Gate radio intel behind configurable site discovery; record Workshop 0.0.119

docs/evidence/radio-intel-2026-10-21/README.md

@@ -0,0 +1,56 @@
+# Radio intel progression — native checks, 21 October 2026
+
+The isolated Workbench mission completed **12 assertions with no script/runtime
+failures**. It used real player inventory, prepared sites, the normal discovery
+service and the actual radio-defence entry point. The runner verified source and
+external inputs and cleanup of its private prefix.
+
+The strict whole-engine reporter exits 1 because of 13 existing resource/world
+errors, matching the earlier standard body-burial run. They are retained without
+filtering in [native/report.json](native/report.json) and the complete console log.
+This is a successful feature fixture, not a clean whole-engine verdict.
+
+The fixture verified:
+
+- Default five-site requirement starts locked after preparation.
+- A fixed radio document in actual player inventory stays unspent and reveals
+  nothing before the threshold, including after four distinct identifications.
+- Native line of sight and the normal site observation tick identify a nearby
+  radio while radio intel remains locked.
+- The real `TuneRadio` / radio-defence service starts while intel is locked.
+- Identifying the fifth non-radio site unlocks intel. The ordinary discovery tick
+  automatically processes the held document and identifies its target, retaining
+  the physical objective document.
+
+Five eligible sites are provided by the historical medical stash/outpost plus
+three fixture-only supply locations. A second real radio entity is the document
+target so early physical discovery of the original radio does not invalidate the
+later document assertion. No installed addon or live server is modified.
+
+`tests/test_radio_intel.py` executes the production gate, recovery, assignment and
+retarget methods against explicit native API doubles. It covers exclusions,
+duplicate IDs, rough versus exact knowledge, sticky unlock, sparse/preparing
+worlds, next-round edits, disabled gating, redirection/evidence, held fixed clues,
+no alternative target and ownership/lifecycle rejection. Settings/menu/feedback
+regressions cover migration, ranges, command routing and non-repeating notices.
+The historical Core smoke arranges its remaining eligible outpost identification
+before its separate document ownership checks, accommodating the new default.
+
+Reproduce using installed dependencies and the existing private Workbench prefix:
+
+```
+python3 docs/evidence/radio-intel-2026-10-21/isolated-runner.py next01
+python3 tests/test_radio_intel.py
+```
+
+Choose an unused alphanumeric suffix. The runner freezes source into `/tmp` and
+uses a disposable profile. This scripted local mission does not establish remote
+multiplayer UI/RPC presentation or ordinary human input timing.
+
+Repository validation covered every check in `tools/check.py`. The full run
+stopped at an older help-line-count assertion; the resumed sequence stopped at
+an older Intel-summary test double. Both were updated for the new setting and
+rerun successfully, then all 134 remaining checks passed. The ordered transcripts
+are retained in [validation](validation/); earlier failures there are test-adapter
+mismatches corrected by the following run, not outstanding failures. Menu source
+regeneration and `git diff --check` also passed.