1.1.0: audit follow-ups (gallery repeats, tutorial coverage, native usage)

[jeremymanning]

Tracking issue for the three audits run on 2026-09-05 before the 1.1.0 re-cut (repeats in the gallery, tutorial coverage of the public API, native-hypertools usage). Everything here ships in 1.1.0. Constraint: the five launch tutorials the Bluesky thread links (market_sectors, weather_decades, painting_embeddings, conversation_shape, morph_shapes_zoo) keep their pages and clips.

A. Gallery repeats and misdescriptions (examples/)

• Delete plot_shape_morph.py (strict subset of animate_morph_zoo.py); fold the "all seven shapes" note into animate_morph_zoo; retarget animate_surface_morph.py's reference
• Merge plot_TSNE.py + plot_UMAP.py + plot_digits.py into one example
• Merge plot_PPCA.py into plot_missing_data.py
• Merge animate_MDS.py into animate.py
• Merge chemtrails.py + precog.py into one trails page (keep animate_trails_mix.py)
• Merge plot_clusters.py/2/3 into one three-cell example
• Fix misdescribed docstrings: plot_story_trajectories.py (execute the pipeline for real), plot_hue.py (categorical hue), animate_surface_morph.py (changelog prose), save_movie.py (hidden align=), plot_datasets_tour.py ("five vs four")
• 'hyper' -> 'HyperAlign' in analyze.py, plot_align.py
• Gallery ordering (within_subsection_order) and the "Story trajectories" entry in docs/tutorials.rst
• Retitle conversation_trajectories and hugging_face_embeddings so they read as distinct from their siblings

B. Native usage rewrites

• load_digits -> hyp.load('digits') (three examples); load_dataset -> hyp.load(...) (streaming_data, hugging_face_embeddings)
• plot_procrustes.py: private procrustes() -> hyp.align(model='Procrustes')
• plot_gensim_text.py: text2mat -> hyp.plot(texts, vectorizer=..., semantic=...)
• plot_autoencoders.py: ax.scatter -> hyp.plot(..., ax=)
• animate_surface_morph.py: drop the hand downsample; set_alpha -> alpha= (c4ed79c + morph alpha fix)
• Remove dead sentence-transformers / pykalman import guards (six files) (examples done; tutorials in progress) now that extras install on demand
• projectile_kalman.ipynb and stock_forecasting.ipynb hand ax.plot panels -> hyp.plot(..., reduce=None, ndims=2, ax=)
• Library gap: isotropic (whole-array) option for Normalize (feefcbe: Normalize(mode='isotropic'), 21 tests), so the three morph files stop hand-rolling normalize_shape(); tests + docs
• Tighten tests/test_examples_are_native.py DEFECT_MARKERS with the signatures the audit listed (with a per-file allowlist for deliberate ax= demos)

C. Coverage: new and rebuilt tutorials (each executed, in docs/tutorials.rst, install cell, gate entry)

• hierarchy.ipynb: column-MultiIndex frame, per-sector traces + mean, names=, per-level alpha=, hierarchical predict= and return_model trace_data

• animate_forecast.py + tutorial: animate=True, predict='Kalman', forecast_trail=True, forecast_hue/palette/fmt

• io.ipynb: hyp.load of local files / URL / HF streaming / DataFrame, hyp.save round trips, save_path formats, HypertoolsIOError

• pipelines.ipynb: return_model on every dispatcher, .transform, pipeline= replay, apply_model, supported_models(), bundle keys, Pipeline directly

• manip.ipynb: hyp.manip Smooth/Resample/ZScore/Normalize and chains, animate='window' + focused=

• Rebuild plot.ipynb: saving executed, animate=, names=, colorbar=, alpha= list, size/elev/azim, labels, font=, label_alpha=, backend='plotly', impute=, resample=, exception classes

• Rebuild align.ipynb: Procrustes, SRM, align= inside plot/analyze, fitted aligner reuse, local image

• Rebuild analyze.ipynb: call every stage it advertises, hypertools figures instead of seaborn heatmaps

• Rebuild reduce.ipynb: describe curve drawn, UMAP, autoencoders, return_model

• Extend stock_forecasting.ipynb (Chronos section) and projectile_kalman.ipynb (sklearn imputers vs Kalman)

• Library gap found by the morph rewrite: alpha= does not reach the travelling morph cloud (matplotlib and plotly); fix + tests, then drop the set_alpha call in animate_surface_morph.py

D. Verification and release

• Every tutorial executed with scripts/execute_tutorial.py; measured output cells recorded in the gate
• Full suite (4118 passed), ruff, sphinx -W gallery build, Colab feature tour (both backends, fresh runtimes)
• Gallery republished from the release commit (3a9b868); artifacts rebuilt; master pushed; - [ ] CI green; v1.1.0 moved; tag CI green; draft release updated

Found while fixing (all fixed for 1.1.0)

• Glyph 8722 warning on caller-supplied ax= panels: tick labels under the sans-serif alias have no per-glyph fallback -> the ax= path sets the tick-label font list (217a2a6)
• hyp.io.lsl_stream() left its inlet open -> LSLStream destroys it on close/with/GC/exit (61f3e7e)
• PPCA divide by zero encountered in log on wide data -> slogdet (54b4b2e)
• hyp.load(DataFrame) raised TypeError -> passthrough for loaded data (a8d073a)
• hyp.plot(x, pipeline=<hand-built Pipeline with raw sklearn stages>) raised -> fit-or-transform replay of raw steps, Pipeline.is_fitted (3340595)
• DEFECT_MARKERS tightened with the audit signatures (6b6bcbc); not added: weather's trailing rolling(12).mean() (hypertools Smooth is centred; visual rewrites deferred to 1.2 per maintainer)

Also fixed on the way (not in the original audit): HF vectorizers crashing with the default semantic= (#285, 1b34b1d), pipeline replay of hand-built Pipelines (3340595), LSL inlet destroy (bb07321), docs-gate literal for the market notebook (d724181), gensim example title overline, stale gallery output pages.