df-data-transform-lenslisted

# Dark Factory — The Data-Transform Lens ## Overview An application is not a machine with features; it is a **data-transformation algorithm**: input data → transforms → output data, carrying state forward in time. Modelling a system this way collapses the design space and concentrates effort where the risk actually lives. Every other `df-*` skill is this model applied at one stage. ## The model: two primitives + two tags - **Data node** = a schema + its single-datum invariants, carrying context: - `location` — which store/system holds it now (not a trust signal). - `origin` — where it came from (web/mobile/scan/API/another service). Travels for audit; **never** a trust signal. - `authority` — is this the system-of-record for this fact? A ranking; the conflict resolver. - `governance` — security class, retention, residency. Travels with the datum. - **Transform** = `(state, inputs) → (state', outputs)`, tagged: - `pure` — replayable, no marker needed. - `effect` — irreversible world-change (send / charge / publish / notify / external write). **Must** carry an **idempotency key** (retry ≠ double-action) + a **compensation** path (the hand-written un-transform). - **Validation rule** (tag over data) = a predicate that must hold. The only axis that changes architecture is **enforcement locus**: - `LOCAL` — checkable at one ingress/commit with all operands present → **reject at the door**. - `GLOBAL` — ranges across records/systems/time → **flag, then reconcile