How Mova does at school
Mova's small language model, the student, goes through the lessons of its primary school one by one: it reads each lesson, learns from its exercises, and takes its tests and its exam — four options a question, chance about a quarter. Nothing here is answered by a large model: every choice is the student's own, and every choice says why.
Tests: 18472 of 49635 right (37.2%) · Exam: 5442 of 15723 right (34.6%) · By stars: plain 3291 of 7658 · ★ 11413 of 33041 · ★★ 2674 of 6077 · ★★★ 1094 of 2859 · Courses: 25 · Lessons: 3932 · Controls: all red in 25 of 25 courses · Run: 2026-10-18, commit f3c23ea72
The stars are the wit of a lesson: plain lessons ask what the lesson says, ★★★ is the olympiad level. The exam asks the same knowledge in stiff, formal, official language. A test the student did not understand — read into no frame that matched what it remembers — is a guess: it takes the first option shown.
The controls
Each course is run with three permanent controls beside the real student, and a score counts only when all three are red — below the real student's score. No lesson: a student of its own that reads no lesson, learns from the same exercises and takes the same tests; if it does as well, the score comes from the questions and options, not from the lessons. Questions shuffled: the real student answers each test's options under the next test's question; if that does as well, it is not reading the question. Key sentences removed: a student that reads each lesson without the sentence that holds each test's answer; if it does as well, the answers do not come from where the lesson says them. Two guessers set the bar beside them: one that always picks the longest option, one that always picks the first option shown (what the student does when it understands nothing).
The courses
The numbers
| course | tests right | plain | ★ | ★★ | ★★★ | exam | no lesson | shuffled | keys removed |
|---|---|---|---|---|---|---|---|---|---|
| Adult life | 246 of 689 (35.7%) | 53 of 132 | 106 of 310 | 76 of 214 | 11 of 33 | 83 of 248 | 175 of 689 | 197 of 689 | 199 of 689 |
| AI | 131 of 331 (39.6%) | 51 of 125 | 45 of 101 | 21 of 62 | 14 of 43 | 42 of 128 | 83 of 331 | 103 of 331 | 115 of 331 |
| Arts | 173 of 477 (36.3%) | 76 of 232 | 60 of 132 | 30 of 77 | 7 of 36 | 67 of 172 | 122 of 477 | 139 of 477 | 128 of 477 |
| Biology | 2238 of 4581 (48.9%) | 791 of 1647 | 695 of 1392 | 551 of 1175 | 201 of 367 | 639 of 1620 | 1208 of 4581 | 1517 of 4581 | 1501 of 4581 |
| Books | 7303 of 24316 (30.0%) | – | 6950 of 23231 | 192 of 572 | 161 of 513 | 2161 of 7451 | 6090 of 24316 | 6340 of 24316 | 6279 of 24316 |
| Chemistry | 200 of 392 (51.0%) | 73 of 126 | 73 of 149 | 36 of 74 | 18 of 43 | 63 of 148 | 101 of 392 | 125 of 392 | 118 of 392 |
| Chinese street | 139 of 353 (39.4%) | 53 of 120 | 44 of 132 | 33 of 56 | 9 of 45 | 44 of 128 | 92 of 353 | 95 of 353 | 109 of 353 |
| English | 202 of 666 (30.3%) | 77 of 235 | 60 of 200 | 31 of 121 | 34 of 110 | 88 of 240 | 173 of 666 | 196 of 666 | 195 of 666 |
| Euphemisms | 199 of 521 (38.2%) | 49 of 126 | 132 of 333 | 10 of 24 | 8 of 38 | 63 of 168 | 138 of 521 | 147 of 521 | 152 of 521 |
| Everyday | 3331 of 7211 (46.2%) | 559 of 1364 | 1889 of 3953 | 746 of 1552 | 137 of 342 | 884 of 1844 | 1829 of 7211 | 2493 of 7211 | 2589 of 7211 |
| French | 176 of 468 (37.6%) | 58 of 164 | 71 of 177 | 26 of 70 | 21 of 57 | 47 of 160 | 124 of 468 | 129 of 468 | 144 of 468 |
| Geography | 327 of 699 (46.8%) | 127 of 262 | 98 of 193 | 66 of 154 | 36 of 90 | 88 of 256 | 175 of 699 | 211 of 699 | 215 of 699 |
| History | 416 of 960 (43.3%) | 152 of 386 | 121 of 259 | 90 of 199 | 53 of 116 | 132 of 348 | 238 of 960 | 283 of 960 | 284 of 960 |
| Language | 221 of 602 (36.7%) | 86 of 216 | 54 of 176 | 46 of 119 | 35 of 91 | 79 of 220 | 155 of 602 | 193 of 602 | 162 of 602 |
| Medicine | 359 of 858 (41.8%) | 133 of 308 | 116 of 264 | 69 of 179 | 41 of 107 | 111 of 308 | 213 of 858 | 266 of 858 | 250 of 858 |
| Money | 175 of 469 (37.3%) | 75 of 178 | 46 of 134 | 31 of 88 | 23 of 69 | 54 of 168 | 122 of 469 | 151 of 469 | 132 of 469 |
| Myths | 175 of 389 (45.0%) | 22 of 51 | 69 of 155 | 64 of 129 | 20 of 54 | 50 of 148 | 101 of 389 | 114 of 389 | 124 of 389 |
| Physics | 241 of 483 (49.9%) | 79 of 163 | 68 of 142 | 59 of 98 | 35 of 80 | 61 of 176 | 123 of 483 | 157 of 483 | 158 of 483 |
| Politics | 201 of 475 (42.3%) | 72 of 179 | 58 of 132 | 42 of 92 | 29 of 72 | 63 of 168 | 123 of 475 | 160 of 475 | 166 of 475 |
| Reflection | 146 of 340 (42.9%) | 36 of 93 | 61 of 117 | 33 of 74 | 16 of 56 | 48 of 128 | 87 of 340 | 102 of 340 | 107 of 340 |
| Religion ethics | 492 of 1141 (43.1%) | 161 of 385 | 148 of 329 | 135 of 291 | 48 of 136 | 156 of 412 | 292 of 1141 | 381 of 1141 | 370 of 1141 |
| Space | 146 of 357 (40.9%) | 56 of 132 | 46 of 99 | 25 of 67 | 19 of 59 | 52 of 128 | 94 of 357 | 123 of 357 | 116 of 357 |
| Technology | 264 of 481 (54.9%) | 82 of 134 | 87 of 155 | 67 of 135 | 28 of 57 | 75 of 172 | 126 of 481 | 166 of 481 | 154 of 481 |
| Ukrainian street | 132 of 357 (37.0%) | 51 of 131 | 38 of 111 | 27 of 68 | 16 of 47 | 41 of 128 | 95 of 357 | 97 of 357 | 107 of 357 |
| World | 839 of 2019 (41.6%) | 319 of 769 | 278 of 665 | 168 of 387 | 74 of 198 | 251 of 656 | 537 of 2019 | 632 of 2019 | 629 of 2019 |
The runner is world knowledge course <course> --trace <file> in Mova's repository (world/src/knowledge.rs): the same run as without the trace, byte for byte, with every test written down. The courses' lessons are on the site too, sentence by sentence with their trees: the school in the corpora.
This page in Markdown → https://md.mova.elamur.ai/free/en/school-run/