processbehavior 0.5.0__tar.gz → 0.6.0__tar.gz

{processbehavior-0.5.0 → processbehavior-0.6.0}/CHANGELOG.md

@@ -7,6 +7,114 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.6.0] - 2026-11-07
+
+### Added
+- **`validation/truth_report.py`.** It reports how closely the methods, as implemented, recover
+  answers known by construction: the PM INERT pure-noise data under all six missing-data
+  patterns, and Dr. Bishop's known-effects data. It is a report, not a gate.
+- **Two design-report lines.** In ADS 3 the report says how many cells hold a single reading,
+  since one is enough to switch R2 for the whole study to the differencing method. When
+  conditions were observed in different periods, it says how far apart their average periods
+  are, since condition effects then average over different times.
+- **`LossResult.reference`: what each loss part shows from noise alone.** The PDC, Time and
+  Interaction parts average squared deviations of estimated means, so with no effects each is
+  still above zero (about 14% interaction on the PM INERT pure-noise data). `reference` gives
+  the expected value of each under no effects for the study's own layout, exact for any
+  pattern of occupied cells and cell sizes, and the printed result shows it as a share of the
+  total. It is in `as_dict()` too. In ADS 2/3 the interaction is a remainder, so its reference
+  is `None`. Loss values are unchanged.
+- **Windowed capability shows the window's own noise.** `CapabilityResult.window_sigma_r2` is
+  σ̂ from the window's R2 values, printed beside the full-study `sigma_hat_r2` that the
+  potential indices use. It appears only on windowed results; the default result is unchanged.
+- **A Provenance sheet in Excel exports.** Written with the Summary sheet: library version,
+  time written, variables, input rows and rows analysed, design states, and a SHA-256
+  fingerprint of the analysed values. `AnalysisDataSet.n_input_rows` holds the input row count.
+- **`validation/chance_signals.py`.** It reports how often charts signal on pure noise: X/mR
+  charts under the default rules (and mR with all eight), and the R3-R5 effect charts, whose
+  limits sit between about 2.2 and 4.4 standard deviations of the plotted points depending on
+  the chart and layout. A report, not a gate; its tables are in the WECO rules reference and
+  the residuals guide.
+
+### Changed
+- **Manual citations follow the 11-7 edition.** Dr. Bishop's manual of 2026-11-07 has 20 chapters
+  (chapters 7 and up are one lower than before) and dot-numbered captions; every equation and
+  chapter citation in the code, docs and tests now uses its numbering, in the hyphen form the
+  manual repository checks ("11-7 manual Eq 13-8"). Seven equations it removed are cited by
+  their replacements. No behaviour changes.
+- **mR charts check Rule 1 only by default.** Wheeler reads the moving range chart for points
+  above its upper limit. The zone and run rules assume symmetric, independent points, and
+  successive moving ranges share a reading, so with all eight rules on, 95% of 100-point
+  pure-noise mR charts showed a signal (59% with Rule 1 alone). X charts still check all eight.
+  `rules='all'` restores the old behaviour. Signal counts on mR charts drop.
+- **The validation page says what is compared with VAS.** It claimed "every analytical
+  output"; the gate compares chart centres and limits, capability indices and loss shares. The
+  page now lists what is not compared (residual values one by one, signals, staged charts,
+  effects tables, calibrations, capability windows, the design report) and points to the truth
+  report. The `detect_signals()` docstring no longer says Rules 1-4 are the default.
+
+### Fixed
+- **Tutorial outputs match the library again.** The docs site shows each notebook's saved outputs,
+  and eight tutorials had drifted: limits from before 0.4.0 in the Design-State-1 analysis, flat
+  limits where they now vary by subgroup size in the reference validation, older process means and
+  signal counts. All were re-executed. The Derived Variables tutorial, saved without outputs, now
+  shows them. Prose that the new outputs contradicted was corrected; the Bishop reference values in
+  the reference validation now quote his VAS rerun of 13 October 2026.
+- **The factor x time interaction chart draws its lines in factor order on every machine.** It
+  sorted by time with an unstable sort, so the legend order, and the colour each line got, depended
+  on the numpy version.
+- **Three signal rules now match their definitions.** Rule 5 (trend) fired only on 7 rising or
+  falling points, not 6; Rule 6 (oscillation) only on 15 alternating points, not 14; and Rule 2
+  counted only points in Zone A, leaving out points beyond 3σ, which Western Electric and Wheeler
+  count as "beyond 2σ". The docs, the rule descriptions and the minimum-observation table already
+  said 6, 14 and "Zone A or beyond". X charts may show slightly more signals; on 100 points of
+  pure noise, the share with a signal under the eight default rules rises from 70% to 79%.
+- **The WECO rules page gives correct chance rates.** Six of its eight per-rule false-alarm rates
+  were wrong (for example 0.003% for 15 points hugging the centre, where it is 0.33%), and its
+  rule-set table had no source. The rates are now exact per-window figures, the table is measured
+  by `validation/chance_signals.py`, and the wording no longer says a signal is "almost certainly"
+  a special cause.
+- **Commas in numbers stored as text are no longer misread.** A comma was removed wherever it
+  appeared, so a decimal comma such as `10,4` became 104; and a lone `10,4` in a column of plain
+  numbers silently became missing. Commas are now read one way per column: as decimal points
+  when the column has no `.` and some comma is not a thousands group, otherwise only as
+  separators between groups of three digits. Any other comma value becomes missing and is named.
+- **The numeric-cleaning warning says what happened.** It counts the values actually reformatted
+  (it used to count every value in the column), gives examples, says when commas were read as
+  decimal points, and names the values that could not be read. A column of plain numeric strings
+  no longer warns.
+- **A factor nested in time is named.** When every time period holds one level of a factor (a
+  stage, a before/after split), the design report says so beside the empty-cell count, and
+  `loss_function()` warns that the PDC and Time parts both contain that factor's effect, so the
+  decomposition counts it more than once. It points to staged charts instead. Loss values are
+  unchanged.
+- **The design report no longer says "Complete structure" when there is nothing to compare.**
+  Without a sampling plan the structure line reads "Not checked (no sampling plan supplied)". It
+  used to say "Complete structure", even beside hundreds of empty cells.
+- **Potential capability says why it is unavailable.** A study with no factors (such as one
+  stream of subgroups) or no time variable got "R2 residuals not available (ADS 1)". It now
+  says that Cp/Cpk use the VAS R2 residuals, which need at least one factor (or a time
+  variable), and that the study has none.
+
+### Removed
+- **`examples/` and `validation/effects_visualization.ipynb`.** These demos predated the 0.2.0
+  API (`ProcessDataFrame`, `study.sds`, moved data files) and no longer ran; nothing linked to
+  them. The maintained, CI-executed tutorials are in `docs/tutorials/`.
+- **`tests/test_bishop_reference.py`.** Its one test compared residual values with a workbook
+  of Dr. Bishop's that is not in the repository, so it skipped on every run. Residual charts are
+  still compared with VAS through their limits.
+
+### CI / Infrastructure
+- **CI checks that docs notebooks match their saved outputs.** `scripts/refresh_notebooks.py`
+  re-executes the docs notebooks; with `--check` (in CI, replacing the nbmake run) it fails when a
+  notebook no longer runs or what a reader would see has changed, comparing printed text, tables
+  and each figure's data and labels. The `[test]` extra now needs nbclient, nbformat and ipykernel
+  instead of nbmake.
+- **Python 3.14 is tested** on Ubuntu and Windows, in CI and in the release gate, and listed in
+  the package classifiers.
+- **The truth report runs in CI.** `validation/truth_report.py` runs on every push (Ubuntu,
+  Python 3.13) and its output is kept as the `truth-report` artifact.
+
 ## [0.5.0] - 2026-10-29
 
 This release adds staged (phased) Xbar and S charts, computed per stage as Dr. Bishop
@@ -1181,7 +1289,8 @@ and may change between minor versions until 1.0.
 - `natsort >= 8.0`
 - `plotly >= 5.18, < 7`
 
-[Unreleased]: https://github.com/cnicholas/processbehavior/compare/v0.5.0...HEAD
+[Unreleased]: https://github.com/cnicholas/processbehavior/compare/v0.6.0...HEAD
+[0.6.0]: https://github.com/cnicholas/processbehavior/compare/v0.5.0...v0.6.0
 [0.5.0]: https://github.com/cnicholas/processbehavior/compare/v0.4.0...v0.5.0
 [0.4.0]: https://github.com/cnicholas/processbehavior/compare/v0.3.3...v0.4.0
 [0.3.3]: https://github.com/cnicholas/processbehavior/compare/v0.3.2...v0.3.3

{processbehavior-0.5.0 → processbehavior-0.6.0}/CITATION.cff

@@ -17,7 +17,7 @@ authors:
 repository-code: "https://github.com/cnicholas/processbehavior"
 url: "https://cnicholas.github.io/processbehavior/"
 license: Apache-2.0
-version: 0.5.0
+version: 0.6.0
 keywords:
   - statistical process control
   - process behavior charts

{processbehavior-0.5.0 → processbehavior-0.6.0}/CONTRIBUTING.md

@@ -74,13 +74,7 @@ The full test suite is the gate: if it fails, the change is not ready.
 The file `validation/PBTESTDATABASE_T100.csv` is ground-truth Minitab output
 for DS 1–6 from Bishop's reference materials. Numerical changes that touch
 residuals (R1–R5), DS detection, limit calculations, or variance
-decomposition must keep the relevant Bishop-reference tests green:
-
-```bash
-pytest tests/test_bishop_reference.py -v   # skips unless Bishop's residual workbooks are present
-```
-
-There is also a full end-to-end check that compares every chart center, control
+decomposition must keep the end-to-end check green. It compares every chart center, control
 limit, capability index and loss component against Bishop's VAS runs for the
 10-12-2026 manual (PM SDS 1–6, PM INERT, his known-effects data and the Medicare data):
 
@@ -97,10 +91,24 @@ summary.
 If a Bishop check fails after your change, the change is wrong — not the
 reference.
 
+## Docs notebooks
+
+The docs site shows each notebook's saved outputs; it does not run them. After a change that
+alters what a notebook prints or plots, refresh the saved outputs:
+
+```bash
+python scripts/refresh_notebooks.py                 # every docs notebook
+python scripts/refresh_notebooks.py docs/tutorials/signal-detection.ipynb
+```
+
+CI runs `python scripts/refresh_notebooks.py --check` and fails when a notebook no longer runs or
+no longer matches its saved outputs. Then read the text around any changed output: the check
+compares outputs, not prose.
+
 ## Citing the methodology
 
 Code and docs cite Bishop's VAS documentation manual by edition and equation number:
-`(10-12 manual Eq 14-8)`, `Eqs 14-4..14-13`, `Eqs 11-16/17`. Name the edition once per file and use
+`(11-7 manual Eq 13-8)`, `Eqs 13-4..13-13`, `Eqs 10-16/17`. Name the edition once per file and use
 the chapter-number form; `tests/test_manual_citations.py` checks both. The citations are read
 against the manual itself, so a new edition reports every place in the library whose equation
 changed or moved.

{processbehavior-0.5.0 → processbehavior-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: processbehavior
-Version: 0.5.0
+Version: 0.6.0
 Summary: Python-native Process Behavior & Control Charts (spec-driven SPC with tidy outputs)
 Project-URL: Homepage, https://github.com/cnicholas/processbehavior
 Project-URL: Repository, https://github.com/cnicholas/processbehavior
@@ -21,6 +21,7 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
 Classifier: Topic :: Scientific/Engineering
 Classifier: Topic :: Scientific/Engineering :: Information Analysis
 Classifier: Typing :: Typed
@@ -33,12 +34,14 @@ Requires-Dist: typing-extensions>=4.0; python_version < '3.11'
 Provides-Extra: dev
 Requires-Dist: filelock>=3.20.3; extra == 'dev'
 Requires-Dist: ipykernel; extra == 'dev'
+Requires-Dist: ipykernel>=6; extra == 'dev'
 Requires-Dist: jupyter-book>=2.0; extra == 'dev'
 Requires-Dist: kaleido>=0.2; extra == 'dev'
 Requires-Dist: lxml>=6.1.0; extra == 'dev'
 Requires-Dist: mypy>=1.10; extra == 'dev'
+Requires-Dist: nbclient>=0.10; extra == 'dev'
 Requires-Dist: nbconvert>=7.17.1; extra == 'dev'
-Requires-Dist: nbmake>=1.5; extra == 'dev'
+Requires-Dist: nbformat>=5.10; extra == 'dev'
 Requires-Dist: openpyxl>=3.1; extra == 'dev'
 Requires-Dist: orjson>=3.11.6; extra == 'dev'
 Requires-Dist: pillow>=12.2.0; extra == 'dev'
@@ -73,7 +76,9 @@ Requires-Dist: pre-commit>=3.7; extra == 'lint'
 Requires-Dist: ruff>=0.5; extra == 'lint'
 Requires-Dist: virtualenv>=20.36.1; extra == 'lint'
 Provides-Extra: test
-Requires-Dist: nbmake>=1.5; extra == 'test'
+Requires-Dist: ipykernel>=6; extra == 'test'
+Requires-Dist: nbclient>=0.10; extra == 'test'
+Requires-Dist: nbformat>=5.10; extra == 'test'
 Requires-Dist: openpyxl>=3.1; extra == 'test'
 Requires-Dist: pyarrow>=14.0; extra == 'test'
 Requires-Dist: pytest-cov>=5; extra == 'test'
@@ -282,7 +287,7 @@ The full report renders to `validation/e2e_bishop_report.html`.
 - **Effects analysis**: Main effects, time effects, and interaction effects
 - **Stratified charts**: Automatic per-stratum limits for grouped individual data
 - **Staged charts**: `phased=True` gives each stage its own center and limits on X/mR and Xbar/S; `phase_statistics()` returns the numbers behind the steps
-- **Signal detection**: all eight Western Electric rules on X/mR; Rule 1 (3-sigma) on Xbar/S, where the run- and zone-based rules need a time order that subgroup comparisons don't have
+- **Signal detection**: all eight Western Electric rules on X charts; Rule 1 (3-sigma) on mR, Xbar and S, where the run- and zone-based rules don't fit: subgroup comparisons have no time order, and successive moving ranges share a reading
 - **Beyond the charts**: capability indices (Cp/Cpk/Pp/Ppk), Taguchi loss decomposition, and maximum-information analysis — validated in the same Bishop-reference CI gate
 - **IDE support**: Column auto-completion via `ProcessBehavior(df).cols`
 - **Self-diagnostic errors**: Helpful messages that say what's available and how to fix it

{processbehavior-0.5.0 → processbehavior-0.6.0}/README.md

@@ -199,7 +199,7 @@ The full report renders to `validation/e2e_bishop_report.html`.
 - **Effects analysis**: Main effects, time effects, and interaction effects
 - **Stratified charts**: Automatic per-stratum limits for grouped individual data
 - **Staged charts**: `phased=True` gives each stage its own center and limits on X/mR and Xbar/S; `phase_statistics()` returns the numbers behind the steps
-- **Signal detection**: all eight Western Electric rules on X/mR; Rule 1 (3-sigma) on Xbar/S, where the run- and zone-based rules need a time order that subgroup comparisons don't have
+- **Signal detection**: all eight Western Electric rules on X charts; Rule 1 (3-sigma) on mR, Xbar and S, where the run- and zone-based rules don't fit: subgroup comparisons have no time order, and successive moving ranges share a reading
 - **Beyond the charts**: capability indices (Cp/Cpk/Pp/Ppk), Taguchi loss decomposition, and maximum-information analysis — validated in the same Bishop-reference CI gate
 - **IDE support**: Column auto-completion via `ProcessBehavior(df).cols`
 - **Self-diagnostic errors**: Helpful messages that say what's available and how to fix it

{processbehavior-0.5.0 → processbehavior-0.6.0}/processbehavior/__init__.py

@@ -47,7 +47,7 @@ its ``.sds`` field. Internally some modules still use the legacy "SDS"
 prefix (e.g. ``sds_detector.py``); the model itself is three-state.
 """
 
-__version__ = '0.5.0'
+__version__ = '0.6.0'
 
 # Result object
 from processbehavior.analysis_result import AnalysisResult

{processbehavior-0.5.0 → processbehavior-0.6.0}/processbehavior/analysis.py

@@ -254,7 +254,7 @@ def _mr_family_limits(mr_spec: _MRChartSpec, *, mean, mR) -> pd.DataFrame:
     """Limits for one MR-family chart per row of a per-stratum or per-phase table.
 
     The array form of the scalar ``calculate_limits`` these sites used to call through
-    ``DataFrame.apply(axis=1)``: the same formulae, 10-12 manual Eqs 12-10/12-11 (X) and
+    ``DataFrame.apply(axis=1)``: the same formulae, 11-7 manual Eqs 11-8/11-9 (X) and
     12-5 (mR). The scalar call built a ``pd.Series`` per stratum, about 0.1 ms each, so
     it cost 30s per chart at 300,000 strata. ``mean`` is ignored for the mR chart.
     """
@@ -921,7 +921,7 @@ class Analysis:
         """Return the cell-grid columns at which to compute Bishop's grand mean
         for an Xbar chart on this column.
 
-        Each re-centred residual is R + Ybar (10-12 manual Eq 14-26; see
+        Each re-centred residual is R + Ybar (11-7 manual Eq 13-26; see
         `AnalysisDataSet._calculate_centered_residuals` and
         `residual_calculator.calculate_r6_residuals`). The canonical grand mean of an
         RCRk chart averages over the grain its effect lives at — equal weight per
@@ -1216,10 +1216,10 @@ class Analysis:
     def _xbar_limits(self, out: pd.DataFrame, s_bar: float, n_to_use: str) -> pd.DataFrame:
         """Xbar natural process limits for the subgroups in ``out`` (columns center, s, n, N).
 
-        Equal subgroup sizes (or n_mode="average", Eq 11-20): S-bar / c4(N) for every row.
-        Unequal sizes: sigma_hat is the average of each subgroup's S_r / c4(N_r) (10-12 manual
-        Eq 11-5) and each subgroup's limits are center ± multiplier · sigma_hat / sqrt(N_r)
-        (Eqs 11-16, 11-17), so they step with N_r, as VAS draws them. A one-reading subgroup
+        Equal subgroup sizes (or n_mode="average", Eq 10-20): S-bar / c4(N) for every row.
+        Unequal sizes: sigma_hat is the average of each subgroup's S_r / c4(N_r) (11-7 manual
+        Eq 10-5) and each subgroup's limits are center ± multiplier · sigma_hat / sqrt(N_r)
+        (Eqs 10-16, 10-17), so they step with N_r, as VAS draws them. A one-reading subgroup
         has no S_r: it takes no part in sigma_hat and is charted at center ± multiplier · sigma_hat.
         """
         if n_to_use == 'n':
@@ -1283,12 +1283,12 @@ class Analysis:
 
         Every subgroup with a value is charted. A one-reading subgroup has no S, so sigma-hat
         comes from the subgroups with n >= 2, and its limits are center ± 3·sigma-hat (10-12
-        manual Eqs 11-16, 11-17 with N_r = 1), as VAS draws them; the companion S chart keeps
+        manual Eqs 10-16, 10-17 with N_r = 1), as VAS draws them; the companion S chart keeps
         only n >= 2. n counts the limits column, so a reading without one (R2's first value) is
         not charted, as in VAS. Returns None when no subgroup has n >= 2; the caller decides
         whether that raises (one chart) or is recorded (a stratum, a phase). Under
         ``n_mode='average'`` the ``N`` column becomes the mean fitted n and both charts take
-        the constant-N path (Eq 11-20).
+        the constant-N path (Eq 10-20).
         """
         out = out[out['n'].ge(1)].copy()  # n=0: every charted value missing
         fit = out['n'].ge(2)
@@ -2007,9 +2007,9 @@ class Analysis:
         A phase is one combination of the grouping factors: a contiguous run of subgroups in
         time order on the stage-by-stage layout that ``by=None`` already draws. Each phase is
         computed as the per-condition chart of ``by=[time]`` computes one stratum — the
-        unweighted mean of its subgroup means as centre (10-12 manual Eq 10-9 within the
-        phase), its own S-bar and sigma-hat (Eq 11-5), its own limits (Eqs 11-8, 11-9, 11-16,
-        11-17) — and the phases are concatenated in display order with the stepped columns.
+        unweighted mean of its subgroup means as centre (11-7 manual Figure 13.3 within the
+        phase), its own S-bar and sigma-hat (Eq 10-5), its own limits (Eqs 10-8, 10-9, 10-16,
+        10-17) — and the phases are concatenated in display order with the stepped columns.
         Nothing is pooled across phases: Dr. Bishop (2026-10-18), "treat each stage separately
         as if it were a single analysis".
 
@@ -2177,10 +2177,10 @@ class Analysis:
 
         The X and mR charts share >85% of their calculation logic. The behavioral
         differences are encoded in ``mr_spec`` — no boolean flags needed. The
-        statistics are the 10-12 manual's Chapter 12: mR-bar over the T - 1
-        consecutive ranges (Eq 12-1), X centre Y-bar (Eqs 12-7, 12-9), X limits
-        Y-bar ± 3·mR-bar/d2 (Eqs 12-10, 12-11) and mR limits 0 … D4·mR-bar
-        (Eq 12-5), within each stratum or phase when the chart is split.
+        statistics are the 11-7 manual's Chapter 11: mR-bar over the T - 1
+        consecutive ranges (Eq 11-1), X centre Y-bar (Eq 11-7), X limits
+        Y-bar ± 3·mR-bar/d2 (Eqs 11-8, 11-9) and mR limits 0 … D4·mR-bar
+        (Eq 11-5), within each stratum or phase when the chart is split.
 
         Parameters
         ----------
@@ -2895,11 +2895,11 @@ class Analysis:
             return result
 
         # === Global limits path (unchanged) ===
-        # Moving range: mR-bar over the T - 1 consecutive ranges (10-12 manual Eq 12-1). A re-centred
+        # Moving range: mR-bar over the T - 1 consecutive ranges (11-7 manual Eq 11-1). A re-centred
         # R2 chart takes them over R2 (_resolve_mr_source_column), not the plotted RCR2.
         out['mr'] = out[mr_source_col].diff().abs()
         mR = out['mr'].mean()
-        mean_ = out[value_col].mean()  # Y-bar, the X chart's centre (Eqs 12-7, 12-9)
+        mean_ = out[value_col].mean()  # Y-bar, the X chart's centre (Eq 11-7)
 
         # R chart: drop first observation
         if mr_spec.drops_first_mr:

{processbehavior-0.5.0 → processbehavior-0.6.0}/processbehavior/analysis_dataset.py

@@ -53,6 +53,7 @@ class AnalysisDataSet:
         # Store inputs
         self.spec = spec
         self.observed_design_state: int = observed_sds
+        self.n_input_rows: int = len(df)  # Rows given to formulate(), before tidying
 
         # Initialize output containers
         self.interactions = {}
@@ -171,7 +172,7 @@ class AnalysisDataSet:
     def _check_r2_scale_factor(self, r2_method: str) -> None:
         """Record (and warn once) when the layout leaves the R2 scale factor undefined.
 
-        The scaled-difference R2 (10-12 manual Eqs 14-4..14-13) needs at least two process
+        The scaled-difference R2 (11-7 manual Eqs 13-4..13-13) needs at least two process
         design conditions and more than K + 1 observations; outside that it has no value.
         """
         if r2_method != 'ma2':
@@ -243,7 +244,7 @@ class AnalysisDataSet:
 
         # Compute n_per_cell and cell means once
         self._n_per_cell = df.groupby('cell_key', observed=True)[y].transform('size')
-        self._ybar_kt = df.groupby('cell_key', observed=True)[y].transform('mean')  # Ȳ_kt (10-12 manual Eq 10-7)
+        self._ybar_kt = df.groupby('cell_key', observed=True)[y].transform('mean')  # Ȳ_kt (11-7 manual Eq 9-6)
 
         # Handle edge case: empty data or all NaN
         if len(self._n_per_cell) == 0 or self._n_per_cell.isna().all():
@@ -355,12 +356,12 @@ class AnalysisDataSet:
         Calculate the re-centred residuals (RCR) and the R1 cell/marginal means (Rbar).
 
         Re-centring puts a residual back on the measurement scale without changing what it
-        shows: RCR = R + Ȳ.. (Bishop's 10-12 manual Eq 14-26, for R1, R3, R4, R5 and R6; RCR6
+        shows: RCR = R + Ȳ.. (Bishop's 11-7 manual Eq 13-26, for R1, R3, R4, R5 and R6; RCR6
         is built per request in residual_calculator.calculate_r6_residuals). An effect chart of
         RCR5 by condition, RCR4 by period or RCR3 by subgroup therefore plots that effect plus
         R2, centred on the grand mean.
 
-        RCR2 is not part of Eq 14-26; it is kept as Ȳ_kt + R2 (which reconstructs Y in ADS 1).
+        RCR2 is not part of Eq 13-26; it is kept as Ȳ_kt + R2 (which reconstructs Y in ADS 1).
 
         Note: This method intentionally mutates self.analysis_dataset in-place
         (adding columns directly to the DataFrame). This differs from
@@ -381,9 +382,9 @@ class AnalysisDataSet:
         df['Rbar_k'] = df.groupby([self.spec.rsg_var_name], observed=True)['R1'].transform('mean')
         df['Rbar_t'] = df.groupby([self.spec.time_var], observed=True)['R1'].transform('mean')
 
-        # Re-centred residuals, Eq 14-26: RCR = R + Ybar
+        # Re-centred residuals, Eq 13-26: RCR = R + Ybar
         df['RCR1'] = df['Ybar'] + df['R1']
-        df['RCR2'] = df['Ybar_kt'] + df['R2']  # not in Eq 14-26; reconstructs Y in ADS 1
+        df['RCR2'] = df['Ybar_kt'] + df['R2']  # not in Eq 13-26; reconstructs Y in ADS 1
         df['RCR3'] = df['Ybar'] + df['R3']
         df['RCR4'] = df['Ybar'] + df['R4']
         df['RCR5'] = df['Ybar'] + df['R5']

{processbehavior-0.5.0 → processbehavior-0.6.0}/processbehavior/analysis_result.py

@@ -1053,9 +1053,10 @@ class AnalysisResult:
             Specific chart to analyze. If None, analyzes all charts.
         rules : str, list, or RuleSet, optional
             Rules to apply:
-            - 'standard': Rules 1-4 (default)
-            - 'extended': Rules 1-8
-            - 'all': All available rules
+            - None (default): Rules 1-8 on X charts; Rule 1 on mR, Xbar and S charts
+            - 'standard': Rules 1-4, on every chart
+            - 'extended': Rules 1-8, on every chart
+            - 'all': Same as 'extended'
             - List of rule names: ['rule_1', 'rule_2', ...]
             - RuleSet: Fluent API configuration
         config : SignalConfig, optional
@@ -1071,7 +1072,7 @@ class AnalysisResult:
 
         Examples
         --------
-        Simple usage (standard rules):
+        Simple usage (default rules for each chart):
 
         >>> signals = result.detect_signals()
         >>> print(signals.summary)

{processbehavior-0.5.0 → processbehavior-0.6.0}/processbehavior/capability.py

@@ -1,5 +1,5 @@
 """
-Process Capability Analysis — Bishop Chapter 16.
+Process Capability Analysis — Bishop Chapter 15 (11-7 manual).
 
 Computes capability indices given specification limits. Core logic lives as
 pure module-level functions (composition pattern matching ResidualCalculator /
@@ -19,7 +19,7 @@ Potential Capability (achievable by removing assignable causes):
 
 References
 ----------
-Bishop, T.  *Variation Analysis System* — Chapter 16.
+Bishop, T.  *Variation Analysis System* — Chapter 15 (11-7 manual).
 """
 
 from __future__ import annotations
@@ -89,7 +89,7 @@ class SpecLimits:
 @dataclass(frozen=True)
 class CapabilityResult:
     """
-    Result of a process capability assessment (Bishop Ch. 16).
+    Result of a process capability assessment (Bishop Ch. 15).
 
     Raw ``float`` values are stored unrounded.  Use :meth:`as_dict` for a
     rounded dictionary or ``repr()`` for a formatted summary.
@@ -111,7 +111,7 @@ class CapabilityResult:
     pp, ppk_lower, ppk_upper, ppk : float | None
         Current capability indices (from overall sigma).
     sigma_hat_r2 : float | None
-        Unbiased sigma from R2 residuals: S_R2 / c4(N_R2) (10-12 manual Eq 16-10).
+        Unbiased sigma from R2 residuals: S_R2 / c4(N_R2) (11-7 manual Eq 15-10).
     cp, cpk_lower, cpk_upper, cpk : float | None
         Potential capability indices (from R2 sigma), with CPL/CPU measured from
         ``potential_center``.
@@ -138,6 +138,11 @@ class CapabilityResult:
         ``y_bar + mean(R2)``. Equals ``y_bar`` when R2 averages to zero (ADS 1);
         with the scaled-difference R2 of ADS 2/3 it differs by mean(R2), which is
         small. None when potential capability is unavailable.
+    window_sigma_r2 : float | None
+        Windowed results only: σ̂ from the window's own R2 values, by the same
+        estimator as ``sigma_hat_r2``. The potential indices keep the full-study
+        ``sigma_hat_r2``; this shows how the window's noise compares. None when not
+        windowed or potential capability is unavailable.
     """
 
     # Input context
@@ -203,6 +208,9 @@ class CapabilityResult:
     # Potential view centre: mean of (y_bar + R2). Last field so positional construction is unchanged.
     potential_center: float | None = None
 
+    # Windowed only: sigma_hat from this window's own R2 values, beside the full-study one used above.
+    window_sigma_r2: float | None = None
+
     # ------------------------------------------------------------------
     # Visualization
     # ------------------------------------------------------------------
@@ -319,6 +327,7 @@ class CapabilityResult:
             out['time_var'] = self.time_var
             out['n_total'] = self.n_total
             out['window_warning'] = self.window_warning
+            out['window_sigma_r2'] = _r(self.window_sigma_r2)
         return out
 
     def __repr__(self) -> str:
@@ -339,6 +348,11 @@ class CapabilityResult:
             lines.append('  Potential Capability (R2 sigma):')
             lines.append(f'    centre={d["potential_center"]}, sigma_hat_r2={d["sigma_hat_r2"]}')
             lines.append(f'    Cp={d["cp"]}  Cpk={d["cpk"]}  (lower={d["cpk_lower"]}, upper={d["cpk_upper"]})')
+            if self.window is not None and self.window_sigma_r2 is not None:
+                lines.append(
+                    f'    sigma_hat_r2 is from the full study (N={d["n_total"]}); '
+                    f'this window\'s own is {d["window_sigma_r2"]}'
+                )
 
         lines.append('')
         lines.append('  Z-scores:')
@@ -389,9 +403,9 @@ def compute_sigma_hat(values: np.ndarray) -> tuple[float, float]:
 
     Notes
     -----
-    Bishop Ch. 16 — ddof=1 is mandatory.  The c4 correction
+    Bishop Ch. 15 — ddof=1 is mandatory.  The c4 correction
     removes the small-sample bias inherent in the sample standard
-    deviation: sigma_hat = S / alpha_N (10-12 manual Eq 10-12).
+    deviation: sigma_hat = S / alpha_N (11-7 manual Eq 9-9).
     """
     n = len(values)
     s = float(np.std(values, ddof=1))
@@ -546,8 +560,8 @@ def _process_mean(frame: pd.DataFrame, response_var: str) -> float:
     Each experimental condition counts once, whatever its cell size, as for the
     Xbar centre line and the loss function's centring term. VAS prints this as the
     capability charts' PROCESS MEAN (PM SDS 6: 237.86, where the plain average of
-    the readings is 237.834), and the 10-12 manual's one-sided indices (Eqs 16-5, 16-6,
-    16-12, 16-13) centre on it; on balanced data the two are equal.
+    the readings is 237.834), and the 11-7 manual's one-sided indices (Eqs 15-5, 15-6,
+    15-12, 15-13) centre on it; on balanced data the two are equal.
     A study without cells (no factors or time) has one cell: the plain average.
     """
     valid = frame[frame[response_var].notna()]
@@ -690,6 +704,7 @@ def assess_capability(
     potential_unavailable_reason = None
     potential_values = None
     potential_center = None
+    window_sigma_r2 = None
     potential_outside = {}
 
     if ads.r2_unavailable_reason is not None:
@@ -699,9 +714,9 @@ def assess_capability(
         n_r2 = len(r2_values)
 
         if n_r2 >= 2:
-            _, sigma_hat_r2 = compute_sigma_hat(r2_values)  # Eq 16-10
+            _, sigma_hat_r2 = compute_sigma_hat(r2_values)  # Eq 15-10
             potential_values = y_bar + r2_values
-            # Centre the potential indices on the potential values themselves (Eqs 16-11..16-13 as
+            # Centre the potential indices on the potential values themselves (Eqs 15-11..15-13 as
             # VAS computes them): y_bar + mean(R2). Tom's Medicare run of 10/3/2026 prints
             # "PROCESS MEAN 10831.3" on the potential slide (y_bar 10832.4, mean R2 -1.15).
             potential_center = float(np.mean(potential_values))
@@ -711,15 +726,27 @@ def assess_capability(
             cpk_upper = pot['ppk_upper']
             cpk = pot['ppk']
             potential_outside = compute_pct_outside(potential_values, specs)
+            if window is not None:
+                window_r2 = frame['R2'].dropna().to_numpy(dtype=float)
+                if len(window_r2) >= 2:
+                    _, window_sigma_r2 = compute_sigma_hat(window_r2)
         else:
             potential_unavailable_reason = (
                 f'Too few R2 residual values ({n_r2}) for potential capability; need at least 2'
             )
     else:
-        potential_unavailable_reason = (
-            f'R2 residuals not available (ADS {ads.analytical_design_state.sds}); '
-            f'Cp/Cpk require VAS residuals from a factorial+time design'
-        )
+        needs = (('at least one factor', ads.spec.rsg_vars_list), ('a time variable', ads.spec.time_var))
+        missing = [what for what, present in needs if not present]
+        if missing:
+            potential_unavailable_reason = (
+                f'Cp/Cpk use the VAS R2 residuals, which need {" and ".join(missing)}; '
+                f'this study has {"none" if len(missing) == 1 else "neither"}.'
+            )
+        else:
+            potential_unavailable_reason = (
+                f'R2 residuals not available (ADS {ads.analytical_design_state.sds}); '
+                f'Cp/Cpk require VAS residuals from a factorial+time design'
+            )
 
     return CapabilityResult(
         specs=specs,
@@ -757,5 +784,6 @@ def assess_capability(
         n_total=n_total,
         window_warning=window_warning,
         potential_center=potential_center,
+        window_sigma_r2=window_sigma_r2,
         observed_values=y_values,
     )

{processbehavior-0.5.0 → processbehavior-0.6.0}/processbehavior/exporters/excel_exporter.py

@@ -73,6 +73,8 @@ class ExcelExporter:
 
         Creates a multi-sheet Excel workbook with organized analysis results:
         - Summary: Analysis metadata, SDS info, signal counts
+        - Provenance: library version, time written, variables, row counts,
+          design states and a fingerprint of the analysed data (with Summary)
         - Charts: One tab per chart (Xbar, S, stratified X, etc.)
         - Residuals: R1-R5 variance decomposition (if available)
         - Effects: Main effects (if calculated)
@@ -148,6 +150,7 @@ class ExcelExporter:
             # 1. Summary Tab
             if include_summary:
                 self._write_summary_tab(writer, format_cells)
+                self._write_provenance_tab(writer, format_cells)
 
             # 2. Chart Tabs
             if include_charts:
@@ -236,6 +239,35 @@ class ExcelExporter:
             ws = writer.sheets['Summary']
             self._apply_formatting(ws)
 
+    def _write_provenance_tab(self, writer: pd.ExcelWriter, format_cells: bool) -> None:
+        """Write what produced this workbook, so a reader can tell which data and version it came from."""
+        import hashlib
+        from datetime import datetime, timezone
+
+        from .. import __version__
+
+        ads = self.result._ads
+        spec = ads.spec
+        data = ads.analysis_dataset
+        columns = [spec.response_var, *spec.rsg_vars_list, *([spec.time_var] if spec.time_var else [])]
+        row_hashes = pd.util.hash_pandas_object(data[columns], index=False).to_numpy()
+        rows = [
+            ('Library', f'processbehavior {__version__}'),
+            ('Written (UTC)', datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M:%S')),
+            ('Response', spec.response_var),
+            ('Factors', ', '.join(spec.rsg_vars_list) or '(none)'),
+            ('Time', spec.time_var or '(none)'),
+            ('Analysis type', self.result.summary['analysis_type']),
+            ('Input rows', ads.n_input_rows),
+            ('Rows analysed', len(data)),
+            ('Observed design state (ODS)', ads.observed_design_state),
+            ('Analytical design state (ADS)', ads.analytical_design_state.sds),
+            ('Data fingerprint (SHA-256)', hashlib.sha256(row_hashes.tobytes()).hexdigest()),
+        ]
+        pd.DataFrame(rows, columns=['Item', 'Value']).to_excel(writer, sheet_name='Provenance', index=False)
+        if format_cells:
+            self._apply_formatting(writer.sheets['Provenance'])
+
     def _write_chart_tabs(self, writer: pd.ExcelWriter, format_cells: bool) -> None:
         """Write tabs for each chart."""