The programs
The programs behind every number on this site are runnable, and this page says where. Each test on the checked-results page names its program, and each program states its bar in its own first lines. The repository is public at github.com/styopr613/rohonc-codex, and the programs are also served here as plain text.
What they run on
Python 3.10.12 with numpy 2.2.6 and the standard library, from the repository root, as python3 harness/<program>.py. Every test program names its test number and states its bar in its first lines, before any result. Three inputs are not in this repository: the anonymous open transcription of 2014, latest.txt, recoverable from the Internet Archive at the address on the Sources page; Király and Tokai's dictionary, which is their published work and is not served here; and the reference corpora the folios are read against, all public-domain texts named under Sources and fetched from Project Gutenberg and archive.org. With those three in place every saved run below can be reproduced.
From the repository root, python3 harness/reproduce_tests.py verifies the exact input files, runs every published result twice under different Python hash seeds, and requires byte-for-byte agreement with the saved runs. The blind and outside-reader tests score their committed replies without making network requests. Test 4 is explicitly blocked and has no result to reproduce.
Which program made which number
Every test on the checked-results page names its program, and each program names its saved run. This table is read out of the tests document at build time.
| test | what it asks | program | saved run |
|---|---|---|---|
| Test 1 | source agreement after excluding named evidence folios | ktheldout.py | ktheldout.txt |
| Test 2 and 2b | part of speech from context: the instrument is no good | ktpos.py, ktpos2.py | ktpos.txt, ktpos2.txt |
| Test 3 | the blindfold, run clean | ktblind.py | blindfold_1790006519.txt, blindfold_1790006519_score.txt · notes/BLINDFOLD.md |
| Test 4 | the illustrations: blocked | ktscanmap.py | — |
| Test 5 | external truth: sentences K&T translated whole | ktsentence.py | ktsentence.txt |
| Test 6 | word order after excluding named evidence folios | ktorder.py | ktorder.txt |
| Test 7 | take K&T's words away | ktrecover.py | ktrecover.txt |
| Test 8 | from a random 30% of the words, can the rest be got back? | ktbootstrap.py | ktbootstrap.txt |
| Test 9 | the hidden 70% validated against the random 30% seed | ktseedcheck.py | ktseedcheck.txt |
| Test 10, 10b, 10c | the search replayed on null books | ktnullreplay.py, ktnullreplay2.py, ktnullreplay3.py | ktnullreplay.txt, ktnullreplay2.txt, ktnullreplay3.txt |
| Test 11 | can the passage map be recovered from K&T's words alone? | ktpassid.py | ktpassid.txt |
| Test 12 and 12b | Test 5 rescored under a strict rule, then two outside re-glossers | ktstrict.py | ktstrict.txt, ktstrict_reglosser_gemini.txt, ktstrict_reglosser_grok.txt |
| Test 13 | the underdetermination census | ktcensus.py | ktcensus.txt |
| Test 14 | the blind rotated run, with an outside reader | ktblindfill.py | ktblindfill.txt |
| Test 15 | passage identification by an outside reader | ktpassid2.py | ktpassid2.txt |
| Test 16 | a second, independent transcription | ktopen.py | ktopen.txt |
| Test 17 | does Test 15's reader already know the answer? | ktmemory.py | ktmemory.txt |
| Test 18 | two blind models: read from some folios, judged on another chapter | kttwomodel.py | kttwomodel.txt |
The programs and every saved run are in the repository, with their history: github.com/styopr613/rohonc-codex. Nothing is served from here, so nothing on this page can point at a copy that has gone stale.
Licence · How to cite · Sources and data · Repository · Contact