[aaronkanzer]

SerialEM and Thermo Fisher Tomography write a plain-text .mdoc beside every tilt series, montage and saved movie. It holds the acquisition settings (dose, exposure, tilt, defocus, frame count) that the MRC or TIFF itself does not, and compare_acquisition could read EPU but not this.

What it adds

• .mdoc decoder, dispatched by extension with a 1 MiB leading read, also importable as decode_mdoc.
• voltageKv, pixelSizeA, doseEA2, fractions, magnification and exposureS use the EPU decoder's names, so an mdoc goes straight into compare_acquisition.
• Software and version, tilt axis, tilt range, total dose, acquisition times, the first image's stage position, defocus and frame path, and montage settings.
• All four section kinds: [ZValue], [FrameSet] (including the unbracketed title of a per-movie file), [MontSection] (kept out of the per-image values) and IMOD's [Image = name] series.

Verification

• 116 public mdoc files from GitHub: every compared value (16 fields) agrees with mdocfile on the 111 it can read. The other five are deliberately malformed test files; they decode with the bad values left out.
• 46,400 mutations of those files (truncation, byte flips, dropped and shuffled lines, NaN and junk values): nothing raised, all output JSON-safe.
• 12 new tests on synthetic files (no collection data in the repo). Full suite passes on Python 3.9 and 3.14.

Version bumped to 0.7.0. Publishing needs a v0.7.0 tag after merge.