Home › Code

Code Syntology

For each paper whose repository Syntology harvested, every function or class that could be lifted out on its own is one sample. Syntology executes samples on a synthesized fixture and records what happened, in six fixed states. A sample that ran executed on the fixture with no contract checked; when a contract drafted from the code itself was checked, it either ran · honoured contract or ran · violated contract; when the check could not be made, the sample still ran but the record reads ran · our draft was wrong or ran; the contract check could not run (the check raised an error, lacked a dependency, or timed out), failures of Syntology's instrument, not of the code; the rest are unverified, harvested but with no recorded run. A fingerprinted sample has outputs on a shared input battery; a pointer-only copy is one whose text Syntology does not serve, because no licence file was found, the licence was not identified, the licence is recorded as permissive but that copy's record is not marked cleared, or the licence is outside the permissive list Syntology serves text under (some licences outside it permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on that list). This site shows no code text at all: every sample links to its file in the repository. None of this is a claim about a paper's results, and nothing here is from the archive: this page is Syntology's layer alone.

Syntology harvests Python functions and classes from .py files; notebooks and other languages are not read, so a paper whose code is only in notebooks or in another language has no harvested samples here, which does not mean it has no code.

In numbers Syntology

Of the 294,260 distinct samples harvested for 51,887 papers: 211,169 ran (of which 22,388 constructed an object rather than computing a result; 147,384 with no instrument failure: 15,745 honoured, 5,703 violated, 125,936 with no contract checked; 63,785 where Syntology's instrument failed) · 83,091 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. The same code can sit in several papers' repositories, so those samples make 361,576 paper–sample links. One tile per status, counted once per sample:

15,745ran · honoured contract
5,703ran · violated contract
50,332ran · our draft was wrong
13,453ran; the contract check could not run
125,936ran
83,091unverified
58,064fingerprinted

Licence is a property of each paper's copy, so it is counted per link, not per sample: 142,682 of the 361,576 paper–sample links are pointer only, meaning Syntology does not serve that copy's text (no licence file found, a licence not identified, a permissive licence on a copy whose record is not marked cleared, a licence outside the permissive list Syntology serves text under, this paper's copy not located so its licence is not recorded, or a permissive licence on this paper's copy while the stored copy of the same code that Syntology serves is not cleared for its text); 17,699 of the 142,682 are that last case. This site shows no code text for any sample; paper pages and name pages link each sample to its file in the repository.

Per sample; the ran count is every status except unverified (the rule each paper page uses), so ran = with no instrument failure + where Syntology's instrument failed, with no instrument failure = honoured + violated + ran with no contract checked, the instrument figure = our draft was wrong + the contract check could not run, and ran + unverified = samples. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. Per-paper detail is on each paper page under "Code Syntology ran".

Every sample on these pages carries its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. Agents reach the samples through Syntology's MCP tools: get_harvested_code_for_paper(arxiv_id="…") lists a paper's samples and whether each ran, and get_code fetches one. How to connect.

Find a name

Matches by substring on the exact entry name, in your browser, against the 7,868 names that have a page (those in at least 5 papers). Names in fewer papers have no page and are not in this index; they still appear on their paper pages.

Names by number of papers Syntology

50 shown of 7,868 names with a page, ordered by the number of papers whose harvested code contains a function or class with that exact name. Same-named routines are not asserted to be equivalent; the name is a string, not an identity. A census tag links the name's census page, where Syntology ran implementations with that name on one shared input per bucket; it gives how many were compared, from how many papers, and how many distinct outputs they produced.

NamePapersPapers where it ranSamplesUnverifiedFingerprinted
conv3x3 1,376 1,346
169 with a run with no instrument failure, 1,177 where every run was a failure of Syntology's instrument
257 39 1
accuracy
census: 8 implementations from 7 papers, 7 distinct outputs across 6 buckets, one shared input per bucket
872 608
287 with a run with no instrument failure, 321 where every run was a failure of Syntology's instrument
404 82 98
load_data 700 438
228 with a run with no instrument failure, 210 where every run was a failure of Syntology's instrument
726 307 14
conv1x1 691 670
82 with a run with no instrument failure, 588 where every run was a failure of Syntology's instrument
145 22 0
normalize
census: 70 implementations from 82 papers, 51 distinct outputs across 11 buckets, one shared input per bucket
638 577
393 with a run with no instrument failure, 184 where every run was a failure of Syntology's instrument
491 59 294
str2bool 533 532
520 with a run with no instrument failure, 12 where every run was a failure of Syntology's instrument
166 1 0
exists
census: 5 implementations from 103 papers, 1 distinct output on one shared input
528 519
519 with a run with no instrument failure, 0 where every run was a failure of Syntology's instrument
34 10 8
train 522 210
168 with a run with no instrument failure, 42 where every run was a failure of Syntology's instrument
616 382 4
evaluate 514 297
208 with a run with no instrument failure, 89 where every run was a failure of Syntology's instrument
550 258 15
collate_fn 500 461
197 with a run with no instrument failure, 264 where every run was a failure of Syntology's instrument
384 37 18
load_model 479 121
88 with a run with no instrument failure, 33 where every run was a failure of Syntology's instrument
501 381 1
build_model 469 48
34 with a run with no instrument failure, 14 where every run was a failure of Syntology's instrument
365 320 1
resnet18 432 217
143 with a run with no instrument failure, 74 where every run was a failure of Syntology's instrument
393 202 0
default 430 423
412 with a run with no instrument failure, 11 where every run was a failure of Syntology's instrument
55 9 21
rotate_half
census: 24 implementations from 96 papers, 4 distinct outputs across 2 buckets, one shared input per bucket
424 423
36 with a run with no instrument failure, 387 where every run was a failure of Syntology's instrument
51 1 50
apply_rotary_pos_emb
census: 14 implementations from 15 papers, 8 distinct outputs across 7 buckets, one shared input per bucket
393 349
49 with a run with no instrument failure, 300 where every run was a failure of Syntology's instrument
119 30 23
load 392 116
61 with a run with no instrument failure, 55 where every run was a failure of Syntology's instrument
267 150 15
load_config 389 302
104 with a run with no instrument failure, 198 where every run was a failure of Syntology's instrument
367 83 14
get_logger 383 336
177 with a run with no instrument failure, 159 where every run was a failure of Syntology's instrument
287 39 1
test 377 209
167 with a run with no instrument failure, 42 where every run was a failure of Syntology's instrument
411 192 4
window_partition
census: 13 implementations from 17 papers, 5 distinct outputs across 3 buckets, one shared input per bucket
377 371
26 with a run with no instrument failure, 345 where every run was a failure of Syntology's instrument
113 8 61
MLP 375 352
329 with a run with no instrument failure, 23 where every run was a failure of Syntology's instrument
417 25 116
get_model 368 84
59 with a run with no instrument failure, 25 where every run was a failure of Syntology's instrument
356 270 2
parse_args 355 175
66 with a run with no instrument failure, 109 where every run was a failure of Syntology's instrument
415 233 3
Attention
census: 13 implementations from 12 papers, 10 distinct outputs across 4 buckets, one shared input per bucket
344 283
283 with a run with no instrument failure, 0 where every run was a failure of Syntology's instrument
428 84 32
load_json 344 328
96 with a run with no instrument failure, 232 where every run was a failure of Syntology's instrument
290 16 16
drop_path
census: 27 implementations from 61 papers, 5 distinct outputs across 2 buckets, one shared input per bucket
325 314
47 with a run with no instrument failure, 267 where every run was a failure of Syntology's instrument
68 6 24
validate 313 119
84 with a run with no instrument failure, 35 where every run was a failure of Syntology's instrument
306 195 2
gelu
census: 43 implementations from 74 papers, 6 distinct outputs on one shared input
308 294
216 with a run with no instrument failure, 78 where every run was a failure of Syntology's instrument
114 8 95
normalize_answer 304 300
29 with a run with no instrument failure, 271 where every run was a failure of Syntology's instrument
113 4 107
window_reverse 296 288
28 with a run with no instrument failure, 260 where every run was a failure of Syntology's instrument
84 12 0
get_dataset 285 72
54 with a run with no instrument failure, 18 where every run was a failure of Syntology's instrument
292 218 1
preprocess
census: 11 implementations from 13 papers, 10 distinct outputs across 5 buckets, one shared input per bucket
284 222
123 with a run with no instrument failure, 99 where every run was a failure of Syntology's instrument
258 70 57
Encoder 278 183
183 with a run with no instrument failure, 0 where every run was a failure of Syntology's instrument
380 143 56
count_parameters 278 271
265 with a run with no instrument failure, 6 where every run was a failure of Syntology's instrument
90 7 0
load_checkpoint 277 165
126 with a run with no instrument failure, 39 where every run was a failure of Syntology's instrument
232 103 7
load_dataset 276 163
87 with a run with no instrument failure, 76 where every run was a failure of Syntology's instrument
302 130 5
load_image 273 236
215 with a run with no instrument failure, 21 where every run was a failure of Syntology's instrument
185 38 3
get_pairs 260 259
10 with a run with no instrument failure, 249 where every run was a failure of Syntology's instrument
12 1 0
swish
census: 14 implementations from 37 papers, 2 distinct outputs on one shared input
258 254
30 with a run with no instrument failure, 224 where every run was a failure of Syntology's instrument
41 4 28
Block
census: 5 implementations from 5 papers, 3 distinct outputs across 2 buckets, one shared input per bucket
252 173
172 with a run with no instrument failure, 1 where every run was a failure of Syntology's instrument
309 97 21
read_json 250 214
59 with a run with no instrument failure, 155 where every run was a failure of Syntology's instrument
173 14 4
repeat_kv
census: 15 implementations from 60 papers, 2 distinct outputs on one shared input
244 242
7 with a run with no instrument failure, 235 where every run was a failure of Syntology's instrument
33 2 25
softmax
census: 53 implementations from 63 papers, 16 distinct outputs across 8 buckets, one shared input per bucket
240 223
172 with a run with no instrument failure, 51 where every run was a failure of Syntology's instrument
185 15 159
main 235 33
28 with a run with no instrument failure, 5 where every run was a failure of Syntology's instrument
282 251 0
whitespace_clean 230 230
2 with a run with no instrument failure, 228 where every run was a failure of Syntology's instrument
3 0 3
get_optimizer 228 193
119 with a run with no instrument failure, 74 where every run was a failure of Syntology's instrument
171 31 0
Model 227 70
70 with a run with no instrument failure, 0 where every run was a failure of Syntology's instrument
250 171 6
get_data 223 134
100 with a run with no instrument failure, 34 where every run was a failure of Syntology's instrument
248 109 7
split_list 223 222
13 with a run with no instrument failure, 209 where every run was a failure of Syntology's instrument
31 1 25

Which names get a page Syntology

179,672 distinct entry names occur across 294,260 distinct samples. A name gets a page when it appears in at least 5 papers: 7,868 names. The distribution at build time:

Papers per nameNamesPage
1146,890no
216,565no
35,551no
42,798no
5–94,606yes
10–191,897yes
20–49915yes
50–99293yes
100+157yes

Names are matched as exact strings, case-sensitively: 194 lowercase forms are shared by two or more names with a page (for example a class and a function that differ only in case), and 213 page addresses needed a numeric suffix to keep such names apart. Behaviour outputs are not in this export; the graph at syntology.ai holds them.