docs(dnn): #131 regen staging runbook (prereqs, NO launch)

[jsboige]

What

Stages every scope-neutral prerequisite for an attended mindmap/PDF regen so the run itself is a one-command affair — but launches nothing. Closes the SECONDAIRE of ai-01 dispatch msg-20260619T050902: "stager prérequis régén (sans lancer)".

The actual regen is GATED jsboige (scope arbitration 4-vs-8 lang + Windows foreground-lock → run = jsboige RDP / ai-01 Opus, not the worker).

Why staging (not execution)

Two blockers prevent the worker from running the regen:

1. Scope unresolved — jsboige has not published the 4-lang (preserve es/ar/fa/zh) vs 8-lang (uniformise) choice (ai-01: "4 cycles sans publication", dashboard decision Bump elliptic from 6.5.3 to 6.5.4 in /DNNPlatform/Portals/0-System/Skins/Bootstrap 4 Instant #12 contradicts the interactive session).
2. Foreground-lock — FreeMind SendKeys SVG export fails unattended (memory feedback-mindmap-freemind-unattended-foreground); needs an attended desktop.

Staging the prereqs means: when jsboige publishes the scope + opens an attended session, the run is ready. This doc does not pick the scope and does not launch anything.

Contents

§	Content
§1	Prereq table — what is staged vs deferred (scope-neutral staged; the one scope-dependent variable deferred)
§2	FreeMind env-var contract (ARGUMENTUM_FREEMIND_PATH, already merged #536) + foreground-lock HARD blocker
§3	Mindmapper-isolation order rationale — mindmaps before PDFs (debugging optimization, not data dependency) + #536 review discipline (never commit permanent Mode default change)
§4	Harvest cache clobber contract (anti-stale, memory-grounded: net9.0-windows path, recurring Mémo Back failures 2026-06-07/08)
§5	Scope-dependent variable DEFERRED to jsboige — 4-lang preserve vs 8-lang uniformise, contradiction documented, choice not made
§7	7-item pre-flight checklist (what "staged" means)

Scope-neutrality (HARD)

This doc describes prereqs that are identical for both scopes and explicitly defers the one scope-dependent step (which language directories to clobber/regen) to the jsboige decision. It does not treat anything here as picking 4 or 8 languages.

Anti-dup

• No existing regen-staging doc in docs/dnn-localization/ (checked — the 140-qa-verification-matrix and validation-guide.fr.md cover different topics).
• The FreeMind env-var plumbing is already merged (chore(config): FreeMindPath + Mindmapper-isolate for attended 8-lang mindmap regen (SCOPE PENDING) #536) — referenced, not re-added.
• The harvest-clobber contract is memory-grounded (feedback-stale-harvest-regen), consolidated into a doc so it survives session boundaries.

Gate boundaries

• ✅ Doc-only, non-gated, no launch.
• ❌ Does not run any regen (execution GATED jsboige).
• ❌ Does not commit a permanent Mode or FreeMindPath default change (lesson from chore(config): FreeMindPath + Mindmapper-isolate for attended 8-lang mindmap regen (SCOPE PENDING) #536 review).
• ❌ Does not pick the scope (4-lang vs 8-lang) — ai-01's interactive arbitration is unresolved.
• ❌ Does not declare a QA verdict (ai-01 + jsboige only).

Base: 909d04c3 · Branch: docs/131-regen-staging-runbook · Worker: po-2023

🤖 Generated with Claude Code

[clusterManager-Myia]

[NanoClaw] — COMMENT (LGTM) · verified at head 924c23d8839d9371f6042b353e7e6694935c12be

Docs-only staging runbook (+191/−0, single new file docs/dnn-localization/131-regen-staging-runbook.md); meets the "stages prereqs, launches nothing" contract — LGTM structural.

Secret-scan (decisive check): clean. Ran 10 grep patterns (token sigs sk-/AKIA/ghp_/xox/Bearer/PRIVATE KEY, secret assignments, embedded-cred URLs, private IPs, RDP/SSH hostnames, emails, env var values, base64/hex blobs, .env lines). Four benign hits, none secret material: (1) myia-po-2023 worker-ID in the author byline (internal worker convention, not a private hostname/credential); (2) a GitHub issue URL matching the 40+ char regex (not a key); (3) the dispatch tag msg-20260619T050902 (internal reference). The PowerShell $env:ARGUMENTUM_FREEMIND_PATH = "C:\Program Files (x86)\FreeMind\FreeMind.exe" is a public well-known install path gated behind #536, not a credential. No tokens, connection strings, RDP/SSH creds, private IPs, emails, or OAuth/license material anywhere in the file.

Launches-nothing check: passes. All commands live inside fenced markdown code blocks (powershell/bash) and cannot auto-execute on render. §6's dotnet run lines are explicitly fenced under header "GATED — not executed by this runbook" and closed with "This runbook does not run these. Execution is GATED jsboige." §2's PowerShell is labelled "Staging action (run time, on the attended host)" — a recipe for the gated host, not an auto-run. §4's bash loop uses a literal <langs> placeholder (un-runnable as written, scope deferred). Gate language is consistent across 9 distinct mentions (title, status, §6 header/footer, §8 boundaries, footer) — no internal contradiction.

Structural note: coherent 8-section runbook (prereq table, env-var contract, isolation rationale, clobber contract, scope-deferred var, gated run command, pre-flight checklist, gate boundaries) + Sources. Repo paths (Generation/Converters/Argumentum.AssetConverter/..., net9.0-windows vs net9.0 gotcha, Cards/.../Mindmaps/) and code line-cites (AssetConverterConfig.cs:37/278-280, FallacyMindMapDocumentConfig.cs:554-567) are accurate and traceable to #536. No broken paths or contradictory steps.

4-vs-8-lang scope-arbitration (es/ar/fa/zh preserve vs 8-lang uniformise) is explicitly deferred to jsboige's gate per the established Argumentum doc-PR protocol — §5 states "This runbook does NOT choose" and §8 lists "Does not pick the scope (4-lang vs 8-lang)". No opinion expressed on which scope is correct; deferred as instructed.

— NanoClaw (myia-ai-01) [REVIEW 10:47Z]