feat: add eval-mesh-datablock-name and mesh-automasking-settings examples

[TMHSDigital]

Summary

Two pathology cross-version breakage examples, plus a ROADMAP correction of the survey's Claim C (direction was wrong) and a CONTRIBUTING note so version-gated falsifiers are not "fixed" to fail on every binary.

Framing deviation contract and #138 stay queued. Do not merge this PR from the authoring session.

• examples/eval-mesh-datablock-name/ — pathology, no gallery still.
• examples/mesh-automasking-settings/ — pathology, no gallery still.

Authoring-gate conflict (tracked, not fixed here)

docs/new-example-prompt.md wants a render for every example. CLAUDE.md pathology/sidecar examples omit gallery by design. Both examples follow CLAUDE.md. Issue filed as part of this phase (number in Task 5 below).

Claim C, restated (live-run-proven)

The survey said to_mesh().name no longer equals the source. Binaries say the opposite. Witness is evaluated_get().data.name. Source mesh named SourceMesh, SUBSURF levels=1 (8 vs 26 verts).

Accessor	4.5.11	5.1.2	5.2.1
obj.data.name	SourceMesh	SourceMesh	SourceMesh
evaluated_get().data.name	Mesh	Mesh	SourceMesh
to_mesh().name	SourceMesh	SourceMesh	SourceMesh

Code that treats name inequality as "this is evaluated" is silently wrong on 5.2. Nothing raises.

ROADMAP one-line correction is commit 1 of this PR.

Claim D (live-run-proven, no restatement)

	4.5.11	5.1.2	5.2.1
bpy.types.MeshAutomaskingSettings	absent	absent	present (20 RNA props)
Old Brush.use_automasking_* / Brush.automasking_*	present	present	gone
.mesh_automasking_settings	absent	absent	on Brush / Sculpt / Paint / …

Factory-empty has zero brushes. bpy.data.brushes.new("ProbeBrush", mode="SCULPT") yields the relevant attrs. mode="SCULPT" is load-bearing on 5.2: default-mode new(name) leaves mesh_automasking_settings is None.

Subset (not all 18+): use_automasking_topology, use_automasking_cavity (same identifier after the move), cavity factor (Brush.automasking_cavity_factor vs MeshAutomaskingSettings.cavity_factor). Covers the location move and the nested identifier shortening.

Exit codes (live-run-proven)

Local binaries: 4.5.11 LTS 4db51e9d1e1e · 5.1.2 ec6e62d40fa9 · 5.2.1 LTS 9e2066aef7ef.

Example	Flag	4.5.11	5.1.2	5.2.1
eval-mesh-datablock-name default	—	0	0	0
eval-mesh-datablock-name falsifier	--assume-distinct-names	0	0	6
mesh-automasking-settings default	—	0	0	0
mesh-automasking-settings falsifier	--assume-brush-attrs	0	0	5

Each falsifier is red only on 5.2.1, and exits 0 on 4.5.11 and 5.1.2 by design (the old API / old naming still works). Same class as --assume-present / --legacy-ids from #145. Unlike --same-axis.

Task 4

CONTRIBUTING.md § Exit codes, Version-gated falsifiers subsection. Lists the four examples.

Task 5

#146 — docs: new-example-prompt requires a render that the pathology convention forbids. Not fixed in this PR.

Shape (inspection of #145, then live-run of these two)

Scaffolding from cross-version-property-delete (check() returns, argparse naive-API flag, FATAL wrapper). Version gates from gn-modifier-inputs (bpy.app.version tuple). No third approach. No gallery_framing.py, no deviation=, no gallery entries.

Inventory (inspection)

README / plugin.json / catalog / CLAUDE.md / AGENTS.md 57→59. catalog.json rows for the two new examples only. AGENTS.md example count was stale at 55 (missed in #145); brought in line with README as the new examples landed.

Test plan

• needs-5.1 present and Blender 5.1 smoke job appears
• Validate / manifest / counts
• Blender 4.5 smoke
• Blender 5.2 smoke
• Blender 5.1 smoke (label)
• Socket Security
• Do not merge from this session