Learning
What the machine has worked out across every client, and the ledger that keeps two of them from running the same idea.
Pattern library (§7)
0 patterns0Candidate
0Active
0Retired
No patterns extracted yetThe Sunday extraction screens the week's outcomes across every client. A candidate needs evidence from two accounts, or three independent runs, before it becomes active.
A candidate is a deterministic screen, not a claim. It activates on evidence spanning 2 accounts or 3 runs, and retires below a 45% win rate over 6 tagged uses — the library shrinks as readily as it grows.
Vertical benchmarks (§4.2)
0 rowsEvery cell is suppressedBenchmarks pool clients of a vertical, and a cell holding fewer than five is never published — not to this page and not to the Allocator. An empty table here is the suppression rule working, not a missing read.
The Allocator's fallback chain: the client's own baseline first, then this vertical prior, then the platform constant.
Uniqueness ledger (§6.1)
0 live reservationsNo live reservationsThe Allocator reserves a trend cell whenever it plans a fill or experiment item, so another client's plan avoids it for fourteen days. Floor items are cadence-owned formats and are never reserved.
A conflict rejects the second plan — it never carves out territory. No client is promised exclusivity over anything.
Every row on this page names an account by its tenant_ref code name. The §3.2 wall holds here the same as everywhere: this surface reads ops, and no client's content crosses into it.