Home / Mario / Level by level

Mario, level by level

One Cadence brain (cadence-net 0.91.1) plays the levels in order, learning the whole time. Two runs: a nearly blind brain (before), then a fresh brain with better senses (after). Updated 2026-10-31 06:48.

Learning fix: solo clears (updated 2026-10-31 06:50)

After the body-sense fix (LEARNING_FIX.md): one Cadence brain taught by a look-ahead coach's labels (DAgger), playing alone in every test. Start jitter = idle frames before the first button, which moves the enemies. SEL starts pick the checkpoint; TEST starts (32-41) and HELD (TEST plus starts 80-138, 50 in all) are never used for training or picking. Pre-registered bar: TEST 8/10. Stricter bar added later, because neighbouring starts behave alike: HELD 40/50.

Final numbers of the overnight run are on the showcase page. Best per level on 50 held-out starts: 1-1 50/50 (W1_F2ck110), 1-2 29/50, 1-3 45/50, 1-4 50/50 (each its own checkpoint). Best single brain for two levels (one_11_13, 10/30 20:55): 1-1 on 40 of 50 held-out starts and 1-3 on 42 of 50, the same weights for both. Its 1-3 TEST band alone is 6/10, so on the pre-registered 10-start bar it passes 1-1 (8/10) and not 1-3.

levelbrainSELTESTHELDstatus
1-1one_11_13 (st_13d ck_0013)15/248/1040/50consistent on both bars, same brain as the 1-3 row below
1-3one_11_13 (same brain)19/246/1042/50consistent on the stricter bar, not on the 10-start TEST band
1-1joint 1-1 + 1-3 brain (10/30 19:31)12/1710/1044/50consistent on both bars (88% of held-out starts)
1-11-1_steady (10/30)10/1010/1038/50consistent on the pre-registered bar, just short of the stricter one
1-1first clear brain (10/29)6/101/10not consistent: its "8 of 10" was on starts 1-12 only
1-2best (st_12cw ck_0011)1/176/108/50not consistent: clears sometimes, stuck for 3+ hours
1-3joint 1-1 + 1-3 brain (same brain)17/173/1016/50not consistent: the 17 practice starts it clears all share one frame phase (a decision is 4 frames); from the other three phases it stops at x 1,371. The picking starts now cover all phases.
1-41-4 brain (st_14b ck_0010, 10/30 20:46)17/248/1038/50consistent on the pre-registered bar, just short of the stricter one. Beats Bowser. The same checkpoint is only 23/50 on 1-1 and 0/50 on 1-3: not one brain for all levels yet.

What still goes wrong: the brain is fragile. Its fit to the coach's labels barely moves (about 0.9, critical moments about 0.99), but its solo results swing between 10/10 and 2/10 from one 250-lesson step to the next. A level is a chain of dozens of must-get-right moments. Averaging weights, voting across checkpoints, a smaller step, a keep-only-if-better ratchet and longer teaching did not steady it. Saving and testing every step did: the consistent 1-1 brain is one checkpoint that was kept. On 1-2 that is not enough: the brain fits the coach's 1-2 labels only to about 0.85, and even a backprop network predicts only about 85% of them on screens it has not seen, critical moments included. Tested and ruled out as the cause: the coach (it clears 1-2 from every start tried), the eyes (a sharper view around Mario added nothing), the step size, more teaching, a wider or deeper brain.

one_11_13 clears 1-3 alone from held-out start 90 (lifts and all). 15.8 s.
The same brain, same weights, clears 1-1 from held-out start 85. 22.1 s.
The 1-4 brain beats 1-4 alone from TEST start 32 (held-out): firebars, then over Bowser to the axe. Every button is the brain's. 13.5 s.
The joint brain clears 1-3 alone from TEST start 32 (held-out). The same brain, unchanged, also clears 1-1 (next clip). Every button is the brain's. 16.2 s.
The joint brain on 1-1 from TEST start 37. 19.9 s.
1-1_steady clears 1-1 alone from TEST start 35, a start it never trained or was picked on. Every button is the brain's. 19.3 s.

First sight

How far it gets on a level before any practice on it: median x of 3 solo attempts.

0500100015001-1 first sight 5941-1 first sight 40best baseline 4501-11-2 first sight 1831-2 first sight 196best baseline 3381-21-3 first sight 3111-3 first sight 321best baseline 7051-31-4 first sight 2341-4 first sight 234best baseline 5511-42-1 first sight 3222-1 first sight 324best baseline 3842-12-2 first sight 514best baseline 12342-2

blind brain new senses best baseline

What's happening

First sight vs baselines

levelblindsensesrunjumprandomfreshfresh+
1-1594 > pre40 ≤31531445030940
1-2183 ≤196 ≤19619633840177
1-3311 ≤321 ≤32131729970548
1-4234 ≤234 ≤234229331551233
2-1322 ≤324 ≤308306384120322
2-2514 ≤-1625141,2341,199514

Median x, same 3 starts as first sight. run = hold run right; jump = hold run+jump; random = random buttons (10 attempts, fixed seeds); fresh = an untrained Cadence brain with the blind input; fresh+ = an untrained brain with the new senses. > beats the best baseline, ≤ does not. The blind brain's 1-1 is "pre": it began from last night's 1-1 training.

After practice

levelblind bestclearedsenses bestcleared
1-1898no594no
1-2482no196no
1-3746no419no
1-4551no234no
2-1457no338stopped
2-21,496stopped--

Best solo median x over the practice rounds (about 38 minutes per level). Cleared = flag in 2 of 3 solo attempts.

Videos

Blind brain, its best first sight: 2-2 before any practice, x 514 (random buttons reach a median 1,234 here). Every button is the brain's. 9.3 s.
Blind brain. Retention test: back on 1-1 after practising 1-2 and 1-3, x 2,020 alone (overnight best was 1,783). Every button is the brain's. 14.2 s.
New senses, best first sight so far: 2-1 before any practice on it, x 324. Every button is the brain's. 3.0 s.

Learning-fix brains

Solo clips of the later brains, by file name. The Mario page has them with context and the brain roster.

1-4_U3ck9 on 1-4, start 90: flag. Every button is the brain's. 13.5 s.
U3ck83 on 1-1, start 33: flag. Every button is the brain's. 21.9 s.
U3ck83 on 1-3, start 33: flag. Every button is the brain's. 15.1 s.
U3ck83 on 1-4, start 33: flag. Every button is the brain's. 12.7 s.
fix_cad_body3, round 2 on 1-1, start 0: flag. Every button is the brain's. 21.8 s.
joint_11_13_ck8 on 1-1, start 37: flag. Every button is the brain's. 19.9 s.
joint_11_13_ck8 on 1-3, start 32: flag. Every button is the brain's. 16.2 s.
joint_11_14_ck10 on 1-4, start 32: flag. Every button is the brain's. 13.5 s.
one_11_13 on 1-1, start 85: flag. Every button is the brain's. 22.1 s.
one_11_13 on 1-3, start 90: flag. Every button is the brain's. 15.8 s.
U3ck28 on 1-4, start 120: flag. Every button is the brain's. 13.5 s.
1-1_steady on 1-1, start 35: flag. Every button is the brain's. 19.3 s.
W1_F2ck20 on 1-1, start 34: flag. Every button is the brain's. 21.6 s.
W1_F2ck20 on 1-2, start 34: flag. Every button is the brain's. 22.1 s.
W1_F2ck20 on 1-3, start 99: flag. Every button is the brain's. 16.4 s.
W1_F2ck20 on 1-4, start 34: flag. Every button is the brain's. 12.6 s.

Coaching used

New senses

levelcoachedlabelsownold playmin
1-1361,70412,9360%40
1-23057710,89412%40
1-3308368,76411%41
1-4246249,73111%40
2-11887210,53913%40

Blind

levelcoachedlabelsownold playmin
1-1543,37218,6320%40
1-2452,28216,47311%38
1-3702,49520,61713%39
1-4561,85822,69612%38
2-1302,18218,88211%40
2-2121,01314,47011%35

coached: tries where the brain plays and a look-ahead planner writes down the button it would have pressed on each screen; it never takes the controls. labels: buttons it wrote. own: screens from the brain's own attempts that made progress. old play: share of self-play spent on an earlier level. min: practice minutes.

Retention

New senses: solo again after leaving 1-4 (18:47), median x of 3.

levelnowits bestflags
1-13165940/3
1-22581960/3
1-34194190/3
1-42342340/3

Blind: solo again after leaving 2-1 (15:26), median x of 3.

levelnowits bestflags
1-17008980/3
1-23224820/3
1-34197460/3
1-44375510/3
2-14574570/3