DocsRisk Engine

Decision Flow

The exact evaluation order every copy passes through — with the reason strings you will see on the tape.

The flow

The keeper records the same decision in its own vocabulary:

Tape action	Meaning	Example reason
`accept`	Full intended size	`accepted at full size`
`resize`	Capped/clamped	`accepted with size reduced by caps`
`skip`	Rule rejection	`session closed not tradable`, `price stale (181s > 120s)`, `position cap reached for NVDA`, …
`reject`	Failed normalization (before risk)	`UNKNOWN_SYMBOL: …`, `ZERO_NOTIONAL: …`, `BAD_PRICE: …`
`halted: true`	Drawdown halt active	`desk halted by drawdown protection`

The desk's high-water mark is the highest NAV/share ever reached (updated on deposits and copies).
Drawdown in bps: (hwm − navPerShare) × 10_000 / hwm.
At or above maxDrawdownBps (default 2,000 = 20%), evaluation rejects with drawdown halt — new copies stop. Redemptions are unaffected.
The keeper latches its own halted flag (updateDrawdown) so subsequent paper/live evaluations short-circuit.
There is no automatic un-halt: the halt clears only when NAV per share recovers enough that the drawdown is again below the threshold (on-chain, evaluation is stateless), or after operator review of the configuration.

Example numbers — not a live desk or forecast. A 60,000 USDG leader buy of NVDA in regular hours, desk at 5% copy fraction:

Intended: 60,000 × 5% × 100% = 3,000 USDG
Per-fill cap 2,000 (keeper config) → resize to 2,000
Current NVDA position value 4,500; position cap 5,000 → headroom 500 → resize to 500
Gross exposure 9,800; gross cap 10,000 → headroom 200 → resize to 200
Cash 10,000 → no further clamp
Decision: resize, executed 200 USDG, reason accepted with size reduced by caps

The same input on-chain yields (Accept, 200e6, "accepted") with the on-chain caps substituted.