An audit of docs/ against the code at HEAD, concentrating on what landed after
the previous documentation audit: the balanced half-set split and CCanom, the
declared-range completeness denominator, the one rule for a quantity nobody
measured, and the PONI the calibration now refuses to write.
Five statements the code had made false:
- The rotation merge's CCref column shows a dash, not "nan".
- Two pages still promised unweighted 2Fo-Fc / Fo-Fc maps; they have been
sigma_A-weighted since the rigid-body work, and one of the two sat six lines
from a paragraph that said so.
- Inserting the CCanom prose into the SigAno paragraph left the sentence about
the PDBx items and the SigAno column with CCanom as its subject, so it read as
a claim about a quantity that is in neither. CCanom is also rotation-only and
is not in the mmCIF, which nothing said.
- Two cross-references did not resolve - a heading that was renumbered, and a
slug spelled without the hyphen MyST puts in "TCP/IP".
Added where the behaviour is new and a user meets it: the report's contract for
a quantity a run did not measure - no key, and a dash in the table, which is not
the same claim as a measured zero; the half-set rule behind CC1/2, which is why
a CUDA and a non-CUDA build now agree on it and on CCanom; and the second reason
a calibration writes no .poni, a detector whose stored image is mirrored or
quarter-turned, which the PONI format cannot state.
The rc.166 changelog stays a release note. One entry is widened from the shell
table to the rule it is a case of, and two are added for output that was wrong
rather than merely undocumented: FITTED_RESOLUTION was the P1 cross-check's, and
the unmerged MTZ carried the reference setting where the merged file carried the
adopted one.
Sphinx builds clean with -W on the pinned docs/requirements.txt, and every
intra-doc anchor resolves against the generated HTML.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EFEJG6WBQv8th4UJFNe53N