tools/battery: every set runs bare, and an anomalous reference reads the per-hand table

The XDS arms passed -A wherever the reference had been measured with
FRIEDEL'S_LAW=FALSE, so that the merging statistics would be comparable. That
reason is gone: a rotation run decides with Friedel's mates merged whatever -A
says, and now reports itself counted both ways out of that one run, so -A is
presentation only and passing it changes nothing a battery reads. Verified on a
6 keV set: the reflections, the MTZ and the mmCIF of an -A run are byte-identical
to a bare one, and the reports differ in the FRIEDELS_LAW line alone.

Comparability now comes from the report instead. Where a reference kept the
mates apart, the scorer takes uniques, completeness, multiplicity, <I/sigma>,
R_meas and CC1/2 from the run's ANOM_REFRES_* block - the same merge counted
per Bijvoet hand, which is how XDS counted. ISa is not among them: there is one
error model, it is fitted on the hands, and REFRES_ISA is it.

The report reader also learns the third shell table, so it stops taking the
per-hand table's first row for the reference range's low-resolution shell.

Re-scored on six sets. The anomalous rows stop comparing a Friedel-merged
multiplicity against a per-hand one (28.65 against XDS's 12.42 becomes 15.54;
19.69 against 8.85 becomes 11.09), completeness moves 1-2 points onto the
reference's own footing, R_meas ratios move a few percent and the CC1/2 noise
ratio rises, as it must when CC1/2 is read at the reference's multiplicity. No
verdict moves: pass and fail are decided on the lattice, the space group and the
resolution, none of which these numbers reach. The two non-anomalous rows score
exactly as before.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013nW6FNRP1bBJJ8pfHiByAT
This commit is contained in:
leonarski_fandClaude Opus 5 committed 2026-09-21 13:02:40 +02:00
1 parent 14b7fce816
commit 29b5d7cce1
2 files changed
+26 -8

No files matched your search

+8 -7
View File
@@ -26,7 +26,7 @@ The site config (data roots, runs root, lock, gpulock, private manifest, baselin
$JFJOCH_BATTERY_SITE, else site.json beside this script (not committed; site.example.json is).
Every set runs once, with one command per arm (COMMAND_DOC): the open arm with the deposited model
(--model), the XDS arms with -A where XDS was anomalous and XDS's range as a report-only second
(--model), the XDS arms with XDS's range as a report-only second
table (--report-resolution). Neither option changes the processing, so the one run is both what a
user gets and the like-for-like comparison.
@@ -68,8 +68,9 @@ PDB_ID = re.compile(r"^[0-9][A-Za-z0-9]{3}$")
COMMAND_DOC = {
"deposition": "`rugnux --model <deposited coordinates> <input>` (without --model where there is no "
"deposited model: small molecules, unpublished sets)",
"xds": "`rugnux [-A] --report-resolution <XDS d_min>,<XDS d_max> <input>`: -A where XDS kept "
"Friedel mates apart; the second statistics table at XDS's range is report-only",
"xds": "`rugnux --report-resolution <XDS d_min>,<XDS d_max> <input>`: the second statistics "
"table at XDS's range is report-only; where XDS kept Friedel mates apart the scorer "
"reads the run's own per-hand table, which needs no flag",
}
LOCAL_SITE = os.path.join(HERE, "site.json")
@@ -174,10 +175,10 @@ def command(binary, e, opts, model=None):
cmd += ["-N", str(opts["threads"])]
ref = e.get("ref") or {}
if e["reference"] == "xds":
# XDS merged with FRIEDEL'S_LAW=FALSE: merge Friedel mates separately too, or the merging
# statistics are not comparable
if ref.get("anomalous"):
cmd.append("-A")
# No -A, whatever the reference did. A rotation run decides with Friedel's mates merged
# either way, and reports the merge counted BOTH ways from that one run; -A only says which
# of the two tables was asked for. Where XDS merged with FRIEDEL'S_LAW=FALSE the scorer
# reads rugnux's per-hand block (ANOM_REFRES_*) instead, which is the like-for-like table.
# a second statistics table over XDS's own range, from the same merge (report only). XDS's
# own limit, not the derived reference d_min: the manifest's XDS totals are over XDS's range,
# so only that makes the table like for like. Shells past rugnux's own cut print as not
+18 -1
View File
@@ -38,7 +38,10 @@ def read_report(path):
if f[:2] == ["d_min", "N_obs"]:
header = f
elif header and f and re.match(r"^\d+\.\d+$", f[0]):
prefix = "REFRES_" if "REFRES_RANGE" in out else ""
# Three tables in report order: the run's own, the --report-resolution one, and the
# per-hand (FRIEDEL'S_LAW=FALSE) one, whose keys are written just above it.
prefix = ("ANOM_" if "ANOM_UNIQUE_REFLECTIONS" in out
else "REFRES_" if "REFRES_RANGE" in out else "")
out[prefix + "LOWRES_D"] = f[0]
r_meas = f[header.index("R_meas")].rstrip("%") # "3.7%", or "-" for none
out[prefix + "LOWRES_R_MEAS"] = "" if r_meas == "-" else str(round(float(r_meas) / 100, 4))
@@ -199,6 +202,20 @@ def judge(entry, rep, run_note):
# the same merge binned over the reference's range (--report-resolution, XDS arms)
r.update({k: fnum(rep.get(k.upper())) for k in REFRES_KEYS})
r["refres_range"] = rep.get("REFRES_RANGE")
# An XDS reference measured with FRIEDEL'S_LAW=FALSE counts its uniques, completeness,
# multiplicity, R_meas and CC1/2 per Bijvoet hand, so what is like for like is the per-hand
# table of the same rugnux merge - the Friedel-merged one has about twice the multiplicity.
# Both tables come out of one bare run: -A is presentation only and changes nothing that is
# compared here. ISa is not among them - there is one error model, fitted on the hands, and
# REFRES_ISA is it.
r["per_hand_reference"] = False
if ref.get("anomalous"):
for k in ("unique_reflections", "completeness", "multiplicity", "i_over_sigma",
"r_meas", "cc_half"):
v = fnum(rep.get("ANOM_REFRES_" + k.upper()))
if v is not None:
r["refres_" + k] = v
r["per_hand_reference"] = True
r["isa_ratio"] = round(r["refres_isa"] / r["isa_ref"], 4) if r["refres_isa"] and r["isa_ref"] else None
r["r_meas_ratio"] = (round(r["refres_r_meas"] / r["r_meas_ref"], 4)
if r["refres_r_meas"] and r["r_meas_ref"] else None)