[us-commits] [ehb54/ultrascan3] bb9541: perceiver: point the hydration tests at the ionize...

[emre brookes]

  Branch: refs/heads/ehb54-issue-980
  Home:   https://github.com/ehb54/ultrascan3
  Commit: bb9541fcfd5f1ae8137fdde1435768972e4546d9
      https://github.com/ehb54/ultrascan3/commit/bb9541fcfd5f1ae8137fdde1435768972e4546d9
  Author: ehb54 <brookes at uthscsa.edu>
  Date:   2026-08-17 (Mon, 17 Aug 2026)

  Changed paths:
    M us_somo/develop/perceiver/Makefile
    M us_somo/develop/perceiver/tests/builder.cpp

  Log Message:
  -----------
  perceiver: point the hydration tests at the ionized total, and widen `all`

The builder tests have been failing since 2026-08-12 without anything going
red. Two separate stale assumptions, both from the move to emitting both
protonation states.

The pH 7 table and the carboxylate chain-length test read hydration.total,
which is the NON-ionized sum. A protonated carboxyl oxygen carries 0 waters
against the ionized 5, so Asp and Glu both came back at 1.0 -- backbone only
-- against the tabulated 6.0 and 7.0, and the methylene difference the second
test exists to measure collapsed to zero. Reading total_ionized reproduces
both figures exactly: 15/15 residues now, and the methylene is worth 1.0
water on the nose.

The shape of the failure is what made it look like a bug. Only the two
carboxylates broke; Arg and Lys were fine, because an ionizable N stays
hydrated in both states and the two sums coincide for every residue except a
carboxylate. 13 of 15 passing hid the field mistake behind what looked like
selective breakage.

`all` was unit+regress only, which is how five days passed on a pushed branch
with builder red. It now runs everything that has a pass/fail verdict, with
builder last so a failure there does not mask the rest. coverage and sssrreal
stay out: they report rather than assert.

Not fixed here: "uncertain hydration types reach the REVIEW block" still
fails, and is held deliberately. It asserts the observation-agreement
criterion of propose(), but the builder now calls propose_by_rules, where
Ser's hydroxyl is a definite rule rather than a 55%-agreement average.
Pointing it at a residue that merely has a review line would make it pass
while testing nothing; it needs a group that genuinely has no pH 7 rule.
That is a test-design decision, not a rename. `make all` is red until it is
made, which is the honest state.

Co-Authored-By: Claude Opus 5 <noreply at anthropic.com>



To unsubscribe from these emails, change your notification settings at https://github.com/ehb54/ultrascan3/settings/notifications