Final test replay
Held-out test · Curiosity Navcam · sols 950–979
12 traverses, 679 archived frames. POSITION keeps 1/4 of the frames of each traverse at full quality and sends the rest as thumbnail pairs, through the stereo-safe Scheduler V3. The pre-registered criterion (Scheduler V3 + POSITION at 1/4 retention) was PASS.
HISTORICAL REPLAYCuriosity Navcam · sols 950–979 · held-out testarchived (downlinked) frames only — not the rover’s full onboard stream
PLACES landing-frame positions relative to the first frame (m) · sol 967 · 105 frames · path 89 m
Bandwidth meter
3138 kB sent · 25.8% of SEND ALL for this traverse
full-quality stereo pairs 27 · thumbnail pairs 78 · frame 105/105
So far in the replay
5 m coverage (frames seen)1.000
stereo pairs sent whole27
stereo pairs broken0
Stored final-test result for this traverse
bytes / SEND ALL0.258
5 m coverage1.000
largest distance to kept2.05 m
frames kept27 / 105
broken stereo pairs0
source · artifacts/phase3_final/20260926T174018-phase3-final-test-e4e2/results.json → traverse.rows (967:trav00327, 0.25, METADATA_POSITION)
Hollow circles are archived frames sent only as a thumbnail pair; filled squares are kept at full quality. Shaded discs show the 5 m radius around kept frames; thin lines join each thumbnail-only frame to its nearest kept frame — the distance the headline metric bounds.
All frozen strategies on the held-out test
SEND ALL vs the four selection methods
Visual-change coverage is shown for context only; it is not part of the primary criterion. 1/2 and 1/8 are context; 1/4 is the pre-registered operating point.
| retention | method | bytes | 5 m coverage | largest dist. to kept | p95 gap | max gap | visual change | stereo kept / broken |
|---|---|---|---|---|---|---|---|---|
| all | SEND ALL | 1.000 | 1.000 | 0.00 m | 1.21 m | 19.7 m | 1.000 | 679 / 0 |
| 1/2 | EVERY Nth | 0.505 | 1.000 | 2.46 m | 2.21 m | 20.2 m | 0.976 | 342 / 0 |
| 1/2 | UNIFORM DISTANCE | 0.507 | 1.000 | 1.74 m | 2.92 m | 19.7 m | 0.973 | 342 / 0 |
| 1/2 | POSITION | 0.514 | 1.000 | 1.00 m | 2.77 m | 20.2 m | 0.972 | 342 / 0 |
| 1/2 | POSITION + EMBEDDING | 0.501 | 1.000 | 4.31 m | 3.61 m | 20.8 m | 0.980 | 342 / 0 |
| 1/4 | EVERY Nth | 0.261 | 0.996 | 20.04 m | 4.10 m | 39.9 m | 0.954 | 174 / 0 |
| 1/4 | UNIFORM DISTANCE | 0.260 | 1.000 | 3.03 m | 5.89 m | 19.7 m | 0.950 | 174 / 0 |
| 1/4 | POSITION (primary) | 0.261 | 1.000 | 2.05 m | 4.86 m | 20.2 m | 0.950 | 174 / 0 |
| 1/4 | POSITION + EMBEDDING | 0.257 | 1.000 | 4.89 m | 7.77 m | 20.8 m | 0.956 | 174 / 0 |
| 1/8 | EVERY Nth | 0.136 | 0.983 | 20.61 m | 9.02 m | 41.8 m | 0.928 | 89 / 0 |
| 1/8 | UNIFORM DISTANCE | 0.136 | 0.974 | 6.53 m | 11.95 m | 20.2 m | 0.925 | 89 / 0 |
| 1/8 | POSITION | 0.138 | 0.983 | 6.53 m | 11.12 m | 20.2 m | 0.925 | 89 / 0 |
| 1/8 | POSITION + EMBEDDING | 0.135 | 0.947 | 7.35 m | 13.95 m | 22.9 m | 0.926 | 89 / 0 |
source · artifacts/phase3_final/20260926T174018-phase3-final-test-e4e2/results.json → traverse.summary
Figures
Publication figures (drawn from the frozen artifacts)
