Competitive audit · 2026-07-10
Four SUPERIOR. Two PARITY.
Four BEHIND.
An audit that only finds wins is marketing. Nockchain's website runs ahead of its own
careful whitepaper. Octra's marketing is several steps ahead of its runtime. Pearl runs a live useful-work
chain that independent measurement caught doing no useful work. We know because we built their nodes,
ran their PoCs, and read their kernels line by line — ~720,000 words of reverse-engineering dossiers —
and then published the rows where they beat us at full size. That is the brand.
Source: noosphere/10-competitive-audit.md (every cell carries an evidence tag + source path).
Scorecard — one verdict per dimension
| Dimension | Verdict | Why (one sentence, evidence-tagged) |
|---|---|---|
| Consensus | BEHIND | Nockchain runs a live STARK-ZKPoW mainnet (~97k blocks [MEASURED]) and Pearl a live PoUW L1; the Braid is an implementable design with no running network. |
| Useful work | PARITY | Pearl alone ships a live useful-work chain — with conjectural security and a network independently measured doing zero useful inference; MindChain's design plus measured verification primitive fixes exactly that failure mode, without a deployment to prove it. |
| Verification / dispute | SUPERIOR | The only entrant with a measured end-to-end dispute calculus (1.8669 c/MAC, 2⁻⁶⁴/span, 288 B Groth16 receipt, all [MEASURED] 2026-07-10) adding no new cryptographic assumption; Octra's headline NIZK is dead-ended, Pearl's soundness is an explicit conjecture. |
| Execution / VM | BEHIND | NockVM executes a live mainnet and Octra's Circles/OCTB run real contract examples; Grain is a specification plus a lab-scale evaluator. |
| Language | SUPERIOR | Weft is the only language in this field carrying totality, certified cost polynomials, jet obligations, and numeric profiles as types — with a lab compiler passing byte-identity and meter gates; vs Hoon's reproducible toolchain panics, Octra's Solidity-shaped AML, Pearl's nothing. |
| Privacy | BEHIND | Octra ships far more running privacy code (live Bulletproofs confidential transfers, an exercised HFHE PoC); Umbra deliberately ships dark until its falsifiers pass. |
| AI execution protocol | SUPERIOR | NEL is the only consensus-native token-by-token inference protocol with typed finality classes and a measured dispute leaf; Pearl proves matmuls, Octra demos a linear toy, Nockchain's market sits behind two unshipped forks. |
| Data availability | PARITY | MindChain is the only design pricing AI-evidence availability (availability-before-challenge, N-ACT-DA) but ships no storage; Octra ships production-grade chain storage under an unimplemented DSN claim; nobody fields a working AI-evidence DA layer. |
| Economics | BEHIND | Nockchain, Octra, and Pearl all operate live token economies with code-verified issuance; ¤NOOS is a design with modeled lane economics and partially measured proving costs. |
| Evidence discipline | SUPERIOR | The only project that preregisters kill thresholds and has executed them against itself (M-HDF retired, dream market killed); Octra shipped false infrastructure claims, Nockchain's fair-launch framing is contested, Pearl still headlines first-party unreproduced numbers. |
Where MindChain wins
Freivalds + SHA-256 + a standard proof system — no novel hardness conjecture anywhere in the trust chain. 1.8669 c/MAC at the leaf, 2⁻⁶⁴/span, the 1/T law, one real 288 B Groth16 receipt. MEASURED 2026-07-10
The base chain survives a total AI blackout by construction — Ground blocks, transfers,
and checkpoint finality continue when every AI service vanishes. A release test, not a narrative
promise. Pearl's security is its work relation; Octra's liveness rests on a governance
authority set. THEORY, gated by E-BLACKOUT-01
No competitor has ever retired one of its own headline mechanisms in public. M-HDF: retired by exhaustive counterexample. Dream market: killed at 21.7× fee. Both preregistered, both published, neither tuned away. The Scar Museum.
Where MindChain is behind — and the closing step
| Gap | Evidence | Closing step |
|---|---|---|
| No production VM / node | NockVM runs a live mainnet [MEASURED]; Grain is spec + the lab-scale weft-v0 evaluator (weft_all_gates_pass=1 [MEASURED] 2026-07-10). |
No fix inside this program — recorded as a build-phase item. Weft v0.2's dual-VM and tightness gates narrow semantic risk at lab scale only; they are not a production VM. |
| Less running privacy code than Octra | Octra: live Bulletproofs transfers, exercised HFHE PoC, stealth notes [MEASURED]. MindChain: zero running privacy code. | Recorded; Umbra ships dark by policy, not omission — umbra_suite_enabled = false until E-UMBRA-01/E-KEYS-01 pass; the inversion is "seal state, never source." |
| No live proving economics | Competitors operate live token economies [MEASURED]; ¤NOOS is unlaunched. | Partial artifact cited as such: zkvm-port measures the supply side of proving (prover wall-clock, RSS, 288 B settlement, 0.0352 B/span [MEASURED]); no market yet prices the demand side. Live economics remain post-launch. |
Imports adopted — credit where due
| Competitor's best trick | MindChain mechanism |
|---|---|
| Nockchain's genesis ritual (Bitcoin-anchored, undisputed even by critics) | The First Thought — genesis ceremony hashing the honesty-ledger root, conformance vectors, and all-off experimental controls into block zero (ch08 §4.4). |
| Pearl's artifact discipline (papers, tagged releases, public benchmarks) | The papers program — seven papers compiled only from lab artifacts, with the inversion applied: we publish our own negative results first (Papers). |
| Octra's inversion: seal state, never source | Umbra policy — user worlds encrypted; code, parameters, and node public before any value moves (ch01 §8). |
| FAI/Freysa's action firewall | ch03 §8.5 — model output is an untrusted proposal behind a deterministic typed-intent firewall [BUILDABLE]. |
| Virtuals' participation-as-product (16,721 launches) | ch08 §4.6 — community roles mapped to protocol organs: a human-shaped door, a day-one action, an honest earnings statement per organ. |