Skip to content

NotebookWalkAudit: notebooks and the documentation tree follow the walk rename - #23

Merged
p135246 merged 3 commits into
mainfrom
work/notebook-walk-audit
Sep 10, 2026
Merged

NotebookWalkAudit: notebooks and the documentation tree follow the walk rename#23
p135246 merged 3 commits into
mainfrom
work/notebook-walk-audit

Conversation

@p135246

@p135246 p135246 commented Sep 10, 2026

Copy link
Copy Markdown
Contributor

Closes the NotebookWalkAudit work item, split out of WalkFamilyRefactor on 2026-08-19 at the user's request: the refactor was not to touch the notebooks, so the sweep became an item of its own.

Three commits, rebased onto main at 0.13.30. Suite 1404/1404 in a fresh kernel.

T2 — the four tutorials (a2f3cc8)

PathSpace, EuclideanGeometry, ProjectiveGeometry and TropicalConvexity carried code the refactor waves had broken: SelectInfraPath / SelectInfraCycle calls, UniquePencilQ, and count-less calls authored under the repealed count-less = All convention. Every Input and Code cell of all four now evaluates in a fresh kernel with zero messages. PathSpaceTutorial's cycles section moved from the shell 2 to the band {2, 3}FindInfraCircle[GridGraph[{5, 5}], 13, 2, All] is empty at every radius, so three cells were vacuous.

T2b — nothing tropical (7ba7821)

TropicalConvexityGuide.nb and TropicalConvexityTutorial.nb deleted with their four cross-references. Both were built pages with no docs/ source, documenting four MetricAlgebra.wl symbols under a tropical title, which is why no audit had caught either. The user's call: the paclet carries nothing tropical. No symbol lost documentation.

T3 — the documentation tree (3a28a4c)

The tree had never followed the PathWalk rename. It now does, and the three sets coincide for the first time: 187 exported symbols, 187 docs/Symbols/*.md sources, 187 built reference pages. Metadata verdict 188 Build (the 187 plus EuclideanGeometryGuide), 14 Skip — the hand-made guides and tutorials, as always.

  • 10 stale pages deleted with their nine sources: $InfraPathColor, ConcatenateInfraPath, ExtendInfraPath, FindInfraPath, InfraPath, InfraPathQ, SelectInfraCycle, SelectInfraPath, UniquePencilQ, and FindForwardDeformation (page only).
  • 17 stubs built, not the sixteen the inventory counted. $InfraStrikeOutPalette was the seventeenth gap and is the one exported symbol with no ::usage, so its page carries only the generator's fallback line.
  • InfraDeformationSize and InfraDensity built from sources they already had.
  • Cross-references were the larger half. InfraWrapperOntology (refactor(shapes)!: the shape is the kind — delete every Infra* payload wrapper #22) deleted eight wrapper pages and left every reference standing, so the tree carried ten dead SeeAlso entries and nine dead [X]() links across fourteen sources. All ten SeeAlso entries and six of the nine links are fixed: InfraMeasure and InfraEffectivePointInfraDensity, InfraSet → a vertex List or a density, InfraString → the free loop, UniquePencilQUniqueCollinearQ, InfraPath → a walk. EuclideanGeometryGuide's walk vocabulary and predicate hierarchy follow the rename.

What this does not close

  • Three dead links stayInfraLine.md:36, InfraPoint.md:41, InfraSegment.md:49. Each sits inside accessor prose the shape ontology deleted (line[[i]], ["BallVolumes", g] on a set, seg["Start"]), and FindInfraSegment[…, All] now returns a Graph whose seg["Start"] is unevaluated, so no substitution leaves a true sentence.
  • 43 of the 187 sources still describe the wrapper ontology. That is a page rewrite, not a rename.
  • $InfraStrikeOutPalette needs a usage message. It is ColorData[112, "ColorList"], the default colour cycle for strike-out marks.
  • GeodesicExtensionGraph::usage says "wrapper anchors give one DAG per pair". There are no wrappers: a density anchor spreads — infraSpread[<|1 -> 1, 3 -> 1|>] is {1, 3} — and a bare vertex List does not, infraSpread[{1, 3}] being {{1, 3}}. Filed as the fourth row of APISymbolAudit § 1g.

🤖 Generated with Claude Code

p135246 and others added 3 commits September 10, 2026 12:12
NotebookWalkAudit T2. Four of the eight tutorials carried code the refactor
waves had broken. They are hand-made .nb files with no Markdown source, skipped
by build_docs.wls, so they are edited in place.

PathSpaceTutorial: four SelectInfraPath calls and one SelectInfraCycle become
SelectInfraWalk; "ShortestCircumference" remaps to "MinLength", since folding
the cycle head in dropped the cycle-specific "From" values and the cyclic
bundle sets "Cyclic" -> True itself. Length[paths] under a "Cardinality" label
becomes paths["Multiplicity"] -- All returns the geodesic DAG, whose Length is
1. The cycles section ran on FindInfraCircle[g2, 13, 2, All], empty at every
radius on the 5x5 grid, so three cells were vacuous and the selector returned
$Failed; the band {2, 3} has exactly one separating cycle and keeps both the
substrate and the point.

EuclideanGeometryTutorial: the segment and the three triangle sides take All,
which is what the prose already claimed ("the set of all geodesics", "rendered
with intensity following multiplicity"); the enumeration cell reads
seg["Realizations"] instead of repeating the call it now duplicates.

ProjectiveGeometryTutorial: UniquePencilQ, deleted in PR #21, is written out as
its own definition over UniqueCollinearQ.

TropicalConvexityTutorial: Method -> "Weak", an option SegmentHullQ has never
carried, is dropped, and the CodeText that called a grid row "only weakly
convex" is corrected -- SegmentHullQ[g, {1, 2, 3, 4}] is True. The section
title still overstates the contrast; open in the item's Hand-off.

Every Input and Code cell of all four re-run in a fresh kernel against this
branch, each cell its own top-level evaluation so the three-strike message
limit could not hide a repeat: zero messages.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
User's call, 2026-09-08: "we should not have anything tropical."

The tropical kernel layer was removed on 2026-05-10, twice over. Tropical
{Postulates,Predicates}.wl was retired -- four of its six symbols were
rebrandings (FindTropicalSegment was Sort /@ FindInfraSegment realisations,
TropicalT1Q was ConnectedGraphQ) -- and FindSegmentHull / SegmentHullQ were
lifted into MetricAlgebra.wl as graph-metric notions, "not specifically
tropical". Kernel/TropicalOperations.wl was demoted to Code/ the same day. No
kernel file has contained the string since; there is no tropical export and no
tropical test.

What survived was packaging: TropicalConvexityGuide.nb and
TropicalConvexityTutorial.nb, both built pages with no docs/ source, so
build_docs.wls could neither regenerate nor check them, and both documenting
four MetricAlgebra symbols under a tropical title. NotebookWalkAudit T1 diffed
symbol pages against the export set; guides and tutorials were outside that
diff, which is also how the tutorial's dead Method -> "Weak" survived from
2026-08-20 to this morning.

Both pages deleted. FindSegmentHull, SegmentHullQ, MetricInterval and
MedianVertices keep their docs/Symbols sources and their built reference pages,
so no symbol loses documentation -- only the guide-level grouping goes, and
Wiki/Index.md already files those four under Metric Geometry.

Cross-references removed with them: two See-also items and the
Maclagan-Sturmfels tropical citation in SyntheticInfrageometryGuide.nb, one in
PathSpaceTutorial.nb, and "tropical convexity (graph-only)" from the top guide's
layer list. The guide/tutorial pairing it claims still holds at 7 and 7. Nothing
in Documentation/ mentions tropical any more, and PacletInfo.wl lists no page
manifest, so the tree is self-consistent.

Breaking: two documentation pages are gone from the paclet.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The three sets now coincide: 187 exported symbols, 187 docs/Symbols/*.md
sources, 187 built reference pages. Metadata verdict 188 Build (the 187 plus
EuclideanGeometryGuide), 14 Skip -- the hand-made guides and tutorials, as
always. Suite 1404/1404 in a fresh kernel.

Run in the order the link rule forces -- stubs, then deletes, then build:

- 17 stubs, not the sixteen T1 counted. $InfraStrikeOutPalette was the
  seventeenth gap and is the one exported symbol with no ::usage, so its page
  carries only the generator's fallback line (leaving the stub was the user's
  call).
- 10 stale pages deleted with their nine sources: $InfraPathColor,
  ConcatenateInfraPath, ExtendInfraPath, FindInfraPath, InfraPath, InfraPathQ,
  SelectInfraCycle, SelectInfraPath, UniquePencilQ, and FindForwardDeformation
  (page only).
- InfraDeformationSize and InfraDensity built from sources they already had.

Cross-references were the larger half. InfraWrapperOntology T8 deleted eight
wrapper pages and left every reference standing, so the tree carried ten dead
SeeAlso entries and nine dead [X]() links across fourteen sources. All ten
SeeAlso entries and six of the nine links are fixed: InfraMeasure and
InfraEffectivePoint -> InfraDensity, InfraSet -> a vertex List or a density,
InfraString -> the free loop, UniquePencilQ -> UniqueCollinearQ, InfraPath ->
a walk. EuclideanGeometryGuide's walk vocabulary and predicate hierarchy
follow the rename.

Three links stay: InfraLine.md:36, InfraPoint.md:41 and InfraSegment.md:49
carry theirs inside accessor prose the ontology deleted -- line[[i]],
["BallVolumes", g] on a set, seg["Start"] -- and FindInfraSegment[..., All]
now returns a Graph whose seg["Start"] is unevaluated, so no substitution
leaves a true sentence. 43 of the 187 sources still describe the wrapper
ontology; that is a page rewrite, filed in the item's Hand-off.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@p135246
p135246 merged commit 3a28a4c into main Sep 10, 2026
@p135246
p135246 deleted the work/notebook-walk-audit branch September 10, 2026 18:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant