samples: native-bughunt repros cited by the upstream draft (split from #293, 6/6)

[MendixMau]

Split from #293 (6/6). #293's independent review said HOLD, partly because of its 476 sample files (~65k lines). This PR now keeps 4 of them: only the files that something reads.

Direct lane

What changed and why: This PR adds pipelines/mendix-mpr/samples/fsm-template-9.24.2/native-bughunt/: CAPTURE.txt and three minimal repros, D1–D3. The upstream draft in #344 (bug-logs/pending-github-issues/2026-10-12-mxcli-0.26-round-trip-report.md) cites them as the repros for:

• U3: describe entity prints index ( desc); with no column.
• U2: the native Repeater's content slot.
• U14: no microflow or nanoflow can be written to a 9.24.2 project.

The repros are written from scratch in a Repro module and hold no model content from the template. That is why no licence note is needed. The D4 excerpt, which was template content, is not shipped, and CAPTURE.txt now says so.

Dropped, all 472 other sample files of #293. None of them has a citing reader. The field results are stated inline in #346 (skills) and #342 (rehost), and both of those splits say captures are not committed. Dropped in full:

• FSM increments: every CAPTURE.txt and FINDINGS.md for seam-probe, slice-nativenote, slice-assets, slice-assets-core, slice-assets-strangled and slice-dispatch.
• cut-score-findings.md and web-slice-candidates.md.
• Every cut, in-edge and fidelity ledger, every cut-rules*.tsv, crossing-refs.tsv and slice.txt.
• All captures/, mdl/, tools/, journey/ and maps/ trees. These include the ad-hoc fix-*.py, bson-*.py and journey .js scripts, and the screenshots.
• The OData metadata XMLs, dependencies.json, app-layer-map.json, graph-report.md, list-modules.md, cut-score.json and target-architecture.md.
• All GenAI Showcase captures (run 4 and nightly).
• All 59 mpk-export-probe/captures.

A subset of the slice-nativenote ledger inputs survives as the cut-ledger fixture (tests/wave2/fixtures/cut-ledger/, its own split).

Process pages (#293 added 5), all left out. No sample, skill or split branch cites them: process/decompose-and-rehost-overview-2026-10-12.html, process/decompose-and-rehost-plan-2026-10-12.md, process/decompose-and-rehost-report-2026-10-12.html, process/mxcli-nightly-retest-2026-10-13.html, process/rehost-vs-mpk-2026-10-13.html.

Size: 5 files changed, +37 lines. That is 36 lines of samples and 1 CHANGELOG line. The branch history has earlier, larger commits; squash-merge drops them.

Field evidence: D1–D3 were each run alone against a fresh mxcli new Scratch --version 9.24.2 --theme none --skip-init app, with mxcli v0.26.0, on 2026-10-12. The verbatim errors are in the U2, U3 and U14 entries of the draft. Nothing here is an instrument.

• No client data (leak guard clean; "Siemens" neutralised to "a public Marketplace template"; whole-word grep for the denylist terms is clean)
• Size cap: 5 files, 37 lines
• Test tier: T0. Ran render-routing.sh --check (baseline 75038 words before and after, budget 80000), check-scripts, check-portability, check-no-client-data, check-no-private-citations and check-pr-discipline origin/master, all clean. No fixtures target these paths.
• CHANGELOG.md line appended, credited

Impact

Row	Effect	Evidence
Process	None.	diff
Dependencies	No shared files (samples only). Merges clean onto master 868b3cb (git merge-tree). No conflicts with the other PRs in this batch. Cited by #344's upstream draft. Platform: n/a. Reads no artifact.	git merge-tree vs 868b3cb and pairwise
Context cost	Unchanged: baseline 73,522 → 73,522 words (budget 80,000); Stage 5 mdl 30,185 → 30,185, lead 25,820 → 25,820	bin/render-routing.sh --check on master 868b3cb vs this head (4d4f442) merged onto 868b3cb via git merge-tree in a temporary worktree
Behaviour	n/a (sample MDL repros; nothing routed or loaded).	diff
Runtime	None.	diff

🤖 Generated with Claude Code

https://claude.ai/code/session_01VJgWP5vEoAsNsJYqCDMGNw