Track Record
Published record of corrections caught, prediction-resolution items, and tracked positions. The platform's discipline test.
This page is the platform's published record of three things: corrections caught and recorded; prediction-resolution items where the platform held a position before a verifiable outcome; and tracked positions currently carried in the data layer awaiting resolution. Several items below are pending — outcomes may resolve against the platform. Both eventualities will be recorded here without retrospective adjustment. This is the tracking discipline that institutional research products (Wood Mackenzie, S&P, CRU, Benchmark, Fastmarkets) have built over decades; Afrimintel's record is short — the platform's structured tracking begins with the platform itself.
Section 1 — Corrections caught and recorded
CORRECTION
v1.0.45
9 MAY 2026
1.1 — Self-caught hallucination (drift class 18, Audit Log entry #70)
What happenedA v1.0.45 fix to one source citation (Kamoa-Kakula res_source) was extrapolated to a parallel asset (Kipushi res_source) without primary-source verification. The extrapolated text introduced two new fabrications: a non-existent "Kipushi Mineral Reserve and Mineral Resource Technical Report" in the same release cycle as Kamoa-Kakula; a non-existent "2023 NI 43-101 Feasibility Study Update" for Kipushi.
How it was caughtComprehensive hallucination check applied to the build's own corrections. Primary-source verification (Ivanhoe Q1 2026 news release) confirmed Kipushi's controlling NI 43-101 is the separate "Kipushi 2022 Feasibility Study Technical Report" dated 14 February 2022, prepared by OreWin, MSA Group, SRK Consulting, and METC Engineering.
How it was correctedKipushi res_source corrected to verified 2022 FS reference. Same correction propagated to historical migration scripts. Two parallel COUNTRY_PROFILES surface drifts corrected at same fix-class. Build re-tested to 23/23 PASS.
Why this mattersA fix introduced in a build propagated a fabrication parallel to the original. The comprehensive hallucination check on the corrected build caught the new fabrication. The Audit Log records the introduction, the catch, and the correction together rather than only the corrected end state.
SourceAudit Log entry #70 (9 May 2026); Ivanhoe Q1 2026 news release; v1.0.45 pre-deploy hallucination check pipeline.
CORRECTION
v1.0.45
1.2 — Kamoa-Kakula source-citation imprecision (drift class 15)
What happenedAcross multiple platform surfaces, the Kamoa-Kakula NI 43-101 reference was generically labelled "Ivanhoe NI 43-101 IDP 2026" — but the actual document title released 31 March 2026 is "Kamoa-Kakula Mineral Reserve and Mineral Resource Technical Report" (the 2023 IDP is a separate, distinct document).
How it was correctedSource citation precision restored across data layer (js/data.js for both Kamoa-Kakula and Kipushi entries), historical scripts, and DCF Test Battery surface. Document chain now correctly distinguished: 2023 IDP → 31 December 2025 effective date → 31 March 2026 release date → AMC + MSA preparers.
CORRECTION
v1.0.45
1.3 — DRC Country Risk Composite EITI component (drift class 16)
What happenedData layer source string for the Country Risk Composite named only Fraser/TI/RGI as components — but the implementation includes EITI as a fourth component at 20% weighting. Hostile-audit pass against Quality Standard's claim that the 30/25/25/20 weighting was published surfaced the gap.
How it was correctedSource string corrected to include EITI component. New "Country Risk Composite" methodology page section published with formula, source documents, weight re-normalisation worked example, risk-band thresholds, LOW-MEDIUM Fraser top-tier override rationale, and DRC editorial override.
CORRECTION
v1.0.45
1.4 — Pipeline gap: CHECK 23 added (drift class 17)
What happenedCHECK 22 enforced tier-count reconciliation between methodology page and data layer but did not extend to secondary user-facing surfaces — a gap that allowed legacy tier framings to drift undetected.
How it was correctedCHECK 23 (cross-surface tier consistency) added to the pre-deploy pipeline. Self-test confirms CHECK 23 catches injected legacy framing and blocks deploy. Pipeline is now 23/23 PASS.
CORRECTION
v1.0.45
9 MAY 2026
1.5 — Hostile-audit-driven Editorial Judgment addition (Audit Log entry #74)
What happenedTime-cost framings ("12-18 analyst-day breakdown"; "~2 hours of platform interaction time"; "20-30 analyst-day benchmark") had been embedded in Kabanga / Lobito / Manono dossier value-add sections. A senior-reader hostile reading flagged these as marketing-narrative dressing analytical substance and as institutionally-implausible time-saving claims that a credit committee would treat as overreach.
How it was correctedTime-cost claims stripped from Kabanga, Lobito, Manono dossier index.html and dossier.md. Replaced with qualitative value-add framing centred on audit-trail-and-integration-preserved rather than time-saved. Quality Standard extended with Editorial Judgment section explicitly preserving "no fourth state" for numeric and factual claims and adding parallel discipline for qualitative editorial framings (surfaced as judgment not fact; basis named when load-bearing; subject to Correction Velocity SLA when challenged).
Why this mattersThe platform's three-state Quality Standard governs numeric and factual claims. Until entry #74, qualitative editorial framings were not explicitly subject to a parallel discipline. The hostile-audit catch surfaced a structural gap; the Quality Standard now closes it.
CORRECTION
v1.0.45
9 MAY 2026
1.6 — Loulo Benchmark Spread Metric 4 royalty divergence-publication fix (Audit Log entry #77)
What happenedThe Loulo-Gounkoto Benchmark Spread Metric 4 originally framed the post-2023-Code royalty as "9-12% blended gold-price-tranched" without explicit source citations and without surfacing the genuine public-source divergence between mainstream coverage citing flat 10% / 10.5% and Ecofin Agency coverage citing a progressive scale per Mali Decree No. 2024-0396/PT-RM of 9 July 2024. The framing was defensible reasoning but read as opinion rather than as structured surfacing of public-source divergence.
How it was correctedMetric 4 rewritten with full source-citation surfacing: Discovery Alert September 2025 (FLAT 10%); Mining-Technology citing Reuters (FLAT 10.5%); GoldBuyersAfrica July 2025 (~10.5%); Ecofin Agency October 2025 (PROGRESSIVE scale per Decree 2024-0396 with 3% below $1k/oz / 6% at $1.6-2k / 7% up to $2.5k/oz). Reconciliation framing made explicit that progressive-scale extrapolation to current $3-3.5k/oz prices implies 9-12% blended only IF tranching continues above $2.5k/oz. Both readings published as defensible. HIGH/MEDIUM/LOW confidence pill added to most-defensible reading.
Why this mattersThis is exactly the discipline a Benchmark Spread should embody — surfacing public-source divergence transparently rather than producing a single-point platform reading. The discipline is structurally hostile to the incumbent subscription model (Wood Mac / S&P / CRU produce single-point estimates from proprietary data); publishing divergence is the category-creation thesis operationalised on one metric.
CORRECTION
v1.0.45
9 MAY 2026
1.7 — Wodgina ownership fabrication caught on first hallucination check (Audit Log entry #79)
What happenedThe Manono Comparable Benchmarking page (Ship 13 of the v1.0.45 build) asserted Wodgina ownership as "Mineral Resources (post-Aug 2025 100%; previously 50/50 with Albemarle)". This was a fabrication. Verified position per MinRes corporate website + Albemarle 2024 10-K + November 2025 POSCO partnership disclosure: Wodgina remains a 50/50 unincorporated MARBL JV between MinRes and Albemarle; the November 2025 POSCO transaction was at MinRes parent-level for 30% of MinRes' operational lithium business (which holds MinRes' 50% Wodgina + 50% Mt Marion stakes), valuing those stakes at ~$3.9bn — NOT a MinRes acquisition of Albemarle's 50%.
How it was caughtUser-requested hallucination check on Ships 12-19 cluster, immediately following the cluster's completion. Web search verification against MinRes corporate disclosures and Albemarle SEC filings.
How it was correctedWodgina row corrected to "Mineral Resources 50% / Albemarle 50% (MARBL JV); MinRes operates" with explicit POSCO November 2025 parent-level transaction note. Mt Marion ownership confirmed verified at MinRes/Ganfeng 50/50 with POSCO indirect 15% MinRes-side interest noted.
Why this mattersThe Hallucination Guard discipline embedded in the platform constitution explicitly contemplates this category of failure: "AI-generated commercial-strategy docs are particularly prone to fabricating counterparties." The Wodgina assertion was a counterparty-fabrication hallucination of exactly the kind the discipline is designed to catch — and the catch happened on first audit.
SourceAudit Log entry #79; MinRes corporate website Wodgina page; Albemarle Corporation 2024 10-K p.5; mining.com 12 November 2025 POSCO deal coverage.
CORRECTION
v1.0.45
9 MAY 2026
1.8 — Pilgangoora LCE arithmetic error caught on hallucination check (Audit Log entry #79)
What happenedThe Manono Comparable Benchmarking page asserted Pilgangoora at "~10.5 Mt LCE" against a 446 Mt @ 1.28% Li2O resource. Standard chemistry-derived LCE conversion (Li2O × 2.473 conversion factor): 446 × 0.0128 × 2.473 = 14.12 Mt LCE. The 10.5 figure was hallucinated and arithmetically inconsistent with the platform's own cited resource.
How it was caughtHallucination check verification of platform's own arithmetic on cited resource × grade × LCE conversion factor.
How it was correctedPilgangoora row replaced with operator-disclosed framing plus explicit chemistry-implied derivation showing the 2.473 conversion factor and resulting ~14 Mt figure. LCE-comparison observation in Resource Analysis section corrected to use the verified figure.
Why this mattersHallucinated derived numbers that are arithmetically inconsistent with the platform's own cited primary-source figures are the most diagnostically useful catch — they are evidence that the build process can introduce errors that contradict its own source-trail. The discipline of catching its own arithmetic is what makes the platform's primary-source verification claim credible.
CORRECTION
v1.0.45
9 MAY 2026
1.9 — Country Risk Overlay sovereign credit ratings fabrications caught on hallucination check (Audit Log entry #79)
What happenedThe Country Risk Overlay v1.0 (Ship 16) sovereign credit summary table asserted specific point ratings for Mali ("CCC+"), DRC ("B-/B equivalent"), Angola ("B-/B class"), Zambia ("CCC class post-2024 restructuring"), and Mauritania ("B-/B class") without primary-source verification against direct S&P / Moody's / Fitch published rating actions as of May 2026. Web search during the audit confirmed that no public-source aggregation provided specific verified ratings for these jurisdictions with the dates and primary-disclosure attribution the platform's verification standard requires.
How it was caughtHallucination check on Ships 12-19 cluster. Searches for current Mali / DRC / Angola / Zambia / Mauritania ratings against named agency primary disclosures returned variable coverage and no specific point ratings the platform could verify to the standard required for assertion.
How it was correctedSpecific point ratings replaced with verified-context-language-only — "Below-investment-grade context"; "Selective-default-and-restructuring jurisdiction post-2020 default per worldpopulationreview December 2024 SD-list inclusion"; "Coverage and rating action history affected by 2022 sanctions period" — with explicit "this overlay does not assert specific point ratings without direct primary-source verification" provenance note. Tanzania ratings retained because verified to TanzaniaInvest disclosures of Moody's confirmation September 2024 + Fitch reaffirmation March 2026.
Why this mattersA senior credit committee reader will treat asserted-but-unverified specific ratings as a category of hallucination that disqualifies the overlay's other content. The platform's discipline of pulling unverified ratings on first audit and replacing them with framework language preserves the overlay's integrity for the rest of its content (DRC paramilitary force documentation, Mali settlement coverage, AfDB-EIB Mauritania $275M railway precedent — all primary-source verified).
CORRECTION
v1.0.45
9 MAY 2026
1.10 — Loulo AISC and historic royalty unverified figures caught on hallucination check (Audit Log entry #79)
What happenedThe Loulo-Gounkoto DCF input mapping table (Ship 19) and Comparable Benchmarking cost positioning row asserted "~$1,400/oz historic Barrick disclosure" AISC and "~6% historic" royalty without specific primary-source citation in the dossier source-trail. Neither figure was in the original Loulo Benchmark Spread or dossier source materials.
How it was caughtHallucination check verification against original Loulo dossier and Benchmark Spread source materials returned no specific AISC or historic royalty figures with primary-source citation.
How it was correctedBoth figures replaced with "consult Barrick Mining Corporation FY disclosures for asset-level breakdown" framing. Settlement-recovery scenario calculation made explicit: post-settlement 10% royalty per Discovery Alert / Mining-Technology adds approximately $30M/yr cost burden = ~$100/oz at 290 koz attributable steady-state, with the calculation traceable rather than an asserted single figure.
Why this mattersAsserting figures with attribution language ("historic Barrick disclosure") that the platform cannot actually back-cite to a specific Barrick disclosure is the most disciplinarily corrosive form of hallucination — it adds false attribution rather than just an unverified number. The Hallucination Guard catch on first audit pulls the false attribution and replaces it with consult-source framing.
CORRECTION
v1.0.45
9 MAY 2026
1.11 — Protected-name leakage in audit log entry caught on deep hostile audit (Audit Log entry #80)
What happenedAudit log entry #78 (documenting Ships 12-19) contained a self-check note in the form "Self-check: [protected individual's name] FULLY removed from all v1.0.45 platform-side artifacts per standing requirement at session-7 cluster." The meta-note about the protected name's removal itself contained the protected name. The audit log ships publicly. This defeated the entire scrub.
How it was caughtDeep hostile audit second-pass run immediately following the user-requested hallucination check. Bundle-wide grep on the protected name set returned the leakage in audit log entry #78.
How it was correctedSelf-check note replaced with "Self-check: protected individual-name discipline preserved per standing exclusions requirement; full-build grep returns zero on the relevant excluded-name set across all v1.0.45 platform-side artifacts." Bundle-wide grep on protected name set returns zero matches post-correction.
Why this mattersThis is a textbook example of how privacy-preserving meta-notes can leak the very information they purport to protect. The discipline of catching the leakage on hostile audit — rather than concealing it — is itself the institutional credibility surface a senior reader inspects. The two-stage verification architecture (hallucination check → deep hostile audit) caught what the single hallucination check missed; this is exactly the multi-stage discipline that institutional users want to see operating before commercial engagement.
CORRECTION
v1.0.45
9 MAY 2026
1.12 — Zambia "CCC-class exit pathway" residual prose caught on deep hostile audit (Audit Log entry #80)
What happenedThe hallucination check (entry #79) corrected the Country Risk Overlay sovereign credit summary table by removing unverified specific ratings — but missed a residual "CCC-class exit pathway through 2025-2026 reforms" specific-rating reference in the Zambia country block prose. The table now correctly framed Zambia as "restructuring-pathway emergent without specific point ratings asserted" but the prose still contained the unverified CCC-class characterisation.
How it was caughtDeep hostile audit second-pass grep on remaining unverified specific ratings (CCC, Caa, B3-class, B-/B equivalent) caught the leftover where the table-only fix had not propagated to the prose.
How it was correctedZambia country block prose replaced with "restructuring-exit-pathway through 2025-2026 reforms ongoing" framing plus explicit consult-agencies-directly direction. Bundle-wide grep on unverified specific ratings returns only audit-log historical-correction-context references.
Why this mattersDemonstrates that the two-stage verification architecture catches gaps that a single-stage check would miss. The hallucination check missed this; the deep hostile audit caught it. The discipline of the second-pass-catch is exactly the kind of multi-stage verification institutional users want to see operating before commercial engagement.
Section 2 — Prediction-resolution items
PREDICTION-RESOLUTION
DIRECTIONAL VALIDATION
2.1 — Kamoa-Kakula reserve-grade evolution prediction-vs-actual
Position the platform heldThrough Q1 2026, the Kamoa-Kakula intelligence-grade dossier carried a published expectation that the post-May 2025 seismic event combined with Kakula 2.0 mine plan, pillar-width adjustments, and depletion would result in material reserve-grade compression at the 31 March 2026 NI 43-101 release.
Resolution (31 March 2026)Ivanhoe published the Kamoa-Kakula Mineral Reserve and Mineral Resource Technical Report (effective 31 December 2025): Probable Reserve 466 Mt @ 2.82% Cu (13.1 Mt contained), down from December 2022's 472 Mt @ 3.94% Cu — a 25% reduction in contained Cu. The reserve grade compression was material and consistent with the post-seismic-event direction the platform had carried.
What this validatesThe platform's discipline of tracking dated operational events (May 2025 seismic) and propagating them through reserve-confidence framing in the IG dossier produced a position that the subsequent NI 43-101 release validated in direction.
What this does NOT validateThe platform did not predict the precise reserve tonnage or grade. The validation is directional confidence framing, not predictive accuracy.
SourceAfrimintel Kamoa-Kakula IG dossier; Ivanhoe NI 43-101 effective 31 Dec 2025, released 31 Mar 2026, prepared by AMC + MSA.
PREDICTION-RESOLUTION
DIRECTIONAL VALIDATION
2.2 — AVZ ICSID Manono trajectory direction signal
Position the platform heldManono Lithium District record flagged through Q1 2026 with dispute_status: "multi_claim_active_arbitration"; multi-claim sub-records explicitly mapping the Zijin/Cominière NE construction track and Dathcom-AVZ South arbitration track separately. Risk catalogue carried AVZ Manono as the canonical HIGH-severity precedent for "DRC CAMI cadastre cancellations."
Resolution (April 2026)ICSID tribunal issued partial decision favourable to DRC and lifted earlier interim measures requiring recognition of AVZ rights. AVZ position legally weakened. The platform's HIGH-severity DRC CAMI risk framing and multi-claim sub-record structure both held through this resolution direction.
What this validatesThe platform's discipline of treating disputed-tenure assets as multi-claim structures (not single-operator) and of carrying jurisdiction-specific risk overrides (DRC HIGH editorial override) produced a position consistent with the contested multi-claim status of the asset, pending ICSID resolution.
What this does NOT validateThe April 2026 decision is partial. The substantive final award is not yet rendered. The platform did not — and does not — predict ICSID arbitration outcomes.
SourceAfrimintel Manono record; AVZ Minerals 19 Nov 2025 Corporate and Arbitration Update; ICSID procedural disclosures April 2026.
PREDICTION-RESOLUTION
FRAMEWORK VALIDATION
2.3 — Mali Mining Code post-settlement durability framing
Position the platform heldThrough 2025, the Loulo-Gounkoto record carried Mali 2023 Mining Code (Law No. 2023-040) as a structural recalibration framework, with Loulo-Gounkoto provisional administration June 2025 onwards flagged as a HIGH-impact governance event. The platform's risk framework treated 2023 Mining Code adoption as a continental AMV-aligned shift not unique to Mali but with country-specific intensity factors.
Resolution (24 November 2025 settlement)Barrick and Mali agreed: ~$430M cash + 10-year permit extension + return of 3 metric tons of gold + ICSID withdrawal + 2023 Mining Code acceptance + 80/20 ownership preserved + revised 9-12% blended royalty. Loulo permit renewed 13 February 2026.
What this validatesThe platform's discipline of treating 2023 Mining Code as a durable structural framework (not a transient negotiation tactic) produced a position consistent with the eventual settlement architecture (Barrick formally accepted 2023 Code rather than securing an exemption).
What this does NOT validateThe settlement framework's long-term durability has not yet been tested. The platform did not predict the precise settlement amount, the gold restitution structure, or the royalty tranching.
SourceAfrimintel Loulo-Gounkoto record; Barrick official press release 24 Nov 2025; AfronomicsLaw analysis 9 Feb 2026.
Section 3 — Tracked positions (currently carried; outcomes pending)
TRACKED POSITION
PENDING
3.1 — Kabanga FID timing and DFC Letter of Interest progression
Position currently carriedKabanga Nickel record carries mid-2026 FID + financial close as operator-disclosed target; DFC Letter of Interest dated September 2024; Société Générale lead financial adviser; EXIM Bank actively engaged; JOGMEC partnership; Standard Chartered Bank engaged for new equity partner search.
Outcome pendingMid-2026 FID occurs, slips, or is restructured. Financial close achieves multi-source DFI participation, slips, or restructures around different financing parties. DFC Letter of Interest converts to definitive board approval, slips, or is withdrawn.
What will be recordedThe resolution (whether to schedule, slipped, or restructured) will be recorded in this Track Record without retrospective adjustment of the original tracked position.
TRACKED POSITION
PENDING
3.2 — Manono ICSID dispute resolution
Position currently carriedManono record carries the active ICSID arbitration (Case No. ARB/23/20; US-mediated suspension lapsed 23 Jun 2025, proceedings resumed) with five conditional resolution pathways recorded (Zijin operating vehicle stabilises; AVZ-KoBold framework activates; ICSID resumption + AVZ favourable; ICSID resumption + DRC favourable; status quo deadlock).
Outcome pendingOne of the five conditional pathways resolves the dispute; ICSID merits undecided as of mid-2026 (proceedings active since the 23 Jun 2025 resumption).
What will be recordedThe path that materialises will be recorded with the date the resolution became verifiable through public primary sources.
TRACKED POSITION
PENDING
3.3 — Loulo-Gounkoto settlement durability triggers
Position currently carriedLoulo-Gounkoto record carries six trigger conditions whose materialisation would invalidate the November 2025 settlement framework: sovereign-continuity event with explicit settlement repudiation; Mining Code interpretive challenge by state on royalty calculation, local content, or expansion approvals; permit-renewal denial for satellite operations or expansions; sustained gold price below $2,200/oz combined with state cashflow pressure; detention or charge of Barrick personnel post-settlement; provisional administration imposed on any other operating asset.
Outcome pendingNone, one, or multiple triggers materialise over the 12-24 month durability test cycle following the November 2025 settlement.
What will be recordedTrigger materialisations (or non-materialisations confirmed by passage of time without invalidation) will be recorded.
TRACKED POSITION
PENDING
3.4 — Botswana Kalahari Copper Belt extension thesis
Position currently carriedKalahari Platform province record carries Northwest Zambia (Ivanhoe licence area) and NW Botswana extension as the geological continuation of KCB into cover. Exploration upside framework treats KCB extension as a structural prospectivity thesis, not a single-asset call.
Outcome pendingResolution of KCB extension prospectivity through drilling results from Ivanhoe or other operators; regional aeromagnetic surveys; discoveries within the cover sequence that validate or invalidate the prospectivity thesis.
What will be recordedDrilling intercepts; regional survey findings; new exploration projects launched into the prospectivity zone.
TRACKED POSITION
PENDING
3.5 — Indonesian nickel oversupply and first-quartile cost screen
Position currently carriedCritical Minerals sub-vertical and Kabanga record both carry the Indonesian laterite oversupply context as a structural variable. Platform's screening discipline treats first-quartile cost position as the critical Class I sulphide nickel filter for African asset evaluation in the current market.
Outcome pendingIndonesian production discipline (or absence) over 2026-2028; tightening of Class I battery-grade nickel supply versus stainless-grade segregation; Western supply-chain premium emergence (or absence); operator-published cost-curve evolution.
What will be recordedOperator FS / cost-curve disclosures that validate or invalidate the first-quartile screen as the appropriate decision filter.
Section 4 — Honest framing
What this record IS
- A published, dated, source-cited record of corrections caught, prediction-resolution items where the platform held a position before resolution, and tracked positions awaiting resolution.
- A discipline test. Most institutional research products do not publish their corrections, do not record their predictions in advance of resolution, and do not name positions whose resolution may go against them. This page does all three.
- A foundation for institutional evaluation. A senior reviewer can audit the record's construction, source-check items, and form their own view of the platform's discipline trajectory.
What this record is NOT
- A marketing document. There is no claim that any item validates the platform commercially or institutionally.
- A complete record of every position the platform has ever held. Items above prioritise sufficient source-trail to be defensible without inference.
- A predictive performance ledger. Several items are tracked positions whose outcomes are pending — outcomes may resolve against the platform.
- A substitute for the published Audit Log (which records every correction at finer granularity).
- A claim of partnership, customer relationship, or commercial conversation with any party named. The Counterparty Extension discipline applies.
Cadence
- New corrections: added to the Audit Log within the build cycle of the correction. Synthesis here updates at quarterly intervals.
- New prediction-resolution items: added when public primary sources confirm a tracked position's outcome.
- New tracked positions: added when the data layer carries a position that is dated, verifiable, and forward-looking enough to resolve within a 6-24 month cycle.
Cross-references
Methodology · Audit Log · Quality Standard · Case Studies
The full markdown
The full Track Record is available as a standalone markdown document for download or sharing: dossier.md.