Skip to main content
Platform readiness requires Supabase truth, UI truth, and compact client persistence to agree.

Supabase truth check

For the 60-run gate, Supabase returned:
ready | 60 | 3000 | 3000 | 3000 | 3000
That maps to:
  • status: ready
  • runs: 60
  • expected prompts: 3000
  • run items: 3000
  • Lucia responses: 3000
  • reviews: 3000
This confirms persisted cloud data matched the expected platform gate count.

localStorage compactness target

The target diagnostic is:
persistedLocalFullPayloadSessionCount = 0
persistedLocalHasItemLevelData = false
persistedLocalItemLevelDataSessionCount = 0
ownedSessionCount = expected run count
otherOwnerSessionCount = 0
ownerlessSessionCount = 0

Final verified diagnostic

sessionCount = 60
persistedLocalFullPayloadSessionCount = 0
persistedLocalHasItemLevelData = false
persistedLocalItemLevelDataSessionCount = 0
ownedSessionCount = 60
otherOwnerSessionCount = 0
ownerlessSessionCount = 0
rawByteSize around 68,815

Interpretation

This proves the tested owner context did not persist full item-level payloads locally after the gate. It also proves no other owner or ownerless sessions were visible in the diagnostic for that tested context. It does not prove:
  • all future accounts are safe
  • backend/RLS authorization is complete
  • external evaluator security boundaries are ready by themselves
Keep this distinction explicit in future release and onboarding work.

Behavioral label persistence check

Behavioral Observatory labels use a separate table:
public.eval_behavioral_labels
This table is not part of the original 60-run platform readiness count. The 60-run gate verified:
eval_runs
eval_run_items
eval_item_reviews
localStorage compactness
Behavioral label verification must separately confirm:
  • the eval_behavioral_labels table exists
  • RLS policies exist
  • the one-label-per-reviewer-per-run-item constraint exists
  • a saved label reloads after refresh
  • failed saves are not counted as persisted labels
Read next: Behavioral Label Persistence.