Line 02 / Data
OrangePeel is built on a custom deterministic-algorithmic engine (D.A.E.), bracketed by two ML models — one that parses raw input, one that shapes output. The math in the middle is deterministic. Same inputs reconstruct the same number, every time. The audit trail terminates in source events.
Same inputs, same number. Every time. The engine itself has no learned weights to drift; two ML models bracket it for input parsing and output shaping, both pinned per release. The math in the middle never moves.
Deterministic core · ML at I/OEvery metric on screen reconstructs from first inputs. Hand the underlying event log to a third party; they get the same numbers, byte for byte. No private weights, no proprietary scoring.
Byte-for-byteOpen the math. Every signal that fed every metric is reachable from the report — drill into the engine, not a model card. The audit trail terminates in source events, not “trust us.”
Audit trail → source eventsModern analytics is mostly probabilistic — useful, but the cost is opacity.
Deterministic engine in the middle, bracketed by two ML models for input parsing and output shaping. Versioned, pinned, replayable.
Append-only event stream from across the Suite. No silent mutations.
Deterministic-algorithmic engine with two ML models bracketing it — one parses raw input into engine-ready features, one shapes the deterministic output into human-ready reports. Engine in the middle is pure.
Reproducible end-to-end. Every cell drills back to the source event that produced it.
Numbers you defend
OrangePeel works best for operators who have to explain a number to a regulator, an auditor, a board, or themselves. If “because the model said so” doesn’t cut it, the engine is for you.