Corrections
Linkmerica publishes versioned scores that institutions may cite months after publication. That only works if the record is honest about its own errors. This page states the correction policy and logs every correction issued.
Correction policy
Locked outputs are never silently edited. Every published score is locked at publication and carries a SHA-256 integrity hash recorded in a public log. A corrected score is issued as a new, separately dated version. The original file remains unaltered, and its original hash continues to verify against its original content — including its error. Corrections are additive records, not retroactive edits.
A correction and a rescore are different things. A rescore reflects new evidence about the subject: the product changed, a vulnerability was disclosed, a manufacturer shipped something new. A correction reflects an error in Linkmerica's own work: a research gap, a factual mistake, a misstatement. Both are dated and published. They are not the same event and are not described as though they were.
A correction is issued when a factual error is identified in a published output, or when the evidence base underlying a score is found to have been incomplete or inaccurate at the time of scoring.
Every correction states plainly whether any score was affected. Where the answer is no, that is stated directly rather than implied.
Correction log
Two wallet pages showed category scores that did not match their score files
An audit of every wallet page against its locked score file found two pages out of step. The Tangem page had shown Operational Security and Recovery Risk transposed since July 12, and Supply Chain Risk as 5.1 rather than 5.4 since August 26. The Foundation Passport Prime page had shown Supply Chain Risk as 2.9 rather than 3.3 since August 26. Separately, the bar drawn beside the Supply Chain Risk figure had not been redrawn when the figure changed — on the Trezor page since September 10, on the Ledger page since its October 21 rescore, and on the Passport Prime page on November 9. In each case the overall LISR score and tier on the page were correct; only the category table was stale. Both are fixed, and the deploy checks now compare every category row on every wallet page against its score file before anything is published.
A research note was revised after a full re-check against vendor pages
Authorised Isn’t Verified, published October 15, contained four errors. It said four of eight manufacturers publish reseller lists; all eight do. It said no manufacturer had published a rule on returned units before the incident; Ledger’s buying guidance states its distribution agreements contractually forbid resellers from restocking returned devices. It said Ledger had not published a root-cause account; Ledger published one on October 15, before the note went live. And it dated Ledger’s guidance to about four weeks before the incident; it was first published in March 2026 and updated September 18. A first correction on October 19 repeated the returned-units error. Separately, Foundation’s setup guide now says its check does not establish the absence of every physical modification or implant, and the note reflects that wording. The cause: parts of the audit were built from search excerpts rather than full pages, and a search that returned nothing was treated as evidence that nothing existed. Every claim in the note has now been checked against the vendor’s full page, and claims that a manufacturer does not publish something have been removed unless confirmed.
Arc’s founding validators were listed from the wrong source
The research note Same Firms, Every Layer, published October 1, listed Arc’s founding validators from the logo strip on arc.io. Circle’s own announcement of August 5, 2026 names a different set: it adds DTCC, MoneyGram and SBI Group, names ICE and Global Payments rather than their subsidiaries NYSE and Worldpay, and does not include Figment. The note now follows Circle’s announcement. No finding changes: Visa, Mastercard, Circle and Standard Chartered are founding validators in both sources. The lesson is narrow and worth stating — a homepage logo strip is not a primary source when a formal announcement exists, particularly on a page already known to carry attribution errors.
A favourable finding was published, and it was wrong
What was published: the research brief The Purchase Record, August 17, reported that a 90-day data retention policy contractually extended to a fulfilment partner had worked — bounding a breach to roughly 13,689 customers and three months of orders, against a comparable incident elsewhere affecting 39,798. It was presented as a finding in that manufacturer’s favour during a week when they were otherwise having a bad time.
What happened: on September 4, 2026 Trezor disclosed approximately 67,000 further customers from orders placed between November 2019 and August 2021, taking the reported total to roughly 80,689. Trezor states that it repeatedly requested and received written assurance from ShipMonk that the data had been deleted in line with the contract and its published policy, and that it had not been. Records reaching back nearly seven years were still held when an attacker exploited a critical Metabase vulnerability.
What that means for the framework: the brief stated its own limitation explicitly — that it assesses disclosed policy rather than verified practice, and that a stated deletion which is not performed is worse than a longer retention which is honoured. That was written as a caveat. It describes exactly what occurred. A disclosed policy, a contractual obligation, and written confirmation of compliance were all present and none of them removed the data.
The brief has been corrected in place with the original wording quoted rather than removed. The supply chain risk assessment that credited this control is being re-examined and any change will be published separately. Publishing a favourable finding on the evidence available was right; correcting it when the evidence changed is the same act.
LISR composites were not computed deterministically — four of seven were wrong
What was wrong: the LISR framework states that scores are produced by applying locked category weights. The scoring pipeline did not do this. It asked a language model to return the weighted average alongside the category scores, and stored whatever came back. No code performed the calculation. Four of seven published composites did not reproduce from their own category values.
Every error ran the same direction. All four published composites were too high by 0.2 to 0.3, meaning those wallets were rated as carrying more risk than their own category assessments supported. Where the score files record per-category weighted values, those values were arithmetically correct — the error was in the final sum.
| Wallet | Published | Corrected | Tier |
|---|---|---|---|
| Ledger | 4.8 | 4.5 | MODERATE — unchanged |
| Trezor | 4.7 | 4.4 | MODERATE — unchanged |
| SafePal | 6.4 | 6.2 | HIGH — unchanged |
| Foundation Passport Prime | 4.1 | 4.0 | MODERATE — unchanged |
| BitBox02 | 3.5 | 3.4 | LOW — unchanged |
| Tangem | 6.8 | 6.8 | Verified correct |
| ELLIPAL | 6.4 | 6.4 | Verified correct |
Two changes are combined here. The composite arithmetic fix above, and a methodology refinement dated 2026-08-26 folding customer data retention into the supply chain risk category as a disclosed sub-factor, following the research brief The Purchase Record. That refinement adjusted the supply chain risk score on four wallets. Both are published together because a reader experiences them as one change. BitBox02’s movement comes from the refinement alone; its composite was arithmetically correct.
What changed in the pipeline: the composite is now calculated in code from the locked weights, every score file is audited against that calculation, and the deploy process refuses to publish any score that does not reproduce from its own categories. The framework’s determinism claim describes what the system does rather than what it intended to do.
No tier changed on any wallet. All original score files remain locked and unaltered with their original integrity hashes; corrections are issued as new dated versions and both remain in the public record.
Quantum Resistance Readiness — systematic defect, all seven scores re-audited
What was wrong: Quantum readiness scoring searched for news about a manufacturer’s post-quantum posture rather than reading what the manufacturer had published. Where a search returned nothing, the score asserted no position existed. That assertion was wrong in four of seven cases. In every one the manufacturer had published relevant material before the scoring date, on its own site or documentation.
Every error ran the same direction — risk overstated. This was a single mechanism producing consistent directional bias, not scatter. Every manufacturer that had published something was mis-scored; the three scores found accurate are the three manufacturers that had genuinely published nothing.
| Wallet | Before | After | Outcome |
|---|---|---|---|
| Trezor | 7.8 HIGH | 4.4 MODERATE | Corrected |
| Foundation Passport Prime | 7.3 HIGH | 6.3 HIGH | Corrected |
| Ledger | 6.8 HIGH | 5.4 MODERATE | Corrected |
| Tangem | 9.1 CRITICAL | 8.0 HIGH | Corrected |
| SafePal | 8.3 CRITICAL | 8.3 CRITICAL | Verified accurate |
| ELLIPAL | 8.7 CRITICAL | 8.7 CRITICAL | Verified accurate |
| BitBox02 | 7.6 HIGH | 7.6 HIGH | Verified accurate |
What changed in the methodology: quantum readiness scoring now begins with manufacturer technical documentation. Search is secondary. A rationale may assert a manufacturer has published nothing only after those sources are checked directly. See the standard.
What did not change: the framework itself. No score moved because the tiers or categories were wrong. The substantive findings largely survived — in each corrected case the manufacturer’s material established that a position existed, not that post-quantum protection reaches the custody layer. Five of seven wallets remain HIGH risk or above. All original files remain locked and unaltered with their original integrity hashes; both versions of every corrected score stay in the public record.
Ledger Quantum Resistance Readiness — incorrect claim about published implementation
Affected output: Ledger Quantum Resistance Readiness score, published 2026-06-28.
What was wrong: Two category rationales stated that Ledger had announced no production post-quantum implementation and published no explicit FIPS alignment statement. Ledger’s Embedded OS team had published a technical implementation guide for ML-KEM (FIPS 203) and ML-DSA (FIPS 204) in June 2026, before the scoring date, documenting constant-time decapsulation and side-channel considerations.
What was not wrong: The implementation sits in the SDK, available to applications, rather than protecting the boot chain, firmware verification, or transaction signing. That distinction caps the credit and is why the score does not move further.
Root cause: Linkmerica assessed Ledger’s quantum posture on 2026-07-24 based on an engineer’s post on social media rather than the manufacturer’s published technical documentation. This is the third correction in three weeks arising from the same failure — searching for news about a manufacturer’s quantum posture instead of reading what the manufacturer has published.
Was any score affected? Yes. Quantum Resistance Readiness moved from 6.8 to 5.4, and the tier from HIGH to MODERATE. Both versions remain in the public record. View the current score.
This correction was later found to be one instance of a systematic defect. See the 2026-08-04 entry above for the full re-audit of all seven quantum readiness scores and the methodology change.
This correction was later found to be one instance of a systematic defect. See the 2026-08-04 entry above for the full re-audit of all seven quantum readiness scores and the methodology change.
This correction was later found to be one instance of a systematic defect. See the 2026-08-04 entry above for the full re-audit of all seven quantum readiness scores and the methodology change.
Foundation Passport Prime Quantum Resistance Readiness — incorrect claim about algorithm disclosure
Affected output: Foundation Passport Prime Quantum Resistance Readiness score, published 2026-07-12.
What was wrong: Two category rationales asserted that Foundation had not publicly named a NIST-standardized post-quantum algorithm. Foundation had named ML-KEM, the NIST FIPS 203 standard, in its general-availability announcement seven weeks before the scoring date, and in its blog and product documentation.
What was not wrong: The score's central judgment stands unchanged — the feature in question protects the wireless transport channel, not the wallet's transaction-signing cryptography or its boot and firmware verification. Transport-layer post-quantum protection is not custody-layer post-quantum protection. The original score identified the feature and referenced it repeatedly; the error was a specific incorrect claim about public disclosure, not a failure to find it.
What changed: A dated correction note was published and a new score version issued on 2026-07-29. The 2026-07-12 file remains locked and unaltered with its original integrity hash.
Was any score affected? Yes. The Quantum Resistance Readiness score moved from 7.3 to 6.3. The risk tier did not change — it remains HIGH. Both versions remain in the public record. View the current score.
Synthesis brief withdrawn and rebuilt before announcement
Affected output: Research brief, "Five Ways an Agentic Protocol Fails Custody."
What was wrong: The first published version contained factual errors introduced during drafting. All three rated protocols were given incorrect formal names. Both Linkmerica framework acronyms were expanded incorrectly. A statement describing a rated party's response to published research was included without any source. A specification publication date was stated without a source. Two of Linkmerica's own prior rescore dates were stated incorrectly, contradicting the locked score files. One transaction-volume figure was overstated, and two cumulative figures were described as monthly rates.
What changed: The brief was withdrawn and rebuilt the same day from the locked score files, with every factual claim constrained to verified sources. It was republished before any announcement was posted. An editorial verification gate was built the same day and now runs against every brief before publication, checking score claims and dates against locked files and flagging any statement attributed to a named third party.
Was any score affected? No. All five agentic protocol scores referenced in the brief were correct, locked, and hashed throughout. The error was confined to the brief's prose.
Trezor Quantum Resistance Readiness — research gap in original score
Affected output: Trezor Quantum Resistance Readiness score, published 2026-06-28.
What was wrong: The score's rationale stated that no public quantum-readiness statement from the manufacturer had been identified. This was incorrect at the time of scoring. Trezor had published quantum-readiness technical content in March 2026, over three months before the scoring date, and it was publicly accessible throughout.
What changed: A dated correction note was published, and a new score version was issued on 2026-07-24 incorporating the manufacturer's disclosures. That version also documents a device-generation change in what the Trezor entry assesses. The original 2026-06-28 file remains locked and unaltered with its original integrity hash.
Was any score affected? Yes. The Quantum Resistance Readiness score moved from 7.8 (HIGH) to 4.4 (MODERATE). Both versions remain in the public record. View the current score.
Corrections are listed most recent first. For score changes driven by new evidence rather than Linkmerica error, see the changelog.
