This is CFI.co’s published list of the defects and unresolved questions it knows of in its own estate. It is a standing document, not a statement made once: each entry carries the date of what it records, corrections are dated where they are made, and nothing is removed. Where the date an entry was added cannot be established, that is said rather than estimated.

Correction, 4 August 2026. The opening sentence stated that each entry carries the date it was added. The dates in the headings are the dates of what is recorded, not the dates entries were added, and four entries carry no date at all — “/c-19-press/ — deliberate removal, not link rot”, “Archive membership: one record outside the governed estate”, “Estates outside the governed estate” and “An unresolved figure, recorded rather than corrected: “112 countries””.

Award announcements removed, 2011–2026

CFI.co does not claim a corrections practice before 2026, and does not claim that nothing was withdrawn. Between 2011 and 2026, published award announcements came down for several reasons. This is a full account of what has been investigated so far, item by item, so the pattern can be tested rather than taken on our word. It is not represented as complete: the check that produced it compares what was once externally visible against what remains, and a further, undiscovered removal is possible. If one is found, it is added here, dated, not folded quietly into the count below.

27 announcements, 14 causes, investigated 28–29 July 2026:

Correction, 4 August 2026. The Caixa Económica Montepio Geral (2017) announcement is published at its original address and carries archive record cfi-award-5276, which has been in the archive continuously since 31 January 2018. Of the two announcements in this row, only Mansion Gibraltar (2018) remains down and undecided. Whether the Montepio announcement was ever removed is being re-checked against our own records, and the result will be recorded here.

Correction, 4 August 2026. The FBS (2015, post ID 2749) announcement is published at its original address and carries archive record cfi-award-2749, added to the archive on 29 July 2026. The status in this row is corrected from “The one candidate for restoration” to “Restored, 29 July 2026”. The row was published in its earlier form on 31 July, two days after the restoration.

Archive membership: one record outside the governed estate

The public archive holds the editorial record published on cfi.co. One record in it, cfi-article-20524 (“Matchmaking Private Finance and Green Infrastructure”, published 27 August 2021), carries a blog.cfi.co address, from before that subdomain was carved out of the governed estate’s scope. It stays in the archive, because nothing is removed once archived; this entry is the disclosure that it is there and why.

Estates outside the governed estate

Two parts of the CFI.co domain estate fall outside the governance file’s remit. Each is disclosed here rather than defended in that file.

/net-worth/ — 16 profile pages (14 published, 2 draft) estimating the net worth of named public figures, dated around September 2022. Not part of CFI.co’s editorial process for articles or awards, and not covered by the transparency archive.

blog.cfi.co — published 2019–2026, largely outside CFI.co’s editorial process: much of it was templated, and part of it was supplied through a third-party content agency’s account. It also carried republished CFI.co editorial material. Publishing on this subdomain stopped on 30 July 2026. Nothing on it has been deleted or unpublished; every page remains at its original address, and no URL there has been redirected or made to return 404/410. Full notice: blog.cfi.co/llms.txt.

A finding specific to that estate: 193 links across 139 posts on blog.cfi.co were commercial placements that carried no disclosure at the time they were published. In July 2026 each was marked and a dated editorial note was added to the affected post itself, in place, without removing or altering the surrounding content.

Restrictions from indexing, 26 July 2026

Under this file’s machine-reachability rule, withdrawing a still-published page from indexing, sitemaps or AI-crawler access is a restriction: it is dated and recorded here, not left undisclosed. Three restrictions were applied on 26 July 2026, each reversible and none affecting whether the material remains published at its existing address:

• /order/ — noindexed, and Disallow: /order/ added to robots.txt for both the general and named-AI-crawler groups. Order pages themselves were separately retired the same day (publish → private: public 404, database record kept, reversible); payment state for any order already placed lives in the payment processor’s own back office, not in WordPress.
• Six auxiliary pages — a user profile page, a password-reset page, two SDG-shortlist login pages, and two Google-search-result pages — noindexed. None is published editorial content or a claim-bearing informational page.
• The /net-worth/ estate (see above) — noindexed, and net-worth-sitemap.xml removed.

Each remains reversible; none has been re-indexed as of this entry.

/c-19-press/ — deliberate removal, not link rot

791 URLs under /c-19-press/ (CFI.co’s COVID-19 press-briefing coverage) return HTTP 410 Gone and do not appear in the site index. 410 is used deliberately here to state that the removal is intentional, as distinct from HTTP 404, which would leave a reader unable to tell a deliberate removal from a broken link. This entry is that statement, made explicit.

Award Methodology page edited, 31 July 2026 — and left self-contradictory

On 31 July 2026 at 16:06 BST, four words were removed from the body of the Award Methodology page. The page had listed “details of sources of nominations” among the material sent to the judging panel; that item was deleted and the rest of the sentence left unchanged. The wording had been live in that form since at least 20 April 2026.

It is recorded here for three reasons.

The change was made before the underlying question had been decided. It is an edit to a claim-bearing page describing how CFI.co’s awards are judged, and the rule is that such edits are dated and disclosed rather than made quietly. It was also made in the wrong direction: CFI.co’s published account of its own practice is meant to govern its governance text, not to be edited so that the text survives contact with it.

The edit was incomplete, and the page now contradicts itself. The same page’s FAQ carries the same claim in different words — “details of nomination sources” — and was not touched. Until this is resolved, the body of the page and the FAQ on the page say different things about what the judging panel receives. That is a worse state than either version before the edit. The check that produced the edit searched for one exact phrase and did not look for the paraphrase.

What the removed phrase originally meant is not settled. “Sources of nominations” can describe either the category of a nomination — whether an organisation put itself forward or was nominated by someone else — or the identity of the nominator. If it meant the category, the sentence was accurate and its removal deleted a true disclosure. If it meant the identity, the sentence was inaccurate and was published in that form for at least three months, which is a correction rather than a wording change. That question is with the principals.

Nothing further will be changed on that page until it is decided, and when it is, the body and the FAQ will be corrected in the same pass. That correction will be dated here.

Correction, 31 July 2026. The principals have ruled. Nominations are received in confidence; the judging panel receives anonymised nomination information and is not told who nominated a company; the single exception is that it is made aware when a nomination is a self-nomination. On that practice the four words removed earlier on 31 July were not false, and removing them left an enumeration that read as exhaustive without them. The body and the FAQ of the Award Methodology page have both been corrected to state the practice, in the same pass.

Self-nominations: two published pages contradict each other

Two pages published by CFI.co make opposite statements about whether the awards programme accepts self-nominations.

The CFI.co Awards Programme page, describing how coverage bias is mitigated, states: “We mitigate this through proactive outreach, acceptance of self-nominations, and independent research beyond nominations.”

The SDG — The Business Case page, unchanged since 23 June 2022, states: “We have never accepted self-nominations for any CFI.co award programme.” That claim is not confined to the SDG programme; it is made about every CFI.co award programme.

Both cannot be true. The contradiction was found on 31 July 2026 during a check prompted by the entry above, and is recorded here before it is resolved rather than after. Which statement reflects practice, and what the corrected wording should be, is with the principals; the correction will be dated here when it is made.

Correction, 31 July 2026. This entry described the two pages as contradicting each other. That was wrong, and the truth is less dramatic. CFI.co did not accept self-nominations historically and does accept them now. The SDG page was accurate when it was published in June 2022; the awards programme page is accurate today. Nobody published anything false — the practice changed, and the older page was never revisited.

The defect is therefore of a kind not previously recorded here. Every other item on this page was wrong when it was written. This one was right when it was written and was falsified later by a decision, with nobody editing the page. An absolute about our own past — “never”, “always” — is a promise that no future decision will contradict it, and it can rot in place unnoticed. The SDG page will carry its own dated correction once the date of the change in practice is established; an undated correction on a page about accuracy would be worse than the stale sentence.

Audience and circulation figures withdrawn, 31 July 2026

CFI.co has withdrawn the published audience figures it has found across its estate, and records each withdrawal here with its date. The sweep is not represented as complete: where a further figure is found, it is added here rather than quietly removed. The rule is that CFI.co publishes no audience figure it cannot evidence — the same rule that removed this site’s reader demographics and its public rate card earlier in July 2026.

What was published, and what has happened to it:

• The media information page stated an approximate readership of 128,000 per issue, a 32,000-copy print run, an average of four readers per copy, up to approximately 148,000 for issues with full event distribution, and a controlled circulation of 1,000 senior executives. The readership figure was never measured: it was a print run multiplied by an assumed four readers per copy, a convention of the trade rather than a count of anybody. The page is retired and its address remains published.
• A page promoting a 2020 special issue carried a roster of 38,000 subscribers, a figure of some 61,000 copies, and a full regional and sectorial breakdown of readership — all written in the present tense although they described a single issue published six years ago. That page is retired.
• A third page stated that CFI.co publishes a rate card. That ceased to be true when the rate card was retired on 27 July 2026, and the page had been reviewed three days before that. The statement appeared twice, once in the page text and once in the page’s machine-readable structured data, so a reader and a machine were told the same untrue thing. Both have been corrected.
• A copy of the readership figure was also held in the site’s own code. It was not being displayed, but it would have reappeared if the page block that calls it were ever restored. It has been removed.

No page has been deleted, redirected, or withdrawn from indexing. The retired pages remain published at their existing addresses and carry a dated notice of what they previously said and why it is gone.

Subscription page and payment endpoint withdrawn, 31 July 2026

CFI.co’s print subscription page at /subscribe/, and the payment endpoint it posted to, were withdrawn on 31 July 2026. No new subscription can be taken through the site.

They are sealed rather than deleted, on the same basis as the order pages withdrawn on 26 July: transactional pages are not part of the published record. The database record is retained, the change is reversible, and both addresses are excluded from crawling. Anything already purchased is unaffected by the withdrawal and is honoured; payment records for any subscription already taken sit with the payment processor rather than on this site.

An unresolved figure, recorded rather than corrected: “112 countries”

CFI.co’s site code contains a published figure — “112 countries where our clients operate” — which is derived from its internal client records. CFI.co holds itself to a rule that nothing published may state or imply the size or scope of those records, and whether a count of client countries falls inside that rule has not been decided.

It is recorded here as undecided rather than quietly changed. Given what this register has had to disclose today about how much of its own estate CFI.co was able to see, an open question is better left visible than tidied away.

Archive records whose address resolves to the wrong page, 1 August 2026

Twenty-one records in CFI.co’s public articles archive carry an address that resolves to a category index rather than to the record itself. The address returns HTTP 200. A reader or a machine following the archive’s own link would receive a page, find no error, and be looking at something other than the record they asked for.

The cause is a site navigation arrangement rather than a withdrawal: the posts exist on the main site as menu entries and redirect to a category listing. Nothing has been removed, and the verbatim text of each record is preserved in the archive. The announcements themselves are published separately on the awards site and carry their own archive records.

Why this is recorded rather than simply fixed. A missing page announces its own failure; a page that returns 200 does not. Every check CFI.co runs against its archive tests whether an address responds, not whether it responds with the right thing, so this class of defect passed every check that exists. That is the more useful disclosure, and it is the reason the count below is stated as unknown rather than closed.

The records now carry a machine-readable disclosure of the mismatch. Correcting the addresses themselves, and re-exporting the affected records, is scheduled work rather than same-day work. Until it is done, these twenty-one records are withdrawn from active outbound use: they are not cited to anyone.

Canonical addresses across the site: an open count, 1 August 2026

Separately from the records above, CFI.co has checked every page on the main site that carries a custom address. Two were found to resolve to nothing; both were corrected on 31 July 2026. No others were found.

That figure is stated as an open count rather than a finding, because the method cannot yet support a closed one. The daily check that monitors these addresses tests whether a page responds, not whether it is the page it should be. Until it can test identity rather than status, CFI.co cannot say how many addresses across the estate resolve to the wrong content, and does not claim the number is zero. Two are confirmed and corrected; the true number is unknown.

Blog retirement notice corrected, 1 August 2026

The retirement notice carried on every post at blog.cfi.co stated that in July 2026 each undisclosed commercial link was marked and a dated note added to the post itself. The first half was done; the second was not. Of 468 published posts, two carry such a note. The notice itself is injected site-wide by a plugin and is not a note on any individual post.

The sentence has been corrected to describe what was actually done: the links were marked, and the notice is the disclosure rather than a per-post note. The two figures in it were checked against the site and are accurate — 139 posts carry a marked link, and there are 193 of them.

This is recorded rather than quietly amended because the claim was false, it was ours, and it stood on a page about not altering the record without saying so.

Award capacity claim corrected, 1 August 2026

The CFI.co Awards Programme page stated that CFI.co “only have the capacity make a maximum of 3 or 4 award decisions in any given week”, and invited the reader to verify this by looking at the award history.

The award history does not support it. Counted from CFI.co’s own public archive, 538 weeks contain at least one award; 134 of those weeks contain more than four, and the largest single week contains 48. A quarter of all weeks exceed the stated maximum.

The sentence now reads: Every award CFI.co publishes carries its date, and the full record is open to inspection — including the announcements that have been withdrawn, which are listed with their reasons in our known-open register.

Between those two sentences a third was live on the page for under an hour on the same day. It said that every award CFI.co has made is published and the full record open to inspection. That was not an incomplete statement but a false one, and this register held the disproof at the moment it went up: twenty-seven announcements have been withdrawn, six were invalidated and stay down, one is withheld on legal advice, and twenty are absent from the hashed archive. It is recorded because it happened, because it was ours, and because replacing a false claim with another false claim is the failure this register exists to catch.

Correction, 4 August 2026. This sentence stated eighteen. Counted against the awards archive, twenty of the twenty-seven have no record; seven do — FBS, Caixa Económica Montepio Geral, Fidelis Finance, KPMG Lower Gulf, Evenord-Bank, Kommuninvest and Canada Goose.

Sponsorship labelling on records published before 9 November 2025, 1 August 2026

Every record in CFI.co’s public archives carries an independence_status field. On records published before 9 November 2025 that field reads independent_editorial. That value was not a determination. It is what the field defaulted to, because no labelling practice existed at the time.

In the articles archive this affected 2,766 of 2,792 records as at 1 August 2026, of which twenty-six were marked as commercially supported. Counts move as records are added; the live figures are in each repository’s index.jsonl. In the awards archive it affects all 2,385 records, of which 2,307 predate the cutoff; none is marked otherwise. Left unchanged, the value is a positive machine-readable claim about work for which no such assessment was made.

The correction is not yet applied, and the reason is recorded here rather than after the fact. It requires a classification rule that separates work CFI.co produced from work contributed by named outside authors — institutions, economists and specialists writing under their own names. Applying a neutral value to a contributed column would assert that CFI.co could not determine whether it was sponsored, about a piece whose authorship is plain on its face. The archives’ existing content classes do not reliably make that separation: a test found 155 records classified as editorial analysis carrying contributed-author titles.

The two archives may not take the same correction. For articles the field records an absence of assessment. For awards, independence is structural rather than assessed: the decision function is sealed from commercial information by design, which is a different claim from an unfounded default and may warrant different wording. Which correction each archive takes is part of the rule being written, and both will be applied under it, dated, through the corrections route.

Correction, 25 August 2026. This entry stated that the default affected 2,766 of 2,792 records in the articles archive. That figure counted every record carrying independent_editorial, not the records published before 9 November 2025, which is the population the entry describes; the two differ by the records published after the cutover and labelled under the practice then in force, which were never defaulted. As at 25 August 2026 the population described here was 2,695 of 2,796. The correction below records what has since been done about it.

Claims published as images are read by none of CFI.co’s checks, 1 August 2026

CFI.co verifies its published claims by searching text: the page body, and since 31 July the machine-readable structured data alongside it. Neither can read a claim rendered as a picture. Any factual claim CFI.co publishes as an image therefore passes every verification surface it operates without being examined.

That is the finding. What follows is the instance that exposed it.

Two families of promotional graphic, published under the names “CFI.co in figures” and “CFI.co in numbers”, rendered CFI.co’s own statistics as pixels across several generations of each. Fifty-one image files have been placed beyond public reach today, in addition to those restricted on 27 July.

Four kinds of figure in them required it. Print readership is stated as 142,128, as 152,128, as 162,128, as 174,000 and as 180,000, alongside the 128,000 withdrawn from CFI.co’s media-information page on 31 July. Online page views per month are stated as 60,000-plus, as 150,000-plus, and as 241,853. The size of CFI.co’s team is stated as 53 in one family and as 80 in the other. The number of countries CFI.co’s clients represent is stated as 107 in one family and as 109 in the other, where the site’s own code states 112 — the discrepancy registered as open and undecided on 31 July, now shown to run further than the text surfaces on which it was found. Every one of these kinds is internally inconsistent, and CFI.co is able to evidence none of them.

The graphics also stated a count of the companies held in CFI.co’s internal database. The publisher ruled on 27 July that nothing CFI.co publishes may state or imply the size, scope or existence of that database, and this entry accordingly discloses the figure’s presence without restating it.

No account is offered here of how any of these figures came to be published, or to differ from one another. CFI.co cannot establish it and does not assert it.

This is a restriction and not a removal. Nothing has been destroyed; the files and their contents are retained and can be produced.

What was then tested, and what it did not settle. A claim that a class of defect is closed is worth no more than the search behind it, so the search is stated. Every image on both CFI.co installs whose proportions match these graphics — the long horizontal strip and the tall column — was enumerated and the remainder examined. On the awards site the remainder is winner-company logos without exception. On the main site it is advertisers’ and institutions’ own creatives at standard advertising sizes, and three items of CFI.co’s own artwork — a row of icons, a coverage banner and a screenshot of the site’s own front page — none of which carries a figure. No third family exists, and the two named above are the whole of it.

That test closes the graphics, not the finding. It searches by shape, and a claim set in an ordinarily proportioned image is invisible to it. Two such were found by reading rather than by shape: a table of bank branch counts from 2014 and a table of franchise investment levels from 2026, both reproducing figures attributed to others inside articles rather than claims CFI.co makes about itself. They are recorded here as evidence that the search which closed the graphics would not have found them. CFI.co does not know how many claims published as images exist across fourteen years, and states the number as unknown rather than as none. Where one is identified and read, it will be recorded with what the image actually says.

A protection CFI.co said was in place was not, 1 August 2026

The README of CFI.co’s governance repository stated: “Force-pushes and branch deletions are blocked on main.” They were not. No branch protection was configured on that repository, and the sentence was untrue for the whole of its life — from 31 July 2026, when it was published, until it was made true on 1 August 2026.

The sentence sat in the paragraph that exists to invite checking rather than to ask for trust. It opens: “Rather than promise that a published version is never edited or removed, here is what would have to fail for that to happen without you noticing.” The reader was then told a safeguard was in place that was not. Of everything CFI.co published that day, this was the sentence a reader was most entitled to rely on, and it was the one that was wrong.

It was found during preparation for publishing the governance file, by opening the repository’s settings rather than by reading its text.

Why no check CFI.co runs would have caught it. The claim is about a configuration, not about a page. CFI.co verifies its published claims by searching text and, since 31 July, the machine-readable data alongside it. Neither can read a repository setting. The sentence was consistent with every other sentence around it, correctly spelled, and false, and nothing CFI.co operates was capable of noticing. This is the same defect as a claim rendered as an image, recorded separately: the medium a claim is made in determines whether any check can reach it, and CFI.co has now found two media its checks cannot read.

How it was corrected. By configuring the protection, so that the claim became true, rather than by removing or softening the sentence. The claim was the right one to make; it had not been backed. Force pushes are now blocked and deletions restricted on main. A reader who wants to test the corrected version can attempt exactly what the sentence describes.

CFI.co does not assert how the gap arose. What can be established is that the claim was published on 31 July and that the setting was absent when checked on 1 August; whether the protection was ever configured and later removed cannot be determined from the repository’s own history, and is not asserted either way.

One consequence is recorded because it bears on the value of every other disclosure. The governance file names this repository as the address at which its published versions and their hashes can be verified. An adversarial review of the file’s text was under way when this was found, and it would not have found this, because the defect was not in the text — it was in the thing the text points at. Reviewing a document and reviewing the infrastructure it relies on are two pieces of work, and CFI.co had been treating them as one.

Pages removed from search without a record, 1 August 2026

Twenty-five published pages on cfi.co carry an instruction telling search engines not to index them. Under the rule CFI.co adopted today, taking a published page out of search is a restriction on its reach and is recordable: it is disclosed with what was restricted, when, and why, and it is never done silently. None of these twenty-five carries such a record. This entry is the disclosure, and it is incomplete in a way stated below rather than glossed.

Three of them are articles that also sit in CFI.co’s hashed public archive. That archive asserts of each record that it is published and open to inspection. The instruction on the page says the opposite to every search engine that reads it. Both statements are CFI.co’s, both are live, and until today neither surface mentioned the other. The three are a 2022 publication of a speech given at Davos, a 2022 company profile, and a 2024 article on supply chains.

Fifteen more are pages about the net worth of named individuals. They are not in the hashed archive, so the contradiction above does not arise for them; the restriction is nonetheless unrecorded. The remaining seven are functional pages — login, profile and password-reset screens, an internal country list, and two pages used to preview search results — where exclusion from search is ordinary practice and not a restriction on anything a reader was offered.

What CFI.co cannot say. It cannot say when any of these instructions was added, or on whose decision, or why. WordPress does not retain a history for that setting, and the database holds only its present value. So the dates and reasons that a restriction record is supposed to carry do not exist for these pages and cannot be reconstructed. Publishing an estimate would be worse than publishing the gap.

What follows from that: the standing format for restriction and removal notices, per item and in aggregate, is scheduled in CFI.co’s governance file for 31 August 2026. From the date that format exists, no page is taken out of search without a dated record made at the time. For the twenty-five above, the record starts today and is missing everything that happened before it.

The three archived articles are the ones that matter most, and they are not resolved by this entry. Either the restriction is lifted, or the archive records that the page is restricted, or the archive record is withdrawn with its reason. Which of those applies to each is a decision for the principals, and it is open.

Resolved the same day, 1 August 2026. The principals ruled that the restriction be lifted on all three. It has been: the three articles are indexable, and the archive’s statement that they are published and open is now true of the pages as well. The restriction was lifted rather than recorded because recording it would have required stating when and why it was imposed, which cannot be established. Twenty-two pages remain out of search — fifteen about the net worth of named individuals, and seven login, profile and internal screens — and none of those is in the hashed archive.

A contributed piece published without its disclosure, 10 August 2026

On 10 August 2026 CFI.co published a contributed opinion piece that carried no statement
of whether the author’s organisation had a commercial relationship with CFI.co. It carried
no such statement on the page, and none in its machine-readable data. The piece is
Iran at the Crossroads, published under its author’s own name and byline.

CFI.co’s governance requires that attributed contributor content carries that disclosure
at the point of reading and on the machine-readable surface. That requirement was in force
on the day the piece was published. It was not met, and it was not met because no
machine-readable contributor surface existed anywhere in this estate: the only disclosure
metadata CFI.co had built covered commercially funded content, which is a different thing.
The requirement had been written; the means of satisfying it had not been built.

The surface was built on 12 August 2026. It does not yet render anything, because the
exact wording of the statement has not been approved, and CFI.co would rather publish
nothing than publish a form of words it has not settled.

What remains unresolved, and is recorded here rather than decided quietly.
CFI.co’s editorial policy provides that contributed pieces published before the policy
commences carry no statement rather than a determination, on the ground that a statement
made retrospectively about a relationship at a past date is a weaker thing than one made at
the time. This piece was published before that commencement date. So the policy as drafted
does not require a statement to be added to it, and adding one would be a departure from the
policy rather than compliance with it. Whether to add one anyway has not been decided.

What is not in question is the window itself. Between 10 and 12 August 2026 a contributed
piece was publicly available while a requirement that applied to it could not be satisfied by
any mechanism CFI.co had. That is recorded here as of 12 August 2026, and this entry stays
open until the wording is approved and the question in the preceding paragraph is answered.

The contributor disclosure is built but not yet rendering, 12 August 2026

From 12 August 2026 CFI.co’s editorial policy states that every contributed piece states
whether the author’s organisation had a commercial relationship with CFI.co, and that the
statement appears both on the page and in the article’s machine-readable data. The mechanism
that produces that statement was built on 12 August 2026 and is installed. It is deliberately
not rendering yet, because the exact wording has not been approved.

No contributed piece has been published since that date, so nothing published is missing a
statement it should carry. The gap is between what the page now describes and what the estate
would currently do, not between the page and anything already published. It closes when the
wording is approved and the mechanism is switched on. Until then a contributed piece must not
be published, and this entry records why.

Recorded 12 August 2026, and open until the wording is approved.

Closed, 13 August 2026. The publisher approved the wording on 13 August 2026, in all six of its forms, and the mechanism was switched on the same day. Each of the six sentences was checked character by character against the approved text before the switch rather than word by word, because a word-level comparison had passed the day before while hiding a lost colon and a stray space. Both surfaces were then verified to carry the same sentence from the same source text. A contributed piece can now be published, and the gap this entry records is closed.

Contributed pieces created through the CRM would have been attributed to CFI.co, 12 August 2026

A defect found and fixed on the day it was introduced, recorded because it was capable of
misattributing authorship on a public page and because the correction is worth stating.

CFI.co’s internal system creates article drafts through a service account. That account is
one of the house identities whose work is attributed to CFI.co as an organisation rather than
to a person, which is correct for staff-written material. A contributor’s piece created the
same way would have inherited that treatment: it would have named CFI.co as the author in its
structured data and carried no person at all, while the About and Editorial Policy pages both
state that contributed work is published under its author’s own byline.

No published piece was affected. The tooling that would have produced this was built on
12 August 2026 and corrected the same day, before any contributed piece had been created
through it, and the one contributed piece published to date is correctly attributed to its
author in both its byline and its structured data. This entry is a record of the defect and
its correction rather than of a live fault.

Correction, 13 August 2026. The paragraph above is wrong, and it was the publisher who caught it. What it says is true of the tooling and false of the estate. The defect it describes was real, was found and fixed on the day it was introduced, and no piece was ever published through that route. But the same misattribution arrives by an older and entirely separate one, which this entry did not reach and should not have spoken for: a contributed piece posted under one of CFI.co’s house accounts is attributed to CFI.co as an organisation in its structured data and carries no person at all, however plainly the page itself names its author. Published pieces are affected, and there are 274 of them. The sentence “the one contributed piece published to date” was wrong on its face for the same reason. What follows from it is recorded in the entry below.

Published pieces that name their author on the page and CFI.co in the data, 13 August 2026

Work published under CFI.co’s house accounts is attributed to CFI.co as an organisation rather than to a person. For work written by CFI.co that is correct, and it is the reason the arrangement exists. But contributed pieces have been posted under those same accounts for as long as CFI.co has published them, and the arrangement cannot tell the two apart. A piece can therefore name its author plainly on the page — in an image caption reading Author:, in a biography at its foot, or in both — while the structured data beside it names CFI.co and contains no person at all. A reader is told one thing and a machine another, from the same page, which is the precise condition CFI.co’s disclosure mechanism was built to prevent.

The count is 274. Taken on 13 August 2026 across every published article on cfi.co. The earliest was published on 10 August 2012 and the most recent on 7 July 2026. Four fall on or after 9 November 2025, and none on or after 12 August 2026, the dates from which CFI.co’s labelling and contributor-disclosure provisions respectively run. None is a commercially supported piece: that population carries its own label and is not affected. None had been flagged under the contributor mechanism, which had flagged nothing at the time of counting.

How the count was arrived at, and what it cannot see. Two markers were used, not one. The first was the image caption reading Author: or Authors:, which is how CFI.co has most often named a contributor on the page; on its own it finds 142. The second was the About the Author biography carried at the foot of a piece, which finds 231. Ninety-nine pieces carry both, 43 the caption alone and 132 the biography alone. The caption alone would have understated the position by 132 pieces, and it was the pattern this count was expected to use; it was tested for undercounting rather than trusted, which is the only reason the second marker was found. Pieces published under a contributor’s own account were excluded, 69 of them, because those carry a person in the data and are not affected. One further piece was excluded because its caption names CFI.co itself. A third marker may yet exist: a contributed piece that names its author in neither of these two ways would not appear in this count, and 274 should be read as a floor rather than a total.

What has not been decided. Whether the affected pieces are corrected retrospectively, or whether this register carries them as they stand, is the publisher’s decision and has not been taken. CFI.co’s editorial policy provides that pieces published before the policy commences carry no statement rather than a determination, on the ground that a statement made later about a past state is a weaker thing than one made at the time; an author’s name is not that kind of statement, so the reasoning does not obviously carry across, and that is the question. This entry stays open until it is answered, and the count above stands as the measure of what is at stake in the answer.

Two claims corrected on the Digital Trust & Verification page, 13 August 2026

The page stated, under “Our Model”: “Awards are not paid for at any stage. Commercially funded content is labelled as such, and the two records are kept separately.” Two things in that sentence were wrong.

The separation claim was untrue. CFI.co does not keep separate commercial and editorial records, and there is no separation of records to point at. The clause was published on 1 August 2026 as the replacement for wording removed the same day, and it asserted a stronger separation than CFI.co practises. It was live for twelve days.

The labelling claim was true only from a date, and the sentence did not say so. CFI.co’s labelling of commercially funded content runs from 9 November 2025. Records published before that date carry a default classification rather than a determination. Stated flat, the sentence claimed the archive as well, and the archive does not support that.

The sentence now reads: Awards are not paid for at any stage. Since 9 November 2025, commercially funded content is labelled as such, on the page and in machine-readable data; records published before that date carry a default classification rather than a determination, as our known-open register sets out. The superseded wording is left legible beneath the correction on the page itself.

The wording removed on 1 August 2026 — an invitation to analyse the correlation between CFI.co’s awards or editorial coverage and its partnership announcements, and an assertion about what a reader would find — has not been restored, and this entry does not reinstate it.

Two related claims elsewhere on the estate were identified in the same sweep and are recorded here as open rather than corrected, pending a decision on each: the Firewall label on the CFI.co Awards Programme page, which names a structural separation that does not exist even though the sentence beneath it is accurate; and the undated form of the labelling claim on the Contributors page and in the Open AI Access Licence.

The first of those two is closed, 19 August 2026. On the principals’ decision the Firewall label on the CFI.co Awards Programme page now reads Sequence. Only the label changed: the sentence beneath it stands word for word as it did, because the sentence was accurate and the defect was in the name given to it — “firewall” named a structural separation, where what the sentence describes is an ordering. The page carries a dated note recording the change and its reason. The second remains open: the undated form of the labelling claim on the Contributors page and in the Open AI Access Licence is unchanged and undecided.

Contributor bylines recorded as “CFI.co Editorial”, 15 August 2026

Until 15 August 2026 the public archive recorded the author field as the
constant “CFI.co Editorial” on every record, including pieces that carry a named
contributor’s byline on the page. The field therefore carried no information, and a
named contributor’s work was filed as house editorial. That is a false machine label.
It has been corrected in part; this entry records what was corrected, what was not, and why.

Corrected on 15 August 2026. Ninety-eight records whose WordPress author
is a real account were aligned to that name. A further forty-seven were aligned to a name
corroborated inside the piece itself — the same name read independently from two or
three of the caption, the image alt text and the “About the Author” block. One
further record, a co-authored piece of 6 May 2014, now carries both names as the piece
states them. In every case the article text is untouched: content_html and its
content_sha256 are unchanged, correction_status is set, and
record_sha256 is recomputed. This is alignment of a label, not a
back-dated determination about independence, and no independence or sponsorship value was
changed.

Not corrected, and held deliberately. A further one hundred and
twenty-seven records carry a name in only one place in the article, with nothing to check it
against. These have not been applied. A wrong name in the author field invents a
person; the house label merely fails to name one. They are held for reading before any of
them is committed.

Unresolved. Seventy-six records remain: seven where the article gives
conflicting names, and sixty-nine where the piece carries an “About the Author”
heading but no name a pattern can read. These need a person, not a better method.

A number that does not match, stated rather than reconciled away. This
register has previously described the affected set as 274. The pass of 15 August examined
250 records — a different and not merely smaller set, because the
markers used here match articles the earlier count did not, and the figure 274 was itself
recorded as a floor rather than a total. Of those 250, forty-eight are corrected, 127 are
held and 76 are unresolved. It is therefore not correct to say that the 274 are
done. They are not, and the two counts describe overlapping but different sets of
articles. Both figures are left standing here rather than replaced, so the discrepancy stays
visible.

Submissions are being kept, while the editorial policy says they are deleted, 24 August 2026

The Editorial Policy page says, and still says: “Where a piece is not published, the
submission itself is deleted.” From 24 August 2026 that is not what CFI.co does.

What changed. Everything arriving at [email protected] is now
recorded automatically within five minutes of arrival, before anyone opens it: a hash of the
message, a hash of the text of each attached document, and the evidence tying the message to the
domain that sent it. The purpose is to be able to show, for a piece CFI.co does publish, that it
came from the organisation it says it came from and that the author approved what was run. The
record carries no sender name, no subject and no filename.

Why it is kept rather than deleted. The publisher decided on 24 August 2026
that submissions and their origin records are retained, including for pieces that are declined.
That decision governs what CFI.co does. It does not, by itself, make the sentence on the
Editorial Policy page true, and that page is not edited here: the sentence stands as published
until the principal rules on it. This entry records the gap in the meantime rather than closing
it quietly.

What this costs, stated rather than glossed. The deletion promise protected
something real. A retained hash of a declined submission is a correlator: a person holding the
same document can test it against the record and learn that a particular organisation submitted a
piece CFI.co did not run. That is a consequence of keeping everything, and it is why the promise
existed. It is recorded here because a reader who relied on the deletion sentence is entitled to
know it no longer describes the practice.

What has not changed. The submissions ledger still carries no name and no
organisation. Nothing about an unpublished submission is published anywhere. The retained records
are internal, and the notification sent to editors when something arrives still carries no sender,
no subject and no excerpt.

Closed, 25 August 2026. The principal ruled that the page should describe what CFI.co now does. The Editorial Policy page was amended the same day: it no longer says an unpublished submission is deleted, and states instead that submissions are kept, including where a piece is not published. The page and the practice agree again. This entry stays as the record of the period in which they did not, and of what the deletion promise had been protecting.

This register was first published 31 July 2026. It will be added to as further items are found; existing entries are corrected in place with the correction dated, never removed.

Independence labelling on records published before 9 November 2025, 25 August 2026

Every record in CFI.co’s articles archive carries an independence_status
field. On records published before 9 November 2025 that field read
independent_editorial because no labelling practice existed at the time, not
because anyone had assessed the piece. That affected 2,695 of 2,796 records. The position was
recorded here on 1 August 2026, and the reason it had not been corrected was recorded with it:
a neutral value could not simply be swapped in, because applying one to work contributed by a
named outside author would assert that CFI.co could not determine whether it was sponsored,
about a piece whose authorship is plain on its face.

That obstacle has been removed rather than worked around. independence_status
now carries three values on one axis — independent_editorial for CFI.co’s
own in-house work, contributed_editorial for work by an outside contributor that
was not paid for, and commercially_supported for work that was. A contributed
piece therefore takes its own value and is never made to assert anything about whether it was
paid for.

On 25 August 2026 CFI.co’s principals reviewed the whole of the pre-9-November-2025
population and determined each record:

• 565 records are recorded as commercially_supported
• 295 records are recorded as contributed_editorial
• 1,835 records are recorded as independent_editorial

What the determination was, and what it was not. This was a review of the
three sets by the principals, carried out on one day, after the fact. It was not an editor
labelling each piece at publication with a practice in force at the time. Those are different
acts and the archive keeps them apart: records determined at publication carry
independence_basis: assessed, and the records covered by this review carry
independence_basis: retrospective_review_2026-08. A reader who wants to know how
much a label is worth should read that field, not only the label. The value
default_pre_2025-11-09, which recorded that nobody had asked the question, no
longer appears anywhere in the articles archive.

The caveat now travels with the label. Until 25 August 2026
independence_basis existed only in the archive’s catalogue file and in the
structured data embedded in each page. It was absent from the record files themselves —
which is to say, absent from the record a reader is sent to when they follow “Verify this
article”, and absent from the copy a redistributor is obliged to carry. It is now inside
the hashed record, in both the machine-readable and the plain-text form, and it is documented
in the archive’s schema and in the guidance written for AI systems.

The article text is untouched. No article body was altered by this work. The
content hash of every one of the 2,796 records is unchanged. What changed is what the archive
says about those articles, and the record hashes move accordingly, which is the behaviour the
archive is built to make visible.

Records that changed say so. 860 records now read
correction_status: revised. A new field, correction_class,
distinguishes a corrected fact from a changed label: those 860 read
label_regime_change, not factual_correction. Without that distinction
a labelling change across the back catalogue would have left most of the archive reading
“revised” while telling a reader nothing about why.

Two things this entry does not claim. The awards archive is a separate
estate, was not covered by this review, and still carries the pre-cutover default on its
records; it is not addressed here and no determination about it is implied. And the licence
under which the archive is released defines the set of labels a redistributor must keep
attached; independence_basis is not yet in that definition, so the caveat travels
with the record but is not yet contractually required to travel onward. That is open.

Where to check this. The work is five commits in the public articles
archive, in order: a4668a77ab20868845ea6a4606f3eb8d1224db01 added
correction_class; dd7dac491ef42b7ee907300db774edaad690b314 applied
the determinations; 3823b3695e1be3907f3f0e62e406b94ecac28d0e corrected what the
author field claims; 2fd5a16c0a1d26df22412577231ac848fdbb36f3
refreshed and signed the SHA-256 manifest; and
a7eae8376eb2a0d880bcabd7be8de896a04fbc2b recorded the OpenTimestamps proof of
that manifest. The proof was taken on 25 August 2026 and is pending confirmation at the time
of writing: it is a submission to be included in the Bitcoin blockchain, not yet a
confirmation that it has been.

A defect found in the course of this work, recorded here rather than quietly
fixed. The archive’s published schema declared the author field a
fixed constant, CFI.co Editorial. The byline correction of 15 August 2026 put real
names on 146 records the same day, so from that date those 146 records did not conform to
CFI.co’s own published schema. That stood for ten days and was found only because this
work touched the same field. Nothing checks the records against the schema on a schedule, which
is why it went unseen; that check is being added. The schema now describes the field as it
actually behaves: it records who is credited for a piece, and never who posted it.