_lane_sim.py --events 100000 --out results.jsonexit 0 with H8 REJECTED — the simulator must be allowed to falsify the market while passing its soundness gates
6Graveyard row 1: M-HDF retirement, M-HDF-ENERGY survivalfrontier-lab/mathpython checker.pyall 10 theorem/counterexample gates pass
7Graveyard row 3: S01 survivor gateinternational-labpython survivors/lineage_quorum_test.pyexit 0; baseline_weight=12 candidate_weight=1, 96 permutations identical
8Graveyard row 3: S02 survivor gateinternational-labpython survivors/molecular_quarantine_test.pyexit 0; double_payouts=0 across 4,324,320 interleavings

Related work

Competitor dossiers (by path). Nockchain: nockchain-octra-ai-research-handoff/L1-plans/NOCKCHAIN/ — found the substrate (a 12-opcode combinator calculus with jets) but launched with a self-described "useless proof of work puzzle" and an unwired privacy library whose only verifier returns true; its zkVM framework (ePrint 2025/1110) is circuit compilation with no security-vs-usefulness bound. Octra: nockchain-octra-ai-research-handoff/L1-plans/OCTRA/ — found the sealed world but shipped parameters with no security reduction and no validator integration (noosphere/research/01-octra-hfhe-consensus-digest.md). Pearl: research/08-pearl-matmul-pow.md and noosphere/research/02-pearl-cupow-kernel-digest.md — found the engine (proofs of useful work from arbitrary matmul) but its security is an explicit conjecture and independent measurement found the network doing zero useful inference.

Verification-economics prior art (references named in the source dossiers). Proof of Necessary Work (ePrint 2020/190) proves the chain's own state — usefulness ≤ 1 by construction. Ofelimos (ePrint 2021/1379) achieves provable security with usefulness structurally capped at roughly half of computing steps. Gensyn Verde/RepOps pins FP operator order at +58% (A100) to +218% (T4) overhead, NVIDIA-only-evaluated; TOPLOC's locality-sensitive fingerprints carry an acknowledged adaptive-adversary gap and never enter consensus (noosphere/research/03-onchain-llm-execution-digest.md, noosphere/research/05-consumer-swarm-survey.md, noosphere/research/nel-b1-determinism-memo.md). MindChain's position against all of these is the exact-integer-witness move: no tolerance surface, sublinear evidence, and a dispute calculus whose constants are measured rather than asserted.

Program-internal sources. The capstone theorem card and pentagon: noosphere/09-the-identity.md. The honesty ledger and kill rules: noosphere/04-experiments-and-falsifiers.md. The narrative register this paper is written in — radical honesty as the differentiator, the containing dream, numbers policy: noosphere/08-memetics-and-launch.md §3, §5.4. Companion papers in this series: 01 (chunk-Freivalds dispute calculus), 02 (vendor-invariant integer inference), 03 (the price of a nightmare), 04 (genealogical quorum), 05 (molecular quarantine settlement), 06 (ω-invariant difficulty).


MindChain is the only chain whose whitepaper ships its own kill conditions. Every chain has a graveyard. We sell tickets.