Record bounded implementation decision gate

docs/HYPERTWIST_MODEL_A_MODEL_B_COORDINATOR_DOCTRINE_2026-05-14.md

@@ -151,6 +151,19 @@ For the current clean-room sequence:
   - clean `Model B` contribution
   - repo-local consolidation
   - bounded implementation one bound at a time
+- one bounded implementation run equals one bound
+- after each bounded run, the coordinating instance must require an explicit
+  next-step decision:
+  - same bound again
+  - next bound
+  - repo-local reconsolidation if the standing map is too thin or stale
+  - targeted cross-lane reconciliation only if the standing note or
+    just-finished bound closeout explicitly says the next bound depends on
+    unresolved overlap
+  - owner-scoped validation-unblock packet only if validation is blocked by a
+    broader first-party issue outside the lane owner surface
+- the coordinating instance must not allow silent multi-bound widening in one
+  pass merely because a clean implementation thread is already open
 - shared cross-lane reconciliation is an exception checkpoint, not the default
   immediate next step after every contribution pass
 - repo-local and shared reconciliation are additive, not reset-oriented

docs/HYPERTWIST_PHASE_5R_CLEAN_ROOM_PROMPT_AND_HANDOFF_PACKET_2026-05-14.md

@@ -48,6 +48,20 @@ Reconciliation rule:
 - they must not silently omit earlier valid effort; any rejection, supersession, or deferral must be named with rationale
 - those dispositions must also be printed in the task summary for the pass, or the summary must explicitly say that no such omission/disposition occurred
 
+Bounded implementation rule:
+
+- one bounded implementation run equals one bound
+- after each bounded run, the next route must be chosen explicitly:
+  - same bound again
+  - next bound
+  - rerun repo-local consolidation if the standing map is too thin or stale
+  - optional targeted cross-lane checkpoint only if the standing note or
+    just-finished bound closeout explicitly says it is needed before the next
+    bound
+  - owner-scoped validation-unblock packet only if validation is blocked by a
+    broader first-party issue outside the lane owner surface
+- do not pre-authorize donor-lane-wide multi-bound widening in a single pass
+
 It does not mean:
 
 - reopen broad `0R` evaluation
@@ -186,6 +200,11 @@ Clean `Model B` bounded implementation:
 - no primer by default
 - consumes the repo-specific reconciled note and, only if one was explicitly
   created for that lane, the shared cross-lane reconciled note
+- safest route is to review after each closed bound and start a fresh clean
+  conversation for the next bound
+- lower-friction route may reuse the same clean bounded implementation
+  conversation for the immediately next bound only after the prior result has
+  been reviewed and the next bound has been explicitly selected
 
 ## Primer rule
 
@@ -214,6 +233,9 @@ Default operator flow:
 4. complete repo-local consolidation for that repo
 5. implement that repo one bound at a time
 
+After each bounded run, stop and choose the next route explicitly rather than
+auto-continuing into additional bounds.
+
 Shared cross-lane reconciliation is not the default next move.
 
 Use it only when the repo-local consolidated note still says a real owner-boundary
@@ -393,7 +415,10 @@ Topology rules:
 - one clean Model B contributor thread per restrictive repo
 - contributor thread gets primer first, then contribution prompt, in the same clean conversation
 - shared consolidation is one fresh clean thread and does not get the primer by default
-- bounded implementation is one fresh clean thread per repo and per bound and does not get the primer by default
+- bounded implementation starts in one fresh clean thread per repo and does
+  not get the primer by default
+- after review, the same clean implementation thread may be reused for the
+  immediately next bound, but one run still equals one bound
 
 Clean-room boundary rules: