# Code with Papers (codewithpapers.app) > Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror. ## Syntology Code with Papers is the community app of Syntology (https://syntology.ai), which builds a graph of AI research papers, their code and their citations; this site shows one part of it (the harvested code and whether it ran) beside the archive, and every page links into the rest. - MCP server: https://syntology.ai/mcp (transport: streamable-http) - MCP server card: https://syntology.ai/.well-known/mcp/server-card.json (17 tools listed when this build read it, 2026-10-05) - Auth: a trial token, no account and no card, minted by POST https://syntology.ai/api/oauth/trial/token; the gate and limits are stated on https://syntology.ai/developers - Free coverage check, no token: https://syntology.ai/api/graph/have?x= (free, answers coverage only) - A paper in the graph: https://syntology.ai/paper/; its place on the Atlas: https://app.syntology.ai/?focus= (for an id the Atlas holds; the paper twins here carry both URLs only when the build confirmed them) - A paper's run record, one paper per request, free, no token, CORS open: https://syntology.ai/api/ran/.json (schema syntology.ran_record/1; measured at request time, so it may be newer than this site's graph read; each paper twin here whose paper has a line in this build's graph layer carries its URL under syntology.ran_record, with the crosswalk from the twin's counts to the record's; a README badge, to link and not to embed: https://syntology.ai/api/ran/.svg) - Syntology's own llms.txt: https://syntology.ai/llms.txt - Calls, as the card lists them: get_harvested_code_for_paper(arxiv_id) lists the samples harvested from a paper, each with its run record; it does not verify the paper; get_code(code_sha256) fetches one stored sample by the code_sha256 printed on the paper and code twins; this site shows no code text; have(x) is the free coverage check above, as an MCP tool; get_reference_implementation(name, min_level) returns runnable Python that Syntology generated from a paper's text (not the authors' code) for a named method, at a level whose meaning the card states; no level compares it against the authors' implementation; 2,032 paper pages here name one in their In Syntology block, with its level, the card's words for that level and the call; list_reference_implementations(query, min_level) browses those implementations by method name, origin paper, year and level, with no source attached Papers, code links, benchmarks, datasets and methods from the archived Papers with Code dataset (pwc-archive on Hugging Face, CC BY-SA 4.0, snapshot 2025-07-28, modified), plus Syntology's own per-sample record of whether harvested code ran (graph read 2026-10-05). - Papers with a code link in the archive: 218,852 - Benchmarks (leaderboards with rows): 11,026 - Datasets: 12,168; methods: 2,289 - Papers Syntology harvested code samples for: 51,887; where at least one sample ran: 48,557, of which 40,099 have a run with no instrument failure and 8,458 have only runs where Syntology's instrument failed (papers in Syntology's layer, with or without samples: 63,466; 2,441 of the papers with samples have no page here, counted in these totals: /not-shown#pages) - Code samples: 294,260 distinct: 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) ยท 83,091 unverified; "ran" means executed on a synthesized input and is not a correctness claim; "with no instrument failure" is honoured, violated or ran with no contract checked; the samples where Syntology's instrument failed (our contract draft was wrong, or the contract check could not run) are failures of Syntology's instrument, not of the code ## Machine-readable Every JSON twin's about.machine_readable lists /llms.txt, /index/manifest.json, /search/manifest.json, /download, /build_manifest.json; the entries below are this build's whole list. - /llms.txt: The machine catalog: every machine-readable file, counted. This file; its Machine-readable section and every twin's about.machine_readable come from one list in the build. This build: 1 file. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: not hashed here, since the hash would be part of the file it hashes - /.json: A page's JSON twin (home: /.json); /llms.txt counts pages without one. An HTML page's JSON twin: the records the page prints, led by the about block (the non-affiliation sentence, the way into Syntology, then this catalog's addresses) and closed by record_sha256 (the hash of the twin without that about block, with this site's per-build clocks masked; the HTML page shares it), record_changed_at (the date of the build that first held the record with that hash) and record_changed_at_basis (changed, added, or first_hashed: the first build that kept hashes, so no earlier date is known); the home page's twin is /.json. This build: 307,771 twins of 312,945 HTML pages: home streams 20 (/.json, /latest.json, /latest/.json, /ran/.json); papers 244,078 (/paper/.json); leaderboards 11,571 (/sota.json, /sota/.json); areas 16 (/area/.json); placed 1 (/placed.json); tasks 18,590 (/task/.json and its paper lists); datasets 14,309 (/datasets.json, /dataset/.json and its paper lists); methods 11,019 (/methods.json, /method/.json and its paper lists); code 7,869 (/code.json, /code/.json); census 297 (/census.json, /census/.json); ledgers 1 (/not-shown.json). Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: 5,174 HTML pages have no twin: dataset facet pages 4,546 (/datasets/task/, /datasets/language/ and /datasets/modality/ pages); dataset index pages 253 (/datasets/, the dataset index after its first page); method facet pages 367 (/methods/area/ and /methods/category/ pages); prose pages 5 (/about, /attribution, /download, /site/terms, /site/data-policy); forms and search 3 (/search, /contribute, /corrections); the twins under /index/ and /datasets/withheld.json have no HTML page and are listed as their own entries - /datasets/withheld.json: Datasets the content rules withheld, with reasons. A ledger with no HTML page: each withheld dataset's address, the rule that withheld it and its paper count. This build: 1 file, 6,011 bytes; sha256 ecde8f1456a6664779a5aaa60bb87e2f919b5f8a2653b49b7a7117a881418d85. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution) - /index/manifest.json: Paper-to-code index by arXiv id, with Syntology's counts. The paper-to-code index: every arXiv id with a paper page here, in files at /index/code/.json ("old" for pre-2007 ids), each listed in the manifest with its sha256, bytes and entry count. Each file: {"about": {...}, "index": {its totals, n_unverified and n_instrument included}, "papers": {"": {"paper": "/paper/", "twin": "/paper/.json", "syntology": {n_samples, n_ran, n_constructed, n_ran_checked, n_instrument, n_unverified, n_pointer_only, n_repos_linked, read_at} or null}}}, ordered by arXiv id, nothing ranked. n_ran = n_ran_checked + n_instrument; null means Syntology's layer holds no run record for the paper, not that it has no code; counts are per paper, so a sample reached from several papers counts under each; n_repos_linked is the distinct repositories the layer links to the paper (any provenance), not the page's "harvested from N repositories", which counts only the repositories its samples came from; per-sample detail is in the paper twin; each id's live run record on Syntology is https://syntology.ai/api/ran/.json (not in these files; the paper twin carries its paper's under syntology.ran_record). This build: 231,180 arXiv ids in 236 files and the manifest (86,381 bytes); sha256 4d520b34ce50012644fcb2bac2c685a0cfb762fdf8c9289c7c85a46e56a75874. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution) - /search/manifest.json: Site search index (titles, authors) and its files. The site search index, in two parts. /search/w/: titles and names (papers, tasks, datasets, methods, benchmarks), sharded by word prefix. /search/a/: the author index, the author names the paper pages print, sharded by surname word; names, not identities. Each file under /search/w/ and /search/a/ is named by the first 12 hex characters of its sha256. This build: 1,037 title files under /search/w/ and 562 author files under /search/a/, 274,609 rows in the title index; sha256 0ffab851ea83e16d2b5329ee78e9e0d0974bc213bfa3c42ef8074a49f677e7d9. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution) - /search/code.json: Names with a /code page, one row each. One row per name with a /code page: [name, papers, papers with a sample that ran, samples with a behaviour fingerprint, page path, papers with a run with no instrument failure, papers where every run was a failure of Syntology's instrument]; the third column is the last two added together, each paper counted once. This build: 1 file, 7,868 rows; sha256 d638e506b9b543e4c0843e90b70cb2cc10a19b735fdf288a810eff0fc5a0b43d. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: a bare JSON array with no about block: /code's script reads it as an array; its licence and attribution are this entry's - /search/arxiv/.json: arXiv id to page address, for ids with a page here. {"": ""} for every unversioned arXiv id this site has a paper page for ("old" for pre-2007 ids); /arxiv/ redirects (HTTP 301) to the same page and other ids answer 404. This build: 231,180 arXiv ids in 236 files. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: each file is a bare id map with no about block: the 404 page's script, the /arxiv/ redirects and the uploader read every key as an arXiv id, so no other key can share them; its licence and attribution are this entry's - /search/syntology/.json: arXiv ids in Syntology's Atlas id list, by month. One file per month: {atlas_read_at, paper_base, atlas_focus_base, ids (in the Atlas id list, no page here), held_here (in the list, with a page here)}; the 404 page and the search page read them. This build: 235 files, 181,530 ids with no page here and 115,203 with one (Atlas id list read 2026-10-05). Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: no about block in these files; their licence and attribution are this entry's - /feeds/papers.xml: Atom: newest papers with a harvested sample that ran. Atom 1.0; each entry links the paper page (rel=alternate, text/html) and its JSON twin (rel=alternate, application/json), and where this build confirmed them the paper on Syntology and its run record (rel=related). This build: 1 file, 100 entries; sha256 d29ed40807ca2bd20c961772da9ff8cc934fda2c2637e51f8f24b505e1af0e40. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution) - /feeds/results.xml: Atom: results Syntology placed on archive leaderboards. Atom 1.0; each entry links the leaderboard's Since the archive section (rel=alternate, text/html), the leaderboard's JSON twin (rel=alternate, application/json) and the paper page (rel=related); none is an archive row. This build: 1 file, 100 entries; sha256 a238e846f1b0f7987083c38c6b3b3b5d1375d38050bc095944a612d65d2436ce. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: written only when the build reads the extractor's verified file - /download/: Bulk files: Syntology's layer and repository reachability. Syntology's layer as gzipped JSON lines, one line per paper, with the status vocabulary the pages use, with its provenance record, and repository reachability (repo_status.json.gz); the files this build wrote are named below; /download describes the fields and links the archive at its source. This build: 3 files: graph_layer.jsonl.gz, graph_layer.jsonl.prov.json, repo_status.json.gz; sha256 /download/graph_layer.jsonl.gz cb2392f49a259f6b440e475929ed90061a662938c7a8b2d164ab4934ac719008, /download/graph_layer.jsonl.prov.json 08124fec174ba532d61dfcfb4691fd907f181fc38777eb8a4846fa0c9525660c, /download/repo_status.json.gz 5364f22c0553b940d50173b80db4a7bdc2e5544ddb941ee9a71d2f01bc44526c. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: /download itself is an HTML page with no twin - /build_manifest.json: The build record: inputs, counts, exclusions, probes. The stamped build record: input hashes, what each module wrote, the exclusion ledger /not-shown prints, every probe of Syntology and this catalog with its counts. This build: 1 file. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: not hashed here: it is written after this file and carries its own provenance stamp - /sitemap.xml: Sitemap index: the HTML pages crawlers may fetch. A sitemap index over /sitemap--.xml, HTML pages only, one file family per group: paper-with-layer (paper pages whose paper has a line in Syntology's layer or an assembly), paper-archive, leaderboard, task, method, dataset, code (the /code and /census pages) and site (the rest); lastmod is the date the page's record last changed for a page whose record a build saw change or appear, and for a page carrying Syntology's layer (a paper page when its paper has a line in the layer or an assembly; a page that prints when its record last changed); the archive snapshot for every other page. This build: 312,929 URLs in 13 files. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: JSON twins, pages marked noindex, /contribute and the paths robots.txt disallows (/search) are not listed - /index/changes.json: Change lists: pages and twins changed, added or removed, per build. The recent builds' change files, newest first, each at /index/changes/.json (a second build on one day is -2): {since_build, changed, added, removed, counts}, compared by each record's hash with this site's per-build clocks masked, so a clock-only change is never listed; a build that changed nothing writes no file. This build: 3 change files listed; this build's: /index/changes/2026-10-05.json; sha256 3dc30c835fc0ec8e79c8d5de9fd9488bcaf27006171a3d57b10a0c32601386e3. Licence: CC BY-SA 4.0, with arXiv metadata of papers newer than the archive under CC0 1.0 (/attribution). Exceptions: the change files are JSON with no HTML page; they are not records, so no change list lists them and they carry no record fields - https://syntology.ai/.well-known/mcp/server-card.json: Syntology's MCP server card: the tools its server lists. No tool is named on this site unless the card this build read lists it. This build: read by this build (2026-10-05): 17 tools. Licence: Syntology's service, not a file of this site; its gate and limits are stated on https://syntology.ai/developers - https://syntology.ai/api/graph/have?x=: Syntology's coverage check: whether its graph holds a method, arXiv id or title. Free, answers coverage only. This build: a live endpoint, not counted by this build. Licence: Syntology's service, not a file of this site; its gate and limits are stated on https://syntology.ai/developers - https://syntology.ai/api/ran/.json: Syntology's run record for one paper, measured at request time. This build: a live endpoint, one paper per request; this build's probe passed and schema syntology.ran_record/1 was read. Licence: Syntology's service, not a file of this site; its gate and limits are stated on https://syntology.ai/developers. Exceptions: the paper twins carry each paper's URL under syntology.ran_record; other twins do not list it - Vocabulary for a sample's status: ran_honours, ran_violates, ran_draft_wrong, ran_fixture, ran, unverified -- see /about ## Licence Archive-derived content: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/legalcode). Syntology's records on this site: CC BY-SA 4.0. Full attribution: /attribution. Built 2026-10-05T01:46:13Z.