Method, gaps and corrections

Compiled 26 July 2026, last revised 14 August 2026. Read this before relying on the corpus for anything consequential.

Method

Social media as a source. A small number of documents in this corpus are social media posts, and they are labelled as such everywhere they appear: a red badge on the browse card, a red badge on the document page, and a standing warning above the text. They are held under one rule — the account must be an institution or a public figure speaking on the record, and the post must be one a reader could check for themselves. No post by a private individual is collected, and no comment thread is collected at all. That line matters for a disaster: the first days of social media around a mass-casualty event contain the names of people reported dead who were alive, and this archive will not fix that in place.

Social media is used mainly as a lead source rather than as evidence. Where a post points at something real, the archive chases it to published journalism and cites the journalism. The Andrew Donald account is the worked example: it was found as a post by a Member of Parliament, and it is recorded here both as that post and as HGPTV's news report of it, with the report carrying the weight and the post recorded as its origin. Nothing self-published is treated as established because it was posted.

A new tier: the expert submission. On 30 July the corpus took in one document that is none of the three things everything else here is. It is not journalism: no outlet commissioned or edited it. It is not a state release. It is not a social media post. It is an unsolicited analysis, written by a named person with stated professional qualifications, sent to this archive by its author and offered as an aid to the Commission of Inquiry. The document is Robert C. Millington's 253-page reading of the Guyanese maritime statute book. It was the only one of its kind for five days. On 3 August the same author sent a second, and the rule written down in advance did its work: the second is admitted on the same four conditions and published in full. It also did something the rule did not anticipate — the second submission states positions inconsistent with the first, without saying so, and the archive has had to withdraw a finding it published from the earlier document. That is recorded on the figures page and on the analysis page, and the lesson is written into the tier: one qualified person's reading is not corroboration of itself, and a second reading by the same person is not necessarily a correction of the first.

The rule for accepting such a submission. Four conditions, all of which this one meets. The author must be named — an anonymous analysis is not admissible here at any length. He must state his qualifications, so a reader can judge what he is qualified to say and what he is not: this author states he is a former merchant marine deck officer and ship pilot and expressly states that he is not admitted to practise law in any jurisdiction, which is the more useful half of the disclosure given that the document is about statutes. The reasoning and the citations must be supplied, so the work can be checked rather than trusted — here, chapter and section from a named authorised consolidation. And it must be published in full, so a reader can audit it against any use this site makes of it, rather than taking the archive's summary on faith. A submission that met three of those four would be recorded as a lead, in the way social media posts are, and not as a source.

How it is labelled and weighed. It is marked as a submitted analysis everywhere it appears — a badge on its browse card and on its document page, and a standing note above the text. Every proposition drawn from it on the record and analysis pages is attributed to its author by name, marked Single source where it rests on him alone, and described as an analysis rather than a record. It is not treated as corroboration of anything, because one person's argument is not a second witness. Its author's own stated limitations are carried with it: the consolidation he read is current only to L.R.O. 1/2012, so any finding of his framed as the absence of a regulation needs checking against later instruments and the Official Gazette. And where it states a fact about the events of 18 July rather than a reading of a statute — it gives the distress signal as 22:43, against the 23:01 this archive carries from the state's own information service — it is treated as any other uncorroborated account would be, and the conflict is recorded rather than resolved. Expertise earns a hearing here. It does not earn the status of record.

A broadcast caption track, and how it is flagged. On 31 July the corpus took in its first broadcast interview captured from an auto-generated caption track: an hour-long podcast in which a former Minister of Public Works is questioned about the disaster. It is held under the same rule as everything else and one extra warning. Machine captions of live speech mangle proper nouns — the host's own name renders wrongly throughout, as does the vessel's — so the file carries capture_fidelity: auto-caption, quotations are marked as substantially accurate rather than verbatim, and anything turning on exact wording has to be checked against the audio. The file also records how much of the programme was captured, because 0:47 to 59:55 of a 1:04:15 programme is 92 per cent and not all of it. A caption track is a usable source. It is not a publisher's transcript and this archive does not treat it as one.

Tooling. The research, the retrieval, the cross-checking and the reasoning on this site were carried out with Claude Opus 5, Anthropic's frontier model, working over the corpus described below. Every claim is nonetheless traceable to a named published document, and the adversarial checks that found and corrected this archive's own errors are logged on the revisions page. The model is a tool; the sources are the authority.

Five parallel research streams were run against the open web, plus a sixth gap-recovery pass: Guyanese domestic press; official and state sources; international wires and regional press; vessel and institutional history; aftermath, accountability and public debate; and finally the Commission of Inquiry cluster and press-capacity context. Each source was retrieved individually and written to archive/ as a markdown file with metadata frontmatter recording outlet, author, publication date, URL, retrieval date, tier, stream, genre, casualty figures cited, and extracted key claims.

The corpus was then deduplicated by canonical URL, keeping the fullest capture in each case. Ten superseded duplicates were moved to _superseded/ rather than deleted, so nothing is lost. Two files had been misfiled as official primary sources when they were in fact press reports of official statements; the correctly attributed versions were kept.

Standing rule applied throughout: no file was written containing invented, reconstructed or paraphrased-as-verbatim text. Where retrieval failed, the failure is recorded here rather than papered over. Where the retrieval tool returned condensed prose despite instructions, the file carries a capture_fidelity warning — in those files direct quotations are reliable and connective prose is the tool's compression, not the outlet's wording. Do not quote a flagged file as verbatim.

Corpus: 338 documents, counted by the date each was retrieved. 152 in the first build of 26 July; 40 Guyana Chronicle articles the same day; 17 on 27 July, being international sources recovered through a browser session and the first Kiskadee Watch reporting; 24 on 29 July, covering 28–29 July, four of them social media posts; and 33 on 30 July — eight in the morning (seven pre-disaster context sources recovered while answering a submitted question about the Fort Island ferry operation, plus one submitted expert analysis) 25 in an evening sweep covering 29 and 30 July; 18 on 31 July, all of them dated that day; one broadcast interview of 30 July, captured on 31 July from its published caption track; ten on 1 August — four dated that day, two dated 31 July that the previous sweep missed, and four state releases of 26 and 28 July that this archive had wrongly recorded as not existing; five on 2 August, being the salvage procurement and its coverage, plus a second outlet on the former master; three on 4 August — a second submitted statutory analysis and two reports of the Commission's published terms; and twenty-seven on 13 August, after a nine-day gap in which this archive was not swept and the story moved a long way without it. Those twenty-seven cover 3–13 August in two passes: nine in a first sweep concentrated on Kaieteur News, then eighteen in a second deliberately aimed away from it, at HGPTV, News Source Guyana, Village Voice News, Demerara Waves, Guyana Times, iNews Guyana, News Room and the regional wire — plus, where nobody else had the story, three further Kaieteur reports and one from the Guyana Chronicle. The second pass is why this archive could say that four outlets held the leaked maintenance records rather than one — and the 14 August sweep made it five, and it is the reason the first pass's single-source flag on that story had to be withdrawn within a day of being written; and eight on 14 August, being the fifth outlet on the defect records, a father's account of what his son said about the vessel before he was lost with it, two Indigenous interventions, the Commission's own progress report, and two items recovered from an outlet this archive had wrongly written off as unfetchable. That sums to 338. 307 are dated 19 July 2026 onward; the remaining 31 are pre-disaster context and undated reference material reaching back to the 1930s builder's record — 30 carrying a pre-disaster date and one undated. (An earlier version of this line said 235 documents were dated from 19 July and 30 were not. The error was a sorting one: one undated file records its date as the words "Undated (living document)", which a naive string comparison sorts after a 2026 date rather than before it, so it was counted on the wrong side. Corrected on 30 July.) Note that this is a split by date and the table below is a split by kind; they are different partitions of the same 338 documents and their rows do not correspond. One undated reference work, for instance, is counted here among the 31 and below among the three reference works.

Corrections and questions

Corrections. If something here is wrong, it should be fixed and the fix logged. Every correction made so far is on the revisions page, including the ones that withdrew claims this archive had published. Send a correction — and if you hold the document, send that.

Questions. The questions page is open to anyone. Send a question and it will be answered there by Claude Opus 5 reading across the whole corpus, on the same terms as everything else here: evidence named, limits stated, and the short answer first. That page is analysis rather than sourced record, and it says so at the top. A question the archive cannot answer will be published as such, because what the record cannot answer is itself a finding.

Both routes go through the same form, which delivers by email. No address is published on this site, which keeps it out of reach of address harvesters; the form is the contact route.

What the count contains

A reader was right to object that "200 sources" overstates what this archive holds. In archival usage a source is a document in the corpus, and each of the 200 is exactly that — a distinct published document with its own publisher, date and link. But a general reader hears the number as 200 independent accounts, and that is not what it is. The count is now given as documents, and the composition is set out here so the number carries its own qualification.

The 29 July build added 24 documents: seventeen pieces of Guyanese reporting on the charges, the wreck, the former captain's account and the minister's first response to the resignation demands; three regional carriages of the same story; and four social media posts, which are labelled as such wherever they appear and are the first of their kind in this corpus. One of those posts was, at the time, the only record here of a picket at the Office of the President. That is no longer so: the 30 July sweep brought in four press accounts of the same picket, and the note recording its absence has been withdrawn from the chronology and the positions page.

The 30 July morning build added eight. Seven are pre-disaster context, recovered in the course of answering the first question submitted by a member of the public — the 2022 Fort Island stelling contract, the 2022 resumption of ferry calls there, the May 2026 Independence ferry operation and its aftermath, the December 2025 commissioning of the MV Konawaruk 1899, and an undated trade review of the MV Ma Lisha, the last of which is counted as a reference work rather than as reporting. The eighth is the expert submission, and it is the first document in this corpus that is neither journalism, a state release nor a social media post; the tier and the rule for admitting it are set out under Method above. None of the eight is original reporting on the sinking by an independent outlet, which is why that row is unchanged and why the independence figure below has not moved.

Three qualifications matter more than the totals.

Syndication multiplies one act of reporting. A single Associated Press dispatch appears here via Afro.com, WRAL twice, and the Citrus County Chronicle. One CMC wire appears via both CBC Barbados and the Jamaica Observer. CBC News's report is Thomson Reuters copy, so CBC and Reuters in this archive are substantially the same reporting counted twice. These variants are kept deliberately — which outlet carried which figure on which day is part of the record, and the archive uses it to show two irreconcilable death tolls running side by side in international copy for about forty-eight hours — but they are not independent corroboration.

About a fifth of the documents are the state's own voice. The state-owned Guyana Chronicle supplies 41 and the Department of Public Information 30: 71 of 338 — a share that has fallen from about a quarter to about a fifth, because the 13 August sweeps added twenty-seven documents of which one came from either. The DPI figure rose by four on 1 August, and not because DPI published anything new — because a sweep found four releases of 26 and 28 July that this archive had missed and had recorded as a silence. That is primary material of real value, being what the government said and when, but it is not independent of the body under scrutiny. The state's share fell slightly on 30 July, because the evening sweep added 24 documents from private outlets and one DPI release.

Four documents are duplicates or republications, including the Guyana Chronicle publishing the same report at two dates and URLs, and two syndicated pairs carrying identical headlines.

Counting only original reporting on the sinking by outlets that are neither state-owned nor carrying another outlet's wire copy, the figure is 180 rather than 338. That count is checked at each build rather than carried forward. The morning build of 30 July added eight documents and none of them qualified — six pre-disaster news and government items, one undated trade review of a vessel, and one submitted analysis — so the numerator did not move at all. The evening sweep of the same day added 25, of which 21 do qualify: original Guyanese reporting on the swearing-in of the Commission, the protests of 29 July, the legal reaction to the charges, the Attorney General's remarks, a funeral, a bereaved mother's account, the Ministry's cargo list and two mental-health assessments. The remaining four are one DPI release and three opinion pieces. So on 30 July the independent numerator moved for the first time since 29 July, and it moved by 21. The 31 July sweep of eighteen documents moved it again, by eleven, to 139: original Guyanese reporting of the defence's court challenge, the wharf-design account, the vigil outside the minister's home, the Assakata Village Council resolution and the unanswered questions about two commissioners. The broadcast interview taken in late on 31 July did not move it, and it is worth saying why, because the temptation to count it was real: it is the most substantive new material the archive took in that day. But it is an interview programme produced by an advocacy organisation, not reporting — the guest is an opposition spokesman and the host sets an agenda — so it is counted in the opinion row above. A numerator that admits material because it is interesting stops measuring what it claims to measure.

The 1 August sweep added ten and moved the numerator by five: two iNews Guyana reports of 31 July on the end of the recovery phase and on the President's reported salvage commitment, and three Kaieteur News and Guyana Times reports of 1 August on the new audit team and on the opposition leader's response to the charges. The four DPI releases do not qualify, being state releases; nor does the opinion column of 1 August. The 2 August sweep added five and moved it by five: the salvage invitation carried by the Guyana Standard and News Room, the memorial climbdown, the opposition's response, and a second outlet on the former master. The 13 August sweeps added twenty-seven and moved the numerator by twenty-two, to 173: the four accounts of the maintenance records, the salvage-procurement reporting from Kaieteur and Village Voice, the Commission's own account of itself, the Attorney General's two positions, the end of the search as carried by News Room and iNews, the Region One outreach, the two protests of 12 August, the Opposition Leader on the tender, and the fisherfolk consultation. Five of the twenty-seven do not qualify: the state-owned Guyana Chronicle report; the CMC wire carried by the Jamaica Observer; two opinion pieces, the "Sham Tender" editorial and one column; and the Grenada Chronicle's report of the 3 August hearing, which is counted as original reporting in the composition table but excluded here because this archive could not establish whether it is that outlet's own work or a carriage. Where a document could fall either way, it is excluded from the numerator. The 4 August sweep added three and moved it by two: both Kaieteur News reports of the Commission's terms and of the objection to its Secretary. The second statutory submission does not qualify, being neither journalism nor independent of the archive that solicited nothing and received it unasked.

Outlet names were deduplicated on 27 July. The archive had been holding 51 distinct publisher strings for what were in fact 42 publishers: HGPTV had been recorded five different ways, and Demerara Waves, iNews Guyana, Democracy Guyana, the Department of Public Information and Wikipedia twice each. Those twenty records now carry a single canonical name per publisher, which both cleans the outlet filter and stops the archive overstating how many publishers it draws on.

Four apparent duplicates were deliberately left alone because they are different publishers: BBC News, Canada's CBC News and the Caribbean Broadcasting Corporation in Barbados; the Jamaica Gleaner and the Jamaica Observer; the Guyana Chronicle and the Guyana Graphic; and iNews Guyana, News Room Guyana, News Source Guyana and Things Guyana. The Associated Press entries are also kept separate by carrier — via WRAL, via Afro.com, via Citrus County Chronicle — because which outlet carried a wire report is part of what this archive records. CBC News is now labelled CBC News (Canada) to distinguish it from the Barbadian broadcaster.

After deduplication three publishers still account for 150 of the 338 documents — Kaieteur News 79, the Guyana Chronicle 41 and the Department of Public Information 30. Two of those three are the state. Kaieteur's share is the concentration risk this archive has to watch: it supplied eleven of the twenty-seven documents added on 13 August, and it broke the two largest stories of the period. The second sweep of that day existed to test whether anyone else had them. On the maintenance records three other outlets did; on the salvage documents one did. A further dedup was needed on 30 July: HGPTV had drifted back into two publisher strings, "HGPTV" and "HGPTV Nightly News", across sixteen files. They are one publisher and now carry one name.

Retrieval failures — what could not be captured

Guyana Chronicle: closed on 26 July. In the first build this was the largest single gap in the corpus. Every fetch attempt, across bare, www. and /amp/ variants, hit a JavaScript redirect wall, and the state-owned Chronicle was absent from the archive entirely. It has since been retrieved through a browser session: 40 articles dated 22–26 July, of which 38 carry full text. Two are flagged thin-capture and marked as incomplete on their source pages. The change is logged on the revisions page.

Kiskadee Watch: retrieved 27 July. The BBC credits this independent Guyanese outlet as the origin of Elena Moonsammy's account. Twenty-one of its reports are now in the corpus, and they carry survivor testimony on the vessel's machinery that is not present elsewhere. Ignite News, credited by CBC as the origin of Leon Murray's account, is now represented by its own 18–21 July chronology. The specific Murray interview was not found on the outlet's site and remains outstanding; CBC's account of it is in the corpus, the original is not.

Village Voice News: resolved on 13 August, and the note that stood here was wrong. This section said the outlet was robots-disallowed and "therefore wholly unrepresented". It fetched without difficulty on 13 August, and three of its reports are now in the corpus — including its independent account of the salvage procurement conflict, which supplied the MARAD Facebook date this archive would not otherwise have. The earlier failure was real; the conclusion drawn from it was not. A single failed fetch had been left standing for a fortnight as a settled fact about an outlet, and a Guyanese newspaper covering this story heavily was absent from the corpus for that whole period as a result. Retrieval failures are re-tested from now on rather than carried forward. The Edghill-suspensions piece named here, which concerns the suspensions of 20 July 2026, is still outstanding.

Guyana Chronicle: rate-limited on 13 August. The state-owned daily returned HTTP 429 on every attempt across two sessions and several URL variants. One report in the 5–13 August window is in the corpus only through a syndicated republication by the Guyana and North America Trade Association, and is marked as a partial capture on its own page; two others — the Chronicle's account of the search concluding and of the inter-agency review, both 5 August — are known to exist by headline and URL and are not in the corpus, because this archive does not record the contents of pages it has not read.

Kiskadee Watch: failed on 13 August, worked on 14 August, and the failure had been costing this archive for a fortnight. Every attempt on 13 August returned a robots.txt read timeout, and the note that stood here said six of its reports were known only by headline. Re-tested on 14 August, the outlet fetched without difficulty. It holds a fifth independent account of the internal defect records, published on 12 August alongside three of the four this site had; the verbatim MARAD statement, which showed that a quotation this archive had shortened the day before was correctly longer; the only account in this corpus of a crew member's family being told the vessel had recurring problems; and the only passenger description of what the voyage was actually like. Six of those are now in the corpus and none of it would be here if the failure had been left standing a second time. That is twice in two days that a retrieval failure recorded as settled turned out to be a fetch that simply needed repeating, and it is now a standing rule: no retrieval failure is carried forward without being re-tested at the next sweep. That matters more than an ordinary fetch failure: this archive has recorded since the first build that Stabroek News, the country's principal independent daily, ceased print in March 2026, and Kiskadee Watch is the independent outlet whose reporting has most often carried survivor testimony absent elsewhere here.

Paywalls and 403s. CBC's ferry-sinks-guyana piece was recovered on 27 July through a browser session, along with the Guardian and Deutsche Welle. Still outstanding: CNN (both cnn.com and edition.cnn.com, where the browser extension has no read permission for the domain), four Washington Post AP-syndications (three recovered via WRAL and Citrus County Chronicle instead), France 24, Telesur in both languages, El Universal (Venezuela), the New York Times, Radio New Zealand, and Stabroek's own 15 March liquidator follow-up.

BBC and Reuters: resolved on 27 July. The first build recorded that no BBC byline and no directly-bylined Reuters article had surfaced through any query, and flagged this as a limit of search reach rather than a confirmed absence. Both exist and are now in the corpus: BBC News, "Death toll from Guyana ferry disaster continues to rise," by Vanessa Buschschlüter and Tom Bennett, 19 July, updated 22 July; and Reuters, "Twenty-seven bodies recovered after ferry capsizes off Guyana," by Kemol King, 20 July. Both were found through the reference list of the English Wikipedia article on the vessel, which is itself now in the corpus. The earlier statement was wrong, and the search method that produced it — keyword queries against the open web — is what failed.

Two Kaieteur pages served submission stubs rather than article bodies: "MV Barima tragedy exposes Guyana's deep governance failures" (Tevin Skeete) and one other letter. The important Kaieteur piece, "Major discrepancies found on MV Barima manifest," failed two subagent attempts, which returned only condensed output, and was then recovered verbatim by the lead researcher — it is now the single most valuable news document in the archive.

Ship registries were largely unreachable: VesselFinder and Clyde-built Ships are robots-disallowed, MarineTraffic returned 403. Vessel provenance therefore rests on the Paxman History Pages engine register, which is fortunately the best available source anyway.

Substantive gaps — documents that do not appear to exist publicly

These are recorded because a reader needs to know which documents were searched for and not found.

No certificate. Nothing in the corpus — not one government release, not one ministerial statement — names a class certificate, a passenger safety certificate, or a load-line survey date for the MV Barima. Seaworthiness is asserted entirely on a docking chronology. The certified capacity that underwrites MARAD's "not overloaded" finding has never been published, and MARAD's Director-General cannot say when or why it rose from the 1938 rating of 150 passengers to 394.

~~No terms of reference.~~ CLOSED, 4 August. The President published the Commission's terms in the Extraordinary Gazette, reported that day: six heads — seaworthiness and safety, passenger and cargo operations, crew qualifications and conduct, search and rescue effectiveness, regulatory oversight by state entities, and human factors in emergency decision-making — with a two-month deadline for final written findings, public hearings, a power to issue interim safety reports, and explicit authority to refer evidence of criminal negligence, official misconduct or statutory breaches to prosecuting authorities. A secretary has also been named, attorney Javed Shadick, closing a second gap this page recorded as open on 30 July. What is still missing: a Statutory Instrument or Gazette number, and a budget figure. The account below is left standing as the record of what was absent for the first seventeen days.

The gap as it stood. This gap had narrowed twice and was not closed. A chair and four commissioners were named on 26 July, and on 30 July all five were sworn in at State House before Chief Magistrate Judy Latchman, the chairman virtually. The Department of Public Information's release of that day states the enabling statute for the first time — section 2(1) of the Commissions of Inquiry Act, Cap. 19:03 — which confirms what Kiskadee Watch reported on 27 July and what the AFC had proposed. What is still not in the corpus: any gazetted instrument or Statutory Instrument number, any published terms of reference as a document rather than a paraphrase, a named secretary (Demerara Waves states flatly on 30 July that none has been named), a reporting deadline, a budget figure or appropriation, rules of procedure, or a call for submissions. President Ali is reported as saying the commissioners will determine their own procedures, which means the procedural rules did not exist as at 30 July. The swearing-in itself was closed to the press and streamed on the President's social media accounts instead.

No official death toll. The state publishes bodies recovered and has never published a death toll. It is not true that the residual never appears in a government document — DPI's own release of 30 July gives "30 could not be accounted for", and the Prime Minister gave 30 unaccounted for on 24 July. What has never appeared is the sum: no government document adds the recovered to the unaccounted for and states what that means, so the roughly 103 people who did not come back have never been counted in one place by the state. (An earlier version of this line said the residual had never appeared in a government document at all. That was wrong and is corrected.)

No Auditor General finding specific to the T&HD was located, despite targeted searching, and no trade union statement of any kind — notable given that the suspended and detained are wharf and marine workers. On 31 July the corpus took in a possible explanation for the first of those, and it is worth separating from the gap it explains. Single source David Patterson, a former Minister of Public Works, stated in a 30 July broadcast interview that the department "has not turned a profit since… 2007 or 2009" and has not been officially audited by the Auditor General's office since "most likely I think 2005, 2006." If that is right, the reason targeted searching found no audit finding is that there has been no audit for roughly two decades — which would convert an evidentiary gap in this archive into a governance finding about the operator. It is one man's recollection, hedged twice in his own words, uncorroborated by any document here, and captured from a machine caption track. It is recorded as a lead, not as a fact, and the Auditor General's annual reports remain on the retrieval list below.

A correction, made on 1 August, to a claim this page carried for six days. This paragraph previously stated that the Department of Public Information published seventeen releases dated 19–25 July, then four days of nothing, then one release on 30 July. That was wrong. A sweep on 1 August found DPI releases dated 26 July and three dated 28 July, all now in the corpus. The four days of silence did not happen; what happened is that this archive's sweeps of 26–29 July did not reach dpi.gov.gy. The claim is withdrawn.

It is at least the eighth claim of absence this archive has published and had to withdraw — the revisions page logs the others — and the pattern in most of them is identical: a gap in the searching was recorded as a gap in the record. The rule stated on 30 July applies again without amendment — an absence claim is only as good as the search behind it. What can now be said, and only this, is that the corpus holds twenty-two DPI releases dated 19–30 July, and that no attempt has been made to establish that this is all of them.

No GDF, MARAD, Police Force, Office of the President or Ministry of Public Works web publication on the disaster, other than through DPI. MARAD issued no advisory, no notice to mariners and no certification statement. parliament.gov.gy press releases stop at April 2026, its motions index returned empty, and no Hansard exists for the July sitting. The state's published output on the disaster runs almost entirely through DPI and through ministers' Facebook posts. The suspension of the loading management team was announced by Facebook post; the declaration of national mourning was published as a DPI release and reported as a proclamation.

No IMO statement. CARICOM and UNDP Guyana issued statements. No statement by the International Maritime Organization appears in the corpus, and TIGI's call of 23 July for an IMO-led investigation is still unanswered — TIGI's president spoke publicly again on 30 July and did not renew it. PAHO is no longer absent: its Country Representative praised the state's response on 30 July, but incidentally, at the signing of an unrelated agreement, and on public-health rather than maritime ground.

No statutory text. The corpus now holds a detailed reading of Cap. 49:01, Cap. 49:04, Cap. 49:07 and Cap. 50:01, with chapter and section cited, in the expert submission described under Method. It does not hold the instruments. No authorised text, no subsidiary regulation and nothing from the Official Gazette has been retrieved, and no post-2012 instrument has been checked — which is the specific gap the submission's own stated limitation points at, since most of its findings are findings of absence.

Four named institutional documents that are not here and may exist. The submission identifies the IMO Member State Audit report on Guyana, Guyana's corrective action plan arising from it, the progress reports on that plan, and the report and recommendations of Professor Duke Pollard on updating Guyana's maritime legislation. Unlike most of the gaps on this list these are not inferred absences: they are named, dated and attributed to a 2019 public statement by MARAD's then Director General. This archive has not established that any of them exists and has not attempted retrieval.

Structural finding that shaped the corpus

Stabroek News, Guyana's principal independent daily for roughly forty years, ceased print publication on 15 March 2026 in voluntary liquidation — four months before the disaster. The consequence for this research was that repeated searches for Stabroek coverage of the sinking returned nothing. Two of the corpus's pre-disaster sources on vessel condition — Capt. R. E. W. Adams's 2015 letter naming this vessel, and the 2012 piece on the Sprostons-built fleet — come from Stabroek's letters and features pages.

Independent print scrutiny in Guyana narrowed sharply with Stabroek's closure. Kaieteur News, at 79 documents, is the largest single contributor here, and civil-society bodies such as the GHRA, TIGI and the APA are issuing statements directly rather than arguing in a newspaper's letter columns. But the state is close behind it and larger in aggregate: the state-owned Guyana Chronicle supplies 41, and with the Department of Public Information's 30 releases the state or a state-owned outlet accounts for 71 of the 338 documents. Kiskadee Watch, added on 27 July, is a further independent Guyanese voice. Readers should weigh that distribution: it reflects which Guyanese outlets were publishing, and which were state-funded, in July 2026.

Priority follow-ups

The Guyana Chronicle hole has now been closed (see above). The remaining highest-value targets are documentary rather than journalistic: the March 2026 tender documents that reportedly show engineers flagging deteriorated hull sections before the sinking; any survey, load-line or capacity certificate for the vessel; the Commission of Inquiry's gazetted instrument, which had not been published when the commissioners were named on 26 July, with attention to whether its terms reach loading authorisation and procurement or stop at loading execution; the Auditor General's reports on the T&HD; and the 19 September 2015 Kaieteur inspection report cited by Adams, which was searched for and not found.

Four more were added on 30 July, all arising from the expert submission and all of a kind this archive has not previously chased: the authorised texts of Cap. 49:01, Cap. 49:04, Cap. 49:07 and Cap. 50:01 together with their subsidiary regulations and anything gazetted since 2012, without which no statutory proposition on this site can be verified; the IMO Member State Audit report on Guyana and Guyana's corrective action plan and progress reports; the Pollard report on updating Guyana's maritime legislation; and contemporaneous 2019 reporting of Claudette Rogers's statement of 27 March 2019, which currently reaches this archive only through the submission that quotes it. Separately and more urgently than any of them: anything bearing on section 23 of the Transport and Harbours Act and whether a six-month limitation period is in fact running against claims arising from this casualty, which is set out on the figures page and is a matter for Guyanese lawyers rather than for this archive.

On the live story, the variables worth monitoring are whether the wreck is righted and what that does to the count, whether the residual of roughly thirty missing is ever officially named, whether charges are laid and against whom, whether Parliament sits before the Commission is constituted, and whether any pre-18-July document surfaces bearing a named official's knowledge of the hull's condition.

The verification pass

After the deliverables were drafted, a separate adversarial pass re-checked every name, figure, date, quotation and claim-of-absence against the archive, with instructions to falsify rather than confirm. It found eighteen substantive errors, all now corrected. They are recorded here because the pattern is instructive for anyone extending this work.

The most common failure was quotation drift — a paraphrase hardening into quotation marks across drafts. "MV Barima was seaworthy" is DPI's headline, not Edghill's words; his actual formulation was the much narrower "there was no report of any kind that suggested the vessel was unseaworthy." "Frequent mechanical failures" was being quoted from a tool-condensed file whose own note warned that only one unrelated sentence in it was verbatim. "Not compensation" was a truncation of a Prime Ministerial statement about sequencing into an apparent legal characterisation the state never made.

The second pattern was attribution slippage. Saiku Andrews, an opposition MP voicing a claim, had become a Jamaica Observer byline. Deon La Cruz, an opposition MP, had become a reporter. Toxicology disclosure had migrated from Edghill to Jagdeo. A Bulkan letter's headline had become "Kaieteur's framing." A letter-writer's argument had become a newspaper's editorial position. In each case the substance was right and the source was wrong, which is the more dangerous kind of error because it survives casual checking.

Third, figures merging. The vessel's capacity had been rendered as a single "284 tons / 394 passengers" when in fact Thomas said 394/126 tonnes, then 394/120 tonnes, and Edghill said 397/284 tonnes — three inconsistent official statements whose inconsistency is itself the finding. A search area "expanded to 32 square miles" was actually expanded by 32, on top of an earlier 400.

Fourth, inference presented as record: the Ma Lisha dated to 2021 rather than its actual 2023 commissioning; "sole vessel serving Region One" where the archive supports only "sole vessel on the Port Kaituma route"; the 2017 refit's "installation of two engines" hardening into a claim that the 1939 machinery was entirely replaced.

And one claim of absence that was simply false: the corpus does contain Reuters copy, indirectly.

Two things the pass surfaced that are now flagged in the analysis rather than fixed. The Prime Minister stated on 25 July that the state has "no such record" of a Mayday from the captain — which contradicts the 23:01 timing on which every published chronology, including this one, depends. And Elena Moonsammy (Guyanese press, lost seven children she travelled with, sole source for cargo at the stern) and Helena Moonsammy (AP and Gleaner, a grandmother who lost four grandchildren and spent nine hours in the water) may be two different people; earlier drafts merged them as one name with variant spellings, which the archive does not support.

Directory structure

archive/ holds the 338 full-text sources, one file per source, plus seven _manifest_*.jsonl files carrying structured metadata and research annotations for 192 of them; the manifests were written per sweep and later sweeps, including this one, are covered by the summary sidecars rather than the manifests, and _synthesis_vessel-dossier.md, a fully cited dossier on the vessel and the T&HD fleet with an explicit unverified-and-contested section.

deliverables/ holds 00_SOURCE-CATALOGUE.md (the annotated listing of all sources with summaries, chronological then contextual, plus a distribution table), 01_TIMELINE.md (reconciled chronology from 1936 to 26 July 2026, with conflicts marked), 02_CONTESTED-FACTS.md (figure-by-figure reconciliation, single-source claims, reliability guidance), 03_SYSTEMIC-ANALYSIS.md (the four-pillar systemic brief), and this file.

_superseded/ holds the ten deduplicated files, retained for audit.