What doesn’t add up

Produced by Claude Opus 5, Anthropic's frontier model, reasoning across every document in this corpus, and compiled 29 July 2026. No human investigator assembled these findings. The model read the whole record, looked for what fails to reconcile, argued both sides of each item, and named the documents behind every claim so the reasoning can be checked rather than trusted.

That is the method and also the warning. This page is inference. It is not the record, and where it and the record pages disagree, the record pages are right.

What an anomaly is here. A gap between two things the record contains: a statement and a document, two statements by the same body, a decision and the assessment said to justify it. An anomaly is not a finding of wrongdoing. Most of the ones below have an ordinary explanation available, and several are probably explained by it. An absence counts as evidence only where this analysis can establish that the document should exist; in several places it cannot, and where that is so the page says so rather than leaning on the silence.

How claims are labelled. Each finding opens with the anomaly, which is documented fact drawn from named documents and dates. The two explanations that follow are inference — that is what they are for. Anything neither documented nor inferable from the documents is marked speculation in bold, and there is exactly one such passage, at finding 4. It is this archive's own, offered because it is a benign reading nobody else has put forward.

On citing an absence. Several findings below rest on something the corpus does not contain. An absence cannot be cited: no document establishes that a document is missing. Where that is the position, the citations point to the documents in which the state asserts the thing the missing paper would prove, and the prose says which part is the absence.

Four things this archive got wrong

An analysis page that cannot correct itself is not worth reading. These four errors were this archive's, they were found by checking the archive against its own corpus on 29 July, and three of them made the record look worse than it is.

1. The "284" is a category error, not a contaminated figure. This archive had flagged Minister Edghill's 19 July figure of 284 tonnes as possibly contaminated from the Kalliopi N's specification, because 592Hub was the only other source carrying it — and 592Hub itself notes that 284 is also the Kalliopi N's passenger figure. That flag was wrong. The Wikipedia capture of 26 July gives the Barima at 284 GT, 198 DWT, capacity 394 passengers and 18 crew. Gross tonnage measures enclosed volume, not cargo weight, so 284 is the vessel's own number and is not a cargo limit at all. Independently, Azruddin Mohamed is reported by Kaieteur News and HGPTV on 29 July citing a 198 deadweight exceeded by about 70 tonnes. Three records here carry the figure 198 — the vessel record and the two reports of Mohamed — and the archive had never connected them.

2. "162 passengers and 17 crew" is not inconsistent with 179. The archive flagged the Wikipedia lede as Contested against the complement of 179. 162 + 17 = 179. It reconciles exactly. The discrepancy was arithmetic that had not been done.

3. The search-area figures reconcile. The figures page carried "expanded by 400 sq mi" and "total area ~2,100 km²" as rival versions. They are the same account of the same search: 1,040 km² plus 400 square miles (1,036 km²) is about 2,076 km², and Kaieteur News of 22 July states both halves in one sentence — an expansion of 400 square miles producing a search area of 2,100 square kilometres. That row was a false discrepancy too.

4. The distress-call contradiction is weaker than this archive claimed. Prime Minister Phillips's "We have no such record" — said at the Friday 24 July briefing and reported by Demerara Waves the next morning — answers a compound proposition: that the captain issued a May Day call and asked passengers to put on their lifejackets. It is a denial about the captain's conduct, and not, on its face, a denial that a signal reached shore. DPI's own release of 19 July states a distress call was received at around 11:01 p.m., and the Guyana Chronicle repeats that on 23 and 24 July. The contradiction is real but narrower than the archive had it. What keeps its full weight is the separate fact that the rescue timeline has been asked for publicly and never published.

Two further changes that are not errors. The March 2026 tender is downgraded from a finding about the hull to a scheduled docking cycle, and is recorded that way at finding 14. And a claim that had circulated in this project's working notes — that a barricade was placed at Charity and only state media admitted — is not in this corpus at all. Every document was searched; Kiskadee Watch, the outlet it was attributed to, has four files here and none mentions it. It is dropped, and should not be repeated unless someone produces a source for it. The four Kiskadee files are cited so that anyone can check that for themselves; the absence itself cannot be cited.

The findings

Ranked by severity, as in the analysis. Four are critical, most of the rest are high, and two are qualified downward. Where the answer to "which does the record support" is neither, that is what is written.

1. Three cargo figures are three different quantities — CRITICAL

The anomaly. Minister Edghill said on 19 July that the vessel was licensed for 284 tonnes against 268 tonnes manifested, and described a line-by-line audit he had ordered as giving him 260 early in the day; MARAD's Director-General Capt. Stephen Thomas gave 126 tonnes on 21 and 22 July and 120 tonnes at the briefing of 24 July, reported on the 25th; the Wikipedia capture of 26 July gives 284 GT and 198 DWT; and Mohamed, via Kaieteur News and HGPTV on 29 July, cites a deadweight of 198 exceeded by about 70 tonnes. These are not four versions of one quantity. Gross tonnage is volume, deadweight is total carrying weight including fuel, water and stores, and a cargo rating is a third thing again.

What would explain it innocently. Officials speaking extempore in the first week of a mass-casualty event, each reading a different line of a different document, is the ordinary way this happens. A minister asked on day one what the ship was licensed for could reasonably give the registry's headline tonnage in good faith; nothing about "284 GT" announces itself as a volume measure to a non-mariner. The 120 and 126 pair is consistent with two ratings of different vintage — Kaieteur News reports the same official being unable to say when or why the vessel's certified capacity changed from an original 1938 rating of 120 tons of cargo and 150 passengers — and a Director-General giving a lower figure days later is a regulator correcting himself rather than one hiding anything. And the manifested 268 is itself a soft number: that manifest was wrong about people by 46, the audit Edghill ordered returned 260, and a deadweight is not a cargo limit, because consumables sit inside it.

What would explain it otherwise. The figure that was used publicly to clear the vessel on overloading, on day one, was a volume measure, and neither the ministry nor the regulator has identified that in the ten days since. On the reading the analysis considers most likely, manifested cargo of 268 tonnes exceeded a deadweight of 198 by roughly 70 tonnes.

Which the record supports. The category error is documented: two figures for the same vessel, 284 GT and 198 DWT, sit in this corpus, and the 284 was publicly described as a licensed tonnage. Whether the vessel was in fact overloaded is not settled in either direction, and this analysis does not settle it — the 268-against-198 comparison is inference, and it sets a manifest figure against a deadweight rather than against a cargo limit. Note also that the two accounts of what "120" refers to differ: Demerara Waves has Thomas giving 394 passengers and 120 tonnes as the vessel's capacity, while Kaieteur News has 120 tonnes as the 1938 original he could not explain the change from. What the record does establish is that the people responsible could not state the permitted load consistently, and that the number used to clear the vessel was of the wrong kind.

The document that would settle it. The International Tonnage Certificate or the registry entry stating gross tonnage and deadweight, with the load line certificate.

2. MARAD relied publicly on a certificate it has not produced — CRITICAL

The anomaly. MARAD's "not overloaded" finding of 22 July rests on the load line: the Director-General's stated reasoning is that the load line was only just submerged on departure and that this does not indicate overloading. The paperwork behind that has not been published. On 19 July, day one and three days before the finding, the Guyana Graphic reported that the vessel's gross tonnage, deadweight and certified passenger and cargo limits were not in the public domain, and demanded their release along with inspection records. On 28 July Christopher Ram called for all the Barima's surveys and inspection reports to be secured against alteration or destruction — a demand for preservation rather than publication. On 29 July former captain Andrew Donald called for the vessel's safety certificates and inspection history to be opened to public examination, and asked on what basis its safety certificates had been renewed. Nothing has been produced. Capt. R. E. W. Adams's 2015 letter sets out the survey and certification convention he says applies to this fleet: annual survey by a qualified surveyor, vessels with major deficiencies held from sea until corrected, minor deficiencies noted at the back of the certificate.

What would explain it innocently. Ship's certificates are the vessel's papers, ordinarily held aboard and by the operator rather than published by the regulator — and the ship is on the seabed. A regulator whose evidence may now sit with investigators, the Director of Public Prosecutions or a commission that has not yet sat has an obvious and proper reason to release nothing piecemeal, and a request from a newspaper or a private citizen is not a legal demand. This analysis cannot establish that certificates of this kind are normally published in Guyana, and if they are not, a regulator declining to publish one is behaving normally and the absence carries much less than it appears to.

What would explain it otherwise. A regulator made a categorical public finding on 22 July, said what it rested on, and the document has not appeared across ten days in which two separate public demands for disclosure and one for preservation were made.

Which the record supports. Neither, yet. What is documented is the asymmetry: the finding was published and its basis was not, and the demands are on the record and unanswered. The absence of the certificate is not itself citable; what is cited above is the finding, the reasoning given for it, and the three demands. The archive holds no rule requiring publication, so the absence cannot be read as refusal.

The document that would settle it. The certificate itself, or a statement from MARAD explaining what it holds and on what basis it does not release it.

3. The wreck: the assessment cited as the reason post-dates the document it justifies — CRITICAL

The anomaly. On 23 July, reported the following day, President Ali told relatives that a full salvage might become the only remaining option if divers could not get inside. At the briefing of 24 July the Prime Minister said several boats were on standby to right the vessel and that if it were righted more bodies were expected to be found. On 25 July a draft Declaration of Agreement was circulated to families asking them to concur that the vessel remain undisturbed in its present resting place as a sacred memorial. On 27 July the French dive team told relatives that entering the wreck was too dangerous — the assessment that has since been cited in support of leaving it. On 28 July the Prime Minister said "I don't think we've contacted anybody to raise the vessel" and that the vessel was "likely to remain there as is"; Demerara Waves reported him the same day saying international salvage companies were prepared to raise it, and that the wreck likely contains additional remains and investigative evidence. On the same 28 July he offered the testimony of the crew and those aboard as what would establish whether the engine had failed — and three crew were charged with murder that day.

What would explain it innocently. Nothing here requires the state to have known on 25 July what the divers would find on 27 July. A government facing a hull that had already resisted attempts to right it, in a river mouth, with bodies inside, might reasonably prepare a document for a likely outcome before that outcome was confirmed — and Phillips called the letters "just for discussion", which is what an early draft looks like. The two 28 July statements are not necessarily inconsistent: "we have contacted nobody" describes what the state has commissioned, "salvors were prepared" describes what third parties have offered, and both can be true of the same day. Nothing in this corpus identifies an instrument requiring a wreck to be raised, and the President is reported as assuring families the designation would depend on their wishes. Offering crew testimony is not offering a substitute for the hull if the state does not believe the hull holds decisive evidence.

What would explain it otherwise. A document asking bereaved families to agree that the wreck stay where it is was circulated before the technical assessment later used to justify leaving it, and one day after those families' briefing was told righting was imminent. The single physical object that could test the state's account independently of paper is the object proposed to be left, and the substitute offered is the testimony of men who were charged that same day and may be advised to say nothing.

Which the record supports. The chronology is documented and it is not in dispute: 23rd, 24th, 25th, 27th, 28th, in that order. The intent behind it is not documented anywhere, and this page does not impute one. What the sequence establishes is that the justification post-dates the decision it justifies, which is a fact about order and not about motive.

The document that would settle it. The Declaration of Agreement with its drafting date and authorship, and the French team's written assessment.

4. The charging pattern — CRITICAL

The anomaly. Three men were charged on 28 July, day ten, jointly, with murder contrary to common law. The count is reported as 72 — where every official figure since 24 July gives 73 bodies recovered. Unresolved No official document states any count at all: the police release quoted in the reporting gives no number and neither does the Director of Public Prosecutions' statement. The figure reached the record through court reporting on 28 July and, for its derivation, through defence comment on 29 July. All three charged are employees of the operator. Kaieteur News's account of the 19 July press conference names the two crew who tested positive as the captain and the first engineer; the first engineer is not among the charged and the chief mate is. The charges were laid at a point when, on the last statement in this corpus about it, the Commission of Inquiry's legal instruments were still being formalised.

What would explain it innocently. Charging decisions follow evidence about conduct, not toxicology: a positive drug test is not an element of any offence, and a prosecutor who charges the three people whose acts the file describes — the master, the officer of the watch, the man responsible for goods — while not charging a man who tested positive but whose conduct is not alleged, is doing the job correctly rather than incoherently. The count is not an official figure at all, so 72 against 73 may be an early number repeated in court reporting before the file was complete. And the DPP's office is constitutionally independent under Article 187 and holds that the two proceedings can run simultaneously, citing two Guyanese precedents; on that view an inquiry whose instruments are still being drawn has no bearing on when charges are laid.

POSSIBLE BENIGN RECONCILIATION — this is speculation, and it is this archive's own. No source in the corpus offers it and no document supports it. Prime Minister Phillips is reported by Kaieteur News on 29 July as saying 77 were rescued and one of them died after rescue. If the 73rd recovered body is that person — someone who died ashore rather than at sea — then 72 bodies came out of the water, and a charge of 72 counts is exact rather than one short. It is unconfirmed, it is inference from an arithmetic coincidence, and it should not be repeated as though it were established. It is recorded because it is the most economical explanation available and nobody has put it forward.

What would explain it otherwise. Every person charged works for the operator; in this corpus no charge, suspension, leave or detention touches the regulator, the departmental board or the ministry. The state's two visible counts of the dead, the one in the briefings and the one in the charge, do not agree, and no official document states either the count or its basis.

Which the record supports. On the 72/73 gap: neither, and the speculation above is the archive's own attempt at the benign reading rather than a finding. On the distribution of charges: what is documented is where they fall, not why, because the file is not public. The three accused have not been tried, and nothing here is a view on their guilt or innocence.

The document that would settle it. The police charge sheet with particulars — which would also state what the state alleges actually happened.

5. The cause was named on day one, and the charges landed exactly there — HIGH

The anomaly. On 19 July, before any investigation had reported, Minister Edghill named the failure in one word — "mischief" — and in the same appearance rejected engine or mechanical failure and rejected overloading. Kaieteur News's account of that press conference summarises his position as being that the disaster stems directly from operational breaches at the ground level rather than regulatory failure, because the Transport & Harbours Department is the service provider and MARAD strictly the regulator. That sentence is the newspaper's wording, not a quotation of the minister, and it is recorded here as such. Nine days later the charges fell exactly on that boundary and nowhere else.

What would explain it innocently. A minister on the evening of day one had the manifest audit in front of him, a passenger count that did not match, and crew toxicology; describing that as operational is a fair reading of what he had, and "mischief" is a word about ticketing and loading rather than a verdict on a criminal charge. That the DPP later charged operator employees is not evidence that a press conference directed her — the office is constitutionally independent under Article 187, and the two would coincide anyway if the operational account is simply the correct one.

What would explain it otherwise. The responsible minister publicly excluded his own ministry's layer from the causes on day one, and the eventual charges reproduced exactly that boundary.

Which the record supports. The statements, their dates and the coincidence of scope are documented. Nothing in this corpus connects the two, and the archive holds nothing on how the police file was built. The finding is that the boundary was drawn early and publicly by the layer with an interest in where it fell, and it goes no further.

The document that would settle it. The police file's scope of investigation, or any record of the instructions under which it was opened.

6. The distress call — HIGH

The anomaly. As corrected above, the contradiction is narrower than this archive had it. What survives correction is this: Janette Bulkan asked publicly on 25 July that the Maritime Search and Rescue Coordination Centre publish the rescue timeline, and it has not been published, and no state release in this corpus names the body that received the call. Unresolved DPI's release of 19 July gives the time and not the recipient. The 23:01 timing on which every published chronology rests, including this archive's, is carried by outlets — Deutsche Welle and CBC News naming air traffic control, Kaieteur News naming an air traffic control tower at Timehri, iNews and Guyana Times attributing receipt to aviation authorities — rather than by a state record of receipt.

What would explain it innocently. A rescue log is an operational record, not a publication; no rule in this archive requires its release, and it may now be evidence in a criminal file. That several outlets name air traffic control rather than a maritime centre is consistent with the call having reached shore by aviation channels, which would explain both the timing everyone reports and the absence of a maritime record of it.

What would explain it otherwise. Ten days on, no state release has said who took the call, and the timeline that would show what was tasked and when has been asked for and not supplied.

Which the record supports. The request is documented and so is the state's own account of the time. That no log has been published is an absence and cannot be cited; whether such a log exists in a form that could be released is not established by anything in this corpus.

The document that would settle it. The MSRC log for the night of 18–19 July.

7. Two government sources name different crew as testing positive — HIGH

The anomaly. Contested Kaieteur News's verbatim account of the 19 July press conference has the captain and the first engineer testing positive, and adds, attributed to authorities rather than to the minister, that "the mate was subjected to the same drug panel, but the results were not the same" as the other crew members. DPI on 22 July states that "the captain and mate tested positive" for marijuana. Neither has been corrected.

What would explain it innocently. A summary written three days later by an information service, compressing a long press conference, can transpose one crew role for another without anyone noticing; it is the commonest error in government communication and carries no design. Nothing suggests the discrepancy was ever put to either body for correction.

What would explain it otherwise. The two men are differently placed in the charges that followed, and the state's own accounts of who tested positive do not agree.

Which the record supports. The discrepancy is documented and unexplained; nothing in the corpus resolves it. One qualification belongs here in fairness to everyone named, and it is in the corpus rather than this archive's own: HGPTV states that a positive test does not establish impairment at the time of the capsizing or prove that drug use contributed to the disaster, and that the timing of the tests and any impairment remain undetermined. No document in this corpus asserts that either man was impaired at 23:00 on 18 July.

The document that would settle it. The toxicology reports, with the names, the analytes and the collection times.

8. The Port Kaituma stelling — HIGH

The anomaly. In June 2021 DPI recorded a G$588m saving on the G$2.6bn contract for the new India-built ferry, and said those savings would fund new stellings at Morawhanna, Kumaka and Port Kaituma. In September 2022 a new Port Kaituma stelling was announced, with CEMCO contracted for design and supervision across three locations; construction was awarded on 30 December 2022 to International Import and Supplies at G$1.4bn. The work reached 77% in January 2024 and about 90% in February 2024, when fissures appeared during final casting of pile caps; the initial investigation indicated that the tie-rod strength specified in the design was inadequate, and an independent review found further design deficiencies. Minister Edghill, quoted from the Sunday Stabroek of 11 August 2024: "At this particular time, based upon what we are doing, I have no bill, and take my word — at this particular time, I have no bill for additional costs." As at 27 July 2026 there is no completion date, no final cost and no disciplinary outcome on the record, and per Kiskadee Watch the ministry made no public acknowledgment of the failure until the contractor spoke. Single source The whole project chronology reaches this corpus through one outlet.

What would explain it innocently. Marine works in soft alluvial ground fail in exactly this way, and a design deficiency found at 90% is a genuine engineering event rather than evidence of anything else. A minister saying he has no bill for additional costs is describing a contract in which the defect is the contractor's to remedy, which is what a properly written contract does. Remediation takes years, and no disciplinary outcome may mean a dispute is unresolved rather than that no one looked.

What would explain it otherwise. A capital contract that stalled in February 2024 was still unfinished twenty-nine months later, and it is the constraint the Prime Minister himself identified as the reason the newer Ma Lisha could not run to Port Kaituma — a delay he conceded "has contributed to us not being able to use the MV Ma Lisha to journey from Georgetown to Port Kaituma."

Which the record supports. The chronology and the figures are documented, largely on the government's own record, though the project chronology rests on a single outlet. Whether anyone was held to account for the design failure is not established: the corpus holds no departmental or contractual document either way, and this analysis cannot convert that absence into evidence.

The document that would settle it. The construction contract, the independent review of the design, and any variation or claim filed against the contractor.

9. The Kalliopi N — HIGH

The anomaly. A 2024-model Greek ro-ro rated for 264 passengers, 82 cars and 19 trucks arrived in Guyana in May 2026 and was toured at the Kingston Goods Wharf by Minister Edghill, with the T&HD's General Manager and MARAD's Director-General present. Edghill said "the president himself will do the commissioning at the appropriate time", alongside three new wharves — a New Kingston Goods Wharf, Mabaruma and Manawarin. It was still awaiting commissioning two months later, at the wharf from which the Barima sailed, while the Barima sailed. The Kingston Goods Wharf works were themselves 80% complete in March 2025.

What would explain it innocently. A new vessel does not enter service on arrival: registration, survey, certification, crew familiarisation, route approval and berth compatibility all take time, and a two-month gap is unremarkable. The corpus gives a specific reason for the delay beyond the ceremony — the three wharves the commissioning was tied to were not finished. Nothing in the corpus establishes that this vessel was certified, crewed or capable of the Port Kaituma run, and a ro-ro built for other work is not automatically a substitute on that route.

What would explain it otherwise. A modern vessel lay at the wharf the Barima sailed from, designated for the same region, while an 87-year-old hull ran the route.

Which the record supports. Neither. The commissioning statement is documented; the vessel's readiness for that route is nowhere established in this corpus. This archive previously said the question of whether the Kalliopi N could have sailed had not been asked. That was wrong, and it is corrected here: Christopher Ram put it directly on 28 July, writing that the vessel "remained tied up awaiting commissioning while the 1939-built MV Barima continued transporting passengers", and APNU's Terrence Campbell had asked on 24 July why newly acquired vessels remained unused. The question has been asked publicly. What has not appeared is an answer.

The document that would settle it. The Kalliopi N's certification and route-approval file, and the T&HD's deployment plan for it.

10. What stopped being said — HIGH

The anomaly. On 19 July the government's briefing disclosed that only 35 of the first 67 rescued were on the manifest. At the briefing of 24 July, reported on the 25th, Prime Minister Phillips declined to say how many of the rescued and recovered had been manifested. DPI's releases on the disaster stop on 25 July — nothing on the 26th, 27th, 28th or 29th, including the day charges were laid. Authorities confirmed there would be no press briefing on the evening of Saturday 25 July, as efforts had yielded no new developments since the Friday. The body count has stood at 73 since 24 July. The count of identified bodies stood at 69 from 24 July and is given as 68 by WIC News on 29 July, with no source explaining the change.

What would explain it innocently. A figure volunteered on day one and declined on day six may simply have become harder to state: the complement moved from 133 to 179, the survivor count moved several times, and ten people recorded as found overnight were verified as already home, so a manifested-against-unmanifested split computed on numbers moving that fast is one a briefer would be right to refuse. Communications tapering after a week is what happens when a story moves from a rescue operation to a legal process, and once a file is with the DPP a government has proper reasons to say less. A body count that stops moving is what a recovery operation looks like when recovery has stopped, and 69 to 68 is the size of an ordinary correction — and it appears in one outlet, not in a state release.

What would explain it otherwise. The single figure showing how the manifest failure was distributed between the living and the dead was volunteered on day one and refused on day six, and the state's published output stops the day before the commissioners were named and does not resume for the charges.

Which the record supports. The pattern is documented — the dates, the refusal and the last dated release. The reason for it is not, and both readings remain open. That no DPI release appears after 25 July is an absence; what is cited is the last release itself. The 69-to-68 movement is unexplained by any source here and is recorded as unexplained, not as concealment.

The document that would settle it. The reconciliation of the manifest against the rescued and recovered — the same document that would answer the question Phillips declined.

11. No medico-legal process for these deaths appears anywhere — HIGH

The anomaly. Across every document in the corpus there is no reference to a coroner, an inquest, a post-mortem, an autopsy, a pathologist or a stated cause of death for anyone who died on the Barima. The only such references are in two Guyana Chronicle columns of 25 July, both about an unrelated drowning case. What the corpus does record is a handling process rather than a medico-legal one: every person recovered was to be handed to the Guyana Police Force at Charity, forensic personnel processed remains at the Charity waterfront, relatives viewed photographs at an image review station and then bodies at Suddie, and the Minister of Health and his team oversaw identification and the release of remains. 63 bodies had been released to families by 24 July and more than 30 funerals had been held by 26 July. 72 counts of murder were laid on 28 July. The casualty figures in this corpus come from the Prime Minister's briefings and from the Maritime Rescue Coordination Centre — not from a coroner, and not from any medical determination of how anyone died.

What would explain it innocently. In a mass-casualty drowning with a known event and identified bodies, a jurisdiction may not require an autopsy on each victim, and releasing bodies quickly to grieving families is a humane priority rather than a procedural failure. Medico-legal work is also not usually announced: the absence of the word "pathologist" from press coverage is not the absence of a pathologist, and the corpus does record forensic personnel and the health ministry in the identification chain. Figures coming from the Prime Minister reflects who fronted the briefings, not who compiled them.

What would explain it otherwise. A prosecution for 72 counts of murder proceeds while the corpus holds no record of any medical determination of how anyone died.

Which the record supports. The absence in the corpus is documented and complete on its own terms, and it is an absence, so it cannot be cited: what is cited is the identification and handling process the state described instead. This analysis cannot establish what Guyanese law requires, so it cannot say the missing process is one that should exist — only that if it exists, nothing in this corpus records it, including in the reporting of a murder charge where cause of death is an element.

The document that would settle it. Any post-mortem report, or a statement from the police or the health ministry describing how cause of death was determined for the recovered.

12. The Commission of Inquiry was announced against a parliamentary motion — HIGH

The anomaly. APNU's parliamentary leader wrote to the Speaker on Friday 24 July seeking an urgent adjournment to debate terms of reference at Monday's sitting. President Ali announced on Saturday 25 July that he would name the commission on Sunday 26 July — the day before that sitting. On 27 July the President was reported as saying the legal instruments were still being formalised, and as at 29 July no gazetted instrument, secretary, reporting deadline, call for submissions or published budget appears in this corpus. Single source The statutory basis has been reported once, by Kiskadee Watch, as Section 2(1) of the Commissions of Inquiry Act; the AFC had proposed the same Act by its chapter number, 19:03, and Christopher Ram argues the commission should instead have been constituted under Part XVIII of the Shipping Act and questions the legal basis on which it was constituted. So the basis is single-sourced and publicly disputed rather than settled.

What would explain it innocently. A government facing a national disaster and a parliamentary motion has every reason to move first, and moving first is not the same as pre-empting: an inquiry announced on the Sunday would have been demanded on the Monday in any case. Gazetting takes time, particularly where all five commissioners are based abroad. And the announced terms are broader than the criticism allows: they cover loading, boarding, seaworthiness, maintenance history, compliance with maritime law, vessel management, life-saving arrangements, weather and sea conditions, a review of the search and rescue response, and whether negligence, misconduct, dereliction of duty or institutional failures contributed.

What would explain it otherwise. The timing removed the occasion for the motion, and the instrument that would fix the commission's powers, deadline and statutory basis has not appeared, so the terms stand as announced rather than as enacted.

Which the record supports. The sequence is documented; the reason for the timing is not established. The absences — no gazetted instrument, no deadline, no budget — cannot themselves be cited; what is cited is the announcement, the naming of the commissioners, and the statement that the instruments were still being formalised. Ram's charge that the terms cover the voyage and not the decisions that put that vessel on it is his characterisation, and it is recorded as his: the published terms do reach maintenance history, vessel management and institutional failure, though not procurement or deployment by name.

The document that would settle it. The gazetted instrument constituting the commission.

13. Paperwork put before survivors and families, none of it published — MEDIUM-HIGH

The anomaly. Survivor Donald Modeste is reported on 20 July as having received G$100,000 and been told no further compensation would be paid for roughly G$1.8m of goods lost. At the briefing of 21 July Prime Minister Phillips found it necessary to deny that anyone had been "forced to sign any document for compensation" — a denial that presupposes documents were signed, though no such form appears in this corpus. On 25 July the Declaration of Agreement about the wreck was circulated to families. Neither the compensation paperwork nor the Declaration has been published. Contested Two reputable outlets describe the Declaration in materially different terms: Kaieteur News as a draft "Declaration of Agreement" it had seen and quoted, Demerara Waves as a questionnaire asking families whether they prefer salvage or a memorial site.

What would explain it innocently. Interim payments to disaster victims are routinely handled by simple forms, and a G$100,000 figure described as final at a counter is as likely to be a communication failure as a policy. A denial is not evidence of the thing denied; a Prime Minister asked a direct question answers it. And a draft circulated for discussion is not ordinarily published, least of all one addressed to bereaved families about a memorial.

What would explain it otherwise. Inside a week the state put paperwork in front of people with claims against it, and none of it is public.

Which the record supports. What is documented is the Declaration's existence and, from the denial, that compensation paperwork of some kind exists. Their contents are not, and two outlets describing the Declaration differently shows the cost of that. Modeste's account is of a payment and what he was told about it, not of a document, and it is one survivor's account in one outlet. This page does not characterise documents it has not seen, and it does not treat the number of them as established.

The document that would settle it. The texts — the compensation or assistance form, and the Declaration of Agreement.

14. The March 2026 tender — MEDIUM, downgraded

The anomaly. DPI on 9 March 2026 records docking and overhaul tendered for three vessels: Ma Lisha at G$199.3m, Barima at G$124.5m, MB Hercules at G$124.2m, under National Competitive Bidding. Five contractors bid for the Barima work, from about G$105m to G$124m. Single source The bill of quantities is reported as requiring examination of about 10,000 m² of hull plating and frames "with deteriorated sections to be renewed where necessary" — a conditional specification, reaching this corpus only through unbylined CMC wire copy, voiced by opposition MP Saiku Andrews. Nothing in the corpus says whether the tender was awarded, to whom, or whether any work was done.

Why it is downgraded. This archive had treated the tender as a signal about the hull four months before the sinking. It is better explained as the scheduled docking cycle: on the government's own account the Barima last docked in 2024 and was due again later in 2026, alongside two other vessels. A specification anticipating deterioration is not a survey establishing it, and it has been reported as though it were.

What survives. A narrow live question, and it is the opposition's and Demerara Waves' rather than this archive's: whether the tender was awarded, and whether docking had been deferred. One comparison has to be withdrawn as well. This archive had noted that the three-year-old Ma Lisha was tendered at G$199.3m against the 87-year-old Barima at G$124.5m and said it did not know why. The same DPI release supplies the likely answer: the Ma Lisha is a 70-metre vessel, roughly three-quarters longer again than the Barima, so the larger estimate is what size alone would predict. The comparison is dropped.

Which the record supports. The downgrade. On the narrow question, neither, because the award file is not in the corpus.

The document that would settle it. The March 2026 tender file and the award.

15. The rescue timeline — MEDIUM-HIGH

The anomaly. A signal at 23:01. The first air asset with a time against it in this corpus is private and fixed-wing: Captain Gerry Gouveia airborne at 01:36, two and a half hours after the Mayday, in a Roraima Airways Islander, who saw a distress flare and reported its coordinates. Survivors were in the water from about 22:00 to about 04:00 on Wayne Kitson's account, or nearly eight hours on the account iNews published of Alicia Atkinson. Janette Bulkan's letter of 25 July asks at what times the MSRC relayed the distress to the Exxon offshore supply vessels, to the Omni Helicopters base at Ogle, to the police boat near the mouth of the Pomeroon, to the Coast Guard, and to MARAD's two Ocean Pilot boats. None of those questions has been answered. Contested On the GDF helicopters, the opposition says two — a Bell 429 and a Bell 412 — were out of service for over a month with cockpit display unit failures; GDF officials say the remaining two in the fleet were grounded for routine maintenance while awaiting imported spare parts, giving no duration.

What would explain it innocently. Night search and rescue over water is genuinely constrained: most helicopters are not equipped or crewed for it, and launching an unsuitable asset into a dark river mouth kills rescuers without saving anyone. The GDF was reported mobilised that night, and the absence of times against its assets here is an absence of published detail rather than an established absence of response. The two accounts of the helicopters are not inconsistent — "awaiting imported spare parts" and "out of service for over a month" can describe the same aircraft.

What would explain it otherwise. The only air asset with a time against it was private, and the timeline that would show what the state tasked and when has not been produced.

Which the record supports. Both accounts of the helicopters stand, unreconciled. Times are documented only where private effort supplied them, and the state's own response times are not established in either direction — which is the missing timeline of finding 6, from another direction, and not a citable absence.

The document that would settle it. The MSRC log and the GDF air corps tasking record for the night.

16. The official proximate cause is a sea state for which no document exists — MEDIUM

The anomaly. The state's account of the sinking rests on rough seas at a location mariners are reported to regard as one of the most hazardous on the route: officials describe the vessel meeting rough sea conditions off the Essequibo Coast near the mouth of the Pomeroon, and Edghill's first explanation was that "it had to do with the tide." In this corpus there is no forecast, no marine warning, no sea-state observation and no notice to mariners for 18 July 2026 — none of those words appears in any document here. No advisory or notice to mariners from MARAD appears at any point either.

What would explain it innocently. Routine forecasts are not news, so their absence from a corpus built out of published journalism is expected rather than meaningful. A regulator does not issue a notice to mariners for ordinary weather, and the sea state may have been unremarkable enough that no warning was warranted — which is consistent with the vessel sailing at all.

What would explain it otherwise. The proximate cause named publicly on day one is the one element of the official account that no document in this archive can corroborate.

Which the record supports. Neither. The absence is real, it is not citable, and the archive cannot establish that any of these documents should exist, which is why this sits at medium. What is cited is the state's own account of the sea state. It is recorded because a stated cause with no record behind it is a check that has not been run, not because the absence proves anything.

The document that would settle it. The Hydrometeorological Service's record for the night of 18 July.

17. Shorter items

Each is smaller than the findings above and each is documented.

The T&HD's General Manager. Yolanda Hughes appears exactly once in the corpus — touring the Kalliopi N in May 2026 — and never again: no statement, no suspension, no appearance. The benign reading is that a general manager is not a public spokesman and that a department under investigation would properly route communication through its minister. The other is that the operator's most senior officer has not been heard from in eleven days. The record supports neither; absence from published journalism is not established to be absence of action, and the absence cannot be cited.

Thirteen of the crew died. On the Wikipedia capture, at least 13 crew were among the dead and the captain and three other crew survived, which reconciles with a crew of 17 — though the corpus also carries 16 and 18 for the crew. Most first-hand crew evidence is therefore gone, which bears on how much any inquiry can recover and is nobody's fault.

Cargo capacity and the replacement. Contested The Ma Lisha's cargo capacity is given as 250 tonnes by 592Hub, against 268 tonnes manifested on the Barima, and Stabroek News records residents reporting before the disaster that the Ma Lisha was inadequate for the route's cargo demands. That is not a clean corroboration and this archive previously treated it as one. Stabroek News also records that the Ma Lisha has the larger tonnage capacity of the two, and 592Hub warns on its own page that its capacity figures conflict with primary reporting. The residents' account stands on its own; the 250-against-268 arithmetic does not carry it. The state has only ever given the wharf explanation, which the Prime Minister has conceded delayed the Ma Lisha's use on the route. Both explanations can be true at once, and the figures page records them as compatible rather than rival.

One state rendering of the complement. The state-owned Guyana Chronicle renders the complement as "179 passengers and 17 crew" — 196 — where every other account has 179 as the total including crew. Contested If it were right it would change the residual by seventeen people. It is most likely an error of exactly the kind corrected at the top of this page, and it is recorded rather than relied on.

Stabroek News. The closure of a leading independent daily, which ceased printing in March 2026, four months before the sinking, is recorded by this archive as reduced independent capacity, not as suppressed investigation. What Stabroek would have found cannot be established, and this page does not speculate about it.

The four that cohere

Findings 1, 2, 3 and 10 describe the same thing from four directions, and the pattern in them is not concealment of a known fact. What follows is the strongest inference on this page. It names no one, alleges no offence, and states its own limit in its last sentence.

At every point where the state's account could have been made checkable against a document, the document has not appeared — and the state has repeatedly asserted conclusions that only that document could support. The minister cleared the vessel on overloading using what is very likely a gross-tonnage figure. The regulator cleared it citing a load line and a certification it has not produced. The head of government has since moved off the categorical "not overloaded" position, saying the question of overloading is one for a commission whose instrument had not been formalised when he said it. The one remaining physical object that could settle the question independently of paper is the subject of a document circulated to bereaved families asking them to agree it stay where it is — circulated the day after their briefing was told righting was imminent, and two days before the technical assessment that would later justify leaving it. And the one figure showing how the manifest failure was distributed across the living and the dead was volunteered on day one and refused six days later.

That is what an administration looks like when its early public statements outran its records. It is also, precisely, what a cover-up looks like. The record as it stands cannot distinguish between those two, and the honest thing is to say so.

The document that would change the picture

The MV Barima's current Certificate of Survey and Load Line Certificate, with the registry entry stating gross tonnage and deadweight.

One page decides: whether 284 was a unit error or a licensed limit; whether 120 or 126 is the real cargo figure; whether 268 tonnes was inside the vessel's deadweight or about 70 tonnes outside it; whether the "not overloaded" finding was sound; whether the 394-passenger rating exists on paper at all, and when it rose from the 150 of the original rating. And because minor deficiencies are noted at the back of a certificate — that is how Capt. Adams described the regime in 2015 — it would show whether anyone had recorded defects on this hull before 18 July.

It is the cheapest document on the list. The regulator has already said publicly that it relied on the vessel's load line and certification.

Second: the police charge sheet with particulars. It is the only document that would say what the state actually alleges happened — and the three men it names have not been tried, which is precisely why what is alleged against them should be a matter of record rather than of inference.