Public — correction log, archive
Corrections CC-2026-0021 to CC-2026-0040
Factual errors, misapplied scoring, stale data and claims the site could not back. Each entry says what was wrong, what was fixed, how it was caught and whether the published order moved. An entry whose logged date is later than its corrected date was copied here from the methodology or data changelog that first published the fix. No entry is ever deleted.
This page prints the corrections numbered CC-2026-0021 to CC-2026-0040, in full. It is closed: no entry is added to it or taken from it. 20 corrections, logged 2026-09-14 to 2026-09-30. Earlier page · All archive pages · Later page
Four leverage venues were tagged as available in the United States with no source, while their own terms or app geo-blocks exclude US residents (Aster, Vertex, Drift, 3Jane), so a reader who stated the US could be offered a venue they may not use. OKX, Bitso, CoinDCX and Luno were offered in modelled markets their terms exclude or do not serve, and Stryke’s options vaults were ranked as a live product although Stryke sunset them in February 2024.
Each region tag was read from the provider’s own terms or geo-block, quoted in its availability note, and re-fetched by a second model run; the excluded modelled countries are carried in restrictedMarkets, and Stryke’s row is recorded as sunset from its own announcement.
The full-coverage audit and WP-05 step 9, which counted leverage rows whose region tag no page sourced (7).
Only below the #1: Stryke fell from 45 to 41 in options_vaults. Country pages list 48 fewer rows.
Obligate was ranked #1 among tokenisation platforms with a control-surface score its own pages do not support: the top rung needs a published repository or a deployed contract address, and Obligate names its token standard without linking either. In Lightning payment infrastructure, Breez, Amboss and Greenlight were scored on pricing and status pages that have since been withdrawn or no longer list the service, while Strike and Voltage were under-credited for evidence they already published.
Both categories were re-read from the providers’ own pages on 2026-09-30, every quote re-fetched and matched, every score placed against its ladder by a model run reconciling them adversarially, and the move that changed the tokenisation #1 re-checked by a second run, which upheld it. The scores follow the evidence still published.
A scheduled cadence re-read ahead of the categories’ 2026-10-26/27 review dates (RS-10b).
Yes. One #1 changed: tokenisation_platforms (Obligate to Taurus, by one point on the published scale). In lightning_payment_infrastructure the order changed below the unchanged top three and five published scores moved. Corrected 2026-10-14: every published row that moved. lightning_payment_infrastructure: Blink (Galoy) #5 → #4 at 75; Voltage 70 → 73 (#7 → #5); Amboss 75 → 72 (#4 → #6); Strike 60 → 69 (#13 → #7); Blockstream Greenlight 72 → 65 (#6 → #8); Breez SDK 67 → 62 (#8 → #13). tokenisation_platforms: Taurus 80 → 81 (#2 → #1); Obligate 81 → 80 (#1 → #2); InvestaX 68 → 67; Tokeny 64 → 65; Brickken 61 → 62.
Two providers were scored on documentation that no longer exists. Entendre was credited with a published cost-basis method setting, a balance-reconciliation report and a step-by-step NetSuite procedure; after its acquisition by MoonPay the API documentation was replaced and those pages return 404. Misty Breez was credited with inbound Lightning capacity and a priced exit to chain after Breez had announced that the service it relies on is gone and Lightning and on-chain payments can no longer be made in the app.
Both rows were re-read from the providers’ own pages on 2026-09-30, every quote re-fetched and matched verbatim, and every score placed against its ladder by a model run reconciling them adversarially. The scores follow the evidence that is still published, the tagline that repeated Entendre’s old documentation was rewritten, and Misty Breez is marked “unclear” with its wind-down stated in its watchouts.
A scheduled cadence re-read ahead of the categories’ 2026-10-26/27 review dates (RS-10a).
Yes, below the top three: Entendre 76 to 63, Misty Breez 60 to 44. The #1s (Cryptoworth, Blink) are unchanged. Corrected 2026-10-14: the rows the order moved below the top three, at an unchanged score: in enterprise_accounting Cryptio #5 → #4, Wag3s #6 → #5 and Ledgible #7 → #6; in lightning_wallets Cash App #12 → #11 and Primal Wallet #13 → #12.
Three hardware wallets carried a design fact their makers’ own pages no longer support: Coldcard was shown with a clean record after Coinkite disclosed that a July 2026 firmware bug let attackers steal funds; Keystone was credited with an EAL5+ secure element its own spec table does not claim; and NGRAVE was called the only EAL7-certified device when the EAL7 belongs to its operating system. Coin counts, prices and connectivity were also out of date on several rows.
Every fact was re-read from the makers’ own pages and adversarially reconciled; the Coldcard incident was re-checked by hand on Coinkite’s security-status page before it was applied. The scores follow from the corrected facts, and the prose that repeated the old claims was rewritten.
A scheduled cadence re-read ahead of the category’s 2026-10-20 review date, run for the first time through a structural-facts pipeline built for categories scored from design facts.
Yes, below the top three: Keystone 70 to 63, Coldcard 65 to 50, NGRAVE 59 to 48. The #1 is unchanged. Corrected 2026-10-14: the rows the order moved at an unchanged score: Ledger #5 → #4, Cypherock X1 #7 → #6, Tangem #8 → #7, Blockstream Jade #10 → #8 and SafePal S1 #11 → #9.
30 published criterion scores across the five categories no longer matched what the providers’ own pages, APIs and contracts supported, three of the categories had not been read for 181 days (past the 45-day cadence every methodology page promises), two bridges could not be shown operating as ranked, three typed fee rates had no remaining source, and 76 sentences of pool text stated as current facts that had since changed.
Every row was re-read against primary sources; every quote was re-fetched and matched, every score placed against its ladder, and each move that could change a top three re-checked by an independent skeptical model run, which refuted one move by the reconciling run (it is not applied). The unsourced rates were removed, the rows not shown operating carry dated watchouts, and the false pool text was corrected.
The build gate on the published review cadence (check-review-cadence-is-kept) carried these three categories as named overdue debt; the other two were scheduled ahead of their 2026-10-19 due date.
Yes. One #1 changed: token_launch_platforms (Bags to Raydium LaunchLab). The order changed in 4 of the 5 categories and 36 published scores moved. Corrected 2026-10-14: every published score in bridges, web3_apps, liquid_staking, stablecoin_issuance and token_launch_platforms was recomputed from the re-read. The rows whose score or rank moved are summarised here, not each named.
100 published criterion scores across the twenty-six categories due for re-read no longer matched what the providers’ own pages, filings, registers and contracts supported. One provider had wound down, one had changed its name back, and 333 sentences of pool text stated as current facts that had since changed.
Every row was re-read against primary sources; every quote was re-fetched and matched, every score placed against its ladder, and each move that could change a #1 re-checked by an independent skeptical model run, which refuted one move by the reconciling run (it is not applied). False pool text was corrected, the wound-down provider dated, and three dead links replaced.
A scheduled re-read ahead of the 9–11 October review dates, with a quote-matching pass, a ladder pass and a winner-move verification wherever the #1 could move.
Yes. One #1 changed: business_banking (Bank Frick to AMINA Bank AG). The order changed in 18 of the 26 categories and 106 published scores moved. Corrected 2026-10-14: every published score in subscriptions, business_banking, wallet, crypto_card, staking, lending_borrowing, token_standards, socialfi, trading_tools, incentive_distribution, key_recovery, restaking, creator_mint_platforms, perp_house_vaults, institutional_execution, onramp, portfolio_tracking, paid_research_seats, institutional_custody, exchange, stablecoins, tax_software, crypto_backed_loans, mining_pools, tokenised_bonds and tokenised_equities was recomputed from the re-read. The rows whose score or rank moved are summarised here, not each named.
90 published criterion scores across the seventeen categories due for re-read no longer matched what the providers’ own pages, filings and contracts supported. Two chains had stopped or been rolled back, and several taglines and watchouts stated as current facts that had since changed.
Every row was re-read against primary sources; every quote was re-fetched and matched, every score placed against its ladder, and each move that changes a #1 re-checked by independent skeptical model runs. False pool text was corrected and the two sunset chains dated.
A scheduled re-read ahead of the 3–6 October review dates, with a quote-matching pass, a ladder pass and a winner-move verification on every category.
Yes. Five #1s changed: fixed_rate_yield (Pendle to Exactly Protocol), bill_pay (Living Room of Satoshi to Bitcoin Well), carbon_credits (Carbonmark to Regen Network), proof_of_personhood (Self to Idena) and cross_chain_messaging (Union to Hyperlane). The order changed in 13 of the 17 categories and 90 published scores moved. Corrected 2026-10-14: every published score in cross_chain_messaging, intents, proof_of_personhood, sidechains, relayers, appchains_l3s, liquidity_management, structured_products, carbon_credits, remittances, dev_frameworks, stablecoin_settlement, liquid_lockers, fixed_rate_yield, bill_pay, nft_analytics and automation_keepers was recomputed from the re-read. The rows whose score or rank moved are summarised here, not each named.
50 published criterion scores across the seven categories due for re-read no longer matched what the providers’ own pages, filings or contracts supported. Several August reads had never loaded the page that decided the score, and several watchouts stated as current facts that had since changed.
Every row was re-read against primary sources; every quote was re-fetched and matched character for character, and every score was placed against the criterion’s published description. False present-tense statements in the pool text were corrected or re-dated.
A scheduled re-read ahead of the 29 September review date, with a quote-matching pass and an independent ladder pass on every category.
Yes. Three #1s changed: launchpads (Fjord Foundry to DAO Maker), attestations (Ethereum Attestation Service to Sign Protocol) and account_abstraction_infra (Pimlico to Candide, on the published tie-break). The order changed in 6 of the 7 categories and 39 published scores moved. Corrected 2026-10-14: every published score in account_abstraction_infra, sentiment_analytics, private_credit, flash_loans, attestations, launchpads and money_market_funds was recomputed from the re-read. The rows whose score or rank moved are summarised here, not each named.
78 published criterion scores across the thirteen categories due for re-read no longer matched what the providers’ own pages supported, and nine rows described venues that have since wound down, pivoted or gone quiet. The largest gaps were JPEG’d (11 points), then Bisq, Stryke, NFTfi, Kaiko and Dune (5 each).
Every row was re-read against the provider’s own pages or on-chain state; a mechanical pass re-fetched every cited page and matched each quote character for character, and a ladder pass placed each score against the criterion’s published description. Each changed score is set against a verbatim source; pages that could not be read keep their score and are listed as unverified in the row.
A scheduled re-read ahead of the 28 September review date, with a quote-matching pass and an independent ladder pass on every category.
Yes. One #1 changed: options_vaults (Typus Finance to SOFA.org, by two weighted points). The order changed in 9 of the 13 categories and 52 published scores moved (corrected 2026-10-14 from 51, a miscount: the diff at its commit moves 52); no category outside the thirteen changed. Corrected 2026-10-14: every published score in gift_cards, otc_desks, p2p_exchanges, wrapped_assets, screeners, nft_lending, governance_tools, shared_sequencers, rollup_as_a_service, market_data_apis, seed_backup, options_vaults and tokenised_commodities was recomputed from the re-read. The rows whose score or rank moved are summarised here, not each named.
66 published criterion scores across the eleven categories due for re-read did not match what the providers’ own pages supported, and several availability statements quoted no source. The largest gaps were Reflexer RAI (16 points), Beefy and Unrekt (11) and Aztec (10).
Every row was re-read against the provider’s own pricing, terms, docs and status feeds, with a second model run re-fetching every quote; each changed score is set against a verbatim source, and eleven availability notes now quote the terms they rest on.
A scheduled re-read ahead of the 25 September review date, with an adversarial model run re-fetching every quote.
Yes. One #1 changed on the re-read: kyc_compliance (Veriff to Sumsub). The same release carried dimension scores to tenths (methodology 2026.09.15), which moved two further #1s — smart_contract_audits (Zellic to Certora) and travel_rule_compliance (Sumsub Travel Rule to CodeVASP) — and the order in 38 of 119 categories.
130 published criterion scores across 14 categories did not match what the providers’ own pages supported, several availability statements did not match the providers’ terms, and several prices were ones the vendors no longer publish.
All 105 rows were re-read and each changed score set against a verbatim source; availability now follows each provider’s terms, prices follow the vendors’ current pages, and unsupported pricing and availability text was removed or replaced with quoted text.
A scheduled re-read ahead of the 25 September review date, with an adversarial model run re-fetching every quote.
Yes. Four #1s changed: smart_accounts (Safe to Squads Protocol), crypto_neobanks (Xapo Bank to Revolut), cross_chain_swaps (Relay to Across) and onchain_monitoring (BlockSec Phalcon to Hypernative); 78 published scores moved, and the order moved in 12 of the 14 categories.
The Rise record published a 3% percentage-of-volume fee, read on 2026-08-03. On 2026-09-14 the Rise pricing page showed a $49 per-seat price and no percentage fee.
The rate was removed rather than kept as a number the vendor no longer states; crypto_payroll now has one structured rate.
The 2026-09-14 re-read of crypto_payroll ahead of its review date (methodology 2026.09.13).
No. The crypto_payroll order is unchanged; the category no longer has a comparable pricing basis, so its cost column and cost-based tie factor no longer apply.
114 published criterion scores across 15 categories did not match what the providers’ own pages supported, and several availability statements did not match the providers’ terms.
All 122 rows were re-read and each changed score set against a verbatim source; availability now follows each provider’s terms, and unsupported pricing and availability text was removed or replaced with quoted text.
A scheduled re-read ahead of the 25 September review date, with an adversarial model run re-fetching every quote.
Yes. Four #1s changed: simple_buy (Relai to Strike), stablecoin_yield (Ondo USDY to Spiko), smart_contract_audits (Certora to Zellic) and money_market_funds (Fidelity Treasury Digital Fund to Spiko US T-Bills Money Market Fund); scores moved in 16 categories.
Personalised dimension weights were rounded to one decimal each, which changed the ratio between dimensions the reader had not touched: a trust priority turned objectiveFit and quantitative 30:18 into 27.4:16.5.
Personalised weights are no longer rounded, so a stated priority can no longer reorder rows on criteria nobody re-weighted.
The decision-board-reweight test (commit 64e82c02).
No. Published weights are whole numbers, so no published order or score changed from this fix.
/methodology presented four named bodies, each "joined 2026-01-01", as the people behind every score, and the methodology changelog named two of them as the approver of published versions. None of them existed.
The roster now lists the people who exist and the build scripts that run, and every approver in the changelog and this log resolves to a person (owner decision O-07).
The comparison audit behind docs/path-to-ten.md, programme W1-10.
No. The plan records W1-10 as not moving any ranking.
/why-trust-us, /editorial-policy, /affiliate-disclosure, /stats and /performance stated things nothing on the site did, among them corrections "within 24 hours", a quarterly accuracy report, a public archive, "Effective: January 2024" and "Methodology version v1.0".
The five pages were rewritten from engine facts, and the translated copies that kept the old claims were deleted so the corrected English renders in every locale.
The comparison audit behind docs/path-to-ten.md, programme W1-09.
No. The plan records W1-09 as not moving any ranking.
Ten /recommended guides printed a hand-typed "#1", answered their question with "Editor’s pick", sent the typed order to Google as ItemList positions and stamped it as uninfluenced by affiliate economics, with rank-1 picks that included providers holding a pending partner programme.
The guides dropped typed ranks, ItemList positions, "#1" and "Editor’s pick", lead with the published #1 line and list options unnumbered; the independence stamp renders only over lists in published order (commit 9aaa7d45).
The comparison audit behind docs/path-to-ten.md, programme W1-06 (refusal R6).
The published ranking did not move; the guides no longer print an order of their own.
/hardware-wallets listed Ledger first and sent that order to Google as "Best Hardware Wallets 2026", while Ledger was the only live paying programme and sixth in the published hardware_wallet ranking.
The list, its ItemList and each use case’s options follow the published score, and the page no longer claims enforced independence for an editorial list (commit c321fbc8).
The comparison audit behind docs/path-to-ten.md, programme W1-06 (refusal R6).
The published ranking did not move; the page’s list now follows it.
rankings.json said for every tie that the published criteria did not separate the two rows and named pricing and evidence depth as the deciders; the criterion scores differed in 18 of the 20 ties, and the evidence-depth factor no longer existed.
Each tie note is computed from the two rows, names its kind, both unrounded weighted criterion averages and the row the criteria alone prefer, and formula.tie
The comparison audit behind docs/path-to-ten.md, programme W1-12.
No. No order or score changed.
formula.step2_absoluteScore in rankings.json still described the removed "<= 20 ? round(40 + weightedTotal)" branch, not the continuous transform the engine has run since commit af882424.
The file states the transform the engine runs and publishes engine
The comparison audit behind docs/path-to-ten.md, programme W1-11.
No. No order or score changed; no published row had reached the removed branch.
This page prints the corrections numbered CC-2026-0021 to CC-2026-0040, in full. It is closed: no entry is added to it or taken from it. 20 corrections, logged 2026-09-14 to 2026-09-30. Earlier page · All archive pages · Later page