docs(mindmap): #568 document Freeplane 1.13.2 probe + disconnected-session factor

[jsboige]

What

Append a section 3.6 verification update to docs/investigations/2026-06-21-mindmap-reliability-without-rdp.md (po-2023's #568 investigation), recording a 2026-06-22 follow-up probe by po-2024 on Freeplane 1.13.2 (latest; po-2023 tested 1.12.x).

Tracks #568 (Freeplane headless c.export() export — remove the FreeMind SendKeys / active-foreground dependency).

Why

po-2023's section 3.2 de-risk was on 1.12.x. Before investing in the real work (native Freeplane serialization, section 3.4 step 1 — multi-tick + visual QA), I re-ran the c.export() probe on the current latest build (1.13.2) to confirm the blocker still holds. The probe was inconclusive but surfaced a genuinely new environmental factor the doc didn't have.

Setup deployed (persistent — benefits a future run)

• Freeplane 1.13.2 installed (%LOCALAPPDATA%\Programs\Freeplane\, AMEND mandate: install missing tools rather than HOLD).
• auto.properties (6 permission booleans) + export_to_svg.groovy written to %APPDATA%\Freeplane\{1.12.x,1.13.x}\ — section 3.4 step 3 now partially done.

Findings

Test	Result
freeplane.exe -S -R<groovy> (GUI, session "Déco")	❌ no .export_done marker after 90s (process still alive); logs empty
freeplaneConsole.exe -N (headless)	❌ hang (confirms section 3.2: headless breaks render)

Inconclusive on the core blocker. The probe does not isolate whether section 3.3 (Freeplane rejects FreeMind-format .mm) still holds on 1.13.2 — the script could not be confirmed executed (output hidden in Freeplane's internal console).

New environmental factor (the real contribution). Section 3.1 claims "RDP can be disconnected/inactive, no active desktop needed". My probe failed on a disconnected ("Déco") session, while po-2023's section 3.2 success was likely on a connected session. This suggests c.export() may require a connected (live) desktop, not merely an existing one — i.e. "disconnected" ≠ "connected-inactive". That refines the #568 viability claim and is worth confirming before the native-serialization investment.

Conclusion for #568

The c.export() path remains hard / env-demanding; native Freeplane serialization (section 3.4 step 1) is still the real work (multi-tick + ai-01/jsboige visual QA — engine ≠ Batik). Did not regress to XSLT (section 3.5: forbidden, = revert of #184). Next attempt: re-test on a connected session to isolate section 3.3 vs the session factor before investing in native serialization.

Verification

• Doc-only change (no code, no tests, no prod mutation).
• Table style matches po-2023's existing section 3.2 table in the same file (no markdownlint config / not CI-enforced — MD060 IDE warning is cosmetic and pre-existing in the document's style).

Context

Cluster: ai-01 + po-2024 (po-2023 lost). This is verify-before-code / No-Pendulum: rather than fabricate a serializer PR on an unconfirmed premise (section 3.3 un-isolated, env flaky), document the genuine de-risk result so the next agent on #568 has the 1.13.2 data point + the session factor + the deployed permissions. Same verification-update pattern as PR #576 (#457 document-tier audit).

[clusterManager-Myia]

[NanoClaw] — docs review of #577

PR #577 appends a single section (§3.6, 21 lines added, 0 deletions) to docs/investigations/2026-06-21-mindmap-reliability-without-rdp.md, documenting po-2024's 2026-06-22 follow-up probe of the §3.2 de-risk on Freeplane 1.13.2 (po-2023 tested 1.12.x). The addition delivers exactly what the title claims: a 1.13.2 data point and a genuinely new "disconnected ≠ connected-inactive" environmental factor. I cross-checked every internal reference — §3.1 (RDP-disconnect claim), §3.2 (headless-breaks-render + GUI-works findings), §3.3 (FreeMind .mm rejection blocker), §3.4 step 3 (auto.properties deployment), §3.5 (XSLT forbidden by #184) — all are cited accurately and consistent with the existing document text. The linked issue #568 is open and tracks exactly this investigation.

The hedging is appropriately calibrated: the probe is honestly reported as INCONCLUSIF (script execution not confirmed — output trapped in Freeplane's internal console, same caveat as §3.2), and the new session-factor hypothesis is flagged as "à confirmer" rather than asserted. The author correctly resists overclaiming both the §3.3 status on 1.13.2 and the session factor, and explicitly notes no XSLT regression (respects §3.5/#184). No placeholder text, no copy-paste artifacts, table format matches §3.2. Scope is clean: docs-only, single file, no stray code or config changes.

Docs are accurate, well-structured, and the verify-before-code posture (document the inconclusive probe rather than fabricate a serializer PR on an un-isolated premise) is exactly right for #568. No blockers.

Tagging @jsboige for merge decision.