fix: implement Nakamura final review conditions (PRODUCTION READY)

Condition 1 — flush.py archive warning:
  warn_if_not_archived() fires when flushing a DB with non-zero keys.
  Prints explicit warning + 5-second Ctrl+C window before destroying data.
  Closes the last gap in the artifact preservation chain.

Condition 2 — dataset checksums in 03_plan.md:
  New "Dataset Checksums" section with sha256 fields for train/test files.
  Added as rule #5 in 03_preregistration.md with explanation.
  Cryptographic anchor for cross-experiment reference distributions.

Protocol is now PRODUCTION READY per Prof. Nakamura's final assessment
(2026-03-04). Health scores: scientific validity 4.5, operational
reliability 4.0, sustainability 4.0, artifact preservation 4.5,
reproducibility 4.0.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: KhrulkovV

tools/flush.py

@@ -22,6 +22,7 @@
 import subprocess
 import sys
 import textwrap
+import time
 
 import redis as redis_lib
 
@@ -74,6 +75,19 @@ def kill_workers(pids: list[int], dry_run: bool) -> None:
         print(f"[workers] Failed to kill (already dead?): {failed}")
 
 
+def warn_if_not_archived(db: int, before: int) -> None:
+    """Warn and pause if DB has data that may not have been archived yet."""
+    if before == 0:
+        return
+    print(f"[warn]  DB {db}: {before} keys present.")
+    print("[warn]  Have you run 'bash tools/archive_run.sh --upload' for this DB?")
+    print(
+        "[warn]  Flushing without archiving destroys all evolved programs permanently."
+    )
+    print("[warn]  Proceeding in 5 seconds — Ctrl+C to abort.")
+    time.sleep(5)
+
+
 def flush_db(db: int, host: str, port: int, dry_run: bool) -> bool:
     """Flush a single Redis DB. Returns True if successful (or dry-run)."""
     r = redis_lib.Redis(host=host, port=port, db=db)
@@ -82,6 +96,7 @@ def flush_db(db: int, host: str, port: int, dry_run: bool) -> bool:
         if dry_run:
             print(f"[flush] DRY-RUN — DB {db}: {before} keys would be flushed")
             return True
+        warn_if_not_archived(db, before)
         r.flushdb()
         after = r.dbsize()
         if after == 0: