fix(examples): gn-bundle-roundtrip exit-5 falsifier and re-verified version table

Exit 5 had no falsifier. --mismatch-mark renames only Separate's Float item, so count and bbox pass and bundle_mark reads 0.0, exit 5. The README now states the RNA per version as a table, re-verified by probe on 4.5.11, 5.0.1, 5.1.2 and 5.2.1: GeometryNodeCombineBundle is undefined on all of 5.x, not only 5.2; on 4.5 it exists with the experimental flag off and evaluates empty. --force-run on 4.5 is documented as exit 1, not under exit 3. Assertions and exit codes unchanged.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Signed-off-by: TMHSDigital <154358121+TMHSDigital@users.noreply.github.com>

Author: TMHSDigital

examples/gn-bundle-roundtrip/README.md

@@ -4,14 +4,35 @@ Combine Bundle / Separate Bundle packing a cube plus Scale, Offset, and
 Mark — witnessing [`geometry-nodes-python`](../../skills/geometry-nodes-python/SKILL.md)
 item-name round-trip, not tree structure.
 
-5.x RNA is `NodeCombineBundle` / `NodeSeparateBundle`.
-`GeometryNodeCombineBundle` is the **4.5 experimental** id and is undefined
-on 5.2. 4.5 LTS keeps the old names behind
-`preferences.experimental.use_bundle_and_closure_nodes` (default **off**);
-unflagged evaluation is empty. This example skips on 4.5
-(`SMOKE_SKIP: Bundles require Blender 5.0+`, exit 77, catalog
-`min_version` 5.0). `--force-run` bypasses the skip and creates the 5.x
-RNA so 4.5 fails for that reason — not a hidden pass.
+## Which RNA, on which Blender
+
+| Blender | Combine / Separate RNA | Legacy `GeometryNodeCombineBundle` | This example |
+| --- | --- | --- | --- |
+| 4.5 LTS | not defined | defined, but evaluates **empty** unless `preferences.experimental.use_bundle_and_closure_nodes` is on (default **off**) | skips, exit 77 |
+| 5.0, 5.1, 5.2 LTS | [`NodeCombineBundle`](https://docs.blender.org/api/current/bpy.types.NodeCombineBundle.html) / [`NodeSeparateBundle`](https://docs.blender.org/api/current/bpy.types.NodeSeparateBundle.html) | **undefined** — `nodes.new` raises `RuntimeError: Node type GeometryNodeCombineBundle undefined` | runs |
+
+If you are porting a 4.5 script to 5.x and `nodes.new("GeometryNodeCombineBundle")`
+raises "undefined", the 5.x id has no `GeometryNode` prefix: use
+`NodeCombineBundle` / `NodeSeparateBundle`. On 4.5 the reverse holds: `NodeCombineBundle` is
+undefined, and the legacy node can be created with the experimental
+flag off — it just evaluates to nothing, which is the easy trap.
+4.5 API pages: [`GeometryNodeCombineBundle`](https://docs.blender.org/api/4.5/bpy.types.GeometryNodeCombineBundle.html),
+[`PreferencesExperimental`](https://docs.blender.org/api/4.5/bpy.types.PreferencesExperimental.html).
+
+Every cell above was re-verified by probe on **4.5.11 LTS, 5.0.1, 5.1.2
+and 5.2.1 LTS**. On 4.5.11 the flag exists and reads `False`; a
+legacy Combine → Separate carrying a cube evaluates to 0 verts / 0 faces
+with it off and 8 / 6 with it on. On 5.0.1, 5.1.2 and 5.2.1 the flag is
+gone from `PreferencesExperimental`.
+
+The catalog floor is `min_version` 5.0 and the example prints
+`SMOKE_SKIP: Bundles require Blender 5.0+` (exit 77) below it.
+`--force-run` bypasses the skip and creates the 5.x RNA, so 4.5 fails
+for that reason — `NodeCombineBundle` undefined, caught by the FATAL
+wrapper as exit 1 — not a hidden pass. On 5.x `--force-run` changes
+nothing and exits 0.
+
+## What it witnesses
 
 Closed form (1 m cube, Scale=2, Offset=`(1.5, 0, 0)`, Mark=`0.314159`):
 
@@ -21,7 +42,9 @@ Closed form (1 m cube, Scale=2, Offset=`(1.5, 0, 0)`, Mark=`0.314159`):
 
 **Count alone is not enough.** A cube that never entered the bundle is
 still 8/6 (`--bypass`). Bbox from the unpacked Scale/Offset is the second
-axis; the named attribute from the unpacked Float is the third.
+axis; the named attribute from the unpacked Float is the third. Each
+axis has its own falsifier, and each falsifier leaves the axes before
+it green, so it fails on the axis it names.
 
 No gallery still. The contract is item names + typed round-trip; the
 visible mesh is one cube and the attribute is invisible. Failure would
@@ -31,17 +54,19 @@ not read at thumbnail.
 
 - exit 77 — Blender < 5.0 and not `--force-run`
 - exit 2 — `GeometryNodeCombineBundle` on 5.x (`--legacy-rna`)
-- exit 3 — pairing/names broken (`--mismatch` → 0/0; `--force-run` on 4.5
-  raises `NodeCombineBundle` undefined, caught as exit 1)
-- exit 4 — count matches but unpacked Scale/Offset did not (`--bypass`,
-  `--pack-scale 1`)
-- exit 5 — bbox matches but `bundle_mark` did not travel through the bundle
+- exit 3 — item names do not pair, so Separate yields no geometry
+  (`--mismatch` renames the geometry item → 0/0)
+- exit 4 — count matches but the unpacked Scale/Offset did not arrive
+  (`--bypass`, `--pack-scale 1`)
+- exit 5 — count and bbox match but the Float item did not travel
+  (`--mismatch-mark` renames only the Float item → `bundle_mark` 0.0)
 
 ## Run
 
 ```bash
 blender --background --python gn_bundle_roundtrip.py --
 blender --background --python gn_bundle_roundtrip.py -- --force-run
+blender --background --python gn_bundle_roundtrip.py -- --mismatch-mark
 ```
 
 ## Exit codes
@@ -52,14 +77,26 @@ against it. `77` is the smoke skip protocol, not a product check.
 | Code | Meaning |
 | --- | --- |
 | 0 | Success |
-| 1 | Uncaught exception (FATAL wrapper) |
+| 1 | Uncaught exception (FATAL wrapper); `--force-run` on 4.5 lands here |
 | 2 | argparse / usage; carrier mesh rewritten; `--legacy-rna` on 5.x |
 | 3 | Evaluated vert/face count off closed form (`--mismatch`) |
-| 4 | Unpacked Scale/Offset x-extent off (`--bypass`) |
-| 5 | `bundle_mark` missing or off closed form |
+| 4 | Unpacked Scale/Offset x-extent off (`--bypass`, `--pack-scale 1`) |
+| 5 | `bundle_mark` missing or off closed form (`--mismatch-mark`) |
 | 77 | `SMOKE_SKIP:` Bundles require Blender 5.0+ |
 
+Measured on 4.5.11 / 5.0.1 / 5.1.2 / 5.2.1:
+
+| Flag | 4.5.11 | 5.0.1 | 5.1.2 | 5.2.1 |
+| --- | --- | --- | --- | --- |
+| (none) | 77 | 0 | 0 | 0 |
+| `--force-run` | 1 | 0 | 0 | 0 |
+| `--bypass` | 77 | 4 | 4 | 4 |
+| `--legacy-rna` | 77 | 2 | 2 | 2 |
+| `--mismatch` | 77 | 3 | 3 | 3 |
+| `--mismatch-mark` | 77 | 5 | 5 | 5 |
+| `--pack-scale 1` | 77 | 4 | 4 | 4 |
+
 The `blender-smoke` workflow runs the check on Blender 5.2 LTS (5.1 on the
 weekly cron, the `needs-5.1` PR label, or manual dispatch) and skips on 4.5
-LTS. Smoke does not pass `--bypass`, `--legacy-rna`, `--mismatch`, or
-`--force-run`.
+LTS. Smoke does not pass `--bypass`, `--legacy-rna`, `--mismatch`,
+`--mismatch-mark`, or `--force-run`.

examples/gn-bundle-roundtrip/gn_bundle_roundtrip.py

@@ -3,7 +3,7 @@
 Witnesses Combine / Separate Bundle from ``skills/geometry-nodes-python``.
 The load-bearing RNA on 5.x is ``NodeCombineBundle`` / ``NodeSeparateBundle``,
 not ``GeometryNodeCombineBundle`` (that id is 4.5 experimental and
-**undefined** on 5.2). Tree-structure checks are vacuous: the nodes can
+**undefined** on 5.x). Tree-structure checks are vacuous: the nodes can
 exist and be linked while Separate looks up the wrong item names and
 evaluates empty.
 
@@ -23,6 +23,7 @@
 bypasses the skip and uses the 5.x RNA so 4.5 fails for that reason.
 
     blender --background --python gn_bundle_roundtrip.py --
+    blender --background --python gn_bundle_roundtrip.py -- --mismatch-mark
 """
 import argparse
 import sys
@@ -61,7 +62,7 @@ def eval_mesh(obj):
     return coords, faces, marks
 
 
-def build_tree(pair=True, match_names=True, pack_scale=PACK_SCALE):
+def build_tree(pair=True, match_names=True, pack_scale=PACK_SCALE, match_mark=True):
     tree = bpy.data.node_groups.new("BundleRoundTrip", "GeometryNodeTree")
     tree.interface.new_socket(
         name="Geometry", in_out="OUTPUT", socket_type="NodeSocketGeometry",
@@ -84,7 +85,10 @@ def build_tree(pair=True, match_names=True, pack_scale=PACK_SCALE):
     sep.bundle_items.new("GEOMETRY", sep_mesh)
     sep.bundle_items.new("FLOAT", "Scale")
     sep.bundle_items.new("VECTOR", "Offset")
-    sep.bundle_items.new("FLOAT", "Mark")
+    # --mismatch-mark renames only the Float item: the mesh, Scale and Offset
+    # still round-trip, so count and bbox pass and only the mark is lost.
+    sep_mark = "Mark" if match_mark else "Marker"
+    sep.bundle_items.new("FLOAT", sep_mark)
 
     tree.links.new(cube.outputs["Mesh"], comb.inputs["Mesh"])
     comb.inputs["Scale"].default_value = pack_scale
@@ -106,19 +110,22 @@ def build_tree(pair=True, match_names=True, pack_scale=PACK_SCALE):
     tree.links.new(xyz.outputs["Vector"], xf.inputs["Scale"])
     tree.links.new(sep.outputs["Offset"], xf.inputs["Translation"])
     tree.links.new(xf.outputs["Geometry"], store.inputs["Geometry"])
-    tree.links.new(sep.outputs["Mark"], store.inputs["Value"])
+    tree.links.new(sep.outputs[sep_mark], store.inputs["Value"])
     tree.links.new(store.outputs["Geometry"], go.inputs["Geometry"])
     return tree
 
 
-def build(pair, match_names, pack_scale):
+def build(pair, match_names, pack_scale, match_mark=True):
     bpy.ops.wm.read_factory_settings(use_empty=True)
     me = bpy.data.meshes.new("Carrier")
     me.vertices.add(1)
     ob = bpy.data.objects.new("Carrier", me)
     bpy.context.collection.objects.link(ob)
     mod = ob.modifiers.new("GN", "NODES")
-    mod.node_group = build_tree(pair=pair, match_names=match_names, pack_scale=pack_scale)
+    mod.node_group = build_tree(
+        pair=pair, match_names=match_names, pack_scale=pack_scale,
+        match_mark=match_mark,
+    )
     bpy.context.view_layer.update()
     return ob
 
@@ -194,6 +201,11 @@ def main():
         action="store_true",
         help="falsification: Separate item name Geom vs Combine Mesh",
     )
+    p.add_argument(
+        "--mismatch-mark",
+        action="store_true",
+        help="falsification: Separate Float item Marker vs Combine Mark",
+    )
     p.add_argument("--pack-scale", type=float, default=PACK_SCALE)
     p.add_argument(
         "--legacy-rna",
@@ -224,6 +236,7 @@ def main():
         pair=not args.bypass,
         match_names=not args.mismatch,
         pack_scale=args.pack_scale,
+        match_mark=not args.mismatch_mark,
     )
     code = check(ob)
     if code: