igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

Mission lane 2, unfinished: proof of useful work (every scheme that passed cheap-verify did so by making the work useless; Filecoin at 29 to 36 percent utilisation), the stalled ASIC-resistance programmes, merged mining, finality on proof of work (every one that held was a second set bought with coins), decentralised pools (only Monero's P2Pool at 7.7 percent; a share sidechain about 40 hours on Igneum), the one-click miner apps against Ember, the three-way verdicts

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-josh 2026-10-07 09:18:20 +01:00
parent 2121b33471
commit d5c94fa89f
1 changed files with 371 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

371
docs/analysis/mission/unfinished.md Normal file
View file

@ -0,0 +1,371 @@
# The last mission, lane 2: started and not finished, across the industry
7 October 2026. Lane 2 of the final research programme. What the industry started and did not finish, group by group, with one verdict per row: Igneum already has the finished version, Igneum could finish it in hours, or Igneum should leave it. Every figure from a source carries its URL and access date in section 9; every figure from memory is labelled approximate. Hours are agent hours (Josh's rule: Claude-side work takes hours, never weeks). Nothing here is a token sale.
What this file does not repeat: the ASIC chip history (`docs/analysis/asic-resistance-history.md`), the useful-work verdict and the stored-state candidate (`new-pow.md` sections 3, 6, 7), the ranked frontier ideas (`frontier.md` section 0, 3.10, 3.11), and Igneum's own pool design (`docs/plans/pool.md`, `docs/spec/09-pool-protocol.md`). Those are cited, not restated.
## 0. Progress (UK time)
| Time | State |
|---|---|
| 08:1x (approximate) | Context read: CLAUDE.md, asic-resistance-history.md, new-pow.md 3, 6, 7, frontier.md 0, 3.10, 3.11, horizon-2026-10.md 1, pool.md, spec 09, polish.md 3.1, 3.6, 3.10 |
| 08:2x (approximate) | Four research lanes launched (A, B, C with D, E). Lane F was refused by the concurrency cap and was researched by this lane directly |
| 08:3x (approximate) | The session's WebSearch budget ran out (200 of 200). Every later fact came from WebFetch of a primary page or is labelled approximate |
| 08:49 | Lane F notes saved (scratch `mission-unfinished/F-notes.md`, file time) |
| 08:50 | Igneum-side pool and Ember comparison drafted (scratch `igneum-side-draft.md`) |
| 08:53 to 08:59 | Lanes A, E, B, C with D landed (scratch file times) |
| 09:09 | All notes read; assembly of this file started |
| 09:17 | File complete: 371 lines, em-dash grep at zero, nothing staged |
Limits of this research: p2pool.observer answered 502 all morning (Monero P2Pool share taken from the miningpoolstats data file instead); Bob Rao's ProgPoW PDF would not extract (his 1.1x to 1.2x figure is cited to EIP-1057); Bitmain's own X9 page was unreachable (the X9 is reported as announced per resellers, unconfirmed shipping); NiceHash's fee and repayment pages are gone (those figures are approximate); Salad's take rate is not published anywhere fetched.
## 1. Group A: proof of useful work
The one question per row: can the useful work be verified cheaper than it is done, and is it sampleable (a nonce picks a random instance the miner cannot steer, with tunable hardness)? The third column the rows keep answering by themselves is whether anyone wanted the output.
### 1.1 The rows
| # | Programme | Attempted, by whom, when | Achieved (numbers) | Left undone | Why it stopped | Verified cheaper than done | Sampleable |
|---|---|---|---|---|---|---|---|
| 1 | Primecoin | Sunny King, 7 Jul 2013: header hash is the origin of a Cunningham or bi-twin prime chain | 13-prime 94-digit chain 10 Sep 2013; first length-15 twin chain 30 Mar 2018; height 7,038,822 today | Any use of the primes; GPU resistance (GPU miners by 2014, approximate) | Nobody needed the output; price USD 0.034, "stopped trading on all exchanges"; last release 26 Jan 2021 | Yes, but polynomial: 10 to 15 modexps to verify against millions of sieve steps (approximate) | Yes: the hash fixes the origin, chain length plus fractional remainder tunes hardness |
| 2 | Gridcoin | Rob Halford, "Proof of Research" live 11 Oct 2014 on BOINC credit | 17 whitelisted projects, 3 greylisted; release 5.5.0.0 on 3 Apr 2026 (78 PRs, 11 contributors) | A verifiable work primitive; consensus moved to pure PoS 2014 to 2015 | Scraper nodes read credit from project servers and sign a superblock; the chain never checks a work unit | Only by trust (an oracle reads a website) | No: the miner picks the project and the unit; the project sets hardness |
| 3 | Curecoin, FoldingCoin | 2014, Folding@home points paid daily (CURE) or monthly (FLDC on Counterparty) | 1 trillion F@h points by 28 Jul 2019, over 10,000 team members | Verification by anyone but Stanford's servers; Curecoin 2.0 roadmap undated; FoldingCoin asset archived, site gone | Tip jar on a centralised stats feed; token value fell under the electricity (approximate) | Only by trusting F@h | No: F@h assigns the work |
| 4 | Aleo PoSW to synthesis puzzle | Proof of Necessary Work (Kattis, Bonneau, 2020/190); testnet 3 coinbase puzzle 2023; mainnet 17 Sep 2024 on AleoBFT with the puzzle outside consensus; ARC-41 synthesis puzzle | 750 M proofs per second and over 44,000 provers on testnet 3 (Messari excerpt); 2/3 of coinbase to provers; about 16 pools took close to 90 percent of puzzle rewards; ARC-46 (31 Jul 2025) demands 100,000 ALEO staked per solution, 2,500,000 by 2027 | The original promise: energy that produces proofs of real state. The puzzle proves nothing the network uses | A BFT validator set gives finality; the puzzle survives only to pay hardware and keep a prover fleet warm; proving centralised on the same path as mining | Yes (SNARK verify in milliseconds) | Yes (epoch hash plus nonce, target tunable). Useful: no |
| 5 | Ofelimos | Fitzi, Kiayias, Panagiotakos, Russell, eprint 2021/1379, CRYPTO 2022: doubly parallel local search, every step a hashed PoW unit | A security proof in the backbone model; WalkSAT variant "competitive with vanilla WalkSAT" | Any deployment; who submits problems and pays; the search's own efficiency (FRLS follow-up 14 Nov 2025 calls DPLS "particularly wasteful") | Economics and operations, not the proof | Yes (each step is a cheap check) | Partly: the step is nonce-bound, the instance comes from a client pool, so hardness is tuned on the hash part |
| 6 | PoW from worst-case assumptions | Ball, Rosen, Sabin, Vasudevan, eprint 2017/203 and 2018/559: Orthogonal Vectors, 3SUM, APSP with a delegation verifier | The conditions: worst-case to average-case reduction, non-amortisability, fast verification | Any implementation or benchmark; a buyer for random OV instances | The instances must be drawn from a distribution no real problem has; polynomial blow-ups | Yes (near-linear verify against quadratic solve) | Yes by construction. Useful: no in practice |
| 7 | Filecoin PoRep and PoSt | Protocol Labs, mainnet Oct 2020: seal a 32 GiB sector (hours of CPU plus 20 to 30 min GPU SNARK, 128 to 256 GiB RAM), then daily WindowPoSt | Utilisation 29 percent (Q1 2025), 36 percent (Q3 2025), 1,110 PiB of real data on 3.0 EiB committed; raw byte power 3.01 EiB (29 Sep 2025) to 1.35 EiB (14 Sep 2026), down 55 percent; 99.5 percent of Q3 2025 fees were penalties | Paid demand; two thirds of proved bytes were zero-filled capacity; the chain now pivots to PDP hot storage (May 2025) that earns no consensus power | Sealing cost was paid by emission, not customers; when rewards fell providers left | Yes (one seal, then a sub-second daily proof) | Yes (random leaf challenges per deadline). The only row that held both for a real good, at hours per 32 GiB |
| 8 | Chia proof of space | Chia Network, mainnet May 2021: plots of random tables, challenge per signage point | 1.5 EiB to 15 EiB in May 2021 alone, peak 36.5 EiB Jul 2021; a k32 plot needs 1.3 TiB of SSD writes; pooling Jul 2021; compression Jan 2023; under 4 EiB effective and under 3 EiB raw in Sep 2026 (about 10 percent of peak); XCH USD 1.36 against a USD 1,934.51 peak; Chia 3.0 (27 Feb 2026) resets plots to 1 GiB k28 with a 256-day phase-out | Anything stored; the space holds nothing | Price down 99.9 percent, reward per TB collapsed, no use for the plots | Yes (a few hashes) | Yes (challenge, plot filter, k). Useful: no |
| 9 | Flux, Akash | Compute markets beside a coin: Flux (2018 as ZelCash, 7,015 nodes on 5 Oct 2026 against a 14,000 peak, no published revenue in six years, "PoUW v2" Oct 2025 = hardware benchmarking); Akash (Q1 2026 lease revenue USD 253,250, 58 providers, 84 GPUs in use of 334) | Node counts, not verified work | A chain that checks the compute; InFlux Technologies entered administration 4 Sep 2026 | Block rewards pay for idle capacity; demand is two to three orders under the subsidy | No (reputation and benchmarks) | No (customers pick the job) |
| 10 | Dynex | 2022 to 2026, "neuromorphic quantum computing cloud", DynexSolve | A WIPO patent (14 Nov 2024); market cap USD 119,386, down 99.9 percent; a "phase-out of the utility token" toward a securities structure | A checkable claim: no public code shows the chain checking that a QUBO instance was solved | The useful-work claim was never made verifiable | Unknown, treat as not verified | Not sampleable |
| 11 | Qubic | 2024 to 2025 "useful PoW" (Aigarth training); from Jun 2025 the work was Monero RandomX hashing sold for QUBIC buybacks | Under 2 percent of Monero in May 2025, over 25 percent by late Jul, 51 percent claimed 11 to 12 Aug 2025; 63 of 122 blocks (3,475,729 to 3,475,850); 6-block reorg 12 Aug (about 60 orphans); 18-block reorg 15 Sep 2025 (117 transactions, past Monero's 10-block lock) | Any verification of the AI claim; no Monero consensus change by Sep 2025; miners moved to P2Pool, exchanges raised confirmations | The "useful" work was a token buyback funded by attacking another chain | The RandomX part: yes, and useless by design. The AI part: no | The AI part: no |
| 12 | Bittensor subnets | 2023 onward, validators score miners off-chain, Yuma consensus; dTAO 13 Feb 2025 | Over 120 subnets (Mar 2026); 24 with commercial income, USD 28 M to 35 M ARR estimated (Sep 2026); validator reward correlates with stake at 0.80 to 0.95, miner reward with performance at 0.10 to 0.30 (64 subnets, 6.66 M events); weight copying fixed by commit-reveal (May 2024); Quasar (SN24) collapsed Aug 2026 on a plagiarised model | Proof of anything; usefulness is judged, not proved | By design: a stake-weighted reputation market with emission attached | No (judged) | Partly (validators pick queries, no nonce, no hardness knob) |
| 13 | Proof of inference, 2024 to 2026 | Gensyn Verde (refereed delegation, about 60 percent overhead on Llama-3.1-1B, needs one honest provider); Hyperbolic proof of sampling (re-execute a fraction, slash); Ambient proof of logits (claimed 0.1 percent overhead, testnet still pending Apr 2026); Inference Labs (self-reported 950 M proofs); zkML (hundreds of seconds per query for billion-parameter models); Boundless PoVW (pays ZKC per proved cycle, a sample proof 2 to 5 min at USD 0.04 to 0.17); Succinct (over 6 M proofs in 2025, 25 provers) | Proving networks exist and are run by a few dozen operators | A job that is both random and wanted; a verifier cheaper than a referee, a chip or a slashing game | zk proving costs 10^3 to 10^4 times the inference; the cheaper schemes trust something | zk: yes at 10^3 to 10^4 overhead. The rest: only by trust | zk: yes if nonce-bound. Nobody has made the job random and wanted |
The 2025 to 2026 literature says the same thing from three sides. The SoK (eprint 2025/1814, over 50 constructions): "PoUW is actually not as useful as expected, since the economic and societal utility do not contribute to the security budget". Pass (arXiv 2606.06700): a majority attack still costs the block reward. Basu (arXiv 2606.04819): Pearl's 24 EH/s network of about 320,000 GPU-equivalents at about 112 MW "produces zero useful AI computation".
### 1.2 The pattern, and what Igneum finishes
Every programme that passed both tests (cheap verify, sampleable with tunable hardness) did so by making the work useless: Primecoin chains, Chia tables, Aleo's synthesis puzzle, Filecoin's capacity sectors, Boundless cycles. Every programme that kept the work useful gave up verification: Gridcoin, Curecoin, Bittensor, Flux, Akash, Qubic, Dynex. Filecoin alone held both for a real good and paid with hours of sealing per 32 GiB and a network two thirds filler.
**What Igneum already finishes (two rows).**
| Row | The unfinished promise | Igneum's finished form | Evidence in the repo |
|---|---|---|---|
| 4, Aleo PoSW and Proof of Necessary Work | Energy that produces succinct proofs of real state, paid from issuance | Every block is proven by miners in chunks and aggregated 20 to 60 s behind the tip; the provers are paid 20 percent of emission by sortition by weight; the proof is of the chain's own execution, not a synthetic circuit | CLAUDE.md design paragraph; lane 4's finding 1: the proof verified in consensus is the fix for the one line where a majority earned more than it spent (11,636 IGN an hour at 51 percent of blocks); switch `proving_consensus_verify_daa` ships off by default in 0.3.16, activation a decision owed |
| 13, Boundless PoVW and Succinct | Pay per verified unit of proving, meter the cycles, let a market set the price | pgas cycle metering through the proving share; external jobs 90/10 provers and burn, priced in dollars, settled in the token | frontier.md 3.10 "PoVW-like cycle metering through pgas"; CLAUDE.md fee lines |
**Why the split is the finished form.** The bound from `new-pow.md` 3.1 is the reason nobody closed the loop: the useful fraction of a proving-as-lottery scheme is (proving work per segment) over (network hashes per segment), 8 percent at 1 GH/s and 0.08 percent at 100 GH/s, because gas sets one and the security budget sets the other. Aleo found the same wall and decoupled on 17 Sep 2024 ("provers do not participate in consensus or produce blocks"), then had to add a stake gate in 2025 because the puzzle centralised on the fastest prover. Igneum's split carries no stake and no coupling: the lottery is a Poisson process on a random program (the Kaspa developer's objection in frontier 3.10 stays satisfied), the proving is paid by sortition by weight, and the one line where the split was NOT yet finished (a majority prover pool earning more than it spent) is closed by verifying the proof in consensus. Until that switch is on, Igneum's split is Aleo's 2024 form; with it on, it is the form Aleo could not reach without a validator set.
**What Igneum should leave.** Rows 1 to 3, 5, 6, 9 to 12: 0 hours. The only adjacent idea with value was taken by lane 8 as scheme C (the dataset is the keyed execution state, the class v5 candidate), which answers rows 7 and 8 by making the stored thing the chain itself, at an unchanged hash rate; it is already judged and not repeated here.
## 2. Group B: memory-hard and ASIC-resistance programmes that stalled
The chip history is done in `asic-resistance-history.md`. This section covers only what was started and abandoned, and ends each row with what Igneum's hourly random program, latency shadow and class ladder already encode, and what they do not.
### 2.1 The rows
| # | Programme | What was planned | What shipped | What was left on the table | Why it stopped |
|---|---|---|---|---|---|
| 1 | Ethash successors | EIP-969 (David Stanfill, 3 Apr 2018): five FNV primes of Hamming weight 7 to 8 against the Antminer E3, fork at block 5,550,000 with 30 days' notice; ProgPoW as "Ethash 2" | Nothing on Ethereum. ETC's Etchash (ECIP-1099, Thanos, block 11,700,000, Dec 2020) halved DAG growth to keep 3 to 6 GB cards | The FNV tweak (Stagnant), the DATASET_PARENTS increase Least Authority asked for, any epoch recalibration on mainnet | No client team adopted EIP-969 for a chip expected to be short-lived; attention moved to ProgPoW; PoW ended at the Merge 15 Sep 2022 |
| 2 | ProgPoW, EIP-1057 | Hold the chip gain to 1.1x to 1.2x by saturating the whole GPU datapath (random math, 16 KB cache, DAG loads, keccak-f800) | Two audits (Least Authority 9 Sep 2019: accurate to design, five suggestions, the light-evaluation attack named; Bob Rao: 100 MB on-die SRAM "possible", DRAM 3 pJ per bit against 0.3 on chip); KawPow (6 May 2020), FiroPoW (26 Oct 2021), Zano, Sero, Quai (29 Jan 2025) | Activation on Ethereum: tentatively accepted 4 Jan 2019, "accepted and final" 21 Feb 2020, petition EIP-2538 27 Feb, kik's 64-bit-seed exploit about 4 Mar, off the fork schedule 6 Mar 2020; authors reposted it 11 May 2020 saying "we do not recommend" deployment | Governance (a revival read as stealth), the PoS roadmap, a bug two weeks after "final", the authorship conflict (Core Scientific, Linzhi's 3x to 8x claim with no evidence) |
| 3 | RandomX v2 | A tweak after seven years: program 256 to 384 instructions, 16 AES mixes in place of XOR, dataset prefetch 2 iterations ahead; work per hash up 52.9 percent (4,456,448 to 6,815,744 ops) to re-match CPU to memory latency | Library PR 317 merged 17 Feb 2026 (SChernykh); commitments PR 265 merged 8 Sep 2023 | Activation: Monero PR 10038 (v17) open since 14 Aug 2025, pending review 19 Sep 2026, no date | The chip came first: Antminer X5 Sep 2023 (212 kH/s, 1,350 W); X9 1 MH/s at 2,472 W listed as "August 2026, delayed" per resellers, unconfirmed shipping; the Qubic episode used plain CPUs so it forced no algorithm change |
| 4 | Kaspa kHeavyHash | A hash for optical hardware (Dubrovsky, Ball, Penkovsky, arXiv 1911.05193): shift cost from OPEX to CAPEX; GPU resistance never a goal; chosen by community vote one day before launch | IceRiver KS0 (100 GH/s at 65 W, 2023), then 15 to 21 TH/s boxes; GPUs at 2.2 GH/s left | A GPU lane (never planned); optical mining (no shipping hardware, approximate) | By design. The GPU fleet forked away: Karlsen (FishHash, 4 GB DAG), Spectre (CPU), Pyrin |
| 5 | Autolykos v2 | Keep the 2 GB table, grow it 5 percent per 51,200 blocks from block 614,400, cap at block 4,198,400 (2,143,944,600 elements) | EIP-0009 at block 417,792 (Feb 2021 approximate); non-outsourceability removed because contract pools bypassed it (Chepurnoy and Saxena, eprint 2020/044) and small miners did not want it | A successor to pool resistance; an answer to HBM FPGAs ("HBM FPGA friendly", Osprey listed ERGO); the growth stops at the cap | Small prize, no chip, the table cap is a policy choice |
| 6 | Grin's two lanes | Cuckaroo29 for GPUs (90 percent of reward falling to 0 over two years, tweaked every six months), Cuckatoo31+ for chips (10 percent rising to 100) | Four hard forks on schedule; HF4 (about 15 Jan 2021) ended the GPU lane; one chip shipped (iPollo G1, 36 GPS at 2,800 W, Dec 2020) | Obelisk GRN1 cancelled 19 Jul 2019 ("lack of interest and funding", full refunds); Innosilicon G32 announced, never documented shipping; the lean-solver memory claim: Tromp's USD 10,000 linear TMTO bounty claimed 11 Apr 2025 (about half an order of magnitude, not tenfold) | The schedule forced chip makers to choose between a single-use C31 chip and a late C32 chip at a USD 2 coin; network today 3.40 KGps on GPUs |
| 7 | Equihash parameter programmes | Zcash: a parameter change "ideally deployed by late 2018"; a 2020 PoW migration (ballot 3 Jul 2018: 38 of 64 for migration by 31 Dec 2020). Beam: forks at 6 and 12 months then an "affordable ASIC" (BeamHash III, 28 Jun 2020). BTG: Zhash 144,5 after the May 2018 51 percent attack (388,000 BTG) | Zcash stayed on 200,9 with chips (2019: BCTV14 forced the choice of Sapling over ASIC work); Beam stayed on BeamHash III with GPUs only; BTG forked and was hit again in Jan 2020 | Zcash's migration; Beam's affordable chip; BTG's real problem (rented hash on a small chain) | Governance by poll chose "reduce chokepoints" over resistance; parameter forks do not fix a rentable chain |
| 8 | Octopus to Ethash | CIP-102 (Chenxing Li, 5 Aug 2022): switch to Ethash "to make it easier for Ethereum miners to switch to Conflux" | Nothing; the CIP has "TBA" for specification and rationale and never left Draft | The AMD support and DAG shrink the forum asked for instead | The community called Ethash "not ASIC resistant" and Octopus "the trademark" |
| 9 | Blake3 on Alephium | ASIC friendly by design, "just like Bitcoin" | Goldshell AL-BOX (360 GH/s, May 2024 per the aggregator), Antminer AL1 (15.6 TH/s, Jul 2024); the project's own gpu-miner and fpga-miner repos stopped mattering | Nothing, by the project's own view | By design |
| 10 | Tensor PoW | Komargodski and Weinstein, eprint 2025/685: PoUW from arbitrary matrix multiplication at 1+o(1) overhead; "this blockchain is currently under construction" | A paper (15 Apr 2025, revised 8 Dec 2025). ethresear.ch has no topic matching "tensor core proof of work" | A chain; a measurement of what tensor work costs the honest card against a chip | Nobody measured it before writing it |
### 2.2 The FPGA question, one line per row
| Row | Was the FPGA the first adversary | What stopped it |
|---|---|---|
| Lyra2REv2 (Vertcoin) | Yes: a Zynq UltraScale+ at 31.25 MH/s and 24.93 W, 4.3x a Titan Xp per joule (arXiv 1905.08792) | A fork to Lyra2REv3 |
| X16R (Ravencoin) | Yes: all 16 hashes on one UltraScale+, about 5x a 1080 Ti per joule at 3 to 4x the price | X16Rv2 then KawPow |
| Equihash 144,5 and 200,9 | Promised in 2018 (a vendor asked a USD 1,000,000 minimum order), no hashrate ever posted | The 2.5 GB table; HBM boards matched GPUs per dollar at best (approximate) |
| Ethash and ProgPoW | HBM boards existed in 2019; Least Authority: a 10-block period makes "building and loading a bitstream in such a short period (about 2 minutes) impractical" | DRAM latency on cheap boards, HBM cost on fast ones, bitstream churn |
| Autolykos v2 | Osprey listed ERGO on an 8 GB HBM board; no share figure published | Nothing stopped it; nothing measured it |
| kHeavyHash | Osprey E300, 14 GH/s at 250 to 500 W, USD 4,999, three VU35P with HBM2, announced 13 Dec 2022 | Overtaken within months by 100 GH/s to 1 TH/s ASICs |
| RandomX, Cuckoo | No FPGA product ever shipped | A CPU-like core with 2 MB per thread; 512 MB to 1 GB of fast memory per graph |
### 2.3 What Igneum already encodes, and what it does not
| Row | Encoded in Igneum (where) | Not encoded (the gap) |
|---|---|---|
| 1 Ethash successors | A dataset that grows on a genesis-fixed schedule with the cache doubling (256 MiB, 512 MiB year 4, 1 GiB year 12); the 4 GB cliff lesson lives in the tier rule ("every number carries its consequences", 8 GB to 32 GB cards); EIP-969's human tweak against a chip is replaced by automatic era draws | The DATASET_PARENTS idea is Igneum's mixer x8 (8 dependent reads per item), which has the fixed shape the chip history calls the 3x factor; the random item derivation (asic-history addition 2) is still open |
| 2 ProgPoW | The governance death is designed out: every layer is a genesis rule or a reserve, no adoption vote (asic-history lesson 7); the 64-bit-seed class: Igneum's seed is 256 bits and the program is the epoch's; the light-evaluation attack is priced (chip-model-v3, 0.31x bare, 0.92x with the 3x factor); the datapath target is the whole GPU | ProgPoW's 2-minute period was its FPGA defence; Igneum's epoch is 1 h, so a soft-overlay bitstream within the hour is the open lane (asic-history addition 5, the 10 min to 2 h epoch reserve, design on `ca2-epoch`) |
| 3 RandomX v2 | The lesson is that a tweak after seven years arrives after the chip. Igneum draws era parameters every 6 months from chain state and unlocks families by height, with no human release; the class ladder (95 percent with a floor height, P2) is the only human path and it is for new code, not for parameters | Per-hash program entropy (RandomX's 25x GPU cost, refused on purpose); the random superscalar item derivation; four external audits before launch (RandomX paid USD 141,000; Igneum's cryptanalysis spend of USD 80,000 to 160,000 is a decision owed) |
| 4 kHeavyHash | The loss of the GPU fleet is the metric: hashrate share by card model from the observer and the January 2027 benchmark (asic-history addition 4) | A published vendor-share number does not exist yet |
| 5 Autolykos v2 | Pools are wanted (spec 09), so non-outsourceability is nowhere by choice; the dataset grows by rule | Whether Igneum's growth schedule ever caps is a spec 1.13.3 question this lane did not reopen |
| 6 Grin | Scheduled change without a vote is the shared idea; Igneum's draws are automatic, Grin's were forks | The lean-solver lesson (a memory claim weakened six years later by a bounty) is the reason the mixer and chained cache need cryptanalysis (ledger M7) |
| 7 Equihash | The rented-hash problem is answered by the 30-day weight finality, not by the hash (section 4) | Nothing |
| 8, 9 Octopus, Blake3 | Nothing to take | Nothing |
| 10 Tensor PoW | Measured and retired: scheme B's mm8 shadow costs the honest card 0.056 to 0.70 pJ per multiply-add and moves the chip's edge only from 6.9x to 4.9x, so it is reserve R8 with the two-output correction (new-pow.md 6) | Nothing; Igneum did the measurement the 2025 paper has not |
| FPGA, all rows | Bitstream churn is encoded as the hourly program; the HBM board is the f = 1 stored-dataset chip in the model (5.1x on GDDR7, 7.5x to 9.2x on HBM3) | The soft-overlay FPGA miner with HBM is unmeasured (addition 5); no row in the chip model for a partial-store chip (addition 1) |
Verdict for the group: nothing in B is a programme Igneum should restart. The three unfinished items that do matter are Igneum's own open rows (random item derivation, external cryptanalysis, the epoch-length and FPGA-overlay measurement) and they are already ranked in `asic-resistance-history.md` 4.3.
## 3. Group C: merged mining and multi-algo
### 3.1 The rows
| # | Programme | When, by whom | What it bought (numbers) | The attacks and the costs | Left undone |
|---|---|---|---|---|---|
| 1 | Namecoin AuxPoW | Block 19,200, Oct 2011 (vinced; Mike Hearn's outline) | Bitcoin's hashrate for free: about 95 percent of Bitcoin on 12 Jan 2020, about 69 percent (732 EH/s) in Sep 2026; nine pools find blocks (AntPool 34.78 percent, F2Pool 19.99, ViaBTC 19.69, 14-day window to 20 Nov 2025) | One pool (F2Pool) above 50 percent of Namecoin for most of late 2014 to 2016 (approximate; Judmayer et al.: pools "operated at the edge of, and even beyond, the security guarantees" of Nakamoto consensus); the spec's nonce-clash flaw is documented and a replacement BIP never came | Independence: the child's rules are signalled by the parent's pools; Foundry, the largest Bitcoin pool, does not merge-mine it; the name market never grew |
| 2 | Dogecoin on Litecoin | Hard fork at block 371,337, Sep 2014, after Charlie Lee's Apr 2014 proposal; Dogecoin's own words: "our hashrate has been on a decline" | Hashrate up over 1,500 percent in Sep 2014; DOGE and LTC hashrates correlate at 0.95; about 90 percent of DOGE hashrate from Litecoin pools by 2019; DOGE's share of pool revenue "has long surpassed" Litecoin's (F2Pool data, 7 Apr 2026) | A pool above 50 percent of Litecoin is above 50 percent of Dogecoin by construction | A miner base of its own; any say in the parent's rules |
| 3 | Myriadcoin and Verge multi-algo | Myriad 23 Feb 2014 (five algos, per-algo floating difficulty, 20 percent of blocks each); Verge (five algos) | Hardware diversity | Verge, Apr 2018: timestamps spoofed about an hour back on the Scrypt lane, per-lane difficulty collapsed, blocks seconds apart, 250,000 to about 20 M XVG (reports conflict); 22 May 2018: Scrypt and Lyra2re both at near-zero difficulty, about 25 blocks a minute, 35 M XVG (about USD 1.8 M); the hard fork between them did not remove the bug; a 560,000-block reorg in Feb 2021 (headline only) | Five lanes made the surface five times wider; each lane is cheaper to rent than the whole |
| 4 | DigiByte MultiShield, Odocrypt | DigiShield Feb 2014 (block 67,200), MultiAlgo Sep 2014 (145,000), MultiShield Dec 2014 (400,000), Odocrypt Jul 2019 (9,112,320): per-block retarget on all five lanes from a 10-block average, clamps 16 percent down and 8 percent up; Odocrypt rewrites itself every ten days for FPGAs | DigiShield was adopted by Dogecoin (1.6, Mar 2014) and Zcash (v3 variant, approximate); the "93 percent on one algo and 51 percent on the other four" claim is a model, not a measurement | 28 Jun 2026: a dropped validation rule on the Groestl lane let about 1,356 blocks (about 351,000 DGB) be mined, no double spend | A lane with little hashrate still mints one block in five; the ten-day rewrite assumes FPGA owners re-flash and does nothing against a party with many FPGAs |
| 5 | Elastos, RSK | Elastos with Bitmain 28 Aug 2018 (AntPool, BTC.com); RSK since Jan 2018 (approximate) | Elastos "over 50 percent" of Bitcoin's work (13 Mar 2024); RSK 62.46 percent (Q1 2024), 54.58 (Q3), 56.40 (Q4 2024, AntPool 38.6, ViaBTC 24.3, F2Pool 16.9); Foundry joined | RSK's Armadillo (RSKIP110): detects forks up to about 448 blocks when the attacker has under 9 percent of Bitcoin; "cannot stop a miner creating a blockchain from a past point"; the PowPeg is a federation with HSMs | The attacker rents or bribes three pools, not hardware; Armadillo alerts, it does not finalise; usage stayed small so fees stayed small |
| 6 | The paper | Judmayer, Zamyatin, Stifter, Voyiatzis, Weippl, "Merged Mining: Curse or Cure?", eprint 2017/791 | One line: merge-mined chains concentrate in fewer pools than the parent, with single pools above 50 percent for months in Namecoin, Huntercoin and Myriad (approximate for the per-coin detail) | | |
### 3.2 Verdict: does Igneum want any of it
No, with four reasons. Igneum's ruling is already "no reliance on another chain"; the question left is whether an Igneum-side auxiliary chain or an "algo slot" has value, and it does not.
| Option | What it would buy | Why not | Hours |
|---|---|---|---|
| Igneum as a merge-mined child of another chain | Borrowed hashrate | Ruled out (CLAUDE.md: no other chain in consensus); and the rows show the child inherits the parent's pool concentration and loses its say in rules | 0 |
| Igneum as a parent for other chains | Nothing for Igneum; a coinbase commitment slot for others | The hash is a new program every hour, so a child cannot verify Igneum work without Igneum's program pipeline, VDF seed and dataset; the vote key per operator would not carry to the child; the only cost is a header field nobody asked for | 0 until asked |
| An algo slot (a second hash lane) | Hardware diversity | Myriad, Verge and DigiByte show each lane is a separate difficulty controller and a separate rentable surface; Igneum's 30-day vote weight is counted in blue blocks of ONE work unit, so a second lane splits the weight table and the DAA; GHOSTDAG's k is derived from one block rate on one hash | 0 |
| A share sidechain merge-mined through the coinbase | A pool without an operator | This is the one merged-mining shape with value, and it belongs to section 5 (it is how Monero P2Pool works) | see 5.3 |
## 4. Group D: GPU-friendly finality attempts
The one axis asked: can a renter buy or break finality in a day? Igneum's rule for comparison (CLAUDE.md, finality rule v2): vote weight = blue blocks per vote key over a flat 30-day window, lock at 2/3 of all 30-day weight, so 10 days of 100 percent hashrate reach 1/3 of weight and 20 days reach 2/3; 51 percent never reaches 2/3 while honest miners stay.
### 4.1 The rows
| # | Programme | Attempted | Finished | Left undone | Renter buys finality in a day? |
|---|---|---|---|---|---|
| 1 | Decred hybrid | Feb 2016: tickets vote on every block, 3 of 5 called tickets must sign; 40,960 ticket pool, 28-day average wait | A PoS veto on every block, stakeholder-voted forks, a treasury; stake locked rose from 50 percent (Dec 2019) to 64 percent (Dec 2022) while hashrate fell about 6x; DCP-0011 (BLAKE3 and ASERT) and DCP-0012 (PoW to 1 percent of subsidy) active at block 794,368 (Aug 2023) because PoW was "used to maliciously manipulate Decred markets" | No finality gadget; PoW is a 1 percent stipend | No for reorgs (needs about half the live ticket pool, roughly 30 percent of DCR bought over 28-day cycles, approximate); yes for withholding and censorship, since the hash layer is thin |
| 2 | Horizen delayed-block penalty | After the 2 Jun 2018 51 percent attack (over USD 500,000, 36 fake blocks): a block n late owes about n(n-1)/2 extra blocks (approximate formula); shipped in ZEN 2.0.16, late 2018 | A reorg tax on hidden chains | Partition safety: the smaller honest side looks like a hidden chain; Horizen now calls itself an L3 on Base and the PoW mainchain is winding down (approximate) | Yes under 5 blocks and for public mining; a deep secret reorg costs quadratically more. A tax, not finality |
| 3 | Kaspa merge depth and finality depth | MERGE_DEPTH_DURATION 3,600 s, FINALITY_DURATION 43,200 s, PRUNING 108,000 s in rusty-kaspa constants; at 10 BPS since Crescendo (about 5 May 2025): merge depth 36,000 blocks, finality depth 432,000 | Online nodes refuse a reorg deeper than 12 hours; a block cannot merge a side chain older than an hour | Both rules are local and subjective; a node syncing later follows the heaviest DAG; DAGKnight (eprint 2022/1494) is in PRs (1104, 1124, 1132) and not on mainnet | Partly: under 12 hours yes with over 50 percent of an ASIC-only hash (thin rental supply); over 12 hours the attacker splits the network instead |
| 4 | Conflux Tree-Graph, GHAST, PoS votes | GHAST (arXiv 2006.01072): confirmation in O(d log 1/epsilon), about 3d against about 360d for six Bitcoin blocks; adaptive weight switches to a conservative mode under a liveness attack | CIP-43 (Hydra, about 28 Feb 2022): a PoS chain of staked CFX (1,000 CFX per vote) votes on pivot blocks "to protect against 51 percent attacks from PoW" | GHAST alone is probabilistic; the PoS committee is the finality | No for finalised pivot blocks; yes inside the trailing minutes |
| 5 | Ethereum Casper FFG over PoW | EIP-1011 (20 Apr 2018): 1,500 ETH minimum deposit, 50-block epochs, "never revert finalised blocks"; testnet 31 Dec 2017 | Nothing on mainnet; the Berlin Eth2 workshop (Jun 2018) discarded hybrid Casper for the beacon chain | The hybrid itself | Under the design: no for finalised checkpoints (slashing two thirds of deposits), yes inside an epoch. Dropped because "the limited bandwidth of the EVM constrained the size of the validator set", forcing whale-sized deposits |
| 6 | Bitcoin Cash Avalanche pre-consensus | Sechet, Sep 2018 (approximate) | On BCH: nothing. On eCash (XEC): post-consensus finality 14 Sep 2022, staking rewards 15 Nov 2023, pre-consensus 15 Nov 2025 at block 923,347 ("finalise transactions in under 3 seconds"); staked XEC 69.4 B to 242 B in the year to Jul 2024, 350 B+ by Jul 2026; 33 to 94 nodes; 100 M XEC minimum per stake UTXO with 2,016 confirmations | BCH never got it; no slashing (approximate); on quorum loss the chain falls back to heaviest work | No while the quorum holds; yes if the attacker knocks the about 94 Avalanche nodes offline |
| 7 | Ethereum Classic MESS | ECIP-1100, active at block 11,380,000 (Oct 2020) after three Aug 2020 reorgs (3,594, 4,446 and 7,278 blocks; 807,000 and 465,444 ETC double spent): a capped polynomial, up to 31x the local chain's difficulty for a segment older than about 7 hours | No deep reorg on ETC after Oct 2020 (approximate) | Deactivated by ECIP-1110 at block 19,250,000 (Spiral, Jan 2024) because "the costs now outweigh the risks"; Core-Geth v1.13.0 (15 Sep 2026) re-enabled it with 96 unreviewed commits, miners rolled back the same day, PR 53 (merged 6 Oct 2026) restored the deactivation | With MESS on: no for reorgs older than 7 h, yes for short ones, and an eclipse strands a victim. Off today: yes, plain PoW |
| 8 | Zcash trailing finality layer, Crosslink | ECC's TFL book (2023, Wilcox and Hopwood); Shielded Labs builds zebra-crosslink, "a proposed upgrade"; BFT PoS finalises a trailing prefix, PoW continues if BFT stalls | An incentivised feature net; milestone 4 of 5 in early 2026 (search summaries) | Mainnet: nothing by 7 Oct 2026 | Today: yes (plain Equihash, rentable, approximate). Under Crosslink: no for finalised blocks, yes inside the trailing window |
| 9 | Nervos CKB | NC-Max (eprint 2020/1101): two-step confirmation, 3.0 to 6.6 times shorter latency; Eaglesong "ASIC neutral", first chip four months after launch | NC-Max runs since launch | No finality layer; no hybrid plan | Yes in principle; ASIC-only supply is thin |
| 10 | Dash ChainLocks, Syscoin, Komodo | Dash DIP-0008 since block 1,088,640 (Jun 2019): an LLMQ of 300 to 400 masternodes signs the first block seen, 240 signatures lock it; Syscoin copied it over Bitcoin merged mining (4.2.0, block 1,004,200, 30 Apr 2021); Komodo notarises to Bitcoin then Litecoin every 10 to 25 minutes through 64 elected notaries | Finality in about one block (Dash) | A second validator set bought with coins (1,000 DASH per masternode, approximate) or elected (Komodo's 64) | No for locked blocks |
### 4.2 The comparison on the one axis
Every finality on a PoW chain that holds in practice is a second validator set paid in coins: Decred tickets, Dash and Syscoin masternodes, eCash stake, Conflux PoS, Komodo notaries, Casper FFG's 1,500 ETH deposits. Every rule that only reweights work (Horizen's penalty, MESS, Kaspa's 12-hour depth) is subjective and partition-sensitive, and two of the three have been switched off or are winding down.
Igneum is the only design in the table whose second set is the miners themselves, weighted by past blocks, with no coins staked. On the one axis:
| Attacker | Decred | Kaspa | MESS (on) | eCash | Igneum |
|---|---|---|---|---|---|
| Rents 100 percent of hashrate for one day | Withholds, cannot reorg | Reorgs up to 12 h | Reorgs up to about 7 h | Nothing while the quorum holds | Reorders inside the 90-second window (horizon 1, finding 3), nothing past a certificate; has 1/30 of a window's weight at the end of the day, needs 2/3 |
| Rents 100 percent for 20 days | Same | Splits the network | Same as above | Same | Reaches 2/3 of weight only if every honest miner has left; with honest miners staying, 51 percent never reaches 2/3 |
| Buys coins | Buys tickets over 28-day cycles | n/a | n/a | Buys stake | Nothing; weight is not for sale |
What is unfinished in Igneum's own finality is Igneum's, not the industry's: the signed LEAVE item (0.3.16) after the 6 October pause (42.7 percent of the frozen voter table left in three minutes and the pause held a window), the 10-percent-per-hour removal rule, weight-gated deep fork choice and vote-or-burn (horizon 1, findings 2 and 3). The one lesson this group adds: every subjective reweighting rule was eventually turned off by its own chain (MESS twice), so the "no hidden-block n^2 penalty" removal in Igneum's rule v2 was the right side of the history.
## 5. Group E: decentralised pools
### 5.1 The rows
| # | Programme | When, by whom | Design | What it achieved (numbers) | Why it stalled or died | Left undone |
|---|---|---|---|---|---|---|
| 1 | Bitcoin P2Pool | Forrest Voight, 2011 | A share chain with a 30-second share period; window = shares worth 3 blocks of work or 8,640 shares (72 hours), whichever is smaller; the generation transaction pays the window; 0.5 percent to the finder, no operator | 350 GH/s in Apr 2012 (about 2 to 3 percent of Bitcoin, approximate); 1.4 TH/s by Jul 2013; today 915 TH/s on 14 payout addresses (about 0.0001 percent of about 950 EH/s, my division); last commit 19 Sep 2018 | Variance for small miners (a small miner rarely lands a share in the window), dust (one output per miner per block), a 20x faster chain so stales are "common and expected", a 2012 orphan flaw, a Python 2 node on a full bitcoind | Payout aggregation, per-miner vardiff inside the chain, a maintained node (c2pool's C++ rewrite says "v37 is design-stage, do not run in production") |
| 2 | Monero P2Pool | SChernykh, Aug 2021 | A sidechain merge-mined with Monero: 10-second blocks, PPLNS up to 2,160 blocks (6 hours), uncles at 20 percent less, payouts in the Monero coinbase, 0 percent fee, about 0.00027 XMR minimum, every miner runs monerod; three chains (main, mini, nano since v4.7, 30 May 2025); Tari merge mining since the 12 Oct 2024 fork | Today: main 441 MH/s and 561 miners, mini 25.3 MH/s and 1,997, nano 4.8 MH/s and 962; 471 MH/s and 3,520 miners in all, 7.7 percent of 6.13 GH/s (my division); against SupportXMR at 2.35 GH/s (38 percent) and HashVault at 672 MH/s (11 percent); the Monero GUI bundles it | A local node is mandatory; below about 15 MH/s of pool hashrate "not all shares will result in payouts"; three chains fragment the hashrate; share has fallen from double digits (approximate) to 7.7 percent | A light mode; cross-chain vardiff; payout smoothing for mini and nano miners |
| 3 | Stratum V2 | Braiins (Moravec, Capek) with Matt Corallo, 2019, from Corallo's BetterHash draft (12 Mar 2018, never past specification) | Binary, Noise-encrypted; Mining, Job Declaration (renamed from Job Negotiation) and Template Distribution sub-protocols; the SRI translation proxy | SRI v1.0.0 tag 23 May 2023, v1.12.0 17 Sep 2025; native V2 pools: Braiins (JD "since about 2020"), DEMAND (public Nov 2025), ckpool (31 Jul 2026); firmware: Braiins OS, Bitaxe ESP-Miner v2.14.0 (Jun 2026), Auradine; stock Antminer and WhatsMiner are V1 only; the first known Job Declaration block is 955,318 on 25 Jun 2026 (DMND, GoMining's template); seven pools holding about 75 percent of hashrate joined the working group on 7 May 2026 "still in testing" | No primary number for SV2 hashrate; the two V2-native pools mined 753 and 1 of 52,297 blocks in the last year, so under 2 percent on SV2 at all and far under 0.1 percent on JD (approximate conclusion) | Stock firmware, JD at any top-five pool, Windows template building ("not yet supported"), a payout scheme that rewards miner-built templates, any measurement of adoption |
| 4 | Braiins Pool | Slush Pool, Nov 2010; the first anti-hopping score (Rosenfeld 2011, section 3.1) | FPPS "0 percent fees with Braiins OS", 2 percent otherwise; on-chain or Lightning payouts | 13.91 EH/s, 177,359 workers, 13,112 users today; 753 of 52,297 blocks (1.44 percent) in the last year | The oldest pool and the SV2 author holds 1.4 percent | SV2 adoption did not become hashrate |
| 5 | Ocean | Luke Dashjr, 28 Nov 2023, USD 6.2 M seed led by Jack Dorsey | TIDES: a share log eight times the difficulty deep, each share rewarded about eight times, 99.9665 percent chance of at least once, paid in the coinbase itself; DATUM (29 Sep 2024): the miner runs a gateway and a full node and builds its own template, 50 percent fee discount (2 percent, 1 percent with DATUM) | First block's coinbase came from a test server and paid nobody (4 Dec 2023 post-mortem); today 28.97 EH/s, 2,522 users, 1,553 blocks of which 1,316 DATUM (85 percent, my division), 1,073 of 52,297 blocks (2.05 percent) in the last year; Tether committed hashrate Apr 2025 | The template fight: Bitcoin Knots filtered inscriptions and Samourai transactions at launch; Dashjr resigned 29 Aug 2026 after the "mining divide" | A single operator still runs TIDES accounting, share verification and the generation transaction; non-DATUM miners get the pool's policy |
| 6 | DEMAND | Guru Protocol Ltd (company 15235937), Alejandro De La Torre; launched Mar 2025 on the SRI, public Nov 2025 | "The world's first Stratum V2 mining pool"; SLICE pays subsidy by hashrate and fees "by the value you built"; Rootstock merge mining | 1 block in 19 months | Scale | SLICE at size |
| 7 | Kaspa pools | Today: 334.9 PH/s; F2Pool 98.14 PH/s (29.3 percent), HumPool 92.95 (27.8), ViaBTC 37.89 (11.3), K1Pool 16.74 (5.0), WhalePool 13.67 (4.1); top two hold 57 percent | All PPLNS or PPS+ with 0.5 to 4 percent fees | No non-custodial or P2Pool-style pool; solo bridges only (K1Pool Solo, Kaspa-Pool SOLO, HeroMiners SOLO); no pool page explains which DAG blocks pay (approximate: blue blocks in the mergeset, red blocks earn nothing) | n/a | A decentralised pool on a DAG chain has never been built |
| 8 | SmartPool | Luu, Velner, Teutsch, Saxena, eprint 2017/019, USENIX Security 2017 | Shares in batches ("1 transaction can claim 1 million shares") in an augmented Merkle tree; the contract samples k random paths; a caught cheater is paid nothing | ETC first ("over 30 blocks"), Ethereum closed beta Jun 2017; 105 blocks across both, peak 30 GH/s from 2 miners (about 0.05 percent of Ethereum, approximate); cost 0.6 percent of block rewards in transaction fees | Gas prices rose about ten-fold in 2017 and the claim cost passed the fee it was meant to beat; the beta never opened; the team founded Kyber; PoS removed the target (all approximate: smartpool.io is dead) | Any successor with hashrate; the opposite approach (Autolykos v1's non-outsourceable puzzles) was bypassed with collateralised contracts (eprint 2020/044) and removed |
| 9 | Non-custodial on GPU chains, 2024 to 2026 | Ravencoin: 2Miners 292.67 GH/s with "51 percent of recent blocks", Hiveon 621 GH/s, no P2Pool; Ergo: 2Miners 303.14 of 491.76 GH/s (61.6 percent), no P2Pool; SoloPool.org sells solo mining at 1.5 to 2 percent "no pool wallet, no balances" | | Outside Monero every "non-custodial" offer is solo with a stratum bridge, not pooled variance reduction | | |
Share verification cost, from the sources: a Scrypt proxypool spends "around 50 percent of the server's CPU time" checking shares; RandomX light mode on an i9-9900K verifies 1,160 H/s on 16 threads, so a core checks about 70 shares a second (fast mode about 700); Bitcoin SHA256d is microseconds per share (approximate, 10^5 to 10^6 per core). Igneum's pool v0 checks every share on the CPU warp verifier at 1.35 ms isolated and 2.1 ms under load, 480 to 740 per core (`pool.md` section 5), which is RandomX-fast-mode class, 1,000x a SHA256 share.
The payout pattern: FPPS won on Bitcoin because large pools carry variance and sell certainty (Foundry about 30 percent, AntPool about 19, ViaBTC 14, F2Pool 10, d-central 2026); every non-custodial design (P2Pool, Eligius CPPSRB, TIDES, SLICE) is PPLNS-shaped because a coinbase cannot pay a debt. That is the structural reason non-custodial pools stay small, and it is the reason Igneum's pool is PPLNS.
### 5.2 What Igneum's pool lacks against the best of these
Igneum's pool v0 (`docs/plans/pool.md`, spec 09): newline JSON over plain TCP, mode A templates, every share CPU-verified, PPLNS over 2 blocks of weight, 1 percent fee, payouts by EIP-1559 transfer from the pool's coinbase key after blue confirmation, `state.json` every 15 s, the member's vote key in every header it hashes, the pool holds no key, the member checks seeds, class and era against its own node.
| Axis | Best in the field | Igneum v0 | The gap |
|---|---|---|---|
| Operator trust for payout | Monero P2Pool and TIDES pay in the coinbase itself; nobody holds a balance | The operator holds the 80 percent in its coinbase address and pays by transfer at most 16 per round; a crash loses up to 15 s of shares; a dishonest operator can misaccount or withhold | The whole of the trust. Mode C (declared templates) and vote carriage are designed, not built; neither removes the operator from payout |
| Operator trust for template and vote | DATUM and SV2 Job Declaration let the miner build the template; neither touches governance | The member's vote key rides in every header (no Stratum pool does this); mode C is designed, not built | Transaction choice is the pool's in mode A; the vote is already the member's |
| Variance | P2Pool main: 6-hour window, 561 miners; Ocean: a window 8 difficulties deep | PPLNS over 2 blocks of weight at 1 block a second: a 2-second window of work. Each share is 2^-s of a block, so a 100 MH/s card at one share per 10 s is paid on every block it has a share in | The window is short because blocks are frequent; variance is a block-time question, not a pool question, and v0 has no number for the income variance of a 100 MH/s member at devnet scale |
| Share verification | Full recompute everywhere; sampling at high difficulty (spec 9.8 item 5, allowed above shift 8, not built) | 1.35 to 2.1 ms per share, one core per 480 to 740 shares a second, no sampling | One template fetch per member per second is the real limit: 1,000 members is 1,000 `getBlockTemplate` calls a second and 10 MB/s (pool.md section 3) |
| Transport | SV2 Noise encryption, binary framing | Plain TCP, `binding` sent empty, against spec 9.3's TLS 1.3 | Q67 (polish 3.6): 8 hours |
| Stats and alerts | 2miners charts, offline alerts, luck | 15-minute samples not persisted, no `/metrics`, no payout history on the page, luck defined inverted | Q68 to Q73: about 17 hours |
| Finality | No pool in the field has a finality concept | Pays on blueness, says nothing during a pause | Q73: 1 hour |
### 5.3 "Shares as first-class protocol objects, no operator": already in Igneum, partly
| Piece | In Igneum already | Where | What is new |
|---|---|---|---|
| Governance weight follows the hasher, not the pool | Yes: the vote key per operator in the header (spec 9.6), pool concentration is not vote concentration | spec 2.2 fork point a5, spec 9.6 | Nothing; this is further than SV2 or DATUM went |
| An operator-less payout by weight | Yes for the 20 percent: the proving share is paid by sortition by weight from consensus data, no operator | CLAUDE.md, `executor.rs` (pool.md "The 20%") | Nothing for provers |
| A share the chain can see | No: a share is a pool message, verified by the pool, paid by the pool | spec 9.8 | This is the new thing |
The finished form is Monero P2Pool's, not SmartPool's. SmartPool's on-chain claims died on gas at Ethereum's share rate; at Igneum's 1.35 ms per share verification, a contract that samples k paths per million-share claim would still need the zkEVM to re-derive the program's warp unit per sampled share, which is the one cost the design keeps off the chain. A share sidechain merge-mined through the coinbase extra data is the shape that works: Igneum's coinbase already carries the member's key reveal and the pool's address (pool.md section 2), so the commitment slot exists.
| Finish | What it is | Hours (agent) | Gate |
|---|---|---|---|
| A P2Pool-style share sidechain in the pool crate | Shares form a 10-second chain committed in the coinbase extra data; the window's payout is the block's own coinbase split (the execution layer already credits by rule from consensus data for the 20 percent, so the 80 percent split follows the same path); no balance, no operator key; every member runs a node (the design already assumes one, spec 9.1) | about 40 (sidechain 16, coinbase split in the executor 8, member client 8, Devnet 2 crossing 8) | a 3-node fast-time network pays a window with no operator; an operator that drops every share from one member cannot stop that member's payout |
| Mode C declared templates | The miner builds the block (DATUM, SV2 JD) | 6 (O-9.4 RPC 2, messages 4) | a declared template is paid like any other |
| TLS and the `binding` field | spec 9.3 | 8 (Q67) | a replayed `authorize` on a second connection is refused |
The verdict for the group: finish the share sidechain, because it is the one unfinished thing in the field that Igneum's existing objects (vote key in the header, coinbase extra data, execution-layer credit by rule, a node per member) make cheap, and nobody has built one on a DAG chain.
## 6. Group F: mining app UX and the "one click" promise
### 6.1 The rows
Minutes to first share are approximate unless a source is named; they count from the download to the first accepted share on a fresh machine with a driver installed.
| # | Product | Install to first share (min, approximate) | Custody | Fee | What it did that nobody else has | What it left undone | State 2026 |
|---|---|---|---|---|---|---|---|
| 1 | NiceHash (2014; QuickMiner 2020, approximate) | 5 to 10 | NiceHash wallet, BTC only; 2.5 M users (infobox) | Seller 2 percent, buyer 3 percent (approximate); BTC deposits at or above 0.00005 BTC free, Lightning free above 0.0000001 | A hashpower marketplace: the miner never picks a coin | Custody: 6 Dec 2017 spear-phishing hack, about 4,700 BTC (USD 64 M), Lazarus indicted 17 Feb 2021; repayment completed Dec 2020 (approximate); no node, no wallet of the user's own | Alive |
| 2 | HiveOS (2017) | 20 to 40 (flash an image on a second machine or drive, make a web account) | Hiveon pool pays the user's wallet | 2 workers free (3 days of stats); USD 0.50 per card per month to USD 3.00 at 6 or more cards; ASIC USD 2 or free with Hiveon firmware; 10 to 50 percent off at 50 to 1,000 workers | Farm-scale remote management, charts over hours, Telegram and Discord alerts, per-GPU clocks | A second machine or a flash drive; a web account; no local-only mode, no wallet, no node | Alive |
| 3 | Minerstat, Awesome Miner | 15 to 30 | User's pool and wallet | minerstat free to 2 rigs, about USD 1.46 per rig per month on larger plans (approximate); Awesome Miner free to 2 miners, paid from about USD 4 per month (approximate) | Awesome Miner: "up to 200,000 ASIC and 25,000 GPU miners" from one Windows console | Management only; no node, no wallet, no pool | Alive |
| 4 | Salad (2018) | 5 | Salad Balance, redeemable for PayPal, gift cards, games | Not published | Gaming-PC "earn while idle"; 450,000+ GPUs since 2018; 16,000+ daily active GPUs in 190+ countries today | Mining itself: the site no longer mentions crypto; SaladCloud sells GPU-hours from USD 0.015; the user's take rate is not disclosed; payout in gift cards for years | Pivoted to a compute marketplace |
| 5 | Honeyminer (2018) | 3 to 5 | Honeyminer balance, paid in BTC | 8 percent for one GPU, 2.5 percent for two or more (approximate) | The first "download, sign in, earn" miner with no settings | Everything else; the site refuses connections on 7 Oct 2026 | Dead (about 2020 to 2021, approximate) |
| 6 | Cudo Miner (2017) | 5 to 10 | Cudo balance | 1.5 to 3 percent by tier (approximate) | Profit switching on Windows, macOS, Linux | The pivot to Cudo Compute (cloud GPU); docs host unresolvable on 7 Oct 2026 | Pivoted |
| 7 | Kryptex | 5 ("Download, Launch, Earn"; first payout 1 to 8 hours) | Kryptex balance; withdraw BTC, USDT, ETH, USDC, LTC, BNB, TRX, Volet, Amazon gift cards; 0.00025 BTC minimum | "A small pool fee", unstated | "153,000 GPU and ASIC miners", "10 million+ downloads"; the pool: BTC 4,945 miners at 6.63 EH/s, XMR 37,576 miners at 1.08 GH/s; PPS+ | A balance, not a wallet | Alive |
| 8 | unMineable | 5 to 10 | unMineable balance, paid in 80+ assets | 1 percent, 0.75 with a referral code (approximate) | Mine any algorithm, be paid in any coin | A balance; conversion risk; no node | Alive |
| 9 | Bitcoin Core "generate" (2009 to 2016) | 0 (one setting) | The user's own wallet, in the node | None | Node, wallet and miner in one process, the only time it has shipped at scale | Removed in 0.13 (2016) once GPUs and chips made it pointless | Dead |
| 10 | Monero GUI mining | 0 after sync (Advanced, Mining, Start mining; thread count; local node required) | The user's wallet | None | A wallet with the daemon's CPU miner, and P2Pool bundled and updated per release (nano sidechain 26 Aug 2023, P2Pool 4.18.1 on 6 Oct 2024); the nearest shipped "node plus wallet plus miner" | The GPU (RandomX is CPU work); no tuning, no alerts; the sync wait | Alive |
| 11 | Kaspa community miners, the big three | 5 to 15 (a command line, a pool, an address) | The pool's | lolMiner 0.75 percent kHeavyHash, 1.0 Karlsen, 0.7 Ethash, 1.5 Autolykos (v1.96a); BzMiner 1 percent most algorithms, 0.5 ETC, 2 on Pearl and Quantus (v100.45, Windows, Linux, macOS arm64, Docker); GMiner 1 percent kHeavyHash, 2 Ethash and KawPoW, 5 Cortex, "charged continuously"; T-Rex 1 percent (2 on Octopus and Autolykos), 730 open issues, no release date shown; NBMiner 1 to 3 percent, last release 42.3 on 2 Sep 2022 | The per-GPU accepted, rejected, invalid and fault line; kernel speed | No node, no wallet, no tuning against a goal, no signed updates; two of the big three stopped in 2022 (approximate for T-Rex) | lolMiner, BzMiner, GMiner alive; T-Rex and NBMiner stale |
Nobody in the table ships a miner that is also a verifying node and a wallet with a GPU worker: Bitcoin Core did it for CPUs in 2009 and removed it in 2016; the Monero GUI does it for CPUs with P2Pool bundled; every GPU miner is a kernel plus a pool address, and every "one click" app is a custodial balance.
### 6.2 Ember against the field
Ember's state is from `polish.md` 3.1 and 3.10: a Rust engine runs `igneumd` and one GPU worker per card, serves a tokenised local dashboard, Welcome, Cards, Address, the one-time key sheet, Start mining; Ember Tune measured 37 to 41 percent more hashes per watt on PC 1; signed, hash-checked, rolled-back updates with a safe-moment rule; no dev fee in pool mode, a 1-in-100 template solo; pool mode is design only (pool.md section 7).
| Axis | The field's best | Ember today | Past or behind |
|---|---|---|---|
| Node, wallet and miner in one process, GPU | Nobody (Monero GUI for CPU) | Yes: `igneumd`, the worker, the key and the address in one app | Past |
| Custody | Direct-to-wallet only at P2Pool and HiveOS; every one-click app holds a balance | The coinbase pays the user's own address; the vote key never leaves the machine | Past |
| Governance | Nobody | The vote key in every header the card hashes | Past |
| Updates | Unsigned zips (the kernel miners); NiceHash and HiveOS auto-update inside an account | Ed25519-signed manifests, sha256, staged, rolled back after 90 unhealthy seconds, a safe-moment rule | Past |
| Tuning | HiveOS per-GPU clocks by hand; Nanominer's watchdog | A goal with the money consequence on screen, hill climb, 37 to 41 percent per watt | Past |
| Fee | 0.5 to 3 percent dev fees; 1 to 8 percent custodial takes | 1 percent template dev fee solo, 0 in pool mode | Past |
| Install to first share | NiceHash, Honeyminer, Kryptex: about 5 minutes with one interstitial | Windows: SmartScreen interstitial, per-user install, a firewall UAC prompt 20 to 50 s in; Mac: an ad hoc signature that macOS 15 refuses without Privacy and Security, Open Anyway (Q3, Q14, Q15); the Windows installer pipeline is dead on GitHub billing (Q80) | Behind: three prompts against one, and no Windows build until Q80 |
| Earnings | NiceHash's fiat per day on the main screen | "£0.00 earned" first; mined IGN never shown (Q18) | Behind, 2 hours |
| History and alerts | HiveOS hours of charts and Telegram alerts | A 10-minute strip; no outbound alert (Q16) | Behind, 4 hours plus an alert hook |
| Per-card faults | lolMiner's status line | "N blocks" per card; `mismatched=` and `faults=` never reach the row (Q13) | Behind, 2 hours |
| Remote view | HiveOS, NiceHash Rig Manager | None for the owner (the console is the operator's) | Behind; not ranked here |
| Pool mode | Every miner | Design only | Behind; the share sidechain of 5.3 is the finish |
| Finality on the surface | Nobody | Nothing during a pause (Q2) | Behind, 3 hours, and nobody else has the concept |
The honest line: Ember is the first app to ship the thing the table says nobody shipped (node, wallet, GPU miner, own key, signed updates, no custody), and it loses to a 2018 Honeyminer on the sixty seconds between download and first share. The first-run fixes are one Developer ID and one Authenticode certificate (Q3, a Josh decision) plus the Q80 installer builder; the rest of the behind rows are 11 hours.
## 7. Verdict table
| Group | Unfinished item | Who could finish it | Igneum has it | Hours on Igneum | Recommend |
|---|---|---|---|---|---|
| A | Energy that produces proofs of real state, paid from issuance (Aleo's 2019 promise) | A chain with a prover network and no validator set | Yes once the proof is verified in consensus (`proving_consensus_verify_daa`, 0.3.16, off by default) | 0 beyond the activation decision | Already done; activate |
| A | Pay per verified unit of proving at a market price (PoVW) | Boundless, Succinct, Igneum | Yes (pgas metering, 90/10 external jobs) | 0 | Already done |
| A | Useful work verified cheaper than done AND sampleable AND wanted | Nobody in 13 years | No, and the bound says why (0.08 percent useful at 100 GH/s) | 0.5 (the ledger row beside F13, new-pow.md rank 8) | Skip |
| A | Stored data as the work (Filecoin, Chia) without filler | Lane 8's scheme C | Partly (class v5 candidate, judged) | per new-pow.md rank 1 (16) | Not this lane's call; already ranked |
| B | A tweak that lands before the chip (RandomX v2 merged, unscheduled; ProgPoW dead on governance) | A chain whose parameters move without a vote | Yes (era draws every 6 months, families by height, the class ladder) | 0 | Already done |
| B | The random item derivation and external cryptanalysis of the mixer | Igneum | No | per asic-history 4.3 additions 2 and 3; the spend (USD 80,000 to 160,000) is a decision owed | Finish, through the existing ranking |
| B | The FPGA overlay measurement and the epoch-length reserve | Igneum | Partly (reserve designed on `ca2-epoch`) | per asic-history addition 5 | Finish, through the existing ranking |
| B | Tensor-core PoW measured against a chip | Igneum did it (scheme B) | Yes, retired to reserve R8 with the two-output correction | 1 (new-pow.md rank 5) | Already done |
| C | A merged-mining BIP that replaces the 2011 spec; any child chain with a say in its parent | Namecoin, RSK | n/a (ruled out) | 0 | Skip |
| C | An algo slot or an Igneum-side auxiliary chain | Nobody should | No | 0 | Skip, with the reasons in 3.2 |
| D | Finality on PoW without a coin-bought validator set | Zcash Crosslink (not shipped), Kaspa DAGKnight (PRs) | Yes: miner-only 30-day weight, 20 days at 100 percent hashrate for 2/3 | 0 beyond Igneum's own open rows (LEAVE item, 10 percent rule, weight-gated deep fork choice, vote-or-burn: 0.3.16 and horizon 1) | Already done; finish the own rows |
| D | A subjective reweighting rule that survives its own chain (Horizen, MESS) | Nobody | Removed on purpose (no hidden-block n^2 penalty) | 0 | Skip |
| E | A pool with no operator on a DAG chain | Nobody has; Monero P2Pool is the finished form on a chain | Partly (vote key in the header, the 20 percent paid by sortition, coinbase extra data) | about 40 (section 5.3) | Finish |
| E | Miner-built templates (DATUM, SV2 JD: one block in 52,297) | Igneum's mode C | Designed, not built | 6 | Finish after the sidechain |
| E | Encrypted member transport | SV2 Noise | No (plain TCP, Q67) | 8 | Finish |
| E | A pool-as-contract with sampled share verification (SmartPool) | Nobody since 2017 | No | 0 | Skip (1.35 ms per share makes it worse than it was for Ethereum) |
| F | Node, wallet and GPU miner in one app, own key, signed updates | Nobody; Ember | Yes | 0 | Already done |
| F | Sixty seconds from download to first share | Honeyminer did it in 2018 | No (three prompts, no Windows build) | 6 plus certificates (Q3); 0 or 4 (Q80) | Finish; the certificates are Josh's |
| F | Earnings in IGN, per-card faults, an hour of history, a finality notice | HiveOS, lolMiner | No (Q18, Q13, Q16, Q2) | 11 | Finish |
| F | Payout in something a gamer can spend (Salad's gift cards) | Salad | No, and the chain says "nothing is bought or sold on devnet" | 0 | Skip until there is a market |
## 8. Three headline findings for the coordinator
1. Every proof-of-useful-work programme that passed both tests (cheap verify, sampleable) did so by making the work useless, and the one exception, Filecoin, ran at 29 to 36 percent utilisation and lost 55 percent of its raw byte power in the year to 14 Sep 2026; Igneum's split already finishes Aleo's 2019 promise, and the only unfinished line (a majority prover pool earning 11,636 IGN an hour at 51 percent) closes when `proving_consensus_verify_daa` is switched on.
2. No pool without an operator exists on any chain but Monero (P2Pool at 7.7 percent, 3,520 miners, 0 percent fee; Ocean at 2.05 percent of blocks with 85 percent of them from DATUM; Stratum V2 job declaration at 1 block in 52,297), and Igneum already holds the three objects that make one cheap (the vote key in the header, the 20 percent paid by sortition, the coinbase extra data), so a share sidechain is about 40 agent hours and would be the first on a DAG chain.
3. Every finality on a PoW chain that held in practice is a second validator set bought with coins (Decred 64 percent of supply staked, eCash 350 B XEC, Dash 1,000 DASH per masternode), and every rule that only reweights work has been switched off by its own chain (MESS twice, Horizen winding down, Kaspa's 12-hour depth subjective); Igneum's work-weighted set needs 20 days of 100 percent hashrate to reach 2/3, so no renter buys it in a day, and the unfinished items are Igneum's own (the LEAVE item and the 10 percent rule), not the industry's.
## 9. Sources
All accessed 7 October 2026 unless stated. Lane notes with the full per-item citations sit in the scratchpad under `mission-unfinished/` (A-notes, B-notes, CD-notes, E-notes, F-notes).
Group A
- https://en.wikipedia.org/wiki/Primecoin ; https://www.johndcook.com/blog/2026/01/10/primecoin-primality-test/ ; https://github.com/primecoin/primecoin/wiki/Primecoin-Reaches-13-primes-Milestone ; https://primecoin.io/ ; https://chainz.cryptoid.info/xpm/ ; https://www.coingecko.com/en/coins/primecoin ; https://bitinfocharts.com/primecoin/
- https://github.com/gridcoin-community/Gridcoin-Wiki/wiki/zArchive~Proof-of-Research ; https://gridcoin.us/assets/docs/scraper-summary.pdf ; https://gridcoin.us/guides/whitelist.htm ; https://github.com/gridcoin-community/Gridcoin-Research/releases/tag/5.5.0.0
- https://curecoin.net/knowledge-base/folding-for-curecoin/how-do-i-start-folding-for-curecoin-quick/ ; https://curecoin.net/news/curecoin-team-reaches-1-trillion-ppd-on-foldinghome/ ; https://curecoin.net/curecoin-roadmap/ ; https://tokenmarket.net/blockchain/counterparty/assets/foldingcoin/
- https://eprint.iacr.org/2020/190 ; https://aleo.org/post/aleo-mainnet-faq/ ; https://docs.aleo.org/participate/run-a-node/prover ; https://github.com/ProvableHQ/ARCs/discussions/97 ; https://provable.com/blog/introducing-arc-0046 ; https://aleo.org/post/announcing-snarkos-v4.0.0/ ; https://messari.io/report/state-of-aleo-q2-2025 (search excerpt; direct fetch 429) ; https://followin.io/en/feed/11596817
- https://eprint.iacr.org/2021/1379 ; https://eprint.iacr.org/2025/2091 ; https://eprint.iacr.org/2017/203 ; https://eprint.iacr.org/2018/559
- https://filecoin.io/blog/posts/a-guide-to-filecoin-storage-mining ; https://lotus.filecoin.io/storage-providers/get-started/hardware-requirements/ ; https://lotus.filecoin.io/storage-providers/operate/benchmarks/ ; https://filecointldr.io/article/key-trends-and-takeaways-from-filecoin-q1-2025 ; https://www.kucoin.com/news/flash/filecoin-q3-2025-report-capacity-drops-10-network-utilization-rises-to-36 ; https://thetradersspread.com/crypto/filecoin-rose-24-percent-ahead-of-75-percent-supply-cut ; https://github.com/filecoin-project/FIPs/discussions/1009 ; https://filecoin.io/blog/posts/introducing-proof-of-data-possession-pdp-verifiable-hot-storage-on-filecoin/
- https://www.chia.net/2022/03/19/chia-mainnet-year-one/ ; https://www.chia.net/2021/05/24/chia-and-ssd-endurance/ ; https://www.chia.net/2021/08/03/chia-and-ssd-endurance-big-progress-less-waste/ ; https://www.chia.net/2023/01/21/plotting-chias-future/ ; https://www.chia.net/2026/02/27/changes-coming-to-3-0/ ; https://lowendbox.com/blog/the-crypto-that-ate-all-the-hard-drives-whatever-happened-to-chia/ ; https://xch.today/2026/02/09/top-10-chia-predictions-for-2026/ ; https://en.wikipedia.org/wiki/Chia_Network
- https://ownyourmind.ai/projects/flux/ ; https://runonflux.com/ ; https://messari.io/report/state-of-akash-q1-2026-final (search excerpt) ; https://akash.network/blog/akash-network-q1-2026-report/
- https://github.com/dynexcoin/Dynex ; https://www.coingecko.com/en/coins/dynex ; https://www.coinlore.com/coin/dynex/news
- https://qubic.org/blog-detail/qubic-s-useful-proof-of-work-the-future-of-ai-compute ; https://qubic.org/blog-detail/the-next-evolution-of-useful-proof-of-work-(upow) ; https://www.halborn.com/blog/post/explained-the-monero-51-percent-attack-august-2025 ; https://www.theblock.co/post/366535/monero-faces-chain-reorganization-fears-after-qubic-says-it-controls-51-of-hashrate ; https://cointelegraph.com/news/monero-qubic-selfish-mining-51-percent-attack ; https://www.bitget.com/news/detail/12560604967511 ; https://wublock.substack.com/p/monero-hit-by-a-51-hashrate-attack
- https://arxiv.org/html/2507.02951v1 ; https://www.theblock.co/news/web3/2026-03-05-dcg-yuma-bittensor-report-392351 ; https://abittensorjourney.com/p/navigating-bittensor-september-2026 ; https://beincrypto.com/bittensor-subnets-reach-ath-in-may/
- https://arxiv.org/abs/2502.19405 ; https://www.gensyn.ai/research/verde-verification-system-in-production ; https://arxiv.org/abs/2405.00295 ; https://blockeden.xyz/blog/2026/04/01/ambient-ai-l1-proof-of-logits-pow-blockchain-decentralized-inference/ ; https://inferencelabs.com/ ; https://arxiv.org/abs/2603.19025 ; https://arxiv.org/pdf/2512.20176 ; https://blockeden.xyz/blog/2026/01/14/boundless-risc-zero-decentralized-proof-market-zk/ ; https://blog.succinct.xyz/succinct-2025-recap/
- https://arxiv.org/abs/2209.03865 ; https://eprint.iacr.org/2025/1814 ; https://arxiv.org/abs/2606.06700 ; https://arxiv.org/abs/2606.04819 ; https://arxiv.org/abs/2510.09729
Group B
- https://eips.ethereum.org/EIPS/eip-969 ; https://eips.ethereum.org/EIPS/eip-1057 ; https://ethereum.org/en/developers/docs/consensus-mechanisms/pow/mining-algorithms/ethash/ ; https://github.com/eth-classic/etchash ; https://2miners.com/blog/how-to-mine-ethereum-and-ethereum-classic-on-4gb-gpus/
- https://leastauthority.com/static/publications/LeastAuthority-ProgPow-Algorithm-Final-Audit-Report.pdf ; https://github.com/ethcatherders/progpow-audit (PDF text not extractable) ; https://github.com/kik/progpow-exploit ; https://github.com/ethereum/EIPs/issues/2640 ; https://hudsonjameson.com/posts/2020-03-02-progpow-the-ethereum-community-speaks/ ; https://cointelegraph.com/news/the-history-of-the-bitter-debate-over-ethereums-progpow ; https://www.theblock.co/linked/57061/ethereum-community-members-submit-dissenting-progpow-petition ; https://www.coindesk.com/tech/2020/03/05/ethereums-progpow-debate-is-about-much-more-than-mining ; https://www.coindesk.com/tech/2020/03/06/ethereums-progpow-call-features-frustration-but-little-progress ; https://github.com/Souptacular/linzhi ; https://linzhi.io/docs/LWP15-Posts-Against-ProgPoW-05092019.pdf ; https://cryptoslate.com/progpow-authors-steps-down-core-scientific-ethereum/ ; https://ecips.ethereumclassic.org/ECIPs/ecip-1070 ; https://2miners.com/blog/kawpow-new-ravencoin-mining-algorithm/ ; https://en.wikipedia.org/wiki/Firo_(cryptocurrency) ; https://www.qu.ai/blog/quai-network-mainnet-launching-january-29th
- https://github.com/tevador/RandomX ; https://github.com/tevador/RandomX/pull/265 ; https://github.com/tevador/RandomX/pull/317 ; https://github.com/monero-project/monero/pull/10038 ; https://github.com/monero-project/research-lab/issues/136 ; https://xmrig.com/docs/algorithms ; https://github.com/xmrig/xmrig-cuda/issues/67 ; https://www.asicminervalue.com/miners/bitmain/antminer-x5 ; https://www.asicminervalue.com/miners/bitmain/antminer-x9 ; https://millionminer.com/news/monero-mining-guide-2026-antminer-x5-x9-pinecone-r1x-randomx (reseller, unverified) ; https://www.coindesk.com/business/2025/08/12/monero-s-51-attack-problem-inside-qubic-s-controversial-network-takeover
- https://arxiv.org/abs/1911.05193 ; https://kaspa-lens.com/kaspa-wiki/kaspa-technology-and-features/kheavyhash-proof-of-work-algorithm/ ; https://kaspa-lens.com/kaspa-wiki/getting-started-with-kaspa/mining-kaspa/ ; https://www.asicminervalue.com/miners/iceriver/ks0 ; https://github.com/karlsen-network/karlsend ; https://github.com/spectre-project/rusty-spectre ; https://github.com/Pyrinpyi/pyipad
- https://docs.ergoplatform.com/mining/autolykos/ ; https://www.ergoforum.org/t/autolykos-v-2-details/480 ; https://github.com/ergoplatform/eips/blob/master/eip-0027.md ; https://eprint.iacr.org/2020/044 ; https://x.com/RedPandaMining/status/1697408861014196711 ; https://cryptoage.com/en/3019-osprey-e300-fpga-miner-for-kaspa-cryptocurrency.html
- https://github.com/mimblewimble/grin/blob/master/doc/pow/pow.md ; https://docs.grin.mw/about-grin/proof-of-work/ ; https://forum.grin.mw/t/grin-v5-0-0-network-upgrade-hard-fork-4-january-2021/7895 ; https://2miners.com/blog/grin-version-4-hard-fork-what-will-change-and-how-to-get-ready/ ; https://forum.grin.mw/t/grn1-cancellation-announcement/5624 ; https://voskcointalk.com/t/new-grin-miner-coming-out/6373 ; https://www.asicminervalue.com/miners/innosilicon/g32-500 ; https://www.asicminervalue.com/miners/ipollo/g1 ; https://2cryptocalc.com/cuckatoo32-algorithm ; https://github.com/tromp/cuckoo
- https://coinbureau.com/mining/battle-against-asics-antminer-z9-zcash ; https://github.com/ZcashFoundation/zfnd/blob/master/_posts/blog/2018-05-08-statement-on-asics.md ; https://raw.githubusercontent.com/ZcashFoundation/zfnd/master/_posts/blog/2018-06-07-asic-equihash-study.md ; https://raw.githubusercontent.com/ZcashFoundation/zfnd/master/_posts/blog/2018-07-03-governance-results.md ; https://forum.zcashcommunity.com/t/will-zcash-fork-to-be-asic-resistant/30829 ; https://crypto.news/zcash-doubles-efforts-on-investigating-asic-resistance/ ; https://github.com/BeamMW/beam/wiki/BEAM-Mining ; https://www.minerupdate.com/news/trending-news/beam-hard-fork-executes-to-deter-asic-miners ; https://u.today/privacy-coin-beam-introduces-new-proof-of-work-algorithm ; https://cryptoage.com/en/2095-beam-cryptocurrency-hard-fork-new-beamhash3-mining-algorithm.html ; https://en.wikipedia.org/wiki/Bitcoin_Gold ; https://www.tweaktown.com/news/62048/bitcoin-gold-hit-51-attack-up-18-million-gone/index.html
- https://github.com/Conflux-Chain/CIPs/blob/master/CIPs/cip-102.md ; https://forum.conflux.fun/t/cip-102-change-pow-mining-algorithm/16068 ; https://f2pool.io/mining/guides/how-to-mine-conflux/ ; https://www.asicminervalue.com/coins/conflux-cfx
- https://docs.alephium.org/frequently-asked-questions/ ; https://docs.alephium.org/mining/ ; https://www.asicminervalue.com/miners/goldshell/al-box ; https://www.asicminervalue.com/miners/bitmain/antminer-al1
- https://eprint.iacr.org/2025/685 ; https://ethresear.ch/search.json?q=tensor%20core%20proof%20of%20work
- https://arxiv.org/abs/1905.08792 ; https://medium.com/@dave_46855/altered-silicon-x16r-mining-on-the-xilinx-fpga-dbc290c56909 (search excerpt; fetch 403) ; https://forum.zcashcommunity.com/t/equihash-mining-in-fpga/31054 ; https://cryptoage.com/en/3195-all-hashaltcoin-fpgas-get-support-for-the-kheavyhash-algorithm.html
Group C
- https://github.com/namecoin/wiki/blob/master/Merged-Mining.mediawiki ; https://www.namecoin.org/docs/faq/ ; https://en.bitcoin.it/wiki/Merged_mining_specification ; https://metrics.namecoin.org/namecoin/period-timestamps-14-days/pool/charts/latest.txt ; https://forum.namecoin.org/viewtopic.php?f=6&t=2421 (search excerpt; fetch 500) ; https://cryptorank.io/news/feed/7620f-while-bitcoins-hashrate-remains-sky-high-merge-mined-crypto-asset-networks-benefit (search excerpt; fetch 403) ; https://eprint.iacr.org/2017/791
- https://www.coindesk.com/markets/2014/08/04/dogecoin-to-allow-litecoin-merge-mining-in-network-security-bid ; https://www.binance.com/en/research/analysis/merged-mining (search excerpt) ; https://www.bitdeer.com/learn/dogecoin-the-logic-of-auxpow-and-stable-profits ; https://www.mexc.com/learn/article/dogecoin-mining-2025-how-doge-is-mined-merge-mining-and-miner-economics/1
- https://myriadteam.github.io/ ; https://www.pxdojo.net/2017/04/myriadcoin-untold-story-of-invincible.html ; https://thenextweb.com/news/hackers-verge-blockchain-steal-1-7m ; https://bitcoinist.com/verge-xvg-hacked-35-million-xvg-tokens-reportedly-generated-hacker/ ; https://en.wikipedia.org/wiki/Verge_(cryptocurrency) ; https://cryptonary.com/verge-suffers-reorg-attack-200-days-of-transactions-wiped-away/ (headline only)
- https://solofury.com/blog/digibyte-dgb-explained-for-miners/ ; https://www.gemini.com/cryptopedia/digibyte-mining-digibyte-coin-dgb-coin
- https://cryptobriefing.com/elastos-bitmain-merged-mining/ ; https://blog.elastos.net/news/ela-queen-of-bitcoin/ ; https://rootstock.io/blog/merged-mining-insights-report-q1-2024/ ; https://rootstock.io/blog/rootstock-x-bitcoin-merged-mining-insights-report-q4-2024/ ; https://rootstock.io/blog/the-security-architecture-of-rootstock-and-the-principle-of-defense/ ; https://github.com/rsksmart/RSKIPs/blob/master/IPs/RSKIP110.md
Group D
- https://docs.decred.org/proof-of-stake/overview/ ; https://docs.decred.org/research/hybrid-design/ ; https://github.com/decred/dcps/blob/master/dcp-0011/dcp-0011.mediawiki ; https://github.com/decred/dcps/blob/master/dcp-0012/dcp-0012.mediawiki ; https://xaur.github.io/decred-news/journal/201912 ; https://xaur.github.io/decred-news/journal/202012 ; https://xaur.github.io/decred-news/journal/202112 ; https://xaur.github.io/decred-news/journal/202212
- https://www.coindesk.com/tech/2018/10/10/a-solution-to-cryptos-51-attack-fine-miners-before-it-happens ; https://blog.horizen.io/horizens-51-attack-solution/ (search excerpt; fetch 403) ; https://www.horizen.io/ ; https://docs.horizen.io/
- https://raw.githubusercontent.com/kaspanet/rusty-kaspa/master/consensus/core/src/config/constants.rs ; https://raw.githubusercontent.com/kaspanet/rusty-kaspa/master/consensus/core/src/config/bps.rs ; https://raw.githubusercontent.com/kaspanet/rusty-kaspa/master/consensus/core/src/config/params.rs ; https://eprint.iacr.org/2022/1494 ; https://github.com/kaspanet/rusty-kaspa/pull/1104 (search excerpt)
- https://arxiv.org/abs/2006.01072 ; https://www.usenix.org/conference/atc20/presentation/li-chenxing ; https://doc.confluxnetwork.org/docs/general/hardforks/v2.0 ; https://doc.confluxnetwork.org/docs/general/conflux-basics/consensus-mechanisms/proof-of-stake/pos_overview
- https://arxiv.org/abs/1710.09437 ; https://eips.ethereum.org/EIPS/eip-1011 ; https://eth2book.info/latest/part2/consensus/overview/ ; https://hackmd.io/@tvanepps/Beacon-Book ; https://ethresear.ch/t/convenience-link-to-casper-sharding-chain-v2-1-spec/2332 ; https://github.com/ethereum/pm/blob/master/AllCoreDevs-EL-Meetings/Meeting%2041.md
- https://e.cash/blog/ecash-day-2024-celebrating-three-years-of-ecash-xec ; https://e.cash/blog/eCash-day-2026 ; https://e.cash/blog/preconsensus-launch ; https://e.cash/blog/staking-guide ; https://e.cash/blog/avalanche-faq
- https://meowsbits.github.io/51-percent-docs/ ; https://ecips.ethereumclassic.org/ECIPs/ecip-1100 ; https://ecips.ethereumclassic.org/ECIPs/ecip-1110 ; https://www.kucoin.com/news/flash/etc-miners-revert-to-argos-after-core-geth-v1-13-0-controversy ; https://github.com/ethereumclassic/core-geth/issues/44 ; https://github.com/ethereumclassic/core-geth/pull/53
- https://electric-coin-company.github.io/tfl-book/design/crosslink.html ; https://shieldedlabs.net/crosslink/ ; https://zechub.substack.com/p/arborist-call-88-r-and-d-updates
- https://eprint.iacr.org/2020/1101 ; https://www.nervos.org/mining (search excerpt; fetch 429)
- https://docs.dash.org/projects/core/en/21.0.0/docs/guide/dash-features-chainlocks.html ; https://syscoincore.org/en/releases/4.2.0/ ; https://komodoplatform.com/en/docs/start-here/core-technology-discussions/delayed-proof-of-work/
Group E
- https://en.bitcoin.it/wiki/P2Pool ; https://github.com/p2pool/p2pool ; https://github.com/p2pool/p2pool/commits/master ; https://github.com/treib-holdings/p2pool.org/issues/1 ; https://github.com/frstrtr/c2pool
- https://github.com/SChernykh/p2pool ; https://github.com/SChernykh/p2pool/blob/master/README.md ; https://github.com/SChernykh/p2pool/releases ; https://monero.observer/schernykh-releases-p2pool-v4.7-support-nano-sidechain/ ; https://www.getmonero.org/resources/moneropedia/p2pool.html ; https://data.miningpoolstats.stream/data/monero.js?t=1791359405 ; https://xmrchain.net/api/networkinfo ; https://supportxmr.com/api/pool/stats ; https://api.hashvault.pro/v3/monero/pool/stats ; https://p2pool.observer/ (502 on every try)
- https://github.com/TheBlueMatt/bips/blob/betterhash/bip-XXXX.mediawiki ; https://github.com/stratum-mining/sv2-spec ; https://github.com/stratum-mining/stratum/releases ; https://stratumprotocol.org/ ; https://braiins.com/stratum-v2 ; https://braiins.com/pool ; https://d-central.tech/data/stratum-protocol-matrix/ ; https://d-central.tech/bitcoin-mining-pool-comparison-2026/ ; https://www.spark.money/research/bitcoin-stratum-v2-mining-decentralization ; https://cryptobriefing.com/demand-pool-first-stratum-v2-block/ ; https://www.tftc.io/stratum-v2-job-declaration-first-block-dmnd-gomining/ ; https://www.dmnd.work/ ; https://mempool.space/api/v1/mining/pools/1w ; https://mempool.space/api/v1/mining/pools/1y
- https://ocean.xyz/docs/tides ; https://ocean.xyz/docs/datum ; https://ocean.xyz/dashboard ; https://bitcoinmagazine.com/technical/an-ocean-launch-post-mortem ; https://crypto.news/bitcoin-mining-divide-pushes-luke-dashjr-out-of-ocean/
- https://data.miningpoolstats.stream/data/kaspa.js?t=1791359405 ; https://2miners.com/kas-mining-pool ; https://woolypooly.com/en/coin/kas ; https://f2pool.io/mining/guides/how-to-mine-kaspa/ ; https://kaspa.herominers.com/?lang=en ; https://solopool.org/
- https://eprint.iacr.org/2017/019 ; https://blog.acolyer.org/2017/09/07/smartpool-practical-decentralized-mining/ ; https://eprint.iacr.org/2020/044
- https://miningpoolstats.net/coins/ravencoin/ ; https://2miners.com/erg-mining-pool
- https://github.com/dogestreet/proxypool ; https://arxiv.org/pdf/1112.4980 ; https://arxiv.org/pdf/1402.1718 ; https://arxiv.org/abs/1411.7099 ; https://eprint.iacr.org/2015/155 ; https://en.bitcoin.it/wiki/Eligius
Group F
- https://en.wikipedia.org/wiki/NiceHash ; https://www.nicehash.com/support/general-help/nicehash-service/fees-and-pricing (deposit fees only; the seller and buyer fee pages returned 404)
- https://hiveon.com/pricing/
- https://www.awesomeminer.com/ ; https://minerstat.com/ (no pricing on either page fetched)
- https://salad.com/ ; https://support.salad.com/article/213-how-much-can-i-earn-with-salad (no take rate published)
- https://www.kryptex.com/ ; https://pool.kryptex.com/
- https://unmineable.com/
- https://www.getmonero.org/resources/user-guides/solo_mine_GUI.html ; https://github.com/monero-project/monero-gui/releases
- https://github.com/Lolliedieb/lolMiner-releases ; https://github.com/bzminer/bzminer ; https://github.com/develsoftware/GMinerRelease ; https://github.com/trexminer/T-Rex ; https://github.com/NebuTech/NBMiner
- Unreachable on 7 Oct 2026: honeyminer.com (connection refused), docs.cudominer.com (no DNS), cudominer.com (429), web.archive.org (blocked for this tool)
Igneum documents cited
- `CLAUDE.md` (the design paragraph, finality rule v2, the fee lines, the 6 October 2026 rules)
- `docs/analysis/asic-resistance-history.md` (rows 3, 17 to 25, 30; lessons 5 to 9; sections 4.1 to 4.3)
- `docs/plans/pool.md` sections 2, 3, 5, 7; `docs/spec/09-pool-protocol.md` sections 9.1 to 9.8
- Scratch inputs: `mission-inputs/horizon/new-pow.md` sections 3.1, 3.3, 6, 7; `mission-inputs/horizon/frontier.md` sections 0, 3.10, 3.11; `mission-inputs/horizon-2026-10.md` section 1; `mission-inputs/horizon/polish.md` sections 3.1, 3.6, 3.10