Three confirmatory Martian volcanic plains, two cratered highlands, seven Earth candidates, 60 km windows, supports at or above MOLA’s effective resolution. The frozen rules returned one clear answer — and it retires the last standing explanation of the analog claim the project had already withdrawn.
A confirmatory result is only worth the regime it was measured in. Every one of the 15 confirmatory honest cells was verified in-regime before the verdicts were read.
| Check | Result |
|---|---|
| Comparability gate — honest cells leaving the honest regime | 0 of 15 |
| Supports used (all ≥ the 1389 m MOLA effective-resolution floor) | 1500 / 2000 / 3000 m |
| Footprint available per unit; extent ratio | 65.8–66.4 km · ≤ 1.01 |
| Confirmatory references per class | 3 volcanic plains · 2 highland |
| Reference contaminated by a ≥25 km IAU crater within 30 km | none |
Rule: at least 2 of the 3 confirmatory plains references must put a smooth_control candidate in Pareto layer 1 at every honest support.
| Reference | 1500 m | 2000 m | 3000 m | Outcome |
|---|---|---|---|---|
| mars_arcadia | TIE {llano_estacado, nullarbor} | RECOMMEND {llano_estacado} | RECOMMEND {llano_estacado} | holds |
| mars_cerberus | RECOMMEND {nullarbor} | RECOMMEND {nullarbor} | RECOMMEND {nullarbor} | holds |
| mars_solis | RECOMMEND {llano_estacado} | TIE {deccan, llano_estacado} | TIE {deccan, llano_estacado} | holds |
3 of 3 hold (2 required) → S SUPPORTED. A smooth non-volcanic terrain — Nullarbor limestone (σ ≈ 5 m) or the Llano Estacado sedimentary plain (σ ≈ 14 m) — sits on the Pareto front for every confirmatory Martian volcanic plain at every honest support. What v2 could only infer, v3 measures.
Rule: at least 2 of 3 plains references must have some basaltic candidate Pareto-dominating every smooth control, and at least 1 of the 2 highland references must not.
| Reference | Class | V outcome |
|---|---|---|
| mars_arcadia | volcanic plains | fails |
| mars_cerberus | volcanic plains | fails |
| mars_solis | volcanic plains | fails |
| mars_terra_sabaea | highland | holds |
| mars_terra_sirenum | highland | holds |
0 of 3 plains hold V — and V holds for every highland reference, so the rule's second condition fails as well. V NOT SUPPORTED. A basaltic lava plain is not a better analog for a Martian volcanic plain than a smooth limestone or sedimentary plain is.
Arcadia (σ 7 m) and Cerberus (σ 5 m) are matched by the two smoothest Earth candidates: Nullarbor (σ 5 m, limestone) and Llano Estacado (σ 14 m, sedimentary).
Terra Sabaea (σ 148 m) and Terra Sirenum (σ 256 m) are matched by Harrat Khaybar (σ 102 m) — a basaltic field. The basalts win where the terrain is rough, not where it is volcanic.
And the one smooth basalt behaves as a smooth candidate: Deccan Traps (σ 32 m, flood basalt) enters layer 1 for Solis Planum — the roughest of the three plains references — precisely when the window is large enough to average its relief. It competes because it is comparatively smooth, not because it is volcanic. Within this panel and lens, origin does no work that roughness does not already do.
The two references that carry S — Arcadia (σ 7 m) and Cerberus (σ 5 m) — sit at the smooth end of this ladder, exactly where the two non-volcanic controls sit. The basalts cluster higher, and they win only where the target is rough.
All 15 confirmatory honest cells were classified raw versus resolution-matched (every unit low-passed to the same 1389 m effective resolution).
| unchanged | stabilised | destabilised | reversed | membership changed | error |
|---|---|---|---|---|---|
| 11 | 0 | 1 | 0 | 3 | 0 |
No reversal → R SUPPORTED. This is where the v2 lesson paid off: v2's over-broad rule (“any change”) fired on a mere stabilisation. v3's classifier separates the cases, so mars_solis @1500 m loosening from RECOMMEND to TIE is recorded as destabilised, not as a reversal. Re-running S and V on the matched panel gives the same verdicts (differs: false).
Seven Earth candidates at 60 km, with the measured roughness that turns out to do the work. Classes were assigned from the literature before the run.
| Candidate | Class | 60 km σ (m) | Role |
|---|---|---|---|
| nullarbor | smooth control | 5 | limestone plain — non-volcanic |
| llano_estacado | smooth control | 14 | sedimentary high plain — non-volcanic |
| deccan_traps | basaltic | 32 | flood basalt |
| snake_river_plain | basaltic | 95 | basaltic plain (contains Craters of the Moon) |
| harrat_khaybar | basaltic | 102 | basaltic volcanic field |
| odadahraun | basaltic | 182 | basaltic lava plain, Iceland |
| nevada_basin | other | 562 | basin-and-range near-miss |
With MOLA + Copernicus GLO-30, real 60 km windows, honest supports and matched footprints, the Earth terrain that best matches a Martian volcanic plain is a smooth one — and at least as likely to be non-volcanic. The withdrawn Craters-of-the-Moon preference is explained by smoothness, and the explanation is now tested rather than asserted.
That no Earth terrain resembles a Martian plain — the morphometric similarity is real and measured. That basalts are irrelevant to Mars analog work — they matter where the target is rough. Or anything about traversability, geologic equivalence, site access, or mission readiness.
Falsifier for the next iteration: a pre-registered panel in which a smooth specifically volcanic terrain beats the smooth non-volcanic controls for Martian volcanic plains but not for highlands. Deccan Traps is the closest case here and it does not do that.
# the Earth candidates come from the public ESA Copernicus GLO-30 COG mosaic on AWS
python scripts/fetch_earth_candidates.py --config config/matched_footprint_v3.yml --source cog
python scripts/run_matched_benchmark.py --config config/matched_footprint_v3.yml --out artifacts/mf_panel_v3
The COG path is verified bit-identical to the OpenTopography COP30 product (maximum absolute difference 0.0 m). The Mars side is the single global MOLA 463 m GeoTIFF. Both need rasterio.