The standard
A standard written before the readings
Igneum does not claim that it resists specialist hardware. It states a standard, registered before the measurements, and reports each reading against it with its grade. The standard has a history on the record: version 1 carried a line that no supplier may reach a third of the work, and that line read FAIL against the strongest credible adversary on 10 October 2026. The failure is not rewritten. On the morning of 11 October 2026 the project retired the line, on the auditor's ruling that retiring it "is sound and honest" as "a versioned amendment, not a result under the original preregistered containment gate", because no proof-of-work rule on the record bounds one actor's share by hardware class without identities or a capital price on identities. Version 2 replaces the line with three guards.
The chain is labelled hybrid proof of work and bonded capital from block one.
The three guards
| guard | the condition | the reading on 11 October 2026 | grade | on what | state |
|---|---|---|---|---|---|
| G1, a small edge | the strongest jointly feasible specialist for the launch relation reads at or under 3x the GPU's energy per accepted hash, centrally, at the gate point | 2.59x central, 3.25x governing upper on the registered programme-wide figures; 2.38x and 2.96x at the 5090's 450 W row of record; the published-band upper 3.15x; a simultaneous minus-50-percent stress 3.70x, a sensitivity | provisional | the GPU side measured (board power on NVIDIA, board telemetry on AMD; Apple's rate measured, its energy estimated until the wall); the specialist side a model, "modelled pending an external placed design" | provisional pass; AMD family qualification passed on the RX 9070 XT at telemetry grade and wall confirmation remains; full qualification awaits Apple wall power and production repeats |
| G2, a long clock | the modelled time for the only chip that runs the relation to reach a third of the work is at least twice the entry's length | 30.32 months at the all-corners attacker-favourable edge against the 12-month entry; the band across the lead and energy corners 25 to 46 months; the exposure from the earliest actionable disclosure 13 to 14 months, about seventeen months inside the clock | model | the supply model on a GPU-class die: 35 to 150 million USD of non-recurring engineering, 24 to 36 months, no incumbent; a hypothesis on every chip term | provisional pass, modelled; re-read on every landing of the band |
| G3, control bounded, not share | a rule under which proof of work determines share-lane rewards, bonded draw units select signers where units are occupied, and the open-seat proof-of-work rule supplies liveness while units remain empty, is consensus text with frozen parameters, live from block one, and security-closed on its three closure rows before genesis | Prototype mechanics agree on two node pairs; the three security closure rows remain open. The mechanics: every keyed-proof refusal, the draw and the fork weight, the payout list paid exactly, liveness with every bonded sealer dark, key rotation under the bond, two independent builds agreeing, 200 invalid keyed blocks refused at 13.6 ms | in closure | a prototype: "attribution and payout mechanics proven, not containment or non-outsourceability" | research evidence today; it counts as launch containment only once its closure rows read prevented and the auditor verifies them, and there is no launch before that |
The complete three-item standard is not provisionally met. In the auditor's words: "Guards (a) and (b) provisionally pass; succession and warm transitions pass; guard (c) remains research evidence." The third guard is the launch's gating item; it has its own page, Work authority.
What each guard shows and does not show
G1 shows that the strongest jointly feasible specialist reconstructed in the current model reads 2.59x centrally and 3.25x at the governing upper, with a simultaneous 50 percent-favourable sensitivity of 3.70x beside it. That is comparable with late-Ethereum coexistence ratios (a chip at 4.3x the then-current card held ten months with no rule change; the 2.5 to 5.5x band, Reported); that precedent does not guarantee Igneum's outcome and is context only. It does not show that no such chip is built, nor that a built chip reads what the model says: the specialist side is modelled until an externally placed design exists, and the figure is a property of the relation, read again at every successor.
G2 shows that the earliest such chip, even at every corner in its favour, takes about two and a half years to reach a third of the work, more than two entries of the 12-month menu, so the succession machinery has two ratifications in hand before the clock runs out. It does not show what the chip does after that, nor that succession resets its clock (a programmable die runs every in-menu successor; the silicon-reset question is unresolved as ruled), nor that the lead is as long as the model says: the lead is the one term that moves the clock, and it is Hypothesis.
G3 shows, today, that a rule exists, frozen and prototyped, under which an unbonded share worker cannot alter the signed template without invalidating its work and the selected bonded key supplies the block seal: what a block contains, where its reward goes and which parent it extends rest with the bonded key that signs the template, and a bond is per key and cannot be transferred on the chain. It does not yet show that this holds against an outsourcing contract, nor that a key measures anything but a payout key, nor that a pool cannot mediate control, nor that hash share is bounded (it is not: the hasher with the least cost wins the hashing market). Frozen parameters alone do not establish that the mechanism works; the closure rows do or do not.
Proven, provisional, research
| item | grade | state | where it is read |
|---|---|---|---|
| exact implementations of the relation on NVIDIA, AMD and both tested Apple compiler classes, one shared source text, frozen vectors for every gate point | measured | proven: exactness on every class, no failed recompute | the vector sets and the per-vendor exactness rows |
| the consensus round trip from the launch document to the IgneumHash version 8 form and back, warm, on two nodes, with the CPU reference, no hand, no rejected or invalid work | final | on the tested pair (the auditor, gate 8) | Succession |
| the same round trip with a CUDA worker on a rented 5090: warm both ways, zero boundary downtime, zero invalid, with a recorded 12 percent preparation cost over the 32-block lead | final | on the tested worker and node pair (gate 10); the cold pair beside it as the bounded fallback | Succession |
| operational succession: seven height-activated transitions on 10 October 2026, both nodes agreeing, zero refused, the ratification margin exercised with no manual intervention | passed | under the auditor's wording; silicon-reset succession unresolved (the machinery is proven; that a transition obsoletes an incumbent specialist is not) | Succession |
| G1, the small edge | provisional | provisional pass, the specialist side a model | this page |
| G2, the long clock | model | provisional pass, modelled | this page |
| AMD inside the vendor-relative loss | passed | family qualification passed on the RX 9070 XT at telemetry grade (13.92 uJ per hash across two exact windows, 1.14x the 5090 pod), the RX 9060 XT beside it as corroboration; wall confirmation remains; the improvement over the original kernel is a corrected implementation cost, not a changed relation | IgneumHash |
| Apple | estimated | rate measured, energy estimated; the mini a provisional product pass at 2 GiB after thirty leak-free transitions; the M5 Max open | IgneumHash |
| the RTX 2060 on IgneumHash | pending | an unresolved cohort row, kept visible; not a passed card and not a launch gate | IgneumHash |
| the production worker at the bench rate in repeated windows | provisional | the parity row stored on the packaged workers under the corrected knee configuration (the old default tune superseded); stable repeats retained as a pre-release requirement | Ember |
| G3, work authority: the mechanics | prototype | on two node pairs | Work authority |
| G3, work authority: outsourcing resistance, Sybil attribution, pool-mediated control | in closure | the three rows ordered, none run; the launch gates | Work authority |
| the entry length (12 months), the card floor (2 GiB the working default) | model | ratified decisions; the entry carried in models only, nothing relocked | the decision register |
What the standard does not claim
Not shown, and not claimed on any page of this site: that any supplier stays under a third (version 1's failure stands on the record); that the specialist's share is bounded by anything but the ratio and the revenue; that work authority contains anyone, resists an outsourcing contract, attributes suppliers, or survives a pool that holds keys (the closure rows, unproved); that the early-concentration guard holds against a party behind many keys; that succession resets a programmable die's clock; that custody is prevented; that the chip's figures are physical (an external placed design is the line between modelled and demonstrated); that Ethereum's coexistence record proves IgneumHash will coexist (the precedent is context only).
The auditor's wording for the amended standard, held
The permitted wording of the ruling on the amendment, with the internal code name replaced by the public name; written before the route decision, it stays the held text until the auditor's wording for that route replaces it: "IgneumHash provisionally meets Igneum's amended specialist-coexistence standard: the strongest jointly feasible reconstruction reads 2.59x centrally, 2.38x at the efficient 5090 point, and the modelled earliest time to one-third share is approximately 30 months against a 12-month entry. Automatic succession has passed live consensus and GPU-worker round trips. The separate work-authority mechanism has passed its functional prototype but has not yet proved resistance to outsourcing, Sybil attribution or pool-mediated control; it therefore remains a preimplemented upgrade candidate and is not counted as launch containment. Full qualification still awaits the RX 9070 XT, Apple wall power and production repeats."
Under the route the project chose, the sentence about the mechanism is read as: it is the genesis rule, not counted as launch containment until its closure rows pass, and there is no launch until they do.
Where the standard is written
The launch text (the amendment, the guards, the frozen rule, the launch gates), the plan's gate table (version 1 beside version 2), the launch objective (the two verdicts and the class ruling), and the auditor's rulings. Each is linked from this page the day its public URL is named; the rows you can read without this site are listed on Verify it yourself.