Guide an agent through the end-to-end SNAP powder-diffraction reduction sequence and its key decision points. Use when deciding the sequence of preprocessing, calibration, normalization, and output products for SNAP diffraction data.
Instalação
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Guide an agent through the end-to-end SNAP powder-diffraction reduction sequence and its key decision points. Use when deciding the sequence of preprocessing, calibration, normalization, and output products for SNAP diffraction data.
version
2
review
{"status":"human-reviewed","reviewer":"Malcolm Guthrie","reviewed_on":"2026-05-05T00:00:00.000Z","basis":["docs","code","instrument-science-review"],"notes":"Reviewed and approved for v2 publication. Confirmed workflow updates: clarified lite-mode definition, added hook-aware scope/provenance language, split SEE handling into explicit pre-reduce and post-reduce gates, and expanded verification/provenance checks for SEE-dependent decisions.\n","approved_commit":"review/sns-snap-reduction-workflow-overview-v2","prior_review":{"status":"human-reviewed","reviewer":"Malcolm Guthrie","reviewed_on":"2026-04-30T00:00:00.000Z","basis":["docs","code","instrument-science-review"],"notes":"Clarified reduction-level context requirements, including sample-environment assembly context and cycle-strict calibration matching policy, and added explicit sample-environment handling in the recommended workflow sequence.\n","approved_commit":"review/sns-snap-reduction-workflow-overview-v1"}}
This skill frames the complete SNAP powder-diffraction reduction sequence,
identifies which software layer handles each step, and defines the quality
gates that must pass before outputs are used downstream. It is the entry
point for all SNAP reduction work; subordinate skills handle calibration
detail, diagnostics, and sample-environment special cases.
Notes: VULCAN cross-correlation text used as concept-only precedent, not implementation policy.
Overview
This skill produces a reproducible reduction sequence for a SNAP dataset: a
confirmed calibration context (difcal + normcal), an appropriate grouping and
masking strategy, any sample-environment-specific adaptations, and labelled
output workspaces (reduced_ or diagnostic_) ready for export and
downstream analysis.
Software provenance:
snapwrap (user interface — first): users call reduction and export workflows
by run number; user-visible choices include grouping scheme, masking inputs,
output formats, and continue flags; workflow customizations in the form of hooks passed to snapred.
snapred (backend): facilitates reduction workflow,resolves state-dependent calibration context; missing
calibration triggers approximation pathways and labels outputs diagnostic. Accepts instructions to customize workflows via hook calls from snapwrap.
Mantid (framework): executes calibration, focusing, unit conversion,
masking, and normalization algorithms orchestrated by snapred.
SNAP-specific conventions:
Lite mode (downsampling of pixels to 8 x 8 arrays of native pixels)is default; native mode is expert/special-case.
Grouping is fully general: 3 built-in defaults (all, bank, column) plus custom schemes.
Both mask types are first-class: pixelmask (exclude entire pixels) and binmask
(exclude bin ranges define in any unit: TOF, wavelength, Q, d-spacing within a pixel).
Sample environment equipment (SEE) assembly type (assembly.pe, assembly.dac, cylinder,
or none) from SEEMeta or documentation.
Where required for SEE type, identify specified customization of workflows (e.g. via hooks or post processing) and acquire necessary artifacts (e.g. pixel masks, wavelength notches) that these require.
Target output format (gsa, xye, csv).
Whether cycle-strict calibration matching must remain enabled (requireSameCycle).
Any known run exclusions or masking requirements.
Verify input runs and metadata consistency — Confirm run numbers exist,
IPTS path is accessible, and run metadata (sample, title, configuration)
is present and internally consistent. Flag any mismatches before
proceeding.
Resolve calibration context — For the instrument state derived from the
run, confirm that both a diffraction calibration (difcal) and a
normalization calibration (normcal) exist and that their appliesTo
fields cover the run number. Use snapwrap.snapStateMgr.checkCalibrationStatus().
If either is absent, decide explicitly: obtain the missing calibration
(preferred) or set the appropriate continue flag and accept diagnostic
output.
[CHECKPOINT]: Calibration status is confirmed. The decision to proceed
with full or diagnostic output is recorded before running reduction.
Choose grouping and masking strategy — Select a pixel grouping scheme
(all, bank, column, or custom) based on the resolution/statistics tradeoff
for the experiment. Identify and document any pixelmasks or binmasks
required. Record the scientific rationale for each choice.
Handle sample-environment special cases (pre-reduce) — If a high-pressure
device is present, branch to
sns-snap-sample-environment-reduction-special-cases
to build all environment-specific inputs. Return here only when that skill's
exit criteria are met.
[CHECKPOINT before step 5]: The following are confirmed ready before
reduce() is called:
Assembly type is confirmed from SEEMeta or documentation.
All required mask objects are constructed and named with correct unit
suffixes (pixel mask and/or binMaskList items as required by device class).
Any attenuation or background workspace required by the device class is
prepared and named.
Post-reduce processing obligations for this device class are noted
(see step 6).
Run reduction — Call snapwrap.utils.reduce(runNumber, ...) with the
confirmed calibration policy, grouping, and mask inputs. Check the output
workspace name prefix (reduced_ vs diagnostic_) immediately on
completion. Do not continue to export if the label is unexpected.
[CHECKPOINT]: Output workspaces exist in the Mantid tree with the
expected prefix and pixel-group suffixes. Any diagnostic output has an
explicit documented reason.
Handle sample-environment post-reduce processing, then perform background
and normalization checks — If a high-pressure device was present, consult
sns-snap-sample-environment-reduction-special-cases
for any required post-reduce steps (for example, manual background
extraction and subtraction for DAC data, or cylinder-cell container
treatment). Complete those steps before the normalization check below.
Then inspect reduced spectra for normalization stability across detector
groups. Compare baseline behavior against a reference or prior cycle.
Flag anomalies for the diagnostics skill.
Export for downstream analysis — Convert and export to the required
format (gsa, xye, csv) using snapwrap.io.exportData(). Confirm
exported files exist on disk and are non-empty.
Capture diagnostics and provenance — Record in the reduction record
(analysis notes and/or run log): run numbers, calibration identifiers and
versions, grouping/masking rationale, SEE assembly type, SEE pre-reduce
inputs (for example masks, notches, attenuation/background workspaces),
SEE post-reduce handling performed (if any), software versions, continue
flags used (if any), and output label with justification.
Exit criteria: Reduced outputs are labelled correctly, exported files
exist, and provenance is recorded. If output is diagnostic, its intended
usage (exploratory only, or pending recalibration and rerun) is explicitly
stated.
Rationalizations
Rationalization
Why it is wrong
"The calibration is probably fine — it worked last cycle."
Cycle-strict matching (requireSameCycle=True) exists because of instrumental drift between cycles. An out-of-cycle calibration produces no warning; it silently maps TOF to wrong d values. Always run checkCalibrationStatus().
"The output says diagnostic_ but the peaks look reasonable, so I'll use it for final results."
diagnostic_ means an approximation pathway was used. Publishing these as final results misrepresents the data provenance. Use diagnostic outputs for exploratory decisions only, then recalibrate and rerun.
"I'll document the masking choices after the analysis."
Undocumented mask rationale cannot be reproduced and cannot be reviewed. Write it down at step 3 and step 8, before the script runs.
"Native mode gives more pixels so it must be better."
Native mode is 64× more expensive and intended for special cases only. Lite mode matches the instrument's diffraction resolution. Using native mode routinely wastes compute and can destabilize calibration fits.
"I'll skip the background/normalization check — the data looks clean."
Normalization instability across groups is not visible in a single spectrum. Step 6 is a cross-group comparison; skipping it means you may export data with a systematic group-dependent error.
Red Flags
Reduction aborts with missing calibration while continue flags are unset
(default) → expected behavior. Check logs for ContinueWarning and revisit
step 2 to resolve calibration availability or intentionally enable a
continue pathway.
No output workspaces appear in the Mantid tree after reduce() completes →
reduction aborted (likely missing calibration with continue flags at
default False). Check logs for ContinueWarning. Revisit step 2.
Reduced spectra show systematic offsets between pixel groups (e.g., bank
peaks do not align with column peaks) → calibration mismatch across
groups. Invoke sns-snap-reduction-diagnostics
and revisit step 2–3.
Exported files are empty or zero-length → export step failed; check
IPTS write permissions and save=True flag. Revisit step 7.
Provenance record missing calibration version or software version →
results cannot be reproduced; complete step 8 before sharing outputs.
Verification
Before marking this skill complete:
Both difcal and normcal status confirmed via checkCalibrationStatus();
result recorded.
Output workspace prefix matches expectation: reduced_ for full
calibration, diagnostic_ only when intentionally accepted.
Grouping, masking, and SEE-specific choices (assembly type, pre-reduce
inputs, and any post-reduce handling) are recorded with scientific
rationale in the reduction record (analysis notes and/or run log).
Exported files exist on disk in the target format and are non-empty.
Reduction record (analysis notes and/or run log) includes: run numbers,
calibration identifiers + versions, software versions, any continue
flags used, output label with justification, and any SEE-dependent
decisions/handling that affected the reduction result.
If output is diagnostic_: explicit statement of whether it is for
exploratory use only or will be superseded by a reduced_ rerun.
