vector-mtrl 0.3.0__tar.gz → 0.4.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/CHANGELOG.md +103 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/PKG-INFO +56 -42
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/README.md +54 -41
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/05-lineage-and-the-recorded-recipe.md +6 -1
- vector_mtrl-0.4.2/docs/user/concepts/units-of-measure.md +62 -0
- vector_mtrl-0.4.2/docs/user/how-to/align-mixed-resolution-members.md +238 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/anchor-verification-with-receipts.md +12 -8
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/attach-author-disclosure.md +18 -12
- vector_mtrl-0.4.2/docs/user/how-to/author-an-ingestion-loader.md +159 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/check-external-evidence.md +13 -8
- vector_mtrl-0.4.2/docs/user/how-to/control-the-operation-registry.md +254 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/handle-missing-extras.md +8 -6
- vector_mtrl-0.4.2/docs/user/how-to/handle-selection-and-data-errors.md +288 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/index.md +13 -3
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/inspect-a-damaged-archive.md +12 -7
- vector_mtrl-0.4.2/docs/user/how-to/prepare-an-ml-dataset.md +340 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/record-scientific-assertions.md +35 -20
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/replace-a-checkpoint.md +20 -9
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/replay-a-committed-chain.md +27 -7
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/how-to/resample-with-scipy.md +22 -13
- vector_mtrl-0.4.2/docs/user/how-to/use-scientific-python-libraries.md +220 -0
- vector_mtrl-0.4.2/docs/user/reference.md +1274 -0
- vector_mtrl-0.4.2/docs/user/start-here/01-your-first-record.md +148 -0
- vector_mtrl-0.4.2/docs/user/start-here/02-save-load-verify.md +123 -0
- vector_mtrl-0.4.2/docs/user/start-here/03-transform-and-plot.md +142 -0
- vector_mtrl-0.4.2/docs/user/start-here/04-handles-and-the-core.md +135 -0
- vector_mtrl-0.4.2/docs/user/start-here/05-fix-a-mistake.md +121 -0
- vector_mtrl-0.4.2/docs/user/start-here/06-replicates-and-the-mean.md +108 -0
- vector_mtrl-0.4.2/docs/user/start-here/07-get-your-data-out.md +107 -0
- vector_mtrl-0.4.2/docs/user/start-here/08-record-a-claim.md +120 -0
- vector_mtrl-0.4.2/docs/user/start-here/09-one-complete-analysis.md +289 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/__init__.py +7 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_alignment_report.py +12 -6
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_archive_serialise.py +1 -3
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_run_management.py +16 -2
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_source_capture.py +1 -1
- vector_mtrl-0.4.2/mtrl/_summaries.py +664 -0
- vector_mtrl-0.4.2/mtrl/_summary_text.py +161 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_transform_validation.py +15 -8
- vector_mtrl-0.4.2/mtrl/_unit_summary.py +151 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/archive/validation.py +7 -7
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/committed_values.py +13 -6
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/exports.py +38 -2
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/frames.py +196 -12
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/grids.py +58 -18
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/__init__.py +5 -2
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_adapters.py +65 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_dataset.py +169 -10
- vector_mtrl-0.4.2/mtrl/ml/_flat.py +201 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_preprocessing.py +76 -23
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_specs.py +31 -9
- vector_mtrl-0.4.2/mtrl/ml/_summaries.py +225 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/models.py +41 -11
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/mtrl.py +34 -97
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/operations/__init__.py +1 -1
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/operations/_validation.py +12 -1
- vector_mtrl-0.4.2/mtrl/operations/arithmetic.py +80 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/operations/base.py +53 -37
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/operations/resampling.py +9 -2
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/operations/resampling_scipy.py +7 -4
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/operations/spectra.py +18 -4
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/processing/registry.py +63 -129
- vector_mtrl-0.4.2/mtrl/sample.py +1535 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/pyproject.toml +6 -12
- vector_mtrl-0.3.0/docs/user/how-to/align-mixed-resolution-members.md +0 -104
- vector_mtrl-0.3.0/docs/user/how-to/control-the-operation-registry.md +0 -123
- vector_mtrl-0.3.0/docs/user/how-to/prepare-an-ml-dataset.md +0 -157
- vector_mtrl-0.3.0/docs/user/reference.md +0 -370
- vector_mtrl-0.3.0/docs/user/start-here/01-your-first-record.md +0 -98
- vector_mtrl-0.3.0/docs/user/start-here/02-save-load-verify.md +0 -163
- vector_mtrl-0.3.0/docs/user/start-here/03-fields-and-units.md +0 -103
- vector_mtrl-0.3.0/docs/user/start-here/04-transform-a-spectrum.md +0 -106
- vector_mtrl-0.3.0/docs/user/start-here/05-fix-a-mistake.md +0 -136
- vector_mtrl-0.3.0/docs/user/start-here/06-replicates-and-the-mean.md +0 -137
- vector_mtrl-0.3.0/docs/user/start-here/07-get-your-data-out.md +0 -221
- vector_mtrl-0.3.0/mtrl/_ingestion_shortcuts.py +0 -99
- vector_mtrl-0.3.0/mtrl/operations/arithmetic.py +0 -207
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/.gitignore +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/LICENSE +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/NOTICE +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/01-samples-runs-checkpoints.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/02-the-commit-boundary.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/03-two-kinds-of-identity.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/04-corrections-vs-replacements.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/06-revisions-the-archive-timeline.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/07-inside-a-mtrl-archive.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/08-trust-over-time.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/index.md +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_archive_documents.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_archive_integrity.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_archive_loading.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_assertions_io.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_assertions_ops.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_assertions_validation.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_candidate_matching.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_cli.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_disclosure.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_install_targets.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_optional.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_provenance.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_receipts.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_record_resolution.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_replay_plan.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_replay_run.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_table_plan.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_table_provenance.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_table_values.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_transform_apply.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/_uuid7.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/archive/__init__.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/archive/canonical.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/archive/codecs.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/archive/dtypes.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/archive/reader.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/assertions/__init__.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/assertions/_resolver.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/connectors/__init__.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/connectors/protocols.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/disclosure.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/__init__.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/archive_limits.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/assertions.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/frozen_containers.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/identities.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/provenance.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/records.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/revisions.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/source.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/domain/verification.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/excel.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/exceptions.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/features.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/integrity/__init__.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/integrity/canonical.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/integrity/digests.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/integrity/manifest.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/integrity/receipts.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/integrity/verification.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_canonical.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_group_keys.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_metadata_view.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_provenance.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_source_walk.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/ml/_splits.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/operations/_identity.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/processing/__init__.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/processing/_table_capture.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/processing/limits.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/protocols.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/py.typed +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/replay.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/tables.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/transformations/__init__.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/transformations/base.py +0 -0
- {vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/mtrl/units.py +0 -0
|
@@ -12,6 +12,109 @@ break, and every breaking change is named in the release entry.
|
|
|
12
12
|
|
|
13
13
|
### Added
|
|
14
14
|
|
|
15
|
+
- Data-first `Sample` workflows, including explicit selection and correction helpers and
|
|
16
|
+
protocol-based `Sample.ingest` for connector-produced runs.
|
|
17
|
+
- `Sample.verify()` returns the core `VerifyResult` through the facade, preserving the separate
|
|
18
|
+
structural and hash facts and creating no revision; an existing receipt is consumed, while
|
|
19
|
+
receipt creation stays on `Mtrl.export_receipt()`.
|
|
20
|
+
- `Sample.export()` writes one committed result through the core suffix dispatch after it
|
|
21
|
+
validates the handle owner and exact active version, so a stale or foreign handle fails before
|
|
22
|
+
any write. The write is staged, no format fallback exists, and the call creates no revision; a
|
|
23
|
+
CSV or workbook projection is not a `.mtrl` substitute.
|
|
24
|
+
- `Sample.claim()` commits one scientific claim version and returns an immutable
|
|
25
|
+
`ScientificAssertion`, with `Sample.claims()` reading the active non-retracted heads in commit
|
|
26
|
+
order. One call is one core commit: each claim adds exactly one revision, and a default-subject
|
|
27
|
+
claim needs an existing sample identity because the facade never assigns one — the
|
|
28
|
+
`assign_sample_uid` prerequisite is reported, not repaired.
|
|
29
|
+
- `Sample.summary()` recaps the sample's active records and claims as bounded, deterministic text
|
|
30
|
+
in commit order, with declared units and real totals, and `Sample.history()` prints the record
|
|
31
|
+
graph behind that state: each active record's producing step and, one hop only, its exact input
|
|
32
|
+
versions, with corrected and replaced versions marked as superseded. Both calls are read-only —
|
|
33
|
+
no revision, no archive read, no verification, no optional import — and neither text carries a
|
|
34
|
+
generated identifier or a timestamp. Both contracts are recorded in `lat.md/interactive-api.md`.
|
|
35
|
+
- Detached `MLDataset.metadata_frame()`, `signal_frame()` and `auxiliary_frame()` views,
|
|
36
|
+
with channel/coordinate columns and a stable auxiliary encoding basis across splits.
|
|
37
|
+
- `MLDataset.summary()` recaps one prepared dataset as bounded, deterministic text: the row,
|
|
38
|
+
group, and channel totals, the target with its task type and class vocabulary, each channel's
|
|
39
|
+
point count, coordinate field, span, alignment mode, and resample flag, the auxiliary basis,
|
|
40
|
+
the metadata detail, the group kind, and the excluded candidate count, closed by that
|
|
41
|
+
dataset's canonical digest. It creates no revision and reads no archive; `limit` bounds the
|
|
42
|
+
channel listing from 1 to 50; and no payload value, sample identity, record identity, or
|
|
43
|
+
source path appears. The recap contract is recorded in `lat.md/ml-datasets.md`.
|
|
44
|
+
- `mtrl.ml.build_dataset()` builds one dataset from flat keyword arguments: `channel`,
|
|
45
|
+
`x_field`, `y_field`, `target_grid`, and `target` are mandatory, while the checkpoint
|
|
46
|
+
stage, alignment mode, task type, group kind, group property, and metadata detail default
|
|
47
|
+
to the guide's recipe. It composes one `DatasetSpec` and delegates to `build_ml_dataset()`,
|
|
48
|
+
so both routes produce one `spec_summary` and one `dataset_digest` for one logical
|
|
49
|
+
dataset, and unknown enumeration text fails with `DatasetPreparationError` naming the
|
|
50
|
+
received value and the valid choices.
|
|
51
|
+
- Bounded interactive summaries: `Mtrl.summary()` returns a bounded text summary of a
|
|
52
|
+
container, `_repr_html_()` renders the same escaped text for notebooks, and
|
|
53
|
+
`MaterialData` and `InstrumentRun` gain the matching text and HTML forms. The summary
|
|
54
|
+
contract is recorded in `lat.md/interactive-api.md`.
|
|
55
|
+
|
|
56
|
+
### Changed
|
|
57
|
+
|
|
58
|
+
- **Breaking:** `MaterialData` and `InstrumentRun` no longer print their previous pydantic
|
|
59
|
+
default representation. Both render a bounded summary from `__repr__`, `__str__` and
|
|
60
|
+
`_repr_html_()` that never shows payload values, and `repr()` of a container returns
|
|
61
|
+
`Mtrl.summary()`. Code that parsed the old representation, or expected it to carry field
|
|
62
|
+
values, must read the documented accessors instead. The interactive-summary contract is
|
|
63
|
+
recorded in `lat.md/interactive-api.md`.
|
|
64
|
+
- **Breaking:** the default operation catalogue contains eight neutral version-two
|
|
65
|
+
contracts. Their input roles are `data` and `members`. Operations require the field
|
|
66
|
+
parameters their schemas declare: `y_field` for scale/offset; `x_field` and `y_field`
|
|
67
|
+
for resampling/mean/median; neither for identity. No field names are guessed.
|
|
68
|
+
- **Breaking:** `GridAlignmentSpec` requires keyword `x_field` and `y_field`, and
|
|
69
|
+
`FeatureChannelSpec` requires keyword `alignment`. `grid_from_record` and `shared_range`
|
|
70
|
+
require an axis-field argument, accepted positionally or by keyword.
|
|
71
|
+
- **Breaking:** alignment reports no longer assume a spectral axis when provenance does
|
|
72
|
+
not record one. Pass `x_field` explicitly in that case; conflicting recorded roles fail.
|
|
73
|
+
- **Breaking:** remove the version-one execution profile and `include_legacy` factory
|
|
74
|
+
argument, the retired peak-normalisation/zero-at-point implementations and aliases, and
|
|
75
|
+
the unused `mtrl.models.SPECTRUM_DTYPE` and `mtrl.models.CHECKPOINT_FILE_TYPE_SUFFIXES`
|
|
76
|
+
constants. The eight current contracts are defined
|
|
77
|
+
directly, not derived from retained historical descriptors.
|
|
78
|
+
- **Breaking:** direct arithmetic, resampling and aggregation functions require explicit
|
|
79
|
+
field arguments too. The read-only operation views reflect only the current catalogue;
|
|
80
|
+
detached registered `apply` validates that catalogue's parameters before execution.
|
|
81
|
+
- Historical archives load as data and stay verifiable, with one precise boundary: a
|
|
82
|
+
payload entry written under the retired `.spa` suffixed convention fails structural
|
|
83
|
+
validation loudly instead of loading. Removed operations are unavailable for replay, and
|
|
84
|
+
version-one general steps refuse the current version-two contract rather than silently
|
|
85
|
+
substituting a different implementation.
|
|
86
|
+
- **Breaking:** remove `Mtrl.load_spa` and `Mtrl.load_tri` and their lazy suite imports. The
|
|
87
|
+
core references no instrument suite: pass an explicit loader to
|
|
88
|
+
`Mtrl.load_instrument_file` or `Sample.ingest`. `LoadedInstrumentFile` and `SourceFile`
|
|
89
|
+
are new top-level exports so a loader needs only `from mtrl import ...`.
|
|
90
|
+
- **Breaking:** bytes payloads are always stored as `.bin`; the archive format no longer
|
|
91
|
+
derives payload suffixes from `file_type`. Archives written with a `.spa` payload entry
|
|
92
|
+
fail structural validation loudly. `file_type` remains recorded manifest provenance.
|
|
93
|
+
- **Breaking:** whole-container frame projection is strict by default. `runs_to_frame`,
|
|
94
|
+
`Mtrl.to_pandas` and `InstrumentRun.to_pandas` refuse with one `UnexportableCheckpointError`
|
|
95
|
+
listing every unconvertible checkpoint instead of warning and returning a partial frame.
|
|
96
|
+
Pass `strict=False` to keep the diagnostic partial frame. Contested label aliases such as
|
|
97
|
+
`type` beside `type_1` now reject in the CSV and labelled-frame paths rather than emitting
|
|
98
|
+
a duplicated column header.
|
|
99
|
+
|
|
100
|
+
## [0.3.1] - 2026-10-27
|
|
101
|
+
|
|
102
|
+
### Changed
|
|
103
|
+
|
|
104
|
+
- Point the PyPI front door at the documentation site: the README leads with
|
|
105
|
+
`mtrl-docs.vector-labs.app` and keeps the shipped guide paths as the offline route, and
|
|
106
|
+
the project URLs name the same documentation URL for the PyPI sidebar. The source-tree
|
|
107
|
+
version fallback also names `0.3.1` when installed distribution metadata is absent.
|
|
108
|
+
|
|
109
|
+
- The post-publish resolution check no longer races PyPI index propagation: a bounded,
|
|
110
|
+
read-only wait (`scripts/public_core_resolution_wait.py`) polls the same simple index
|
|
111
|
+
the resolver reads until both published archives appear, with offline-tested refusal
|
|
112
|
+
handling, before the single-shot resolution runs. The wait never re-uploads.
|
|
113
|
+
|
|
114
|
+
## [0.3.0] - 2026-10-26
|
|
115
|
+
|
|
116
|
+
### Added
|
|
117
|
+
|
|
15
118
|
- Scientific assertions: claims, versions, subjects, evidence, and supersession as a core
|
|
16
119
|
feature, plus the portable UUIDv7 physical-sample identity `sample_uid`
|
|
17
120
|
(`Mtrl.sample_uid` and `Mtrl.assign_sample_uid()`).
|
|
@@ -1,7 +1,8 @@
|
|
|
1
1
|
Metadata-Version: 2.5
|
|
2
2
|
Name: vector-mtrl
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.4.2
|
|
4
4
|
Summary: Materials research tools for spectra and archive workflows.
|
|
5
|
+
Project-URL: Documentation, https://mtrl-docs.vector-labs.app
|
|
5
6
|
Author-email: Fuzzy Labs <info@fuzzylabs.ai>
|
|
6
7
|
License-Expression: MPL-2.0
|
|
7
8
|
License-File: LICENSE
|
|
@@ -33,8 +34,8 @@ immutable record with exact lineage.
|
|
|
33
34
|
- Machine-learning dataset preparation across archives: checkpoint state selection, in-memory
|
|
34
35
|
preprocessing, grid alignment, leak-free group splits, and canonical provenance digests.
|
|
35
36
|
- Bounded `.mtrl` archives with an integrity manifest, external receipts, and forensic inspection.
|
|
36
|
-
- NumPy-only operations: identity, scale, offset,
|
|
37
|
-
|
|
37
|
+
- NumPy-only operations: identity, scale, offset, and linear resample, beside mean and median
|
|
38
|
+
aggregates. The `interp` extra adds SciPy PCHIP and cubic resampling.
|
|
38
39
|
- Explicit analysis grids and a read-only alignment report.
|
|
39
40
|
- A read-only replay plan, a capability view, and a detached rerun for committed chains.
|
|
40
41
|
- Author disclosure that stays inert: stored environment records and passive source bundles that
|
|
@@ -54,12 +55,14 @@ uv pip install --python .venv/bin/python vector-mtrl
|
|
|
54
55
|
|
|
55
56
|
## Quick start
|
|
56
57
|
|
|
57
|
-
|
|
58
|
-
|
|
58
|
+
Start with the `Sample` facade: one object that records, transforms, selects, corrects,
|
|
59
|
+
exports, and saves. It is the routine route for notebook work, like pyplot over the object
|
|
60
|
+
model. `sample.core` returns the underlying container whenever a task outgrows the facade.
|
|
59
61
|
|
|
60
|
-
|
|
62
|
+
Record one synthetic spectrum, normalise it with your own function, save, verify, reload,
|
|
63
|
+
and export CSV. The base install alone runs every step.
|
|
61
64
|
|
|
62
|
-
|
|
65
|
+
<!-- example: Quick start -->
|
|
63
66
|
|
|
64
67
|
```python
|
|
65
68
|
from pathlib import Path
|
|
@@ -67,61 +70,72 @@ from tempfile import TemporaryDirectory
|
|
|
67
70
|
|
|
68
71
|
import numpy as np
|
|
69
72
|
|
|
70
|
-
from mtrl import
|
|
73
|
+
from mtrl import Sample
|
|
71
74
|
|
|
72
|
-
|
|
73
|
-
run = InstrumentRun(run_name="replicate-01")
|
|
75
|
+
sample = Sample("sample-01")
|
|
74
76
|
axis = np.array([2000.0, 2100.0, 2200.0])
|
|
75
77
|
content = np.empty(axis.size, dtype=[("wavenumber", "<f8"), ("absorbance", "<f8")])
|
|
76
78
|
content["wavenumber"] = axis
|
|
77
79
|
content["absorbance"] = [0.1, 0.2, 0.3]
|
|
78
|
-
|
|
79
|
-
|
|
80
|
-
|
|
81
|
-
|
|
82
|
-
|
|
83
|
-
|
|
84
|
-
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
|
|
88
|
-
|
|
89
|
-
|
|
90
|
-
|
|
91
|
-
|
|
80
|
+
raw = sample.record(content, name="spectrum")
|
|
81
|
+
|
|
82
|
+
|
|
83
|
+
def normalise_peak(data, *, target=1.0):
|
|
84
|
+
"""Scale absorbance so its largest signed value equals the target."""
|
|
85
|
+
peak = float(data["absorbance"].max())
|
|
86
|
+
if not np.isfinite(peak) or peak == 0.0:
|
|
87
|
+
raise ValueError("absorbance has no finite, non-zero maximum to normalise")
|
|
88
|
+
scaled = data.copy()
|
|
89
|
+
scaled["absorbance"] = data["absorbance"] * (target / peak)
|
|
90
|
+
return scaled
|
|
91
|
+
|
|
92
|
+
|
|
93
|
+
normalised = sample.transform(raw, normalise_peak, name="normalised")
|
|
94
|
+
print(f"records: {len(sample.records)}; latest: {normalised.name}")
|
|
92
95
|
|
|
93
96
|
with TemporaryDirectory() as directory:
|
|
94
|
-
archive_path =
|
|
95
|
-
verification =
|
|
97
|
+
archive_path = sample.save(Path(directory) / "sample.mtrl")
|
|
98
|
+
verification = sample.verify()
|
|
96
99
|
print(verification.structural_status.value, verification.hash_status.value)
|
|
97
100
|
|
|
98
|
-
loaded =
|
|
99
|
-
|
|
100
|
-
csv_path = loaded.
|
|
101
|
-
run.run_id,
|
|
102
|
-
normalised.checkpoint_id,
|
|
103
|
-
Path(directory) / "signal.csv",
|
|
104
|
-
)
|
|
101
|
+
loaded = Sample.load(archive_path)
|
|
102
|
+
current = loaded.one(name="normalised")
|
|
103
|
+
csv_path = loaded.export(current, Path(directory) / "signal.csv")
|
|
105
104
|
exported = np.genfromtxt(csv_path, delimiter=",", names=True)
|
|
106
|
-
np.
|
|
105
|
+
assert np.allclose(exported["absorbance"], [1 / 3, 2 / 3, 1])
|
|
107
106
|
print(csv_path.is_file())
|
|
108
107
|
```
|
|
109
108
|
|
|
110
|
-
|
|
111
|
-
|
|
109
|
+
```text
|
|
110
|
+
records: 2; latest: normalised
|
|
111
|
+
valid consistent
|
|
112
|
+
True
|
|
113
|
+
```
|
|
114
|
+
|
|
115
|
+
You should see the two records and the transform's name, the two verification facts, and
|
|
116
|
+
`True` for the exported CSV. `sample.record()` commits a detached copy, so later changes to your arrays cannot alter the
|
|
117
|
+
record. Every committed change, including the transform above, creates one new revision, and
|
|
118
|
+
the transform records your own function as a custom step. The function scales by the largest
|
|
119
|
+
signed absorbance value; use `np.abs(...).max()` instead when the spectrum can go negative and
|
|
120
|
+
you want the largest magnitude. `sample.one(name=...)` requires exactly one match, and
|
|
121
|
+
`sample.export` writes one active result through the registered writer for the file suffix — a
|
|
122
|
+
projection of the data, not a substitute for the archive.
|
|
112
123
|
|
|
113
124
|
## Guides
|
|
114
125
|
|
|
115
|
-
|
|
116
|
-
|
|
126
|
+
Read the guides online at https://mtrl-docs.vector-labs.app. The same pages ship in
|
|
127
|
+
the source archive under `docs/user/` for offline reading: download the archive for a release
|
|
128
|
+
from https://pypi.org/project/vector-mtrl/#files, then read:
|
|
117
129
|
|
|
118
|
-
- `docs/user/start-here/` for the
|
|
119
|
-
|
|
120
|
-
|
|
130
|
+
- `docs/user/start-here/` for the tutorial path: installation, first records, the save,
|
|
131
|
+
load, verify loop, fields and units, transforms, corrections, aggregates, exports, and
|
|
132
|
+
one continuous notebook analysis, which the documentation website also offers as a
|
|
133
|
+
downloadable notebook.
|
|
121
134
|
- `docs/user/concepts/` for the ideas underneath, in plain language.
|
|
122
135
|
- `docs/user/how-to/` for task guides: replacements, resampling, mixed-resolution alignment,
|
|
123
136
|
replay, receipts, forensic inspection, disclosure, registry control, missing extras,
|
|
124
|
-
scientific assertions, external evidence,
|
|
137
|
+
scientific assertions, external evidence, machine-learning datasets, and
|
|
138
|
+
scientific-Python interoperability with pandas, NumPy, SciPy, and matplotlib.
|
|
125
139
|
- `docs/user/reference.md` for every public route, the stable typed imports, the exact
|
|
126
140
|
package-root export list, registry names, format features, and the archive safety contract.
|
|
127
141
|
|
|
@@ -12,8 +12,8 @@ immutable record with exact lineage.
|
|
|
12
12
|
- Machine-learning dataset preparation across archives: checkpoint state selection, in-memory
|
|
13
13
|
preprocessing, grid alignment, leak-free group splits, and canonical provenance digests.
|
|
14
14
|
- Bounded `.mtrl` archives with an integrity manifest, external receipts, and forensic inspection.
|
|
15
|
-
- NumPy-only operations: identity, scale, offset,
|
|
16
|
-
|
|
15
|
+
- NumPy-only operations: identity, scale, offset, and linear resample, beside mean and median
|
|
16
|
+
aggregates. The `interp` extra adds SciPy PCHIP and cubic resampling.
|
|
17
17
|
- Explicit analysis grids and a read-only alignment report.
|
|
18
18
|
- A read-only replay plan, a capability view, and a detached rerun for committed chains.
|
|
19
19
|
- Author disclosure that stays inert: stored environment records and passive source bundles that
|
|
@@ -33,12 +33,14 @@ uv pip install --python .venv/bin/python vector-mtrl
|
|
|
33
33
|
|
|
34
34
|
## Quick start
|
|
35
35
|
|
|
36
|
-
|
|
37
|
-
|
|
36
|
+
Start with the `Sample` facade: one object that records, transforms, selects, corrects,
|
|
37
|
+
exports, and saves. It is the routine route for notebook work, like pyplot over the object
|
|
38
|
+
model. `sample.core` returns the underlying container whenever a task outgrows the facade.
|
|
38
39
|
|
|
39
|
-
|
|
40
|
+
Record one synthetic spectrum, normalise it with your own function, save, verify, reload,
|
|
41
|
+
and export CSV. The base install alone runs every step.
|
|
40
42
|
|
|
41
|
-
|
|
43
|
+
<!-- example: Quick start -->
|
|
42
44
|
|
|
43
45
|
```python
|
|
44
46
|
from pathlib import Path
|
|
@@ -46,61 +48,72 @@ from tempfile import TemporaryDirectory
|
|
|
46
48
|
|
|
47
49
|
import numpy as np
|
|
48
50
|
|
|
49
|
-
from mtrl import
|
|
51
|
+
from mtrl import Sample
|
|
50
52
|
|
|
51
|
-
|
|
52
|
-
run = InstrumentRun(run_name="replicate-01")
|
|
53
|
+
sample = Sample("sample-01")
|
|
53
54
|
axis = np.array([2000.0, 2100.0, 2200.0])
|
|
54
55
|
content = np.empty(axis.size, dtype=[("wavenumber", "<f8"), ("absorbance", "<f8")])
|
|
55
56
|
content["wavenumber"] = axis
|
|
56
57
|
content["absorbance"] = [0.1, 0.2, 0.3]
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
|
|
61
|
-
|
|
62
|
-
|
|
63
|
-
|
|
64
|
-
|
|
65
|
-
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
|
|
69
|
-
|
|
70
|
-
|
|
58
|
+
raw = sample.record(content, name="spectrum")
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
def normalise_peak(data, *, target=1.0):
|
|
62
|
+
"""Scale absorbance so its largest signed value equals the target."""
|
|
63
|
+
peak = float(data["absorbance"].max())
|
|
64
|
+
if not np.isfinite(peak) or peak == 0.0:
|
|
65
|
+
raise ValueError("absorbance has no finite, non-zero maximum to normalise")
|
|
66
|
+
scaled = data.copy()
|
|
67
|
+
scaled["absorbance"] = data["absorbance"] * (target / peak)
|
|
68
|
+
return scaled
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
normalised = sample.transform(raw, normalise_peak, name="normalised")
|
|
72
|
+
print(f"records: {len(sample.records)}; latest: {normalised.name}")
|
|
71
73
|
|
|
72
74
|
with TemporaryDirectory() as directory:
|
|
73
|
-
archive_path =
|
|
74
|
-
verification =
|
|
75
|
+
archive_path = sample.save(Path(directory) / "sample.mtrl")
|
|
76
|
+
verification = sample.verify()
|
|
75
77
|
print(verification.structural_status.value, verification.hash_status.value)
|
|
76
78
|
|
|
77
|
-
loaded =
|
|
78
|
-
|
|
79
|
-
csv_path = loaded.
|
|
80
|
-
run.run_id,
|
|
81
|
-
normalised.checkpoint_id,
|
|
82
|
-
Path(directory) / "signal.csv",
|
|
83
|
-
)
|
|
79
|
+
loaded = Sample.load(archive_path)
|
|
80
|
+
current = loaded.one(name="normalised")
|
|
81
|
+
csv_path = loaded.export(current, Path(directory) / "signal.csv")
|
|
84
82
|
exported = np.genfromtxt(csv_path, delimiter=",", names=True)
|
|
85
|
-
np.
|
|
83
|
+
assert np.allclose(exported["absorbance"], [1 / 3, 2 / 3, 1])
|
|
86
84
|
print(csv_path.is_file())
|
|
87
85
|
```
|
|
88
86
|
|
|
89
|
-
|
|
90
|
-
|
|
87
|
+
```text
|
|
88
|
+
records: 2; latest: normalised
|
|
89
|
+
valid consistent
|
|
90
|
+
True
|
|
91
|
+
```
|
|
92
|
+
|
|
93
|
+
You should see the two records and the transform's name, the two verification facts, and
|
|
94
|
+
`True` for the exported CSV. `sample.record()` commits a detached copy, so later changes to your arrays cannot alter the
|
|
95
|
+
record. Every committed change, including the transform above, creates one new revision, and
|
|
96
|
+
the transform records your own function as a custom step. The function scales by the largest
|
|
97
|
+
signed absorbance value; use `np.abs(...).max()` instead when the spectrum can go negative and
|
|
98
|
+
you want the largest magnitude. `sample.one(name=...)` requires exactly one match, and
|
|
99
|
+
`sample.export` writes one active result through the registered writer for the file suffix — a
|
|
100
|
+
projection of the data, not a substitute for the archive.
|
|
91
101
|
|
|
92
102
|
## Guides
|
|
93
103
|
|
|
94
|
-
|
|
95
|
-
|
|
104
|
+
Read the guides online at https://mtrl-docs.vector-labs.app. The same pages ship in
|
|
105
|
+
the source archive under `docs/user/` for offline reading: download the archive for a release
|
|
106
|
+
from https://pypi.org/project/vector-mtrl/#files, then read:
|
|
96
107
|
|
|
97
|
-
- `docs/user/start-here/` for the
|
|
98
|
-
|
|
99
|
-
|
|
108
|
+
- `docs/user/start-here/` for the tutorial path: installation, first records, the save,
|
|
109
|
+
load, verify loop, fields and units, transforms, corrections, aggregates, exports, and
|
|
110
|
+
one continuous notebook analysis, which the documentation website also offers as a
|
|
111
|
+
downloadable notebook.
|
|
100
112
|
- `docs/user/concepts/` for the ideas underneath, in plain language.
|
|
101
113
|
- `docs/user/how-to/` for task guides: replacements, resampling, mixed-resolution alignment,
|
|
102
114
|
replay, receipts, forensic inspection, disclosure, registry control, missing extras,
|
|
103
|
-
scientific assertions, external evidence,
|
|
115
|
+
scientific assertions, external evidence, machine-learning datasets, and
|
|
116
|
+
scientific-Python interoperability with pandas, NumPy, SciPy, and matplotlib.
|
|
104
117
|
- `docs/user/reference.md` for every public route, the stable typed imports, the exact
|
|
105
118
|
package-root export list, registry names, format features, and the archive safety contract.
|
|
106
119
|
|
{vector_mtrl-0.3.0 → vector_mtrl-0.4.2}/docs/user/concepts/05-lineage-and-the-recorded-recipe.md
RENAMED
|
@@ -30,6 +30,11 @@ that supplies hand-built data, like the lesson 5 correction, records what it cor
|
|
|
30
30
|
operation recipe, because no operation produced the new values. Only records an operation
|
|
31
31
|
produced carry the full recipe described below.
|
|
32
32
|
|
|
33
|
+
You can read those edges for one sample without inspecting the archive internals:
|
|
34
|
+
`print(sample.history())` lists each active record's producing step and, one hop only, the exact
|
|
35
|
+
versions it consumed, marking a version that a later correction superseded — see
|
|
36
|
+
[one complete analysis](../start-here/09-one-complete-analysis.md).
|
|
37
|
+
|
|
33
38
|
## The provenance envelope
|
|
34
39
|
|
|
35
40
|
The lineage says which step ran. A second, richer object says everything about how it ran: the
|
|
@@ -74,4 +79,4 @@ digest to compare against. The how-to guides cover replay in detail.
|
|
|
74
79
|
Each sealed recipe becomes one step in a longer history —
|
|
75
80
|
[Revisions: the archive timeline](06-revisions-the-archive-timeline.md) shows how those steps
|
|
76
81
|
accumulate into a verifiable archive. To record a recipe of your own, revisit
|
|
77
|
-
[Transform a spectrum](../start-here/
|
|
82
|
+
[Transform and plot a spectrum](../start-here/03-transform-and-plot.md).
|
|
@@ -0,0 +1,62 @@
|
|
|
1
|
+
# Units of measure
|
|
2
|
+
|
|
3
|
+
Think of a lab notebook. When you write "peak at 2500 cm⁻¹", the number and the unit travel
|
|
4
|
+
together, and a colleague can read both. If the notebook silently rewrote your "cm⁻¹" as
|
|
5
|
+
"1/cm", you would wonder what else it changed. MTRL treats units the same careful way: it
|
|
6
|
+
keeps the text you declared, understands the spellings it knows, and never rewrites a unit
|
|
7
|
+
behind your back.
|
|
8
|
+
|
|
9
|
+
## What you declare
|
|
10
|
+
|
|
11
|
+
Every field of a record can carry one unit, written exactly as your instrument or notebook
|
|
12
|
+
reports it. A wavenumber axis might say `"cm^-1"` or `"cm⁻¹"`; a time axis might say `"s"`;
|
|
13
|
+
a normalised intensity often says `"a.u."`. The declaration is part of the record, so it
|
|
14
|
+
survives save, load, correction and export unchanged.
|
|
15
|
+
|
|
16
|
+
MTRL does not guess. A field with no declared unit stays undeclared, and a unit it does not
|
|
17
|
+
recognise is preserved verbatim rather than rewritten or dropped.
|
|
18
|
+
|
|
19
|
+
## The standard behind the text
|
|
20
|
+
|
|
21
|
+
Behind the scenes MTRL matches the spellings it knows against **QUDT** — Quantities, Units,
|
|
22
|
+
Dimensions and Types, a public standard vocabulary for measurement. QUDT separates two
|
|
23
|
+
ideas that are easy to blur:
|
|
24
|
+
|
|
25
|
+
- The **quantity kind** is the property being measured: an inverse length, a time, a
|
|
26
|
+
dimensionless ratio.
|
|
27
|
+
- The **unit** is the measure of that property: per centimetre, seconds, unitless.
|
|
28
|
+
|
|
29
|
+
A recognised spelling carries both, so a check can compare meaning rather than text. Two
|
|
30
|
+
fields that disagree fail their check with the disagreement named, instead of one being
|
|
31
|
+
converted to match the other.
|
|
32
|
+
|
|
33
|
+
## The spellings MTRL recognises today
|
|
34
|
+
|
|
35
|
+
The recognised set is deliberately small. Each row shows the spellings that map to the same
|
|
36
|
+
standard terms:
|
|
37
|
+
|
|
38
|
+
| You write | QUDT unit | Quantity kind |
|
|
39
|
+
| --- | --- | --- |
|
|
40
|
+
| `cm^-1`, `cm⁻¹`, `1/cm` | per centimetre | inverse length |
|
|
41
|
+
| `s`, `second` | seconds | time |
|
|
42
|
+
| `a.u.` | unitless | dimensionless |
|
|
43
|
+
|
|
44
|
+
Anything else — `"K"`, `"MPa"`, `"counts"`, your own domain unit — is kept exactly as you
|
|
45
|
+
wrote it, with no standard terms attached. It still records, exports and round-trips; it
|
|
46
|
+
plays no part in unit-checked comparisons until the recognised set grows.
|
|
47
|
+
|
|
48
|
+
## Where units matter
|
|
49
|
+
|
|
50
|
+
Declared units do real work in three places. A transform preserves the declarations of the
|
|
51
|
+
fields its output keeps, so derived records stay labelled. Machine-learning dataset
|
|
52
|
+
specifications can require an `expected_unit` for a channel, feature or target, and a
|
|
53
|
+
mismatch fails the build with the field and its declared unit named. And exports carry the
|
|
54
|
+
declared text, so a colleague reading your CSV or workbook sees the same unit you wrote.
|
|
55
|
+
|
|
56
|
+
## How to write units well
|
|
57
|
+
|
|
58
|
+
Use the spelling your instrument reports; that is the honest record. When you are free to
|
|
59
|
+
choose, prefer the canonical forms from the table above (`cm^-1`, `s`, `a.u.`) so the
|
|
60
|
+
standard terms attach. Keep one spelling per field across a study, because comparisons read
|
|
61
|
+
the declaration you wrote. And when a check fails on units, change the declaration or the
|
|
62
|
+
data deliberately — MTRL will not choose for you.
|