Cover geometry at Linka

1. The prediction

The two models being separated are simple and they disagree. If the ground surface rises 20 m but the buried surface stays at the same elevation, you must drill about 20 m further to reach it. A constant-depth model says you always drill 16.8 m. Those predictions diverge, they are measurable, and one of them is wrong.

2. What is already known, and therefore earns nothing

• The RL 1771.2 relationship itself. It was observed during the earlier GS-PRED-002 work and is classified training evidence. It can never be scored.
• The four known points: LKRC0013 (the sole calibration), LKRC0001, and the two MRDS shafts.
• At target T01 specifically, that cover exceeds about 36 m — the operator's own abandoned hole established that. 42% of T01's stated band was dead before it was written. Only the 45.2 m ceiling is prospective there.

What is prospective is the depth to the base of cover at locations not yet drilled, logged and published.

3. Targets, frozen before analysis

N01 is the one that points the other way, and the only one a shovel can falsify.

4. How it fails

• F1 any qualifying hole more than 15 m from RL 1771.2.
• F2 a regression of base-of-cover elevation on collar elevation with a slope closer to 1 than to 0.
• F3 more than 10 m of cover in the sub-datum half of N01.
• F4 a hole within 25 m of the LKRC0001 collar reaching basement shallower than 29 m or deeper than 45 m.
• F5 published evidence of more than 30 m of relief on the cover base inside the domain.
• F6 baseline B3 reproducing this model within the required margin — in which case the case added nothing, and the record says so.

5. Counter-evidence, in our own words

• The model rests on one two-sided measurement, LKRC0013. There are zero independent two-sided measurements of the buried surface.
• LKRC0001 gives only a one-sided constraint: at about 36 m it was still in cover. It measures nothing.
• The MRDS 10095727 shaft, sunk in meta-limestone at 1772.7 m, requires the buried surface to be at or above 1772.7 — and so contradicts a perfectly constant 1771.2 m surface by 1.5 m.
• Those two constraints cannot both hold for a strictly constant surface. Metre-scale palaeorelief is already implied by the evidence used to build the model.
• Twenty-five million years of Basin-and-Range extension are unaccounted for. One normal fault of a few tens of metres inside the domain breaks a constant-elevation surface, and the best available geological map cannot resolve one.
• Inside T02 — the zone the operator has said it will return to, and therefore the likeliest to be drilled — a naive topographic baseline reproduces this model to within 6.2 m, below the margin needed to claim it beat a baseline. Confining the test there can leave this case unable to beat a baseline even if every depth is right.
• Whether “volcanic cover from surface” is logged consistently is unknown. Weathered tuff and tuff-derived colluvium are exactly where a several-metre error would come from.
• The 75 m exclusion radius around the three historic workings is arbitrary. The model is already known to fail there by 17.4 m, 19.4 m and 6.4 m.
• Confidence was frozen at MEDIUM. The pre-seal analysis found LOW was the better call. The frozen value stands.

6. Retrospective performance is not a predictive result

Against the four already-known points, excluding the calibration point, this model's mean absolute error is 1.10 m and the baselines' are 15–18 m. That is training-only and carries zero prospective credit. It establishes that sealing a prospective test is defensible. It establishes nothing else, and n = 3 with one of those a one-sided bound and one marginally violated.

7. Verification

The complete original is published now, encrypted, as GS_PRED_003_PREDICTION_V1.enc. Anyone can hash it today. When the resolving drillholes are published we release the key, and anyone can decrypt it and confirm the plaintext hashes to the value above — so the file cannot have been written afterwards.

8. Independent timestamp

The hash of the encrypted prediction was submitted to an independent third-party timestamping service at 2026-09-12T08:44:56Z, the day of sealing. We do not operate that service and cannot move the date it recorded. The receipt is retained as an internal verification artefact and is available on request.

What you can check today without trusting us at all. The encrypted original is published above and its SHA-256 is fixed. When the key is released, decrypting it must reproduce the prediction hash exactly. A file published before the resolving drillholes existed cannot have been written after them.

Being improved. The public verification step is being replaced by a timestamp token from an independent Time Stamping Authority under RFC 3161, so that the receipt can be downloaded and verified here rather than requested. Until that is in place this page claims no more than the line above.

The full public record and the frozen target geometry are both downloadable.

9. What this does not claim

• No tungsten, scheelite, skarn, grade, tonnage or economic claim of any kind. This case predicts cover geometry only.
• No claim about what lies beneath the cover base. One of four known points logged limestone there; the operator's own interpretation expected none.
• No claim that GeaSpirit discovered or verified anything at Linka. The drilling, logging and assaying are the operator's work.
• Sealing is not validation. One sealed case is an anecdote, and this one rests on a single calibration point.

← Back to the Prospective Validation Registry