Model report

Experimental learned terrain similarity

The physics fingerprint drives the rankings; our encoder runs alongside it in your browser.

Predicted Estimated by a fitted statistical modelExperimental learned view; the Derived physics fingerprint ranks every candidate.

Encoder architecture from input window at the bottom to the training loss at the top: a stride 2 convolution stem, a repeated BasicBlock (two per stage, four stages, widths 32 to 256) with a residual skip, global average pooling to a 256-value embedding that is L2 normalised for retrieval and exported to ONNX, and three training heads: a projector with the VICReg loss, a physics head with a masked loss and a body head behind gradient reversal. — Encoder architecturepaper Figure 2, drawn from the model code

Parameters

2,795,040

from model/model_card.json

Training windows

19,393

earth 11,147 · mars 1,398 · moon 6,848

Embedding

256-d

L2-normalised, from the ONNX export record

Browser parity

0.999995

min cosine WASM vs Python over 256 windows (threshold 0.999)

Training curves

Each loss term over training: mean over seeds with the min to max range shaded.

totalepochs 0–11 · 28.1 to 41.6

Values as a table

Loss term total: mean over seeds with min to max band, 12 epochs
Epoch	Mean	Min	Max	Seeds
0	41.50677826058151	41.44047653980744	41.62139887687488	3
1	37.25583829635229	36.7406185345772	37.62504592308631	3
2	33.27784878983457	32.95392515720465	33.49444496937287	3
3	31.494304991175984	31.30400144136869	31.62047674717047	3
4	30.415230245671722	30.30880448757074	30.47035046112843	3
5	29.715602980719677	29.6430841592642	29.761225260221043	3
6	29.17697907309247	29.13440024547088	29.261907063997707	3
7	28.78077439365224	28.741534599891075	28.856770026378143	3
8	28.488055082467884	28.485075388199245	28.489633755806164	3
9	28.305782831632172	28.257693877586952	28.356210708618164	3
10	28.180835740179077	28.156184832255047	28.219304353762897	3
11	28.143654733641537	28.1307976551545	28.151172320048016	3

vicregepochs 0–11 · 27.6 to 39.7

Values as a table

Loss term vicreg: mean over seeds with min to max band, 12 epochs
Epoch	Mean	Min	Max	Seeds
0	39.65257233839768	39.62533730726976	39.67959961524377	3
1	36.409063714182274	35.949581244052986	36.75350179427709	3
2	32.63453228453286	32.337810565263794	32.855380767431015	3
3	30.89777957068549	30.730738297486916	31.01900535974747	3
4	29.84588995550433	29.752389883383728	29.912920218247635	3
5	29.16318189180814	29.104502408932415	29.194803678072414	3
6	28.642725545116978	28.609156486315605	28.70490683042086	3
7	28.25140585451045	28.22610114170955	28.29869933006091	3
8	27.958112496596115	27.945543851607884	27.96784811753493	3
9	27.780924601432606	27.743316039060936	27.82043009537917	3
10	27.657077862666203	27.63995045882005	27.675695786109337	3
11	27.62012373280321	27.6106290328197	27.62895596333039	3

vicreg_invepochs 0–11 · 0.0119 to 0.186

Values as a table

Loss term vicreg_inv: mean over seeds with min to max band, 12 epochs
Epoch	Mean	Min	Max	Seeds
0	0.1784952204578962	0.1715787758047764	0.18590677701509917	3
1	0.10671754659782362	0.08925985891180925	0.12066783832433896	3
2	0.02666985594595854	0.025053596673294518	0.028417390651809864	3
3	0.019028773670618098	0.018547520781747807	0.019879897065365162	3
4	0.015948966227105666	0.015298359155750427	0.016766380936576005	3
5	0.014622388899119362	0.01401021501694161	0.015112229246789446	3
6	0.014160812227453431	0.013901928093475409	0.01438994805376308	3
7	0.01347178340746233	0.01323033857326477	0.013628225958643433	3
8	0.012937980552769115	0.012776483930886174	0.013228832594811534	3
9	0.012943981858129555	0.012499243206081864	0.01337014901666687	3
10	0.012592534363492688	0.012478726086182853	0.012658249646520767	3
11	0.012440709088745916	0.011869948673754549	0.01281736372038722	3

vicreg_varepochs 0–11 · 0.822 to 1.33

Values as a table

Loss term vicreg_var: mean over seeds with min to max band, 12 epochs
Epoch	Mean	Min	Max	Seeds
0	1.3263970409703052	1.3244435038322058	1.3274926864183867	3
1	1.2259266697443447	1.214719859453348	1.234089528902983	3
2	1.0815202512292779	1.068138145483457	1.0905846097530463	3
3	0.9935751032625508	0.98530068840736	0.998774610268764	3
4	0.9388462371296353	0.9337071806956561	0.941713062616495	3
5	0.902684724993176	0.9004231080030783	0.9038609633078942	3
6	0.8751115452530037	0.8731373938230368	0.8784947540515509	3
7	0.8550276802136348	0.8530450302820939	0.8582496719482617	3
8	0.8388182199918307	0.8383070757755866	0.8396650262368031	3
9	0.8298266642114037	0.8282818694909414	0.8311060957419567	3
10	0.8245585099754171	0.8230498601228763	0.8267397360923963	3
11	0.8228080570697784	0.8220572838416467	0.8238147115096067	3

vicreg_covepochs 0–11 · 1.87 to 6.76

Values as a table

Loss term vicreg_cov: mean over seeds with min to max band, 12 epochs
Epoch	Mean	Min	Max	Seeds
0	2.0302659082107053	1.8665804542027986	2.20875722857622	3
1	3.0929584141470428	2.8845678048256116	3.350088445039896	3
2	4.929779708894909	4.88033094161596	5.008016995894603	3
3	5.582682619746933	5.552642657206609	5.6317486090537825	3
4	5.976009809053861	5.934979090323815	6.0272514575567	3
5	6.2305040257608795	6.220473729647123	6.243669485434507	3
6	6.410916497564724	6.387773795005603	6.424002775779138	3
7	6.538919247113742	6.5116990223909035	6.562585806235289	3
8	6.664207425891843	6.6527590445983105	6.683052900509956	3
9	6.711658565407125	6.702663323818109	6.723788322546543	3
10	6.728301843007405	6.695234329272539	6.747688489082532	3
11	6.738904590280647	6.718512535095215	6.757089883853228	3

physepochs 0–11 · 0.0865 to 0.946

Values as a table

Loss term phys: mean over seeds with min to max band, 12 epochs
Epoch	Mean	Min	Max	Seeds
0	0.869710285948892	0.8258125663567812	0.9459811777640612	3
1	0.20477008109546116	0.17789760594948745	0.22081230045893255	3
2	0.1206177442183352	0.11779912035816754	0.12203100209052746	3
3	0.10901405592250009	0.10619649472527015	0.1136465766108953	3
4	0.10250733542837139	0.10070241443239726	0.10427695121138524	3
5	0.09936759783289371	0.09712942278920075	0.10066380275365634	3
6	0.09373498397568862	0.09263093005388211	0.09477404419046181	3
7	0.09039259509334707	0.08708286786881778	0.09328079940034793	3
8	0.09329881633703525	0.09070055248836677	0.09462127810678421	3
9	0.09105554695885915	0.08994383135667214	0.0916542519743626	3
10	0.09022078892359366	0.08814050524662702	0.09321166226306023	3
11	0.08926823615836792	0.08651914963355431	0.09120351945360501	3

bodyepochs 0–11 · 0.418 to 1.02

Values as a table

Loss term body: mean over seeds with min to max band, 12 epochs
Epoch	Mean	Min	Max	Seeds
0	0.9844954849308373	0.9415221710999807	1.022637663743435	3
1	0.6420045226302921	0.6131397153322513	0.6621422721789434	3
2	0.5226987591411313	0.4983153549524454	0.5527476863219187	3
3	0.4875113047086275	0.4670665623285832	0.5076425014398037	3
4	0.4668328944179747	0.45387195088924503	0.4898989960933343	3
5	0.45305341112817454	0.44145222619557994	0.4706761210392683	3
6	0.440518502623607	0.4267166256904602	0.4622260760038327	3
7	0.4389759675305114	0.42503454211430675	0.4647899506947933	3
8	0.43664378437221557	0.42955528161464596	0.4492908796438804	3
9	0.43380260429321194	0.4228092271548051	0.44583668731726134	3
10	0.43353707107723266	0.418219003539819	0.45546798408031464	3
11	0.43426272464104193	0.42349236248395383	0.45046113507869917	3

Training run details: compute, invariance trade-off and probe metrics

Compute hours

Total: 0

GPU hours per run
Run	Hours	Source
lambda_0_seed0	0	measured: 0 GPU hours, CPU only; 3.203 CPU core-hours (threads x wall-clock), measured
lambda_1_seed0	0	measured: 0 GPU hours, CPU only; 1.9259 CPU core-hours (threads x wall-clock), measured
main_seed0	0	measured: 0 GPU hours, CPU only; 3.1872 CPU core-hours (threads x wall-clock), measured
main_seed1	0	measured: 0 GPU hours, CPU only; 3.1973 CPU core-hours (threads x wall-clock), measured
main_seed2	0	measured: 0 GPU hours, CPU only; 3.1962 CPU core-hours (threads x wall-clock), measured
no_phys_head_seed0	0	measured: 0 GPU hours, CPU only; 3.1937 CPU core-hours (threads x wall-clock), measured

Invariance trade-off over lambda_max

Chance level of the body probe: 0.3333333333333333.

Body-probe balanced accuracy against physics R² per lambda_max, 95% intervals
lambda_max	Body-probe balanced accuracy	Physics R²
0	Predicted Estimated by a fitted statistical model 0.7208404140536292 [0.6178125240816351, 0.8492146627553249]	Predicted Estimated by a fitted statistical model 0.9674416496488536 [0.9465855359285612, 0.9688704458371695]
0.1	Unavailable Observation or supported inference is absent no completed run with this lambda_max was supplied Source: report/curves.json
0.3	Predicted Estimated by a fitted statistical model 0.707647609232532 [0.5835224565110707, 0.8468193365903072]	Predicted Estimated by a fitted statistical model 0.965846356486989 [0.9436878161361877, 0.9673803703117905]
1	Predicted Estimated by a fitted statistical model 0.7187629658158302 [0.5950960435287735, 0.8477212966783378]	Predicted Estimated by a fitted statistical model 0.9678879902023947 [0.9411949126871846, 0.9700786100638019]

Validation probe metrics

Unavailable per-epoch validation probe metrics were not computed during the reduced-scope CPU training; the probes ran once after training (evaluate.py), and the validation-split evaluation chose probe penalties on the same rows, so it is a dry run and is not reported here Source: report/curves.json

Retrieval examples

Nearest neighbours from the experimental learned encoder (Predicted), shown beside the physics-fingerprint top 5 (Derived) that drives the rankings.

moon-sps-x2-y2

Learned encoder neighbours (experimental)

1. earth-n34-00-w119-00-5004-00497Predicted Estimated by a fitted statistical modelmodel_embedding: 0.361
2. earth-n34-00-w119-00-5004-00446Predicted Estimated by a fitted statistical modelmodel_embedding: 0.425
3. earth-n34-00-w119-00-5004-00168Predicted Estimated by a fitted statistical modelmodel_embedding: 0.430
4. earth-n34-00-w119-00-5004-00502Predicted Estimated by a fitted statistical modelmodel_embedding: 0.449
5. earth-n34-00-w119-00-5004-00409Predicted Estimated by a fitted statistical modelmodel_embedding: 0.452

Physics-fingerprint rankings are shown on the evidence cards.

Model card

Show the model card fields

model_id: terraluna-encoder-experimental
version: 0.1.0-experimental
architecture: ResNet-18 layout (BasicBlock, [2, 2, 2, 2]) at half width (32, 64, 128, 256); 3 x 3 stem at stride 2 (a CPU-scope deviation from the stride-1 stem of PRD 11.3); global average pool to a 256-d embedding, L2-normalised for retrieval
parameter_count: 2795040
training bodies: moon, mars, earth
training products: copernicus_glo30:Copernicus_DSM_COG_10_N34_00_W115_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N34_00_W116_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N34_00_W119_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N35_00_W115_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N35_00_W116_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N36_00_W117_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N36_00_W118_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N37_00_W117_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N37_00_W118_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N38_00_W111_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N38_00_W112_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N42_00_W113_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N42_00_W114_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N43_00_W113_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N43_00_W114_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N64_00_W016_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N64_00_W017_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N64_00_W018_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N64_00_W019_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N65_00_W016_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N65_00_W017_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N65_00_W018_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N65_00_W019_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_N75_00_W091_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S23_00_W069_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S23_00_W070_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S23_00_W071_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S24_00_W069_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S24_00_W070_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S24_00_W071_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S25_00_W069_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S25_00_W070_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S25_00_W071_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S26_00_W069_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S26_00_W070_00_DEM, copernicus_glo30:Copernicus_DSM_COG_10_S26_00_W071_00_DEM, ctx_dtm:B18_016650_2206_XN_40N174W__D21_035427_2205_XN_40N174W, ctx_dtm:D04_028795_2302_XN_50N185W__F03_037155_2303_XN_50N185W, ctx_dtm:D15_033146_2247_XN_44N176W__D15_033225_2247_XN_44N176W, ctx_dtm:D16_033291_2253_XN_45N174W__D18_034359_2249_XN_44N174W, ctx_dtm:D18_034187_2276_XN_47N156W__D19_034622_2276_XN_47N156W, ctx_dtm:D22_035981_2253_XN_45N176W__F01_036337_2234_XN_43N175W, ctx_dtm:F18_042930_2210_XN_41N158W__F19_043220_2210_XN_41N158W, ctx_dtm:G03_019590_2280_XI_48N163W__G05_020078_2267_XN_46N163W, ctx_dtm:G20_025906_2242_XN_44N155W__F19_043299_2242_XN_44N155W, lola_pgda90:LDEM_80S_20MPP_ADJ, pgda78:Site01_final_adj_5mpp_surf
window counts per body and split: earth: test 244, train 11147, val 2760; mars: test 726, train 1398, val 364; moon: test 2937, train 6848, val 2418
split grouping: earth: spatial blocks (eval_block); train/val/test by block; mars: spatial blocks (eval_block); train/val/test by block; moon: spatial blocks (eval_block); train/val/test by block
seeds: 0, 1, 2
intended_use: A second opinion only. The encoder embeds 5 km windows at 78 m sampling of Moon, Mars and Earth terrain into 256 numbers and offers a third, always Predicted, distance family beside the deterministic terrain fingerprint. It is EXPERIMENTAL, stays outside every ranking until gate G5 passes, and never replaces the fingerprint on any card.
limitations: Every encoder statistic is Predicted, a model output and not an observation. The model makes no claim about radar, trafficability, regolith mechanics or ice, and a close embedding is not evidence of an analog. It was evaluated only on 5 km windows at 78 m sampling from spatially held-out blocks; results do not transfer to other scales. The morphology probe is Unavailable because no crater labels exist in this dataset. Frozen ImageNet ResNet-18 and DINO ViT-S/8 baselines are Unavailable until their weights are fetched. Windows overlapping a target footprint are excluded by the dataset build. Training ran at a reduced CPU scope (no GPU) that departs from the PRD training recipe in data volume, epochs, batch size, precision and stem stride; each departure is listed per run in report/curves.json.
licence: Apache-2.0
onnx_sha256: 931bfc36393a35ca76a61b84a2681057a055043845bb5d254200c45516e47f21

Metrics with 95% intervals

Show all 56 test metrics (Predicted, 95% intervals)

Test metrics on the test split, model terraluna-encoder-experimental, with 95% intervals
Test	Measure	Target and body	Value (95% interval)	Baselines
physics_probe	r2	log_nu_relief_270; pooled	Predicted Estimated by a fitted statistical model 0.965846 [0.943688, 0.96738] n_blocks 12, 1000 bootstrap draws	random_init: 0.969199 [0.936641, 0.971953]; paired difference [-0.0109164, 0.00820212] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_relief_270; moon	Predicted Estimated by a fitted statistical model 0.943503 [0.934431, 0.951106] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_270; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_relief_270 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_relief_270; earth	Predicted Estimated by a fitted statistical model 0.95379 [0.609401, 0.962849] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_360; pooled	Predicted Estimated by a fitted statistical model 0.959983 [0.938973, 0.964645] n_blocks 12, 1000 bootstrap draws	random_init: 0.968783 [0.941648, 0.970507]; paired difference [-0.0210346, 0.00665334] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_relief_360; moon	Predicted Estimated by a fitted statistical model 0.945743 [0.937023, 0.952443] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_360; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_relief_360 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_relief_360; earth	Predicted Estimated by a fitted statistical model 0.950773 [0.612315, 0.9593] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_540; pooled	Predicted Estimated by a fitted statistical model 0.945982 [0.907822, 0.965034] n_blocks 12, 1000 bootstrap draws	random_init: 0.966119 [0.944653, 0.967751]; paired difference [-0.0428541, 0.00177529] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_relief_540; moon	Predicted Estimated by a fitted statistical model 0.944163 [0.935249, 0.94972] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_540; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_relief_540 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_relief_540; earth	Predicted Estimated by a fitted statistical model 0.951836 [0.569463, 0.961472] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_720; pooled	Predicted Estimated by a fitted statistical model 0.938313 [0.893867, 0.96064] n_blocks 12, 1000 bootstrap draws	random_init: 0.963343 [0.944043, 0.965684]; paired difference [-0.0505335, -0.00184201] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_relief_720; moon	Predicted Estimated by a fitted statistical model 0.93989 [0.930367, 0.945316] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_720; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_relief_720 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_relief_720; earth	Predicted Estimated by a fitted statistical model 0.942241 [0.550788, 0.952456] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_1080; pooled	Predicted Estimated by a fitted statistical model 0.926277 [0.873882, 0.950315] n_blocks 12, 1000 bootstrap draws	random_init: 0.957374 [0.93379, 0.961308]; paired difference [-0.0597957, -0.00680009] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_relief_1080; moon	Predicted Estimated by a fitted statistical model 0.930006 [0.919753, 0.935383] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_relief_1080; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_relief_1080 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_relief_1080; earth	Predicted Estimated by a fitted statistical model 0.918372 [0.485118, 0.930584] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_270; pooled	Predicted Estimated by a fitted statistical model 0.937174 [0.828408, 0.953023] n_blocks 12, 1000 bootstrap draws	random_init: 0.943751 [0.883242, 0.950489]; paired difference [-0.0549836, 0.0118984] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_texture_270; moon	Predicted Estimated by a fitted statistical model 0.827045 [0.809421, 0.83805] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_270; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_texture_270 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_texture_270; earth	Predicted Estimated by a fitted statistical model 0.94176 [0.561774, 0.949913] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_360; pooled	Predicted Estimated by a fitted statistical model 0.952404 [0.862756, 0.964064] n_blocks 12, 1000 bootstrap draws	random_init: 0.956487 [0.905069, 0.961212]; paired difference [-0.0425074, 0.00853636] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_texture_360; moon	Predicted Estimated by a fitted statistical model 0.862123 [0.844963, 0.872813] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_360; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_texture_360 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_texture_360; earth	Predicted Estimated by a fitted statistical model 0.949407 [0.567925, 0.956899] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_540; pooled	Predicted Estimated by a fitted statistical model 0.966228 [0.902388, 0.973783] n_blocks 12, 1000 bootstrap draws	random_init: 0.968439 [0.925708, 0.972954]; paired difference [-0.0245616, 0.00408824] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_texture_540; moon	Predicted Estimated by a fitted statistical model 0.902184 [0.887039, 0.911745] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_540; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_texture_540 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_texture_540; earth	Predicted Estimated by a fitted statistical model 0.958797 [0.584018, 0.965666] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_720; pooled	Predicted Estimated by a fitted statistical model 0.973157 [0.924764, 0.978742] n_blocks 12, 1000 bootstrap draws	random_init: 0.974156 [0.937898, 0.978118]; paired difference [-0.0155115, 0.0027782] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_texture_720; moon	Predicted Estimated by a fitted statistical model 0.924351 [0.910449, 0.932937] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_720; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_texture_720 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_texture_720; earth	Predicted Estimated by a fitted statistical model 0.964786 [0.614058, 0.971132] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_1080; pooled	Predicted Estimated by a fitted statistical model 0.98011 [0.94749, 0.983841] n_blocks 12, 1000 bootstrap draws	random_init: 0.97974 [0.951655, 0.983197]; paired difference [-0.00571416, 0.00264048] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	log_nu_texture_1080; moon	Predicted Estimated by a fitted statistical model 0.947914 [0.936764, 0.954787] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	log_nu_texture_1080; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for log_nu_texture_1080 Source: report/metrics.json	none recorded
physics_probe	r2	log_nu_texture_1080; earth	Predicted Estimated by a fitted statistical model 0.970264 [0.67693, 0.975715] n_blocks 2, 1000 bootstrap draws	none recorded
physics_probe	r2	H_relief; pooled	Predicted Estimated by a fitted statistical model 0.710049 [0.446142, 0.727835] n_blocks 12, 1000 bootstrap draws	random_init: 0.813307 [0.663616, 0.824645]; paired difference [-0.220365, -0.0951481] imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched
physics_probe	r2	H_relief; moon	Predicted Estimated by a fitted statistical model 0.553843 [0.359227, 0.669016] n_blocks 9, 1000 bootstrap draws	none recorded
physics_probe	r2	H_relief; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial test blocks for H_relief Source: report/metrics.json	none recorded
physics_probe	r2	H_relief; earth	Predicted Estimated by a fitted statistical model -0.0392614 [-0.104025, 0.176036] n_blocks 2, 1000 bootstrap draws	none recorded
morphology_probe	r2_gain_over_fingerprint_ridge	no target; no body	Unavailable Observation or supported inference is absent crater labels do not exist in this dataset: no Robbins catalogue is ingested, so no log crater density can be formed for Moon or Mars windows Source: report/metrics.json	imagenet_resnet18: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched dino_vits8: Unavailable Observation or supported inference is absent LIC-01: ImageNet ResNet-18 / DINO ViT-S/8 weights not fetched fingerprint_only_ridge: Unavailable Observation or supported inference is absent crater labels do not exist in this dataset: no Robbins catalogue is ingested, so no log crater density can be formed for Moon or Mars windows
body_invariance	balanced_accuracy	no target; pooled	Predicted Estimated by a fitted statistical model 0.707648 [0.583522, 0.846819] n_blocks 12, 1000 bootstrap draws	lambda_0: 0.72084 [0.617813, 0.849215] random_init: 0.865778 [0.782785, 0.90852] chance: 0.333333
physics_agreement	spearman_rho	rms_log_nu_10; pooled	Predicted Estimated by a fitted statistical model 0.197789 [0.0764087, 0.361369] n_blocks 12, 1000 bootstrap draws	random_init: 0.847421 [0.804733, 0.87884]
physics_agreement	spearman_rho	rms_log_nu_relief_5; pooled	Predicted Estimated by a fitted statistical model 0.189818 [0.0678233, 0.354569] n_blocks 12, 1000 bootstrap draws	random_init: 0.824605 [0.789971, 0.851198]
physics_agreement	spearman_rho	rms_log_nu_texture_5; pooled	Predicted Estimated by a fitted statistical model 0.202733 [0.0907371, 0.346276] n_blocks 12, 1000 bootstrap draws	random_init: 0.817345 [0.771654, 0.84553]
target_retrieval	jaccard_at_20	no target; earth	Unavailable Observation or supported inference is absent the deterministic M0 Earth ranking of T4 is not an input of this build Source: report/metrics.json	random_ranking: Unavailable Observation or supported inference is absent the deterministic M0 Earth ranking of T4 is not an input of this build
seed_stability	jaccard_at_20_mean	earth_top20; earth	Predicted Estimated by a fitted statistical model 0.318326 [0.318326, 0.370264] n_blocks 2, 1000 bootstrap draws	random_ranking: 0.0437393
held_out_body	r2	no target; mars	Unavailable Observation or supported inference is absent no Moon+Earth-only run was supplied Source: report/metrics.json	none recorded
out_of_distribution	fraction_above_train_p99	mahalanobis; pooled	Predicted Estimated by a fitted statistical model 0.0432557 [0.00373721, 0.101037] n_blocks 12, 1000 bootstrap draws	training_expectation: 0.01
out_of_distribution	fraction_above_train_p99	mahalanobis; moon	Predicted Estimated by a fitted statistical model 0.0102145 [0.00225659, 0.025913] n_blocks 9, 1000 bootstrap draws	training_expectation: 0.01
out_of_distribution	fraction_above_train_p99	mahalanobis; mars	Unavailable Observation or supported inference is absent fewer than 2 spatial blocks Source: report/metrics.json	training_expectation: 0.01
out_of_distribution	fraction_above_train_p99	mahalanobis; earth	Predicted Estimated by a fitted statistical model 0.00819672 [0.00662252, 0.0107527] n_blocks 2, 1000 bootstrap draws	training_expectation: 0.01