D297 - Learned policy is a file of measurements
Status: decided Date: 2026-09-26
learned.toml holds [[measurement]] entries
- function, the title it was measured on, date, tool version, how it is
known, evidence, answer, writes and assumes - and the
policy a run uses is derived from them. --verify re-derives
an entry in a fresh data directory with no learned file.
Why: a measurement from a binary the submitter owns
is reproducible and falsifiable by a command, so it can be received from
anyone without a provenance question. The title field stops a value
measured on one guest reading as a platform fact. A machine that has
applied a measurement no longer hits the wall it came from, so
verification must run from the state the measurement was taken in. An
entry nobody can verify locally is accepted as assumed.
Rejected:
- A settings cache: cannot be sent or checked.
- Verifying against the accumulated state: the applied answer hides the wall and confirms itself.