Weak-lensing convergence of a cluster-scale dark-matter halo
NOT CLAIMED — failed: recovery. Numbers below are DIAGNOSTIC ONLY and must not be used to support a decision about this dark-matter map.
Gates
Check
Result
PASS
licensing
0 (< 95.0 pct) observation must be plausible under the prior
PASS
calibration(MC)
0.01027 (median rel err < 2%) closed form vs 8k Matheron samples
PASS
stability
0 (< 1e-06) sd ratio 1.000
PASS
adequacy
0.02062 (0.01 < pp < 0.09) the only gate that tests reality
FAIL
recovery
0.3819 (cover >= 0.90 and worst miss <= 4.10 sigma (n=576)) worst 4.9 sigma — the gate the other four cannot be
What the data supports
THE ONE NUMBER THAT SURVIVES — 95% exclusion
kappa < 0.042 where nothing is detected. Every bias measured here pushes the estimate DOWN, which makes an upper limit conservative rather than optimistic. This is the quantity a search publishes, and it is the same array a mining report calls drill risk
halo detected at
14.6 sigma (peak kappa 0.472 +/- 0.032) — the DETECTION is solid; the VALUE is not
pixels with a >2 sigma detection
68 of 576
gate 5 (recovery) FAILED — and that is why nothing above is a map
only 38% of pixels have the truth inside their 95% interval, worst miss 4.9 sigma
peak value, DIAGNOSTIC ONLY
0.472 against a truth of 0.597 (21% low). Widening the prior does NOT fix this: at a prior of 1.0 the peak still returns 0.476. It is the lensing operator smoothing a cusp it cannot resolve, not shrinkage
aperture mass, DIAGNOSTIC ONLY
11.66 +/- 1.39 against a truth of 19.86, missing by 5.9 sigma. THIS one is shrinkage, and a prior of 1.0 would cut the miss to 1.8 sigma — at the price of degrading the exclusion limit 2.2x. Not taken
The same code, at 10^22 times the length scale
posterior engine
src/posterior.py — written for Utah gravity stations, run here with zero modifications
why that works
prisms->gravity and mass-sheet->shear are the same linear potential-field operator
the customer's number
aperture mass here, tonnes-in-the-block there — one function, src/product/verdict.py
the physicist's number
an exclusion limit here, a drill-risk interval there — the same posterior sigma
This is the premise the project was founded on, tested rather than asserted. It could have failed; the honest-negative clause was written before the run. It passed.
The hole this run found, and the gate that now catches it
core gates passed
4 of 4
gate 5, recovery
FAILED — 38% coverage, worst miss 4.9 sigma
aperture mass vs known truth
misses by 5.9 sigma — interval does not cover
true peak against the declared prior
0.597 is a 2.0-sigma excursion under a prior of 0.3
what each gate was actually testing
licensing: is the DATA plausible. calibration: is the posterior self-consistent for truths drawn FROM the prior. stability: do two runs agree. adequacy: does the answer reproduce the data. None asks whether a REAL truth lands inside the interval
This is the same species of blind spot that created gate 4, found the same way — by checking against a truth we happened to know. The exclusion limit survives it because shrinkage makes an upper bound conservative. A mass estimate does not, so it is not claimed here. Open question, deliberately left open: whether the suite needs a fifth gate for coverage at a fixed realistic truth, or whether the prior class must be declared wide enough to contain the object being searched for.
What the standard method gives you instead
Kaiser-Squires inversion
a convergence map
its error bars
none
its statement about what was NOT seen
none
The second panel of the figure is the industry standard. It is a picture. The fourth panel is what a search actually needs: the most mass that could be hiding without this survey noticing.
Where the model is real
Panel 4 is the deliverable. In a mining report the same array is the drill-risk map; here it is an exclusion limit.
