chore: moved from delay-based tests to deterministic gated server tests

for chat

.claude/projects/-Users-gb-Projects-bDS2/memory/MEMORY.md

@@ -1 +1,2 @@
-- [Fix all test failures](feedback_fix_all_failures.md) — Never dismiss failures as pre-existing; fix everything
+- [Fix all test failures](feedback_fix_all_failures.md) — Never dismiss failures as pre-existing or flaky; investigate and fix
+- [Debug targeted](feedback_targeted_debugging.md) — Analyze the code and fix; don't brute-force with repeated suite runs

.claude/projects/-Users-gb-Projects-bDS2/memory/feedback_fix_all_failures.md

@@ -1,11 +1,13 @@
 ---
-name: Fix all test failures
-description: Never dismiss test failures as pre-existing — if tests fail after changes, fix them
+name: Fix all test failures including flaky ones
+description: Never dismiss test failures as pre-existing or flaky — investigate root cause and stabilize
 type: feedback
 ---
 
-All test failures after changes must be fixed, even if they appear unrelated. The test suite was clean before, so any failure is the responsibility of the current change.
+All test failures must be fixed, even if they appear unrelated to current changes. The test suite was clean before, so any failure is my responsibility.
 
-**Why:** The user confirmed the suite was green before. Dismissing failures as "pre-existing" is wrong and wastes time.
+Flaky tests are deeper problems waiting to surface. Running a test in isolation and seeing it pass is never enough — must find out why it was flaky in the full suite run and make it stable.
 
-**How to apply:** After making changes, if any test fails, investigate and fix it before reporting the task as done. Never stash/skip/ignore failures.
+**Why:** Dismissing failures as "pre-existing" or "flaky" is wrong. Flaky tests indicate real issues (race conditions, test pollution, shared state) that will bite harder later.
+
+**How to apply:** After making changes, if any test fails: investigate the root cause, fix it, and verify it passes reliably in the full suite. Never stash, never skip, never re-run and hope. Never dismiss ordering-dependent failures — find and fix the shared state or race condition.

.claude/settings.local.json

@@ -6,7 +6,8 @@
       "Bash(mix dialyzer *)",
       "Bash(mix ecto.migrate)",
       "Bash(git add *)",
-      "Bash(git push *)"
+      "Bash(git push *)",
+      "Bash(git -C /Users/gb/Projects/bDS2 status)"
     ]
   }
 }