Save learning
Record a learning. The app computes its evidence and returns the insight check.
statement is the why; scope the Values it applies to. Status starts as hypothesis, or supported / refuted when test_id names a concluded Test. check.looks_like_observation never blocks the save.
Requires a key with write scope or higher.
MCP: meta_save_learning.
bearerAuthAuthorizationBearer <token>An API key from Settings → API keys, sent as Authorization: Bearer rsk_….
Authorization has two independent axes.
The scope is ranked — a key satisfies any requirement at or below its own tier:
read— every GET.write— launch ads PAUSED, sync, pull history, edit thresholds, Values, briefs, tests, swipes and research.admin— everythingwritecan. No operation requires more thanwritetoday.
There is no approve scope. It was a rung once; it is not one now, and a key requested with it is rejected.
The approval grant (can_approve) is a separate boolean, not a rung. Changing spend on Meta — activating a launch, pausing or resuming ads, retargeting an ad set, or rebuilding a launch's live creative — needs write and the grant. Keeping them on separate axes is what makes the review gate a control rather than a convention: a key that may propose is not automatically a key that may approve its own proposal.
A key reaches the one workspace — one Meta ad account — it was minted in, and stops working when the operator who minted it leaves the workspace. The same key authenticates the MCP endpoint (/api/mcp) for a headless agent, with the same scope and grant.
Idempotency-Key?stringAny unique string up to 256 characters. A retry with the same key and body replays the first success (Idempotency-Replay: true) instead of doing the work twice; the same key with a different body is refused.
length <= 256application/json- body
statement*string1 <= length <= 600observation?stringlength <= 1200implication?stringWhat to make differently because of it. Briefs quote this line.
length <= 800scope?array<>items <= 8[]proposal_key?stringA learning proposal's key: test:<id> or trait:<trait id>.
length <= 300test_id?string^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$uuidagent?stringYour name, e.g. "Claude" — shown as "Added by Claude over MCP".
length <= 40Created. The body is the tool's own result — the same payload the MCP tool returns.
application/json- response
curl -X POST 'https://winandscale.com/api/v1/learnings' \ -H 'Content-Type: application/json' \ -d '{ "statement": "string"}'{}