<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><atom:link href="https://afrimintel.com/feed.xml" rel="self" type="application/rss+xml"/><title>Afrimintel — record of changes</title><link>https://afrimintel.com/audit-log/</link><description>Every build entry on Afrimintel's append-only audit log: scored register rows, divergence records, drops and corrections, each dated. Generated from /api/v1/audit-log.json.</description><language>en</language><lastBuildDate>Sun, 18 Oct 2026 00:00:00 +0000</lastBuildDate><item><title>v2.137.281 — OPEN DEPOSIT REGISTER: SIX ROWS FOUND WRONG ON 12 OCTOBER CORRECTED OR WITHDRAWN INSIDE THE SEVEN-BUSINESS-DAY WINDOW; THREE POSITIONS CORRECTED (ONE TO A FILED COORDINATE, TWO TO THE NAMED TOWN);…</title><link>https://afrimintel.com/audit-log/#v2-137-281</link><guid isPermaLink="false">afrimintel-v2.137.281</guid><pubDate>Sun, 18 Oct 2026 00:00:00 +0000</pubDate><description>OPEN DEPOSIT REGISTER: SIX ROWS FOUND WRONG ON 12 OCTOBER CORRECTED OR WITHDRAWN INSIDE THE SEVEN-BUSINESS-DAY WINDOW; THREE POSITIONS CORRECTED (ONE TO A FILED COORDINATE, TWO TO THE NAMED TOWN); TWO ROWS THAT MATCHED NO SINGLE PROJECT RETIRED; MBAMBA BAY WITHDRAWN; /cite/ NOW DISCLOSES THE SIX ROWS AS STILL CARRIED BY THE DEPOSITED ZENODO FILE. Handshake: v2.137.280 deployed. WHY. SLA-2026-10-12-open-deposit-positions (material; identified 12 Oct 2026 by the adversarial passes on v2.137.277; due 21 Oct 2026) required each of six open-deposit-register positions to be corrected against two named sources or the row withdrawn, with the change dated on this log. Closed on 18 Oct 2026. (1) CORRECTED. Kolomela: -27.7, 22.9 -&gt; -28.383, 22.987, envelope 5 km (was the 30 km default); the prior point was about 76 km north of the mine. Sources: the Kolomela pit point filed on SAHRIS (Butler,…</description></item><item><title>v2.137.280 — HOMEPAGE AND CLAIMS REGISTER NOW STATE WHAT IS FREE TO READ AND WHAT IS OPEN-LICENSED; /pricing/ STOPS LISTING THE PROVINCE SCORING WITHDRAWN ON 12 OCTOBER.</title><link>https://afrimintel.com/audit-log/#v2-137-280</link><guid isPermaLink="false">afrimintel-v2.137.280</guid><pubDate>Sat, 17 Oct 2026 00:00:00 +0000</pubDate><description>HOMEPAGE AND CLAIMS REGISTER NOW STATE WHAT IS FREE TO READ AND WHAT IS OPEN-LICENSED; /pricing/ STOPS LISTING THE PROVINCE SCORING WITHDRAWN ON 12 OCTOBER. Handshake: v2.137.279 deployed. WHY. On 17 Oct 2026 the operator asked Google Gemini for an analysis of Afrimintel from the public website. The answer said: 'their data is locked behind a paywall or sits quietly in an audit log.' Neither is the record. The licence-register, traceability and deposit datasets are deposited on Zenodo under CC BY 4.0 (concept DOI 10.5281/zenodo.22664745), and the claims register is a public page with a public feed. The site gave a reader grounds for the paywall reading. The homepage's visible text carried the Pro offer at $149/month and the $1,850 single-asset offer, and stated the CC BY 4.0 licence once, inside the deposit card, where it mentioned a permanent DOI without giving it; it did not name…</description></item><item><title>v2.137.279 — 28 SEPTEMBER RIGHT-OF-REPLY WINDOWS CLOSED WITH NO REPLY; THREE HELD OUTCOMES PUBLISHED AS STATED IN THE LETTERS; PERSEUS REVISED-GUIDANCE ROW RELABELLED MET TO MATCH ITS LETTER, WITH THE…</title><link>https://afrimintel.com/audit-log/#v2-137-279</link><guid isPermaLink="false">afrimintel-v2.137.279</guid><pubDate>Thu, 15 Oct 2026 00:00:00 +0000</pubDate><description>28 SEPTEMBER RIGHT-OF-REPLY WINDOWS CLOSED WITH NO REPLY; THREE HELD OUTCOMES PUBLISHED AS STATED IN THE LETTERS; PERSEUS REVISED-GUIDANCE ROW RELABELLED MET TO MATCH ITS LETTER, WITH THE PRODUCTION-GUIDANCE CONDITION AMENDED; FOURTEEN ROWS OPENED FROM THE 12 OCTOBER LETTERS, OUTCOMES HELD TO 27 OCTOBER; KONÉ GAP DECLARED; v2.137.278 RETIRED UNUSED. Handshake: v2.137.277 deployed. v2.137.278 was assembled on 15 Oct 2026 and never deployed; a contrarian audit found a defect in its rule amendment (11 below), and its number is retired unused. Everything it carried is in this build, corrected. (1) REPLIES. The letters of 28 Sep 2026 to West Wits Mining, Lotus Resources, Lifezone Metals, Allied Gold and Perseus Mining asked for any response by 14 Oct 2026. None was received (operator confirmation, 14 Oct 2026). Each of the twelve rows those letters covered now records that a right of…</description></item><item><title>v2.137.277 — HOMEPAGE NOW DESCRIBES ALL FOUR RECORDS; AFRIMINTEL STOPS PUBLISHING DISCOUNT RATES, PROBABILITIES AND COVERAGE GRADES OF ITS OWN; NEW OPEN DEPOSIT REGISTER PAGE AT /deposits/, WITH ITS KNOWN…</title><link>https://afrimintel.com/audit-log/#v2-137-277</link><guid isPermaLink="false">afrimintel-v2.137.277</guid><pubDate>Mon, 12 Oct 2026 00:00:00 +0000</pubDate><description>HOMEPAGE NOW DESCRIBES ALL FOUR RECORDS; AFRIMINTEL STOPS PUBLISHING DISCOUNT RATES, PROBABILITIES AND COVERAGE GRADES OF ITS OWN; NEW OPEN DEPOSIT REGISTER PAGE AT /deposits/, WITH ITS KNOWN ERRORS DISCLOSED ON THE PAGE; THREE CONTRACT-SUITE FAILURES SHIPPED IN v2.137.276 CLOSED, ONE OF THEM ON THE PUBLIC CLAIMS REGISTER. Handshake: v2.137.276 deployed. Copy, pages and links only; no data feed changed. WHY: the platform holds four records side by side (which of 40 African countries publish a mining licence register; the open deposit register; mineral traceability schemes in the same 40 countries; and what companies promised against what their own later filings show), but the homepage described only the last of these. /tenure/ and /traceability/ existed and were not linked from it, the deposit register had no page of its own, the right of reply was not mentioned, and the navigation…</description></item><item><title>v2.137.276 — KAMOA-KAKULA Q3 2026 PRODUCTION RECORDED; KURMUK COMMISSIONING STATUS UPDATED; FORWARD-CLAIM ROW HELD, NOT SCORED.</title><link>https://afrimintel.com/audit-log/#v2-137-276</link><guid isPermaLink="false">afrimintel-v2.137.276</guid><pubDate>Sun, 11 Oct 2026 00:00:00 +0000</pubDate><description>KAMOA-KAKULA Q3 2026 PRODUCTION RECORDED; KURMUK COMMISSIONING STATUS UPDATED; FORWARD-CLAIM ROW HELD, NOT SCORED. Handshake: v2.137.275 deployed. Two verified events, each from the operator's own release, in the week to 11 Oct 2026; trade-press reports were used only to locate the releases and never as the basis for a change. (1) Kamoa-Kakula (intelligence-grade): Ivanhoe Mines' Q3 2026 production news release, dateline 8 Oct 2026, reports 76,401 t Cu in saleable products for Q3 2026 (61,506 t anode + 2,347 t toll-treated blister + 5,645 t slag concentrate + 6,904 t copper fines, Ivanhoe noting rounding may affect totals (components as reported sum to 76,402 t)), a record Phase 1&amp;2 concentrator recovery of ~89% following completion of Project 95 modifications in July 2026, and the 500,000 tpa smelter operating at ~60% of design capacity; FY2026 guidance affirmed at 290,000–310,000 t…</description></item><item><title>v2.137.275 — KONÉ FIRST GOLD POUR RECORDED — FORWARD-CLAIM ROW SCORED MET.</title><link>https://afrimintel.com/audit-log/#v2-137-275</link><guid isPermaLink="false">afrimintel-v2.137.275</guid><pubDate>Sun, 04 Oct 2026 00:00:00 +0000</pubDate><description>KONÉ FIRST GOLD POUR RECORDED — FORWARD-CLAIM ROW SCORED MET. Handshake: v2.137.274 deployed. Montage Gold Corp announced first gold at its Koné mine in Côte d'Ivoire (dossier-grade) in a news release dateline Abidjan 27 Sep 2026: first gold was poured from the oxide circuit on 26 Sep 2026, on budget and ahead of schedule; oxide-circuit commercial production is expected in Q4 2026 and the hard-rock comminution circuit remains on schedule for Q2 2027. Applied across surfaces: (1) the deposit-index record kone now records the pour and its date in status_detail with last_reviewed 2026-10-04; dev_stage is held at development pending a declaration of commercial production, an editorial reclassification left to a later build. (2) The per-record export api/v1/deposits/kone.json appends a dated, sourced sentence to res_source and preserves the prior construction-stage history unchanged; its…</description></item><item><title>v2.137.274 — THE POST-1-OCTOBER BUILD — FIRST LETTER OUTCOME RECORDED, REGISTER AMENDED TO MATCH THE LETTERS SENT, RIGHT-OF-REPLY CELLS SPLIT FROM THE BACKLOG, CHECKING DESK OPENED, AND THE 26 SEPTEMBER…</title><link>https://afrimintel.com/audit-log/#v2-137-274</link><guid isPermaLink="false">afrimintel-v2.137.274</guid><pubDate>Fri, 02 Oct 2026 00:00:00 +0000</pubDate><description>THE POST-1-OCTOBER BUILD — FIRST LETTER OUTCOME RECORDED, REGISTER AMENDED TO MATCH THE LETTERS SENT, RIGHT-OF-REPLY CELLS SPLIT FROM THE BACKLOG, CHECKING DESK OPENED, AND THE 26 SEPTEMBER RULINGS APPLIED. Handshake: v2.137.273 deployed. (1) IVANHOE. The right-of-reply letter of 17 Sep 2026 to Ivanhoe Mines gave ten business days, to 1 Oct 2026. No reply was received. The Kamoa-Kakula row now says so: a right of reply was offered and not taken up, a neutral fact about Afrimintel's procedure and not a finding about the company; the channel stays open. The reply cell keeps absence_basis not-yet-verified and records reply_received false, dated 2 Oct 2026; a right-of-reply basis is held for the ratchet ruling. The Kamoa guidance row is unchanged and stays scoreable only at 2026 year-end. /right-of-reply/ now lists every letter sent, its date, its clock and its status. (2) KABANGA…</description></item><item><title>v2.137.273 — RULINGS APPLIED — SIX OPERATOR RULINGS OF 29 SEP 2026, ONE UNDECLARED GAP DECLARED, AND ONE NEW FINDING HELD FOR A SECOND SOURCE.</title><link>https://afrimintel.com/audit-log/#v2-137-273</link><guid isPermaLink="false">afrimintel-v2.137.273</guid><pubDate>Tue, 29 Sep 2026 00:00:00 +0000</pubDate><description>RULINGS APPLIED — SIX OPERATOR RULINGS OF 29 SEP 2026, ONE UNDECLARED GAP DECLARED, AND ONE NEW FINDING HELD FOR A SECOND SOURCE. The operator accepted the recommendations listed on the v2.137.271 readme. Handshake: v2.137.272 deployed. (1) KABANGA RESOURCE LINE. api/v1/deposit-reserves.json carried the November 2023 estimate (43.6 Mt @ 2.02% Ni) as Kabanga's resource line, while divergence kabanga-ni::resource_line_vintage::2023-carried-2024-exists had been marked resolved on 3 Sep with a note that this feed had been corrected. Only the asset record had been. The row now carries the claimant's 4 December 2024 estimate as printed on Lifezone's Kabanga project page (read 29 Sep 2026): 46.8 Mt Measured and Indicated at 2.09% Ni, 0.29% Cu, 0.16% Co, attributable to Lifezone at 69.7% of total, with that basis in the category, the 100% figure marked Derived, contained metal, reporting…</description></item><item><title>v2.137.272 — CORRECTION TO v2.137.271 — THE COUNTRY INDEX WAS STILL A RANKING.</title><link>https://afrimintel.com/audit-log/#v2-137-272</link><guid isPermaLink="false">afrimintel-v2.137.272</guid><pubDate>Tue, 29 Sep 2026 00:00:00 +0000</pubDate><description>CORRECTION TO v2.137.271 — THE COUNTRY INDEX WAS STILL A RANKING. The v2.137.271 entry said rankings are not published and retired /decision/countries.html, a top-10 country table with LOW/MEDIUM bands, with a 301 to /country/. Read against the live site after deploy, /country/ carried the same thing at full length: all 40 jurisdictions sorted by composite, with LOW, MEDIUM, HIGH and VERY HIGH bands, Rwanda and the DRC among them. The redirect sent readers from a 10-row ranking to a 40-row one, and the v2.137.271 statement was wrong while that page stood. Neither the review, the hallucination check nor the two hostile passes caught it; it was found by opening the redirect target on the live site. FIXED: /country/ now lists the 40 jurisdictions A–Z with the composite only, and states that the list is not a ranking and carries no risk band; each composite keeps its stated basis on its…</description></item><item><title>v2.137.271 — CITATION BUILD — THE REGISTER CAN NOW BE CITED ROW BY ROW, EVERY COUNT ON A CITATION SURFACE IS READ FROM A FEED, AND THE LAST RANKING SCREENS AND VERDICT WORDING ARE RETIRED.</title><link>https://afrimintel.com/audit-log/#v2-137-271</link><guid isPermaLink="false">afrimintel-v2.137.271</guid><pubDate>Tue, 29 Sep 2026 00:00:00 +0000</pubDate><description>CITATION BUILD — THE REGISTER CAN NOW BE CITED ROW BY ROW, EVERY COUNT ON A CITATION SURFACE IS READ FROM A FEED, AND THE LAST RANKING SCREENS AND VERDICT WORDING ARE RETIRED. From a read-only review of the deployed v2.137.270 build on 29 Sep 2026, looking at the site as a journalist, a researcher and a buyer would. Handshake: v2.137.270 deployed. WHAT A JOURNALIST COULD NOT DO, NOW FIXED. (1) No row of /claims/ could be linked: the feed carries a stable id per row and the page carried none. Every one of the 30 rows now has a permalink anchor (the # beside its date), placed in the date cell so the row parsers of check-62 and check-71 are unchanged. (2) No claim source could be clicked. A source_url field is added to the claims feed where the claimant document is already cited elsewhere in this build or was opened on 29 Sep: the Lifezone H1 2026 interim report on SEC EDGAR (the source…</description></item><item><title>v2.137.270 — CORRECTION — NGUALLA: A CLOSED DIVERGENCE WHOSE FIX DID NOT PROPAGATE TO THE REST OF THE RECORD, AND TWO FEEDS NAMING THE WRONG ACQUIRING ENTITY.</title><link>https://afrimintel.com/audit-log/#v2-137-270</link><guid isPermaLink="false">afrimintel-v2.137.270</guid><pubDate>Tue, 29 Sep 2026 00:00:00 +0000</pubDate><description>CORRECTION — NGUALLA: A CLOSED DIVERGENCE WHOSE FIX DID NOT PROPAGATE TO THE REST OF THE RECORD, AND TWO FEEDS NAMING THE WRONG ACQUIRING ENTITY. Divergence ngualla-ree::ownership_control::peak-carried-shenghe-scheme was closed on 2026-09-03 (v2.137.106) as own-defect-resolved. Read against the published v2.137.269 record on 2026-09-29, the correction had reached one field only. ownership_control correctly carried the scheme outcome: Shenghe Resources (Singapore) Pte. Ltd. acquires 100% of Peak Rare Earths Limited at not less than A$0.443 cash per share, A$195m for 100% (Peak ASX 5 Sep 2025; reaffirmed 16 Sep 2025, subject to Tanzanian Fair Competition Commission approval). Six sibling surfaces still carried the pre-scheme position: on the deposit record, ownership ('Peak Rare Earths (LSE:PEAK) ~84%' — Peak was ASX:PEK, never LSE-listed), control_jurisdiction (ultimate control 'UK /…</description></item><item><title>v2.137.269 — CORRECTION — LOBITO PAGE: A BROKEN SENTENCE SHIPPED IN v2.137.268, AND FOUR WORDINGS TIGHTENED.</title><link>https://afrimintel.com/audit-log/#v2-137-269</link><guid isPermaLink="false">afrimintel-v2.137.269</guid><pubDate>Fri, 25 Sep 2026 00:00:00 +0000</pubDate><description>CORRECTION — LOBITO PAGE: A BROKEN SENTENCE SHIPPED IN v2.137.268, AND FOUR WORDINGS TIGHTENED. Found by reading the published v2.137.268 page. (1) Section 2 Q1 shipped as 'Binding to date: US$753M for the Angola rail concession (at financial close 3 July 2026), . AfDB's first tranche...' — a stray comma-full-stop left by the same-day rewrite that removed additive phrasing for check-58. Fixed. (2) The AfDB line read 'a first tranche of $255M (ADF-16), with a $10M grant in August 2026, with mobilisation up to $500M envisaged' — two 'with' clauses that left the grant's date doing double duty. Restated: the Board approved a first tranche of $255M (ADF-16) and a $10M grant in August 2026, and mobilisation up to $500M is envisaged in later phases (AfDB release, Aug 2026). (3) Q1 claimed DFC, DBSA, AfDB, AFC and EU+EIB were 'named with verified commitments'. AfDB's ~$500M is a stated…</description></item><item><title>v2.137.268 — FOLLOW-UP TO THE LOBITO SELF-AUDIT, SAME DAY: THE AUDIT'S OWN CORRECTIONS HAD NOT PROPAGATED, AND ONE OF THEM WAS WRONG.</title><link>https://afrimintel.com/audit-log/#v2-137-268</link><guid isPermaLink="false">afrimintel-v2.137.268</guid><pubDate>Fri, 25 Sep 2026 00:00:00 +0000</pubDate><description>FOLLOW-UP TO THE LOBITO SELF-AUDIT, SAME DAY: THE AUDIT'S OWN CORRECTIONS HAD NOT PROPAGATED, AND ONE OF THEM WAS WRONG. The v2.137.267 audit note was read against the published page. Two failures. (A) NOT PROPAGATED. Six places still carried the framing the audit had rejected: section 2 Q1 ('Total cumulative Western and multilateral commitment above $6bn'); the sensitivity table's 'Zambia phase commissioning / Q3 2026 target (Tayali / U.S.-Africa summit)' row, which the audit note said had been corrected and had not been; two funder-stack rows ('$6bn auditable + $3-4bn fundraise mandates'; '$6bn realised; fundraise mandates partial'); the section 5 overlay caveat ('Cumulative &gt;$6bn aggregates publicly-disclosed funder commitments'); and the companion-spread paragraph ('Cumulative committed financing (~$6bn auditable floor)'). The header line also still described the Zambia figure as…</description></item><item><title>v2.137.267 — LOBITO-CORRIDOR RECORD — SCHEDULED SELF-AUDIT RUN ON THE ANNOUNCED DATE; SIX CORRECTIONS PUBLISHED.</title><link>https://afrimintel.com/audit-log/#v2-137-267</link><guid isPermaLink="false">afrimintel-v2.137.267</guid><pubDate>Fri, 25 Sep 2026 00:00:00 +0000</pubDate><description>LOBITO-CORRIDOR RECORD — SCHEDULED SELF-AUDIT RUN ON THE ANNOUNCED DATE; SIX CORRECTIONS PUBLISHED. The platform announced a public self-audit of /case-studies/lobito-corridor-infrastructure/ for 25 Sep 2026 and carried it as an open row in its own register. It was run on the date. RESULT: the record HOLDS WITH CORRECTIONS — no cell was found Sourced to a document that does not support it; six items were wrong, stale or stated more firmly than their sources allow. WHAT WAS CORRECTED. (1) The US$753M Angola concession financing was carried as signed December 2025 only; AFC announced financial close on 3 July 2026 (AFC release; DFC $553M senior debt, DBSA $200M). The close is now on the record. (2) 'Cumulative committed funding. Above $6bn' survived the 22 Sep pass on the page's own summary line and in the commitment-matrix heading. Binding to date is the US$753M; the ~US$6bn is…</description></item><item><title>v2.137.266 — /DATA-SOURCES/ NO LONGER CONTRADICTS THE PLATFORM'S OWN TENURE RECORD, AND A SENTENCE OFFERING ACCESS THROUGH THE CONSULSHIP IS REMOVED FROM A PUBLIC SCRIPT.</title><link>https://afrimintel.com/audit-log/#v2-137-266</link><guid isPermaLink="false">afrimintel-v2.137.266</guid><pubDate>Thu, 24 Sep 2026 00:00:00 +0000</pubDate><description>/DATA-SOURCES/ NO LONGER CONTRADICTS THE PLATFORM'S OWN TENURE RECORD, AND A SENTENCE OFFERING ACCESS THROUGH THE CONSULSHIP IS REMOVED FROM A PUBLIC SCRIPT. Data and copy change; dataset_version advances v2.137.103 → v2.137.104 because js/data.js changed. (1) FOUND: js/deal-flow.js, which is served publicly and loaded by platform.html, carried in its Rwanda cadastre entry the sentence 'Direct relationship available via Honorary Consul standing.' That contradicts the published independence policy (/methodology/independence/: no commercial approach where the Honorary Consulship could reasonably be understood to have facilitated the introduction or the access). The sentence is removed. A sweep of every served .js, .html, .json, .txt and .md file outside the append-only log found no other wording that offers access, a channel, a route or a relationship through the Consulship. The…</description></item><item><title>v2.137.265 — TWO REGISTER OUTCOMES HELD UNTIL THE CLAIMANT'S OWN FILING AND A REPLY LETTER; THE FEED AND /claims/ AGREE AGAIN.</title><link>https://afrimintel.com/audit-log/#v2-137-265</link><guid isPermaLink="false">afrimintel-v2.137.265</guid><pubDate>Wed, 23 Sep 2026 00:00:00 +0000</pubDate><description>TWO REGISTER OUTCOMES HELD UNTIL THE CLAIMANT'S OWN FILING AND A REPLY LETTER; THE FEED AND /claims/ AGREE AGAIN. Register change; dataset_version advances v2.137.102 → v2.137.103. FOUND: at v2.137.264, /api/v1/claims.json carried two rows as scored — qala-shallows (West Wits Mining, MET) and kayelekera (Lotus Resources, NOT MET) — while /claims/ showed both open with the outcome Absent, and the page's opening line counted 7 scored rows while its table showed 5. Each row's own status note said it was held from publication because no right-of-reply letter had been sent, but the feed is public, so the hold did not hold. The standing rule of 17 September is that every company in a row receives the same right-of-reply letter before its row is scored in public. Each row also carried the blocker primary-pending (the claimant's own filing is not on record), which contradicted its Sourced…</description></item><item><title>v2.137.264 — ONE FLAGSHIP — THE SCORECARD TAKES /report/, AND THE BANKABILITY REPORT BECOMES A DATA PAGE.</title><link>https://afrimintel.com/audit-log/#v2-137-264</link><guid isPermaLink="false">afrimintel-v2.137.264</guid><pubDate>Tue, 22 Sep 2026 00:00:00 +0000</pubDate><description>ONE FLAGSHIP — THE SCORECARD TAKES /report/, AND THE BANKABILITY REPORT BECOMES A DATA PAGE. Copy, page-structure and register change; dataset_version advances v2.137.101 → v2.137.102 (see 6). (1) /report/ redirected to 'The State of African Mining Bankability — 2027', labelled 'Flagship annual report · Mining Indaba edition'. The plan of 17 September makes the Africa Mining Guidance Scorecard the report published on fixed dates, Edition 1 in November 2026 and Edition 2 in late January 2027; a second report labelled for the same February conference competed with it, and the page did not appear among the top pages in the operator's analytics for the preceding two weeks. /report/ is now the Scorecard page: the question, the two edition dates, the letter rule, and the live register it is published from. It publishes no MET / NOT MET share. The register's decided company rows are already…</description></item><item><title>v2.137.263 — SHARED DOSSIER — REQUEST-ACCESS PATH FOR UN-PROVISIONED VISITORS.</title><link>https://afrimintel.com/audit-log/#v2-137-263</link><guid isPermaLink="false">afrimintel-v2.137.263</guid><pubDate>Tue, 22 Sep 2026 00:00:00 +0000</pubDate><description>SHARED DOSSIER — REQUEST-ACCESS PATH FOR UN-PROVISIONED VISITORS. Copy-and-affordance change to /dossier/ only; no data feed changed (dataset_version unchanged at v2.137.101). Before this build the shared-dossier page told a visitor with no organisation session to 'sign in', but the page carries no sign-in affordance — the session is issued elsewhere — so an unprovisioned visitor (the likely state for a first-time development-finance officer) reached an instruction with no next step. The three not-provisioned states — no session, a non-work-email session, and the store being switched on — now each render a 'Request access for your team' action: a pre-filled mail to the editor, carrying the asset the visitor was viewing where one was selected. The honest read of the page is preserved: the structure stays visible, nothing claims to save until the store is provisioned per organisation,…</description></item><item><title>v2.137.262 — RULINGS APPLIED — NO-VERDICT CLEAN-UP OF THE DFI SCREEN, AND THE NOT-YET-VERIFIED RATCHET FIXED, NOT NUDGED.</title><link>https://afrimintel.com/audit-log/#v2-137-262</link><guid isPermaLink="false">afrimintel-v2.137.262</guid><pubDate>Tue, 22 Sep 2026 00:00:00 +0000</pubDate><description>RULINGS APPLIED — NO-VERDICT CLEAN-UP OF THE DFI SCREEN, AND THE NOT-YET-VERIFIED RATCHET FIXED, NOT NUDGED. Two operator rulings of 22 Sep 2026, applied in one build. (1) TIERS REMOVED. The shortlist page grouped 59 development-stage projects into a cross-gate 'verdict' — Tier A ('gates cleared, bankable'), B and C — and the DFI screen hub described it as 'collapsed to one verdict'. That is a ranking and a verdict, which the no-verdict rule rules out, and it contradicted the home page ('Afrimintel states no verdict'). Removed from the data (financeability.json: tier and rationale deleted from all 59 shortlist blocks) and from both pages. The page is rebuilt as 'Seven gates, project by project': stage, financing status as disclosed, the facts from the gate feeds, and the development banks whose published mandate covers each project's country and sector, listed by country with no…</description></item><item><title>v2.137.261 — CORRECTION — HOME PAGE NO LONGER CLAIMS EVERY FIGURE IS DATED.</title><link>https://afrimintel.com/audit-log/#v2-137-261</link><guid isPermaLink="false">afrimintel-v2.137.261</guid><pubDate>Tue, 22 Sep 2026 00:00:00 +0000</pubDate><description>CORRECTION — HOME PAGE NO LONGER CLAIMS EVERY FIGURE IS DATED. The home page said each figure is 'dated and sourced', that every figure carries 'a named primary document and a date', and the meta description said 'dated, sourced'. That holds for the fully checked records. It does not yet hold for the two screening records declared at v2.137.260: of 158 environmental and social items with a named source, 45 show at least the year of the document; for climate, 11 of 63. Four edits, home page only: the lede now reads 'each figure sourced'; the three-states section keeps the standard as written ('Sourced — a named document with its date') and adds that the environmental and social and climate records are still being brought to the dating standard, linking both; the citation answer now says every figure in the fully checked records carries a named primary document and a date, and that the…</description></item><item><title>v2.137.260 — CORRECTION — DFI SCREEN SCOPE STATEMENTS, BLANK COUNTRIES, ONE MISFILED CLIMATE CELL, AND SOURCE-DATING STATUS DECLARED.</title><link>https://afrimintel.com/audit-log/#v2-137-260</link><guid isPermaLink="false">afrimintel-v2.137.260</guid><pubDate>Tue, 22 Sep 2026 00:00:00 +0000</pubDate><description>CORRECTION — DFI SCREEN SCOPE STATEMENTS, BLANK COUNTRIES, ONE MISFILED CLIMATE CELL, AND SOURCE-DATING STATUS DECLARED. Found 22 Sep 2026 while preparing the environmental and climate records for an external reader. (1) SCOPE STATEMENTS WERE FALSE. Seven DFI-screen feeds (additionality, climate, integrity, offtake, political, technical, safeguards) carried a coi_screen line saying the screen held no DRC, Rwanda or Great Lakes asset, and six screen pages carried a footer calling the set a '13-asset COI-clean DFI financeability pipeline'. Every one of the seven includes two projects in Uganda; three include projects in the DRC (safeguards and technical: Kamoa-Kakula and Manono; additionality: Manono). The statements now say what each screen holds, and that Great Lakes projects are recorded on the same terms as every other under the published editorial recusal, with no editorial…</description></item><item><title>v2.137.259 — CORRECTION — ALL FIVE CASE STUDIES: VERDICTS, LIKELIHOOD LABELS AND ELIGIBILITY RULINGS REMOVED; MANONO AND KAMOA-KAKULA FACTS CORRECTED ACROSS PAGES AND DATA LAYER; ONE TRACK-RECORD VALIDATION…</title><link>https://afrimintel.com/audit-log/#v2-137-259</link><guid isPermaLink="false">afrimintel-v2.137.259</guid><pubDate>Tue, 22 Sep 2026 00:00:00 +0000</pubDate><description>CORRECTION — ALL FIVE CASE STUDIES: VERDICTS, LIKELIHOOD LABELS AND ELIGIBILITY RULINGS REMOVED; MANONO AND KAMOA-KAKULA FACTS CORRECTED ACROSS PAGES AND DATA LAYER; ONE TRACK-RECORD VALIDATION WITHDRAWN AND ONE ARITHMETIC ERROR CORRECTED. SUPERSESSION: v2.137.258 was assembled on 22 Sep 2026 and never deployed; it carried the Manono page correction only. Its number is retired unused and its content is folded into this build. Found 22 Sep 2026 while tracing answer-engine referrals to /case-studies/manono-disputed-tenure/, then widened to every case study on the operator's instruction. (1) VERDICTS. Each case study carried a 'Screening reading' or 'Directional reading' — Kabanga ABOVE THE BAR, Kamoa-Kakula RE-RATE, Manono HOLD / DFI-INELIGIBLE, Loulo-Gounkoto MONITOR / LIMITED DFI debt eligibility, Lobito MULTI-LAYER / STRONG candidate — and Kamoa-Kakula, Loulo-Gounkoto and Manono…</description></item><item><title>v2.137.257 — CORRECTION — STATE AND LOCAL OWNERSHIP TERMS, SIX SOUTHERN AFRICAN JURISDICTIONS.</title><link>https://afrimintel.com/audit-log/#v2-137-257</link><guid isPermaLink="false">afrimintel-v2.137.257</guid><pubDate>Mon, 21 Sep 2026 00:00:00 +0000</pubDate><description>CORRECTION — STATE AND LOCAL OWNERSHIP TERMS, SIX SOUTHERN AFRICAN JURISDICTIONS. The state_participation entries in api/v1/fiscal-regimes.json (rendered on /fiscal-regime/) were checked on 21 Sep 2026 against current law before external use. Four were wrong or out of date and two needed material context. (1) BOTSWANA — out of date. The record carried only the Government option on up to 15% working interest (Mines and Minerals Act 1999, s.40(1)), and did not say that Government pays its share of costs under that option or that s.40 does not apply to diamond licences (s.40(3); diamond participation is negotiated under s.51). The Mines and Minerals (Amendment) Act No.14 of 2024, in force 1 Oct 2025, inserted s.40(2A)-(2C): where Government does not exercise its option, the holder shall dispose of a working interest of up to 24% to citizens or citizen-owned companies within three years…</description></item><item><title>v2.137.253 — SCHEMA DATASET MARKUP — DEPOSIT DOI AND LICENCE CORRECTED ON THE CITABLE SURFACES.</title><link>https://afrimintel.com/audit-log/#v2-137-253</link><guid isPermaLink="false">afrimintel-v2.137.253</guid><pubDate>Sat, 19 Sep 2026 00:00:00 +0000</pubDate><description>SCHEMA DATASET MARKUP — DEPOSIT DOI AND LICENCE CORRECTED ON THE CITABLE SURFACES. Three defects found 19 Sep against the deployed tree and fixed at the emitter, not the page. (1) /cite/ carried the Zenodo concept DOI as the identifier of the PROPRIETARY Data Feed block — a terms-licensed dataset publicly claiming a CC-BY deposit's DOI — while the CC-BY open-tier block, which IS that deposit, carried no DOI at all. The DOI now sits on the open-tier block as a typed identifier with sameAs; the proprietary block identifies by its own manifest URL. (2) /tenure/ Dataset markup declared the proprietary terms licence on data that is part of the CC-BY Zenodo deposit (African Mining Governance Infrastructure and Open Deposit Register). Licence label aligned to CC-BY 4.0 and the concept DOI added; operator-directed 19 Sep 2026. (3) /traceability/ — the deposit's third jurisdiction file —…</description></item><item><title>v2.137.252 — TWO OVERDUE ROWS SCORED, ONE MET AND ONE NOT MET, AND NEITHER IS EDITION-READY.</title><link>https://afrimintel.com/audit-log/#v2-137-252</link><guid isPermaLink="false">afrimintel-v2.137.252</guid><pubDate>Fri, 18 Sep 2026 00:00:00 +0000</pubDate><description>TWO OVERDUE ROWS SCORED, ONE MET AND ONE NOT MET, AND NEITHER IS EDITION-READY. The report spec names three overdue rows as the third of its next three actions. Two are scored here against the first-production class condition, which is written per class rather than per row and therefore predates both outcomes. QALA SHALLOWS — MET. First gold was poured on 17 March 2026 at a third-party processing plant, within the stated window. A finding in the row itself: the claimant's own framing of the promise was the first quarter of 2026, stated by its chief executive in the first-gold announcement, while the register had recorded the promise as March 2026. The two resolve identically here, and the row now carries both wordings rather than the paraphrase alone. A register that scores other parties' sentences cannot paraphrase them. KAYELEKERA — NOT MET. Steady state of two hundred thousand…</description></item><item><title>v2.137.251 — EDITORIAL RECUSAL PUBLISHED — GREAT LAKES / CONFLICT MINERALS / MUA.</title><link>https://afrimintel.com/audit-log/#v2-137-251</link><guid isPermaLink="false">afrimintel-v2.137.251</guid><pubDate>Fri, 18 Sep 2026 00:00:00 +0000</pubDate><description>EDITORIAL RECUSAL PUBLISHED — GREAT LAKES / CONFLICT MINERALS / MUA. The editor holds two roles in Rwanda: Honorary Consul of the Republic of Rwanda in Mauritius, and Chairman of MUA Insurance Rwanda. The platform carries records for Rwanda, the Democratic Republic of the Congo and Great Lakes conflict-mineral traceability. Until this build the recusal governing that overlap existed as an operating rule and was not published. An undisclosed conflict is a diligence discount; a published recusal is a standard, which is the reason for the build. WHAT IS PUBLISHED. Four numbered rules: records in those jurisdictions carry on the same terms as every other jurisdiction and none is added, removed or weighted because of the roles; the editor publishes no opinion on those jurisdictions or on conflict-mineral policy under his name; the editor solicits no commercial relationships with…</description></item><item><title>v2.137.250 — THE LETTER WENT OUT FIRST, SO THE PLATFORM IS WHAT MOVES.</title><link>https://afrimintel.com/audit-log/#v2-137-250</link><guid isPermaLink="false">afrimintel-v2.137.250</guid><pubDate>Fri, 18 Sep 2026 00:00:00 +0000</pubDate><description>THE LETTER WENT OUT FIRST, SO THE PLATFORM IS WHAT MOVES. On 17 September a right-of-reply letter was sent to Ivanhoe Mines investor relations covering the Kamoa-Kakula records. It stated two things the published surfaces did not: a ten-business-day reply window, and that if no reply were received the record would note that a right of reply had been offered and not taken up, carrying no implication either way. Neither was on any page. THE RULING. Where a letter has already been sent to a company, the company's copy is the fixed point and the platform is amended to match it. The alternative — writing to a company a second time to withdraw something it was told — corrects a document the recipient may never have doubted, at the cost of the platform appearing unable to state its own procedure. Between a platform that adjusts its published procedure to honour what it told a company, and a…</description></item><item><title>v2.137.248 — THE ANSWER WAS ALREADY IN THE CITATION LIST, AND THEN THE SAME MISTAKE WAS MADE AGAIN WITHIN HOURS.</title><link>https://afrimintel.com/audit-log/#v2-137-248</link><guid isPermaLink="false">afrimintel-v2.137.248</guid><pubDate>Thu, 17 Sep 2026 00:00:00 +0000</pubDate><description>THE ANSWER WAS ALREADY IN THE CITATION LIST, AND THEN THE SAME MISTAKE WAS MADE AGAIN WITHIN HOURS. An adversarial review pass — AI-assisted, logged as such, not an independent review — was run against the full Kabanga record as extracted for an external reviewer brief. Three findings; all three verified against the documents; all three applied in this build. THE WARRANT CELL FIRST. The 8 September correction on the bridge-facility warrants said the reason the waiver was granted was not established by the filing that reports the issuance, and named a March Form 6-K — unread, outside the tripwire window — as the document that would settle it. The pass pointed somewhere closer: the H1 2026 interim report, a document this record already cites for its going-concern language, states the waived item in terms at Note 19.2 — consideration for waiving the requirement that the Government of…</description></item><item><title>v2.137.246 — THE SIXTH COPY WAS ON A CREDIT MEMO, WHICH IS THE WORST PLACE FOR IT.</title><link>https://afrimintel.com/audit-log/#v2-137-246</link><guid isPermaLink="false">afrimintel-v2.137.246</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>THE SIXTH COPY WAS ON A CREDIT MEMO, WHICH IS THE WORST PLACE FOR IT. Extending the register-to-records gate from feeds to rendered pages found the superseded net present value on a lender-facing credit memo, carried as the study basis in the same wording the bankability feed used before that feed was corrected. Feeds are read by machines; a credit memo is read by the institution deciding whether to lend. That is six copies of one wrong number across five surfaces now, and each extension of the check has found the next one. THE CHECK WAS WRONG TWICE BEFORE IT WAS RIGHT, AND BOTH ERRORS ARE RECORDED. The first form scanned a whole page wherever the page named the asset. On a multi-asset credit memo that matched a gold price assumption of US$1,350 per ounce against a retired net present value of US$1,350m belonging to a different asset entirely: a false positive, and exactly the kind…</description></item><item><title>v2.137.245 — WIDENING THE CHECK FOUND THE FIGURE IN TWO MORE FEEDS, WHICH IS WHY THE LIMIT WAS WRITTEN DOWN RATHER THAN LEFT IMPLICIT.</title><link>https://afrimintel.com/audit-log/#v2-137-245</link><guid isPermaLink="false">afrimintel-v2.137.245</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>WIDENING THE CHECK FOUND THE FIGURE IN TWO MORE FEEDS, WHICH IS WHY THE LIMIT WAS WRITTEN DOWN RATHER THAN LEFT IMPLICIT. The previous build added a gate joining the claims register to deposit records, and its own entry recorded the limit: a superseded figure could survive in other feeds and that gate would not see it. Widening it to every served feed confirmed the limit was real. The net present value this platform scored as unsupported was still carried in the evidentiary archive, where it was labelled as the figure the preserved primary document supports, and in the bankability feed's basis line. Both are corrected. That is five copies of one wrong number across four surfaces, found in three passes: a typed field, an asset narrative, a provenance note, an archive figure label, and a basis string. HOW THE WIDENING AVOIDS FALSE POSITIVES. Two constraints, both explicit in the check…</description></item><item><title>v2.137.244 — A CHECK BUILT TO CATCH ONE KNOWN DEFECT FOUND TWO MORE OF IT ON ITS FIRST RUN.</title><link>https://afrimintel.com/audit-log/#v2-137-244</link><guid isPermaLink="false">afrimintel-v2.137.244</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>A CHECK BUILT TO CATCH ONE KNOWN DEFECT FOUND TWO MORE OF IT ON ITS FIRST RUN. check-115 joins the claims register to the deposit records. The rule: a figure this platform published and has since scored as wrong in its own public self-correction must not still be live in that asset's record. It is scoped to this platform's own corrections, because a third party's superseded claim may legitimately still appear in a record as their stated position, whereas ours may not. WHAT IT FOUND. The manual correction at v2.137.243 changed the structured net-present-value field and stopped there. Two further copies of the superseded figure survived in the same record: the asset narrative, which still described the study as returning that value, and a provenance note which asserted that the primary filing itself carried it. The second is the worse of the two, because it is a statement about what a…</description></item><item><title>v2.137.243 — THE REGISTER SCORED OUR OWN NPV AS WRONG AND THE FEED UNDERNEATH IT WAS NEVER CORRECTED.</title><link>https://afrimintel.com/audit-log/#v2-137-243</link><guid isPermaLink="false">afrimintel-v2.137.243</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>THE REGISTER SCORED OUR OWN NPV AS WRONG AND THE FEED UNDERNEATH IT WAS NEVER CORRECTED. A scored public self-correction on this platform records that a published Ngualla net present value of US$1,350m appears nowhere in the cited study, which states US$1,483m. The deposit record still carried 1.35. The error was found, published, scored and left in place in the data. That is worse than the original error, and no check caught it: the register-to-feeds guard added at v2.137.235 compares scored rows against open verification items, not against deposit records. The record now carries 1.483, together with capital expenditure of US$321m and an equity internal rate of return of 37.3 per cent, all read off Table 1 of the announcement. THE OPERATING COST WAS THE THIRD FIGURE FROM THE DISCREDITED PUBLISHER. Cash operating cost was carried as US$17.42 per kilogram. Table 1 of the BFS Update…</description></item><item><title>v2.137.242 — NINE MORE DIVIDEND-WITHHOLDING PARAMETERS, EACH CARRYING THE LIMIT OF WHAT WAS ESTABLISHED.</title><link>https://afrimintel.com/audit-log/#v2-137-242</link><guid isPermaLink="false">afrimintel-v2.137.242</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>NINE MORE DIVIDEND-WITHHOLDING PARAMETERS, EACH CARRYING THE LIMIT OF WHAT WAS ESTABLISHED. Dividend withholding is now carried for twenty-eight of forty jurisdictions, up from nineteen. The nine added are Zambia, Chad, Cameroon, Gabon, Republic of Congo, Botswana, Mozambique, Madagascar and Libya, each taken from the dividend column of a single comparative withholding-tax chart published by a major professional-services firm, with that territory's own review date recorded in the provenance rather than the date the chart was read. Libya is carried as not applicable, which is a positive finding and not an absence. WHAT IS NOT ESTABLISHED, AND IT IS WRITTEN INTO EVERY ONE OF THEM. These are general-regime statutory rates. No mining-sector override was checked for any of the nine. That matters because the Democratic Republic of the Congo, already carried on this feed, has a general…</description></item><item><title>v2.137.241 — THE TERRAIN VIEWERS WERE THE ONLY PAGES ON THIS PLATFORM THAT FAILED SILENTLY.</title><link>https://afrimintel.com/audit-log/#v2-137-241</link><guid isPermaLink="false">afrimintel-v2.137.241</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>THE TERRAIN VIEWERS WERE THE ONLY PAGES ON THIS PLATFORM THAT FAILED SILENTLY. Fourteen three-dimensional terrain pages load their rendering library from a public content delivery network and then call it unconditionally. When that network is unreachable — blocked by a corporate or institutional proxy, filtered regionally, or simply down — each page threw an uncaught reference error and presented a blank stage with no explanation. Everywhere else on this platform a component that cannot get its data says so; these said nothing. WHAT THE GUARD DOES. The page script runs inside an immediately-invoked function, so the guard sits at the top of the scene section and returns early when the library is absent, after writing a plain statement into the stage: that the viewer could not start, that its rendering library is served from a content delivery network this browser could not reach, that…</description></item><item><title>v2.137.240 — THE NGUALLA OUTPUT FIGURE CAME FROM THE SOURCE WE HAD ALREADY DISCREDITED ON THIS ASSET.</title><link>https://afrimintel.com/audit-log/#v2-137-240</link><guid isPermaLink="false">afrimintel-v2.137.240</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>THE NGUALLA OUTPUT FIGURE CAME FROM THE SOURCE WE HAD ALREADY DISCREDITED ON THIS ASSET. The platform carried annual output as approximately 24,000 tonnes of rare earth oxide, and separately as approximately 37,200 tonnes of 45 per cent concentrate containing approximately 16,700 tonnes, both attributed to the October 2022 BFS Update, with the discrepancy between them logged as an internal non-reconciliation to be resolved by reading that document. Reading it resolves nothing, because none of those three figures is in it. The announcement states a high grade concentrate containing approximately 16.2 thousand tonnes per annum of TREO over the life of mine, and the company's own summary of the study gives 18 thousand tonnes per annum contained over the first six years and 16 over the life of the project. Independent reporting of the same study describes the update as reducing…</description></item><item><title>v2.137.239 — DIVIDEND WITHHOLDING FILLED FOR THE THREE DEEP-ASSET JURISDICTIONS WHERE IT COULD BE SOURCED, AND LEFT ABSENT FOR THE TWO WHERE IT COULD NOT.</title><link>https://afrimintel.com/audit-log/#v2-137-239</link><guid isPermaLink="false">afrimintel-v2.137.239</guid><pubDate>Wed, 16 Sep 2026 00:00:00 +0000</pubDate><description>DIVIDEND WITHHOLDING FILLED FOR THE THREE DEEP-ASSET JURISDICTIONS WHERE IT COULD BE SOURCED, AND LEFT ABSENT FOR THE TWO WHERE IT COULD NOT. The government-take engine reads this feed. Five jurisdictions carrying deep assets had no dividend-withholding parameter at all: Côte d'Ivoire, Tanzania, Mali, Guinea and the Democratic Republic of the Congo. Three are now carried with their provenance, and two are not, because the sources that cover them are subscription-gated and no substitute was found. Filling three and declaring two is the correct outcome; filling five would have meant writing a general-regime rate into a mining fiscal parameter on the assumption that no sector rule exists. WHY THAT ASSUMPTION WOULD HAVE BEEN WRONG. The Democratic Republic of the Congo carries a standard dividend withholding of twenty per cent and a mining-sector rate of ten. A general-regime figure is…</description></item><item><title>v2.137.238 — AN OPEN ITEM SHIPPED ON A FALSE PREMISE AND IS WITHDRAWN.</title><link>https://afrimintel.com/audit-log/#v2-137-238</link><guid isPermaLink="false">afrimintel-v2.137.238</guid><pubDate>Mon, 14 Sep 2026 00:00:00 +0000</pubDate><description>AN OPEN ITEM SHIPPED ON A FALSE PREMISE AND IS WITHDRAWN. At v2.137.237 a noted item was opened against Balama asking whether two facts were carried on this platform's bankability surfaces: that the published Ore Reserve is wholly Probable with no Proved at either vintage, and that the claimant's force-majeure release discloses events of default triggered under its United States DFC and DOE loans. The item stated that neither had been established. Establishing it required one search of the feeds. Both were already carried, in four places: the deposit record holds the reserve class as Probable with the tonnage, grade and contained graphite, and holds the financing note naming the DFC facility of October 2024, the DOE loan and the default events; the deposit-reserves feed holds the same reserve with a source digest; and the technical and verification feeds both carry the reserve and…</description></item><item><title>v2.137.237 — FIVE OPEN ITEMS CLOSED AGAINST THE CLAIMANTS' OWN FILINGS, AND ONE OF THEM WAS OUR ERROR.</title><link>https://afrimintel.com/audit-log/#v2-137-237</link><guid isPermaLink="false">afrimintel-v2.137.237</guid><pubDate>Mon, 14 Sep 2026 00:00:00 +0000</pubDate><description>FIVE OPEN ITEMS CLOSED AGAINST THE CLAIMANTS' OWN FILINGS, AND ONE OF THEM WAS OUR ERROR. BALAMA FORCE MAJEURE. The deposit dossier read that operations were suspended from mid-2024 under force majeure. Syrah's own ASX release of 12 December 2024 states that protest actions linked to farmland resettlement grievances commenced in late September 2024, that the general election result was announced on 24 October 2024, and that force majeure was declared by subsidiary Twigg Exploration and Mining under the Mining Agreement because a December-quarter production campaign could not be run. Mid-2024 is neither the protest onset nor the declaration. The dossier text is corrected on the feed and on the pro-data mirror that carried the same sentence. WHAT THE SAME RELEASE ALSO DISCLOSES, WHICH WE WERE NOT LOOKING FOR. The duration of the protest actions triggered events of default under the…</description></item><item><title>v2.137.236 — THE NGUALLA CITATION CHAIN WAS NOT IN DISORDER.</title><link>https://afrimintel.com/audit-log/#v2-137-236</link><guid isPermaLink="false">afrimintel-v2.137.236</guid><pubDate>Mon, 14 Sep 2026 00:00:00 +0000</pubDate><description>THE NGUALLA CITATION CHAIN WAS NOT IN DISORDER. ONE DOCUMENT, TWO PRECISIONS, AND A SECONDARY PUBLISHER THAT CONTRADICTS ITSELF. The open verification item held 176 Mt, 214 Mt and 214.9 Mt at 2.15 or 2.24 per cent TREO as an unresolved chain requiring the claimant's annual statements and the partner's exchange disclosures. Read against the primary document, the chain resolves and most of it was never real. Table 4 of the October 2022 BFS Update states Measured 86.1 Mt at 2.61 per cent, Indicated 112.6 Mt at 1.81 per cent, Inferred 15.7 Mt at 2.15 per cent, total 214.4 Mt at 2.15 per cent TREO. The same document's opening summary rounds that same total to 214 Mt. The capital layer carried the rounded form and the archive carried the table form: one source at two precisions, not two competing figures. The capital layer now carries 214.4 Mt and re-verifies against the preserved…</description></item><item><title>v2.137.235 — THE PLATFORM SCORED ITSELF AS CORRECTED ON A FIGURE IT ELSEWHERE HOLDS UNRESOLVED.</title><link>https://afrimintel.com/audit-log/#v2-137-235</link><guid isPermaLink="false">afrimintel-v2.137.235</guid><pubDate>Mon, 14 Sep 2026 00:00:00 +0000</pubDate><description>THE PLATFORM SCORED ITSELF AS CORRECTED ON A FIGURE IT ELSEWHERE HOLDS UNRESOLVED. A scored public self-correction on Ngualla, dated 12 September, states that the published resource figures appear nowhere in the cited document, which reads 214Mt at 2.15% TREO. An open verification item on the same asset, in the same build, holds 176, 214 and 214.9 Mt at 2.15 or 2.24 per cent as a citation chain in visible disorder, unverified until 30 September, pending the claimant's own reserves and resources statements and the partner's exchange disclosures. The archive carries a third form of the same number. A reader taking the scored row at face value would conclude the record was settled. It is not. WHAT WAS AND WAS NOT DONE. The resource figure is NOT resolved here and no figure was chosen: resolving it needs the claimant's own documents, and reasoning to an answer without them is the failure…</description></item><item><title>v2.137.234 — THIRTY-NINE PAGES SCROLLED SIDEWAYS ON A PHONE AND NOTHING HAD EVER LOOKED.</title><link>https://afrimintel.com/audit-log/#v2-137-234</link><guid isPermaLink="false">afrimintel-v2.137.234</guid><pubDate>Mon, 14 Sep 2026 00:00:00 +0000</pubDate><description>THIRTY-NINE PAGES SCROLLED SIDEWAYS ON A PHONE AND NOTHING HAD EVER LOOKED. The v2.137.233 sweep was run at desktop width only, and its result — no page scrolls horizontally — was stated without that scope. A hostile re-read of that claim tested the same pages at phone width and found five failing, including one the same build had just declared fixed. Building the guard properly, across every surface rather than a list of suspects, raised that to thirty-nine. Almost all of it predates this build and none of it was visible to any existing check. Causes, each traced rather than guessed: multi-column tables with no narrow-width scroll box, so the table set the width of the page; flex and grid items taking the default min-width of auto and refusing to shrink below their content; and badge elements held at white-space nowrap that carry a full sentence rather than a short state. FIXED…</description></item><item><title>v2.137.233 — THE DFI SCREEN CARDS WERE NESTED INSIDE EACH OTHER, NOT LISTED BESIDE EACH OTHER.</title><link>https://afrimintel.com/audit-log/#v2-137-233</link><guid isPermaLink="false">afrimintel-v2.137.233</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>THE DFI SCREEN CARDS WERE NESTED INSIDE EACH OTHER, NOT LISTED BESIDE EACH OTHER. Every gate passed while seven DFI-facing pages rendered as a staircase. card() on each of the seven /dfi-screen/ layer pages opened a card div and returned without closing it, so each card became a child of the one before it. Loaded in a real browser at desktop width, the cards nested between sixty-two and sixty-five levels deep depending on the page, each indented by its own padding, until the last of them were about thirty pixels wide with their tables spilling past the right edge of the page. No check saw it: the feeds were correct, the counts reconciled, the markup parsed, and every figure on the page was the right figure. It was found by rendering the pages and measuring the boxes. check-95 already loads a named list of surfaces in headless Chromium and fails on page errors; no /dfi-screen/ page is…</description></item><item><title>v2.137.232 — THE GOVERNMENT-TAKE ESTIMATOR OPENED ON DRC BY DEFAULT.</title><link>https://afrimintel.com/audit-log/#v2-137-232</link><guid isPermaLink="false">afrimintel-v2.137.232</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>THE GOVERNMENT-TAKE ESTIMATOR OPENED ON DRC BY DEFAULT. WHY THAT MATTERS. The editor holds Honorary Consul of Rwanda and Chairman of MUA Insurance Rwanda, and the standing independence position is that Great Lakes jurisdictions are data-layer only — statutory facts, never an editorial read. The tool loaded with DRC pre-selected and its Reset button returned to DRC, so the jurisdiction appeared on screen without anyone choosing it. In private that is a default; demonstrated to an audience it reads as a selection. The existing guard was already correct — a caveat fires when DRC is chosen, stating data-layer only, no editorial or outreach — but a guard on a deliberate choice does not cover an automatic one. FIXED. One DEFAULT_REGIME constant used by both first load and Reset, set to Tanzania: the deepest asset coverage on the platform (Kabanga, the only worked DCF with preserved source…</description></item><item><title>v2.137.231 — THE v2.137.230 FIX ATTRIBUTED ONE ASSET'S FINANCING FACT TO ANOTHER.</title><link>https://afrimintel.com/audit-log/#v2-137-231</link><guid isPermaLink="false">afrimintel-v2.137.231</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>THE v2.137.230 FIX ATTRIBUTED ONE ASSET'S FINANCING FACT TO ANOTHER. WHAT WAS WRONG. v2.137.230 derived a missing mandate-fit statement from another record in the same country, to stop three Côte d'Ivoire assets giving three different answers to the same jurisdiction question. The derivation was applied across all three mandate dimensions. Two of them are jurisdiction-level and share correctly. The third, western_dfi_eca, is asset-specific — it names who is financing THIS asset — so on the IFC and DFC panels Enchi (Newcore Gold, Ghana) displayed 'Control change pending: Huayou Cobalt to sole-fund', which is Ewoyaa's fact, a different operator's project in the same country. A reader screening Enchi would have been told a financing arrangement that does not exist for it. FIXED. Derivation now applies only to dimensions that are genuinely properties of the jurisdiction: EBRD…</description></item><item><title>v2.137.230 — THREE DEFECTS ON THE FOUR-LAYER DFI SCREEN, THE PAGE AN OFFICER SCREENS FROM.</title><link>https://afrimintel.com/audit-log/#v2-137-230</link><guid isPermaLink="false">afrimintel-v2.137.230</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>THREE DEFECTS ON THE FOUR-LAYER DFI SCREEN, THE PAGE AN OFFICER SCREENS FROM. (1) MANDATE FIT DISAGREED WITH ITSELF WITHIN ONE COUNTRY. Assafou, Doropo and Koné showed 'Côte d'Ivoire is an EBRD country of operation'; Afema and Boundiali, also Côte d'Ivoire, showed '—'. Ewoyaa showed Ghana's accession status; Enchi, also Ghana, showed nothing. Root cause: dfi_mandate ships in two shapes — older records carry a structured block with per-institution keys, newer ones a flat {value, state} prose summary with no mapping — and the emitter read the key alone. A jurisdiction-level mandate fact belongs to the jurisdiction, so the country-level statement is now taken from the records that carry it and applied to every asset in that country, marked Derived with the basis in the tooltip. Countries where no record carries a statement still show '—' rather than an invented one. (2) CELLS WERE CUT…</description></item><item><title>v2.137.229 — THE SHARED DOSSIER HAS A SCREEN.</title><link>https://afrimintel.com/audit-log/#v2-137-229</link><guid isPermaLink="false">afrimintel-v2.137.229</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>THE SHARED DOSSIER HAS A SCREEN. WHY IT EXISTS. When an organisation looks at an African mining asset, several people assemble the same folder from the same public sources and none can see the others' work. That duplicated assembly is the largest block of wasted hours in the process, and the errors follow from it: two desks disagreeing on which reserve statement is current, a figure re-keyed and never re-checked, an hour spent finding a filing a colleague found last week, the whole exercise repeated six months later because nobody recorded what was checked. The endpoint shipped at v2.137.217 with no interface, which made it plumbing rather than a product. WHAT SHIPPED. /dossier/ — pick an asset from the platform's own bankability roster, record a figure with its document, that document's date and where in it, and read what colleagues at the same organisation have recorded. A value…</description></item><item><title>v2.137.228 — §11 FRAMEWORK ALIGNMENT PUBLISHED, AND TWO COUNTS CORRECTED ON THE PAGE THAT TEACHES COUNTING DISCIPLINE.</title><link>https://afrimintel.com/audit-log/#v2-137-228</link><guid isPermaLink="false">afrimintel-v2.137.228</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>§11 FRAMEWORK ALIGNMENT PUBLISHED, AND TWO COUNTS CORRECTED ON THE PAGE THAT TEACHES COUNTING DISCIPLINE. WHAT WAS WRONG. §11 shipped as 'DRAFT v0.3 — Pre-Publication' dated 22 May 2026, publicly indexed and in the sitemap three times, while its own verification queue records all seven items closed against primary sources. The page was finished and labelled unfinished. Worse, it carried two wrong counts of exactly the kind its own discipline learning was written about: after the v0.2 error of stating AfDB ISS has eleven Operational Safeguards rather than ten, §11.6 records that 'number-of-X verification requires enumeration of X1 through X[n] from the primary source, not oblique reference to a related-but-distinct count'. The header claimed twenty-two frameworks; enumeration of the §11.2 tables gives twenty-three. §11.5 claimed twenty-two distinct publishers and then listed…</description></item><item><title>v2.137.227 — THE SAFEGUARDS SCREEN TOLD READERS THE FEED WAS UNAVAILABLE WHILE THE FEED WAS FINE.</title><link>https://afrimintel.com/audit-log/#v2-137-227</link><guid isPermaLink="false">afrimintel-v2.137.227</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>THE SAFEGUARDS SCREEN TOLD READERS THE FEED WAS UNAVAILABLE WHILE THE FEED WAS FINE. WHAT WAS WRONG. /dfi-screen/safeguards/ displayed 'Could not load /api/v1/safeguards.json.' on the live site. The feed loads and parses correctly and carries all 62 assets. The page filtered on a.screen.indexOf(...); three rows — balama-graphite, kamoa-kakula and manono-lithium-district — carry no screen field, so the call threw. Because the .catch wrapped the whole chain rather than the fetch, a rendering fault was reported as a failed load. The KPI strip above rendered first, so the page looked half-working rather than broken, and a reader was told the data was unavailable when the data was correct and the view was at fault. This is the E&amp;S screen a development-finance officer opens. FIXED. Every access guarded with String(x||''), rows with no screen state say so rather than throwing, and the catch…</description></item><item><title>v2.137.226 — APPEARANCE AUDIT: CONTRAST, VIEWPORT AND WIDE TABLES.</title><link>https://afrimintel.com/audit-log/#v2-137-226</link><guid isPermaLink="false">afrimintel-v2.137.226</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>APPEARANCE AUDIT: CONTRAST, VIEWPORT AND WIDE TABLES. WHAT WAS CHECKED. Structure had been walked (290 pages, 3,703 links, 121 navigation items) but nothing had been checked for how it looks. This pass computed what is computable without rendering: theme-variable resolution, WCAG contrast ratios, viewport meta presence, asset resolution, fixed-width overflow risk, and per-table column counts. WHAT WAS WRONG. (1) The evidence-depth panel's footnote used --faint (#5E6B7C) at 0.82rem against the #0A0F1A background — contrast 3.53, below the 4.5 WCAG AA needs for body text. Readable on a good monitor, marginal on a projector. Moved to --muted (#8B98A8), contrast 6.53, on all 24 pages. (2) /verification/ shipped with no viewport meta, so it rendered desktop-width on a phone. Added, and its seven-column table given a scroll container with a 640px minimum so columns stay legible rather than…</description></item><item><title>v2.137.225 — THE HONESTY PANEL INFLATED ITS OWN DEPTH CLAIM, AND ITS COUNTS WERE GUARDED BY NOTHING.</title><link>https://afrimintel.com/audit-log/#v2-137-225</link><guid isPermaLink="false">afrimintel-v2.137.225</guid><pubDate>Sun, 13 Sep 2026 00:00:00 +0000</pubDate><description>THE HONESTY PANEL INFLATED ITS OWN DEPTH CLAIM, AND ITS COUNTS WERE GUARDED BY NOTHING. WHAT WAS WRONG. The evidence-depth panel shipped at v2.137.222 read '13 at study depth or better', counting worked-DCF, sourced-study and sourced-headline together. The platform's own tier definition says sourced-headline is 'Headline DFS/FS-grade markers (typically NPV and/or capex) ... NOT the complete study marker set. Scoping/PEA-grade'. The nine sourced-headline assets are explicitly excluded from study depth by the definition the panel links to. The honest figure is 4, not 13 — inflated by nine assets across 24 pages, in the panel built to state depth honestly. SECOND FAULT. The panel hardcoded five counts (roster, worked DCFs, study-level, preserved assets, figures verified) into 24 pages with nothing recomputing them. The canonical-facts registry exists for exactly this and held none of…</description></item></channel></rss>
