Every criticism, answered
or conceded.

The line is "A GPU-secured network for Ethereum-compatible applications and verifiable computation." Igneum is a sovereign GPU-mined network that runs Ethereum-compatible applications and proves their execution; the proof architecture is explained beneath that line, with its three boundaries (B1 to B3), and never leads.

The target is contestable mining, not chip destruction: a manufacturer with a profitable product is not the failure; an exclusive, durable advantage large enough to displace the accessible GPU fleet is. Two measured results already bound the design. The connected-state reorganisation was killed as a class because rearranging the same operations moves neither side (D2). The mixed FP32 branch was killed because deterministic FP32 costs the cards 15 to 26 percent energy per hash against a 10 percent budget and widens the chip's edge to 3.0x to 3.2x (measured cards, modelled chip). Both stay as regression controls.

The cohort is the discrete-GPU population, used cards included. Each class gets two tests: the existing owner (power, wear, fees, alternative use) and the new entrant (purchase, operating cost, resale). The central measure is cost per accepted unit of work: annualised hardware, power, hosting, failures and fees over annual accepted work. Nothing here is measured yet as a population.

The model replaces the capex wall. It covers growing and shrinking networks, reduced issuance, cheap and dear electricity, replacement and resale on both sides, changing proving demand and miners who react. Its mandatory stress case assumes the opponent's development is already paid for.

The four properties must hold with no rescue upgrade assumed. The no-rescue network exercise (D5) is the test: epoch boundaries crossed, signing interrupted, the network partitioned, major operators removed, hostile proof submissions, independently written clients, with specified behaviour and no emergency algorithm change or privileged intervention. That exercise is Open.

One exact generator, verifier, dataset policy, compiler configuration and measurement harness, pinned by digest and served. Mining and proving are measured together on the final configuration, with wall power beside device telemetry, accepted and rejected work, compile time, memory use and sustained thermals. Nobody outside the project has reproduced anything yet.

The critic is right for D2(a). With the same operations reorganised so that live state feeds every load address, the 64-register window is necessary (63 of 64 registers live at every address, measured), but a chip answers with a clock-gated register file that pays per write, so only the window's width reaches it: +1.2 pJ per lane-op at N5 (synthesised), moving the chip's edge 1.10x node for node against a 1.25x gate (modelled). The generator variant and the liveness tool stay behind a flag. D2(b) prices memory sharing, recomputation and data-local execution against class v6; pass when a candidate improves against re-optimised adversaries across the declared population within the preset cost and verification limits, otherwise class v6 stands and the experiment is published as a failure.

The adversary is allowed to survive: whole-system cost minimised across every published family, free to change lane count, register implementation, instruction storage, memory technology, scheduling and support hardware, placed rather than synthesised, with a three-year stress life. Next-generation opponents are in the matrix. Pass when its best supported cost and energy advantage sits inside the chosen competitiveness envelope with uncertainty published; 1.5x energy is a research goal, not the pass criterion. That is Open.

The capex wall is replaced. Tariff advantage is shown apart from hardware advantage, and the economics page resolves the tip-share discrepancy with an explicit user-funded proving payment and congestion pricing, burn treated separately, the hard cap and no development tax kept. A profit-maximising operator simulation (mine, prove internally, prove externally, switch off) must show pricing and capacity rules restoring service without an administrator.

Every producer verifies proofs off the consensus path today, so the network demonstrates proving activity rather than enforcing a permissionless proving economy. The rule and its activation switch exist in the node fork and are off on the devnet. Enforcement in consensus is the prerequisite of the no-rescue exercise, and node 2.0.0 follows it.

Vote keys stay with the miner at protocol level; the pool aggregates payment, not votes. Not built yet.

P2Pool is the precedent, not the code. Neither the payout design nor the latency measurement exists yet.

Agreed that this is the bar. The within-1-percent figure is Firo's own report (claimed, not measured here). Ember is measured against it, not against whitepapers.

An L2 would suit a project whose objective is Ethereum settlement. Igneum's objective is an independent network secured by accessible hardware. Sovereignty has a cost: the network must establish its own consensus security, data availability and credible cross-chain verification, and execution proofs do not remove those duties.

The interface stays replaceable; what the chain accepts does not. Pinning in the protocol is designed and not yet shown on the devnet.

Igneum's mining design accepts that specialised hardware may be built; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. The window costs a GPU under 1 percent of rate at stock (measured on rented RTX 5090 and RTX 4090 cards, 8 October 2026). The long-program and select-tree proposals stay out as regression controls.

Winning the mining lottery and producing proofs stay separate, and they must remain separable during a failure, not only on paper.

The proving stack is judged on the full pipeline: inputs, proving, aggregation, verification, payment, memory footprint and mining income forgone. A fast shard result is not enough when aggregation or memory pressure makes ordinary operators uncompetitive.

The differences are known and are listed; the test suite that makes compatibility a deliverable does not exist yet.

The ladder: prove Igneum's own execution; then one external customer's exact workload with repeat paid jobs; then further workloads only where the fleet has a demonstrated edge. Pilots and announcements do not count; customers repeatedly paying for proofs at prices that carry reliable service and an operator margin do.

Proven execution is not finality. A proof shows a block's execution was computed correctly; finality is a property of the chain's own consensus and checkpoints.

EVM compatibility is not Ethereum security. Igneum is a sovereign network: its security is its own GPU-mined consensus, not Ethereum's validators.

ZK is not privacy. The proofs here verify execution; transaction data is public, and privacy would need additional application or protocol design.

Restating the goal is not enough. Igneum is measured against the operating system, not its whitepaper.

Ergo is a reference point, not a whitepaper competitor; the comparison is owed.

The ceiling of the served ranking language is serious contention for the top of the GPU-mining space on engineering and operator proposition. The tests that would earn more are miners staying through hard conditions, customers repeatedly paying for proofs, and the network running without the founding team (D5); none is met yet.