Route delivered HQ tracks to temporary road interception and refresh playtest packages

docs/packed-external-development.md

@@ -1,5 +1,7 @@
 # External dependency development packages
 
+The latest handoff is [road interception](../.build/playtest-2026-09-20-road-interception/README.md), with [exact build records](evidence/package-2026-09-20-road-interception/handoff.json). Five PAKs/222 payload entries match261 authored files and the editable ZIP. Core packed startup passes; the full external diagnostic now reaches ordinary local prisoner admission but retains its failed localization-error verdict. Packaging remains unqualified. Earlier radio/maintenance records below remain historical.
+
 Current AI/Chat requires the six installed dependencies in `external-addons.json`. Historical vanilla-only Core v1/Chat v2 packed contracts and evidence remain unchanged. Build only owned successor files; dependencies stay in their installed locations and are never bundled or published.
 
 Freeze addon sources and serialize this command with other private native runs. The session must be new:
@@ -12,7 +14,7 @@ python3 tools/native_package.py --project chat \
   --silent-workbench --continue-on-failure
 ```
 
-This uses the established hidden display, muted audio and owned cleanup. Native shutdown failures and the four installed external localization errors remain failures. Diagnostic continuation cannot waive unsafe cleanup or changed inputs. Inspect produced files before testing ordinary consumption; a PAK emitted during a failed build is only an unqualified development artifact.
+This uses the established hidden display, muted audio and owned cleanup. Native shutdown failures and the four installed external localization errors remain failures. Diagnostic continuation cannot waive unsafe cleanup or changed inputs. With explicit `--continue-on-failure`, a naturally completed native exit 0 followed by verified owned helper cleanup may now permit the next addon job after another idle/process and source check. The failed per-addon verdict and nonzero overall exit remain unchanged; a separate `diagnostic_continuations` receipt records this decision. Timeout, interruption, unknown/nonzero native exit, failed cleanup or changed inputs still stop the batch. This replaces the ad hoc remaining-addon scripts used by earlier handoffs. Inspect produced files before testing ordinary consumption; a PAK emitted during a failed build is only an unqualified development artifact.
 
 ## Explicit packed Chat v3
 
@@ -39,7 +41,7 @@ python3 tools/headless_game.py --project chat \
   --packed-addons /tmp/erri-radio-maintenance-packages/packages \
   --packed-session /tmp/erri-packed-chat-radio-maintenance \
   --external-addons '/home/jakub/.local/share/Steam/steamapps/compatdata/1874880/pfx/drive_c/users/steamuser/Documents/My Games/ArmaReforger/addons' \
-  --compat /tmp/erri-headless-compat --timeout 180 --failure-grace 30
+  --compat /tmp/erri-headless-compat --timeout 180 --failure-grace 120
 ```
 
 Default canonical 15-second readiness settling is mandatory. Optional 30-second
@@ -89,3 +91,7 @@ Core's PAK contains exactly 132 expected authored non-metadata files with matchi
 All four separate remaining jobs completed with native exit 0, confirmed cleanup and unchanged source/external inputs. Their failed candidate verdicts remain in [the raw build record](evidence/package-2026-09-20-radio-maintenance/README.md). Exact payload comparison verifies Core 132, AI 58, Intel 4, Map 6 and Chat 12 paths/hashes: 212 total, with exactly fifteen runtime files. The matching development handoff is `.build/playtest-2026-09-20-radio-maintenance`; older playtests are unchanged.
 
 The [explicit v3 packed Chat diagnostic](evidence/raw-packed-chat-external-2026-09-20-radio-maintenance/README.md) verifies all thirteen exact addon paths/mounts/databases and normal modules, with no input violations or script/runtime failures. Four installed localization errors appear twice; the run remains failed. Its bounded failure grace ended after site/prison preparation, before local participant readiness. Sources/packages/external inputs remained unchanged and cleanup was confirmed. No full ordinary-startup or release qualification is claimed; the raw v3 evidence was not relabeled as a legacy v2 archive.
+
+### Bounded external startup diagnostics
+
+`--failure-grace` defaults to zero and now accepts up to 120 seconds, still capped by the original run deadline. This allows the full scenario to finish initialization after known installed localization errors for diagnosis. Repeated errors or later readiness never extend the lease or change a failed run into success. Raw participant/mission markers can be inspected separately; they are not clean packed-startup qualification. The historical thirty-second run and its failed verdict remain intact.