Public — correction log, archive
Corrections CC-2026-0041 to CC-2026-0060
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-2026-0041 to CC-2026-0060, in full. It is closed: no entry is added to it or taken from it. 20 corrections, logged 2026-10-01 to 2026-10-11. Earlier page · All archive pages · Later page
A further 90 refusal reasons were published in the research notes' own voice: "VERIFIED: I fetched …", what an earlier draft had said and how it was corrected, "both analysts", "the judge", "adjudicated", internal ids and file names. Several also quoted things a reader could not find or that had changed. The CertiK Skynet refusal quoted the Smart-contract Audits record with a sentence that record no longer carries; the 0G Storage refusal quoted an older wording of the 0G DA note; the Angle refusal quoted a code comment as if it were the Merkl record. Two refusal names were stale: Pekuna's said the catalog "ranks it today" (it does not), and the Australian Crypto Tax Calculator listing was still called "the pool's most damaging defect" after it had been folded into the ranked Summ row. The Yellow Card refusal repeated an aggregator integration it marked unverified.
Each refusal says what was checked, when, and what it showed, in the published category names, with the quotes replaced by what the records say now and the unverified claim removed. The Wrapped Assets LBTC row's summary and watchout, which still described the Babylon staking LBTC no longer uses (an 8% rewards fee, a slashing exposure, an unbonding queue), now describe the covered-call strategy and the 20% fee on its gains that the row's own 24 September re-read recorded. The list of refusals still in that voice is empty, and a refusal written in that voice now fails the build.
Burning down the list that check-exclusions-published started on 2026-10-11 (CC-2026-0059).
No. No published score or rank changed.
Refusal reasons ("Considered and not ranked") were published as the research notes had them: source-file paths and line numbers, internal ids, and its own vocabulary ("Proposal B", "the judge", "this rebuild"). Some of that text was also out of date or wrong. Trading tools said the category "currently scores" automation capability and exchange-API coverage, two criteria it had stopped using. Liquid staking quoted the Restaking category as excluding "Bitcoin restaking (Babylon, Lombard, SolvBTC)", words Restaking does not publish. Token standards said four messaging protocols carry a token standard ranked here, where five do. Subscriptions said Coinbase Commerce is ranked among payment gateways, where it is compared only for US and Singapore merchants. Two refusals carried statements about firms with no source: a wallet-compromise date for Serenity Shield, and a court claim about Multichain that the text itself said was unverified.
The refusals on those pages say what was checked, when, and what it showed, in words a reader can use, with the wrong statements corrected and the unsourced ones removed. A build check fails any refusal that publishes a source-file reference or that vocabulary. The 90 refusals still written in that voice are listed by name, and the list may only shrink.
Reading the crypto-card refusals for audit row TC-MEDIA-001 on 2026-10-11, then every refusal for the same pattern.
No. No published score or rank changed.
Two published refusal reasons made adverse statements about named firms that no loaded source supported. Crypto cards called Ready Card, Solflare Card and their issuer Kulipa "DEAD OR SANCTIONED, AND UNVERIFIED" beside "the UAB Monavate regulatory notice" (Monavate issues two ranked cards), in a note that called itself quotable nowhere. Mining pools called the cloud-hashrate segment it named "scam-dense" and repeated a criminal charge against HashFlare's operators that its own text said must first be loaded from the Department of Justice's release.
Kulipa's cards are listed as not ranked because their status could not be confirmed from a primary source, and the lead stays in the research notes; the mining-pools refusal gives its structural reason only. A build check fails any published refusal or catalog row that calls a named firm a scam, a fraud, insolvent, sanctioned or criminal without a regulator, court or law-enforcement source, and any refusal reason that reads as a research lead.
Audit row TC-MEDIA-001 named both on 2026-09-30.
No. No published score or rank changed.
The market-signal feed (/account, the landing variants, the command center) printed figures computed from typed constants as market data: "DeFi TVL" was two pools' value plus a typed $170 billion, and hardware-wallet shipments, the Visa stablecoin settlement run-rate and crypto-card daily spend each grew from a typed baseline. It stated unsourced claims and one false one ("licensed platforms also carry deposit insurance"), gave trading instructions ("Stake into the rally", "Deploying idle stablecoins into the highest-yield protocol now captures the spike", and, picked by keyword from a headline, "Stop new deposits to the named venue until the story clears"), and dated every editorial card with an invented "published X ago" time. The provider dashboard printed fees from a table dated 1 April under an undated "static" label.
The feed carries only signals built from fetched data (CoinGecko, DefiLlama, alternative.me, the article feed and exchange status pages), each with its figure or headline and its source, and none tells a reader what to buy, sell, hold or when; a test runs every branch and fails on that language. The dashboard's live block is gone.
Reading every reader of live data for audit rows BC-LIVE-002 and BC-LIVE-006 on 2026-10-11.
No. No published score or rank changed.
The fee term of Measured data read rates and fees from a table dated 1 April 2026 instead of the catalog. In lending it scored each protocol's borrow rate (Morpho 4.9 % to Venus 7.1 %), on a page that says lending is not ranked on rates, and that rate made Morpho the #1; in a browser, once the yield feed loaded, every pool it returned read a borrow rate of 0. In exchanges it scored Kraken at a 0.26 % taker fee and Gemini at 0.40 %, where their current schedules say 0.80 % and 1.20 %.
The fee term reads only the fee a provider publishes, as the catalog records it and re-reads on the 45-day review; lending has no fee term, so it ranks on its published criteria. /methodology/live-data publishes the rule, and on every build a test fills each live feed with extreme values and fails if one score moves.
Reading the fee term for audit row BC-LIVE-002 (which live fields may enter a score) on 2026-10-11.
Lending and borrowing: Venus #7 → #1, Spark #2, Morpho #1 → #3, Compound #8 → #6. Exchanges: Kraken #6 → #9, Gemini #10 → #11, Coinbase, Luno, Bitvavo and VALR up one place each; Bybit stays #1. Corrected 2026-10-14: the scores that moved with this order, some of which this entry left out: exchanges Bitstamp 77 → 79, Coinbase 64 → 69, Luno 64 → 69, Kraken 73 → 60, VALR 56 → 58, Gemini 59 → 48 and Mercado Bitcoin 41 → 42 (Bitvavo #9 → #8 at 62); lending Venus 58 → 81, Spark 78 → 80, Morpho 83 → 78, Euler V2 68 → 77 (#3 → #4), Kamino Lend 65 → 74 (#4 → #5), Compound 58 → 74, Fluid 64 → 73 (#5 → #7) and Aave 62 → 71 (#6 → #8).
The provider dashboard (/dashboard/<provider>) ranked each provider on five dimensions of its own, scored from capability flags and keywords, and printed that rank as its place in the category: 784 of 980 rows showed a rank the published ranking does not hold (Coinbase #4 where it is published #7), beside 'improvements' with effects it estimated itself. The page could be indexed.
The dashboard reads only the published ranking: the published rank and neighbour, each published dimension beside the category's average, and the published gap to the best-placed row on each; it predicts nothing, loads no catalog and is no longer indexed. A test compares every row it shows with rankings.json.
Audit row CT-PRTL-001 measured the gap on 2026-09-30; fixed 2026-10-11.
No published order moved; the dashboard now prints it.
FAQ answers, a quick answer and two headline lines in eight categories named criteria their rankings no longer score: 'What matters most when choosing a crypto card?' was answered 'Regional eligibility, spend utility, funding model, and issuer confidence', where cards are ranked on what a swipe abroad costs, whether the programme survives, whose licence issues the card and whether spending creates a taxable disposal; staking's quick answer and headline led with custody posture and post-staking utility, two criteria the category dropped.
Each is rewritten from its category's current criteria and weights, the FAQ answers state those weights in a form the weight check reads, and the build fails any surface, the decision guide included, that names two or more criteria a specification has retired.
Holding the quick answers, FAQ answers and key questions to their specifications' criteria for audit row AI-EVAL-001 on 2026-10-11.
No published order moved.
The Social-Fi page listed "A failure-history / 'already walked away' axis (Proposal A, weight 18)" among the products considered and not ranked. It was never a product but a criterion the methodology did not adopt, and the category counted it as one of 17 products assessed.
The note sits beside the criteria as the reason for that decision; Social-Fi counts 16 products assessed, 8 ranked and 8 turned away, and check-exclusions-published fails a refusal whose name or reason reads as a note about the criteria.
Reading the refusal index while testing the assistant's answer for products this site turned away (AI-EVAL-004), 2026-10-10.
No published order moved.
When no product in a category served the country entered, the list shown for reference printed scores no page publishes, in an order that was not the published one (35 country runs; for crypto cards in New Zealand it led with the published #3), and the explorer gave Coinbase Commerce, which the published payment-gateway ranking does not hold, its position in the US run as its overall rank.
The reference list is the published order with the published scores, and the explorer prints an overall rank only for a row the published ranking holds; check-wizard-country-excludes fails a fallback out of the published order or with an unpublished score.
Re-reading audit row CT-RANK-017 on 2026-10-10 over every category and every country a reader can state.
No published order moved.
Review pages in eleven categories stated a review standard in an older methodology's criteria (/reviews/metamask: 'Reviewed on custody posture, network fit, operational clarity, and access mode', where wallets are ranked on signing visibility, key export, incident disclosure and recovery), and the headline on the /best pages for fixed-rate yield, automation keepers, stablecoin issuance and remittances, with the remittances category summary, named another category's criteria or an outdated set.
Each of the 49 review standards, comparison standards and summaries the new check flagged is rewritten from its category's own criteria and weights, drafted from the spec and checked against it by a second model run told to refute it, and the build fails a standard that names fewer than half of its category's criteria or copy that reads as another category's.
The coverage and sibling checks added for audit row AI-EVAL-001 on 2026-10-10 flagged them; each was rewritten the same day.
No published order moved.
Provider review pages printed each criterion's weight a hundred times over: the rubric on /reviews/metamask read ×2800%, ×3000% and ×2400% for criteria the methodology weighs 28%, 30% and 24%, and so did every review page that shows it.
The rubric prints the weight the methodology publishes, and a test fails any source that multiplies a methodology weight by 100 outside the two places whose weights are fractions of one.
Tracing every weight the site prints while re-reading audit row AI-EVAL-001 on 2026-10-10.
No published order moved.
The category pages for automation keepers, market-data APIs and stablecoin issuance summarised another category's criteria (the relayers', the indexers' and stablecoin settlement's) and seed backups an outdated set; the restaking and stablecoins answers on /best stated weights their methodology no longer uses; and the assistant told readers that prediction markets, MEV protection, seed backups, key recovery and institutional custody are scored 60% on security and 40% on trust-minimisation.
Each is rewritten from the category's own criteria and weights, as are nine categories' review and comparison standards that no page printed, eight of them another category's; the assistant answers from each category's published promise, inclusion rule and weights; and the build fails when a published sentence states a weight its category's methodology does not carry or reads as another category's.
Re-reading audit row AI-EVAL-001 on 2026-10-10: the check built for it read every weight the site states against the methodology.
No published order moved.
41 products in 26 categories were offered to readers in countries their own quoted terms exclude or do not serve: among them Coins.ph (Philippines only) and Mercuryo, RedotPay's card and MoonPay's off-ramp in India, Gemini Custody in the UK and the EU after its wind-down, and US-only funds such as WisdomTree and Franklin BENJI outside the United States.
Each country the quoted terms exclude is now a restriction the country test reads, and each row whose terms serve only named countries carries them in servedCountries, confirmed by a second model run against the note's exact words; the build fails a restriction or a served country its evidence does not name.
Phase 2 of audit row CT-RANK-011 on 2026-10-10: a reading model run and an adversarial one went through every row whose availability note names a country in an exclusion or exclusive sentence.
No published order moved. In India six categories each lose one row; country pages list 23 fewer rows.
Ten exchanges were offered to readers in countries their own quoted terms exclude or do not serve: Bybit in Singapore, Hong Kong and mainland China, Kraken in Japan and Singapore, OKX in India and Malaysia, Bitso in Venezuela; CoinDCX (Indian residents only), Upbit (Korean residents only), Mercado Bitcoin (Brazil only) and Luno (five countries) across their whole region, Bitvavo in the United Kingdom and Switzerland, and Gemini across Asia. The data changelog of 2026-09-30 said CoinDCX serves India only, while the engine still offered it elsewhere in Asia.
Every country those terms exclude is now a restriction the country test reads, and a row that serves only named countries carries them in servedCountries, which the country test reads before any region. A test fails the build if a row excludes or serves a country its quoted evidence does not name.
Re-reading audit row CT-RANK-011 on 2026-10-10 against the exchange rows’ own availability notes.
No published order moved. In India the exchange shortlist falls from 10 eligible rows to 6, runner-up OKX to Bitstamp; country pages list 20 fewer rows.
The MiCA licence registry behind /mica and the /mica/check pages, a review, five guides, the MiCA hub and the country guides said Nexo holds a MiCA authorisation from CySEC, put Coinbase’s authorisation with BaFin and Bitvavo’s with the DNB, kept six providers transitional after every transitional period had ended, recorded KuCoin, Gate and Relai as unauthorised, and dated most authorisations 30 December 2024. Seven country guides gave the wrong transitional period (Germany 18 months to 30 June 2026; it was 12, to 30 December 2025), and the result text for German readers said Binance held a BaFin licence.
Each licensed record names its entities in a dated snapshot of ESMA’s interim register and takes its dates and authorities from them; Nexo, Bitget, Binance, Wirex, Plutus and Anycoin Direct are recorded as not licensed and KuCoin, Gate and Relai as licensed. The copy follows the registry, and the country guides take their transitional periods from ESMA’s list, with Poland’s vetoed implementing act noted.
Correcting the MiCA readiness index against ESMA’s register on 10 October 2026 (CC-2026-0045) showed the registry and the copy contradicting the same register.
No. No published score or rank changed, and no guide changed its picks.
The MiCA readiness index (v1.0, 18 June 2026) scored Bybit as unauthorised although Austria’s FMA had authorised Bybit EU GmbH in May 2025, called OKX’s and Gemini’s authorisations pending when Malta’s MFSA had granted both, and said Nexo passports via CySEC with no Nexo entity in ESMA’s register. Its answers told readers that Crypto.com, OKX and Bybit were still building toward MiCA authorisation.
Version 1.1 scores each exchange’s authorisation against a dated snapshot of ESMA’s interim register (read 10 October 2026), names each entity it relies on in the published dataset, and answers “Is X MiCA-licensed?” from the register. Three rationales were corrected: Coinbase’s authorisation is Luxembourg’s, Bitvavo’s is the AFM’s, and Binance withdrew its Greek application in June 2026.
The master audit compared the index with the verified evidence corpus on 30 September 2026 (SEO-PR-001); ESMA’s register confirmed it on 10 October 2026.
Yes. Bybit moved from 10th to 8th (45 to 68) and Nexo from 8th to 10th (65 to 50); OKX rose from 61 to 68 and Gemini from 75 to 79. The leader did not change.
In 32 categories, among them hardware wallets and layer-2 chains, the provider pages said “No evidence reading date recorded” and “Score evidence: Unknown, no observation on file”, and the score popovers opened no quote, although every published row had been re-read on every criterion with a quote, a source and a reading date (1,210 cells; hardware wallets read on 2 October 2026, layer-2 chains on 1 October). The research had moved into re-read files that the evidence generator did not publish.
A category whose re-reads cover its whole published table now publishes its evidence from them: each cell’s score, quote, source and reading date opens on the provider page and in the score popover. The 11 categories whose re-reads cover only part of their table are not published from them.
A phone check of the live Ledger page on 7 October 2026: its review stamp said “Reviewed 1 October 2026” while its Evidence section said no reading date was recorded.
No. No published score or rank changed.
Several hardware-wallet facts described a sibling model or the brand rather than the device the row prices: BitBox02 carried the Nova's EAL6+ chip, OneKey the Pro's QR air gap, Cypherock an EAL6+ level its own spec table gives as EAL5+, and launch years were the brand's first device. Blockstream Jade was listed as Bitcoin-only although its firmware signs Liquid assets, Tangem's firmware as partly open although Tangem keeps it closed, and the Ledger, Jade, CoolWallet and Cypherock prices were out of date or belonged to another model.
Every fact was re-read and quoted for the model the row prices (methodology 2026.10.2): the BitBox02's ATECC608B is uncertified, OneKey's Classic 1S Pure has no air gap, Cypherock's ST31H320 is EAL5+, launch years are each model's first shipment, Jade and Tangem are corrected, and the prices are Ledger $69, Jade Classic $67, CoolWallet Go $69 and Cypherock X1 $99.
The receipts-and-verify workflow for hardware_wallet v2, which graded every row on its priced model for the first time.
Yes, together with the four new criteria shipped in the same version: Trezor stays #1; Tangem falls from #7 to #14 and Cypherock from #6 to #10.
The fallback list a reader sees when no product serves their country ranked twelve criterion-scored categories on attribute maps their methodology pages no longer publish: it carried its own copy of the scorer dispatch, which never learned about the criterion path.
The fallback list now uses the same objective as the published ranking (scoreObjective), with its liveness guard and the published situational adjustments.
Reading the dispatch while moving l2_chains onto its published criteria (methodology 2026.10.0).
No. No published score or rank changed; only the order of the no-coverage fallback list.
Scroll, Mantle and Blast were recorded as EVM-equivalent although their own documentation calls them EVM-compatible; Starknet was counted from a 2021 mainnet that only whitelisted developers could use; and the scorer credited Ethereum data availability whenever the availability text contained the word "ethereum", so Arbitrum Nova's committee ("…Ethereum L1 rollup fallback") and Polygon zk
Each fact was re-read and quoted (Blast also probed live over RPC), Starknet is dated from its 2022 public opening, and data availability is now a quoted L2BEAT reading. Arbitrum Nova has in fact posted its data to Ethereum since 31 August 2026, so its score stands on the new evidence; Polygon zk
The l2_chains receipts and adversarial-verify workflow for methodology 2026.10.0, which quoted every score for the first time.
Yes, below the #1: Blast fell from #8 to #13 and Mantle from #6 to #12, together with the four new L2BEAT criteria shipped in the same version. Arbitrum One stays #1.
This page prints the corrections numbered CC-2026-0041 to CC-2026-0060, in full. It is closed: no entry is added to it or taken from it. 20 corrections, logged 2026-10-01 to 2026-10-11. Earlier page · All archive pages · Later page