Capture current mission, AI, map and administration development

Include the current addon, terrain integrations, tools, regression tests and recorded development evidence.

docs/intel-registration.md

@@ -9,3 +9,11 @@ Mortar preparation now owns the group and registers/reserves it directly with HQ
 `tests/test_intel_registration.py` executes the actual Intel registration/tick methods, HQ registration method, and the readiness method's existing early guards plus final qualified handoff with native boundaries represented by adapters. It verifies initially empty groups, later materialized members, duplicate readiness/registration, accepted asynchronous inventory operations, folder removal, death before/after assignment, original negative rolls and no duplicate HQ unit. Source-order checks require publication after the native readiness validations. Adjacent mortar transaction, activity-completion, refresh and report-ledger regressions also pass.
 
 This is source/policy coverage. It does not establish native inventory completion, remote simultaneous carriers, or multiplayer discovery replication for this revision. The existing Chat `mortar-inventory` fixture now observes this handoff after its actual battery-readiness gate and before its existing shell-depletion mutation. A Workbench-only Intel accessor reports whether each exact actor belongs to the existing one-roll set. Chat requires four distinct actual crew identities and a positive latch observation for each, without changing carrying chance, registering again, rolling again, or forcing documents. This works for both positive and negative carrying decisions. Intel has no AI-addon type references and keeps its standalone dependency scope. These new assertions remain native-unverified until the next run. Successful inventory insertion is a separate outcome: capacity refusal must remain visible rather than be counted as successful carriage.
+
+## Repeated intel with larger camp populations (2026-10-12)
+
+New generic carried folders choose among locations the team has not fully identified. Assignment selects randomly among the least-assigned eligible sites, keeping per-round counts even after a document is recovered or retired. This spreads a larger AI population across available targets and lets later generated sites receive assignments. Explicit mission documents retain their target, and previously sealed folders are never retargeted; they can still become outdated before pickup.
+
+The team-wide duplicate-topic and fully-exhausted explanations each broadcast at most once per round, retaining the shared five-second burst limit and the separate useful-discovery queue. Additional redundant folders are still spent once by the existing recovery ledger.
+
+The production sealing test covers 570 assignments across 57 unfinished locations, later site availability, already-identified exclusions, rough clues, fixed authored targets and repeated inventory scans. Feedback tests cover repeated notices over 100 later scans, useful discoveries, transport deduplication and a fresh round. Adjacent registration, reroll and mine-intel recovery tests pass. These are policy tests with engine boundaries substituted; in-game acceptance remains pending.