Validation, published honestly.
Every case study below committed its pass/fail bar in code before the result existed, then recorded the verdict verbatim — whatever it was. The spread is the point: we publish the fails on the same page as the passes, because a validation record you can't lose is a validation record you can't trust.
The record so far: ~24 blind, pre-registered studies across roughly ten commodities and eight countries, run in about three months, with every verdict recorded as it came out. Three of those studies are cleared by our Competent Person and published in full below. The rest stay internal — counted in the spread, never named — until each one is signed off and scrubbed for client confidentiality.
This spread is deliberate: the ledger behind this page is the strictest record we keep — blind, pre-registered, and impossible to lose quietly — while the cleared, fully-published case studies live in the applied section further down.
The full ledger, in numbers
Every count below is computed from the committed ledger — nothing here is hand-written.
By study — case verdicts
By committed metric — outcomes
What each verdict means
No verdict appears on this page without its criteria. The labels below are the ledger's own, assigned by the committed mapping rule — never softer than their source — and defined here so a reader is never left guessing what a PASS or a FAIL is.
- PASS
- The study met its committed bar: the pass/fail metric locked in version-controlled code before any result existed was cleared on the blind run.
- PASS (qualified)
- The committed bar was met, with a caveat recorded beside the result. The qualification is published with the verdict, never dropped.
- MIXED
- Part of the signal, short of the full bar: the primary committed metric passed, but at least one other committed metric did not. Stated plainly — a MIXED case is never rounded up to PASS.
- FAIL
- The committed bar was not cleared. A FAIL is recorded whenever that happens — including when missing the bar is the correct behaviour. The reason is published beside every FAIL, and it is one of four:
- Control held — a blind decoy or placebo control correctly returned no detection. The engine did not invent a target; the scoreboard still reads FAIL because the detection bar itself was not met.
- Limit identified — data — the engine ran clean, but the available public data lacks the layer needed to resolve this system. The limit is external to the method and named in the report; it is the clearest case for new acquisition.
- Limit identified — by design — an underdetermined problem where low recovery was expected and documented up front, together with the planned fixes.
- Bar not met — a genuine miss: the method did not clear its pre-registered threshold on this dataset. It stays on this page, beside the wins.
- NOT-YET-SCORABLE
- The bar is committed, but the input needed to score it does not exist yet — for example drilling that has not happened. The verdict stays pending; it is never guessed.
- DESCRIPTIVE/METHODS
- A methods or synthetic demonstration. It carries no committed bar and claims no verdict, and it never counts toward the pass or fail tallies.
How to read these numbers
Each scored study is pre-registered: the pass/fail decision rule (the "committed bar") is fixed in version-controlled code before any result exists — each study's pre-registration commit is cited in its report, and the record's integrity can be checked at tolun.ai/verify— so a bar can never quietly move to meet a result. Verdict labels map the repo's verbatim result strings onto a small enum (PASS, PASS (qualified), FAIL, NOT-YET-SCORABLE, DESCRIPTIVE/METHODS, MIXED), and a label is never softer than its source: a primary-metric FAIL is a case FAIL no matter how well a secondary did, and a primary PASS with any other FAIL reads MIXED, never PASS. Method and synthetic demonstrations claim no verdict at all. Every count above is computed from the committed ledger — nothing on this page is hand-written.
Why only three studies are named: publication here follows the same rule as a signed report. A study is named in public only after the Competent Person of record signs it off and it is scrubbed for client confidentiality. The remaining studies are counted in the spread above and join the shelf below one by one as each clears sign-off.
Cleared case studies
The three studies our Competent Person has cleared for publication, in full — method, scoring, and verdict included.
Winu blind prospectivity test (Cu-Au) →
A blind prospectivity test over the Winu Cu-Au discovery (Western Australia) on public data — modest, honestly-reported skill, published as-is with the full method and scoring in the report.
Pilgangoora exploration-to-closure study (Li) →
The full asset lifecycle run on public data over the Pilgangoora lithium district (Western Australia): exploration screening, resource, development, operations, and closure — with the uncertainty carried forward at every stage and every demonstration-scale step labelled as such.
NSW Macquarie Arc (Cadia-Ridgeway · Northparkes) →
Run end-to-end, blind, on 100% public NSW field data. It reproduces the documented geophysical reality of both porphyry districts — including, honestly, where regional potential fields do not localize deep or magnetite-destructive orebodies. A methods-and-limits study, published in full.
Further applied studies from client engagements are available under NDA.