Skip to content

chancel

Provable scope isolation for AI retrieval. Firm-level instructions travel everywhere; a client matter's data provably never crosses a matter boundary — enforced below the model boundary, so the guarantee survives swapping the AI provider.

The whole argument is one table, the literal output of chancel demo --no-llm, run cold-clone with no API key and no Docker:

finding                 backend   provider      expected  actual  ok
----------------------  --------  ------------  --------  ------  --
canary-leak             isolated  echo          clean     clean   ✓
unrepresentable-call    isolated  n/a           denied    denied  ✓
deletion-verifiability  isolated  n/a           clean     clean   ✓
canary-leak             filtered  echo          clean     clean   ✓
unrepresentable-call    filtered  n/a           leaked    leaked  ✓
canary-leak             shared    echo          leaked    leaked  ✓

PASS: 12/12 cells matched their predicted color

isolated holds; filtered (the vendor-default multitenancy pattern) leaks on the findings collection separation makes impossible; shared (prompt-only "isolation") leaks first. "Green" means every cell behaved as predicted, reds included — not that nothing leaked.

chancel is a demonstration, not a product. It carries zero real data (the corpus is synthetic and script-generated). Out of scope, named so omission is not mistaken for oversight: authentication, RBAC, multi-user sessions, key management, retention, egress DLP, answer-quality evaluation, any UI, any hosted deployment.