Utah FORGE geothermal site · gravity inversion · DOE GDR 1144 (CC-BY 4.0)
CLAIMABLE — four core gates pass. RECOVERY UNTESTED: no known truth was available, so nothing here has been checked against a right answer. The numbers below for the Utah FORGE density model are supported by the data.
Gates
Check
Result
PASS
licensing
5.333 (< 95.0 pct) observation must be plausible under the prior
PASS
calibration(MC)
0.02305 (median < 0.115 (4x the 600-sample floor)) exact functional sd vs sampled
PASS
stability
0 (< 1e-09) sd ratio 1.001
PASS
adequacy
0.5409 (0.253 < pp < 2.28) the only gate that tests reality
What the data supports
excess mass, 2x2 km box above the blind depth
-346 Mt (95%: -484 to -207 Mt)
is that a detection?
YES — the interval excludes zero
the model is INVENTED below
1,322 m elevation (1,044 m below the top of the mesh)
cells the survey actually informs (>5%)
29.9% of 10,093
best-informed cell
51% of its prior uncertainty removed
What the standard workflow would have told you
error bar from a regularizer tuned to a misfit target (implied prior 0.021 g/cc)
+/- 27 Mt
error bar from a prior declared as rock (0.25 g/cc, correlated over 500 m)
+/- 71 Mt
over-confidence factor
2.6x
The weight that makes an inversion look right is not a statement about rock. Read as a prior it claims density varies by 0.021 g/cc — which no geologist would sign — and it fails the licensing gate at the 100th percentile: that prior could not have produced the anomaly that was measured.
The declared prior
marginal density sd
0.25 g/cc (the published report's own basin contrast)
correlation length
500 m (the scale the structure varies on)
regional field
order-1 surrogate for the undelivered ~50 km padding cells
prior-predictive vs observed
2.83 mGal vs 2.03 mGal
Every number here was read off the site or the customer's own report before any gate was run. Widening a prior until a gate turns green passes every gate and means nothing.
Where the model is real
Red dashed line: below this the survey no longer constrains density and the model is the regularizer, not the rock.
