Shared Strategy Kernels · ttAlgoShared · Zero Copy-Paste Divergence

Strategy Kernels

Battle-tested building blocks shared across every production algorithm — risk gating, position sizing, protection geometry, session circuit breakers, cost modeling, and self-tuning parameter feedback. All kernels live in ttAlgoShared: one tested implementation, JSON-configured per algo instance.

Overview Risk Kernel Sizing Stack Protection Policy Venue Protection Session Guard Staleness Guard Cost Model Value Feedback Order Brackets

One Kernel, Every Strategy

Every production algo links against the same tested implementation. No copy-paste divergence. Configuration is JSON-driven per algo instance, so the same protection engine behaves differently for a scalper vs. a swing strategy. All financial math uses decimal_t. All hot paths are allocation-free.

10+
Shared Kernel Classes
0
Heap Allocations in Hot Path
21
Production Algos Using Them
Fail-Closed by Default
Every kernel defaults to the safe outcome when misconfigured or starved of data. A sizing cap of zero rejects the order rather than inflating it. A stale feed suspends trading rather than trusting old quotes. A missing correlation sample blocks the entry. The system never trades blind.

Risk Kernel

Pre-trade gate enforcing circuit breakers, funding window guards, gross exposure caps, per-asset sleeve limits, EWMA pairwise correlation bounds, and vol-targeted fixed-fraction sizing. Route every entry order through approve() before calling orderCreate.

Configuration (algoKernelRiskLimits_s)

FieldTypeDefaultPurpose
decRiskPerTradePctdecimal_t0.01Per-trade risk fraction
decDailyBreakerPctdecimal_t0.03Daily circuit breaker threshold
decGrossCryptoMultipledecimal_t2.5Max crypto gross notional / equity
ui32GrossIbContractsuint32_t4Max IB open contracts
decPerAssetCapPctdecimal_t0.10Per-asset sleeve cap
decCorrelationCapdecimal_t0.60Pairwise EWMA correlation limit
decVolTargetDailyPctdecimal_t0.016Daily vol target for sizing

API Reference

MethodPurpose
init(limits)Configure limits from struct
approve(request, outSize, outReason)Full pre-trade gate; reason string on rejection
circuitBreakerTripped(equity, sodEquity)Daily hard stop check
inFundingWindow(minuteUtc)Settlement no-entry guard
grossCapBreached(gross, ibContracts, equity, isCrypto)Exposure cap check
perAssetCapBreached(existing, proposed, equity)Sleeve cap check
correlationCapBreached(assetA, assetB)Pairwise correlation gate
volScaledSize(request)Fixed-fraction vol-targeted size
updateCorrelation(assetA, assetB, sample)EWMA correlation update

Sizing Stack

Base-size multiplied by bounded regime/confidence multipliers, then clamped through contract, notional, and participation rate caps. Consolidates the pattern previously duplicated across MIRE, HUB, QIS, DELEV, and RANGE. A multiplier of 0 collapses to 0. Negative multipliers assert. Caps are fail-closed.

algoSizerStack_c Configuration

FieldTypeDefaultPurpose
decProductCapdecimal_t1.5Hard cap on multiplier product
ui32MaxContractsuint32_tUINT32_MAXContract cap (sentinel = uncapped)
decMaxNotionalUsddecimal_tmaxNotional cap (sentinel = uncapped)
decMaxParticipationRatedecimal_t0Participation rate cap (0 = off)

algoSizerStack_c API

MethodPurpose
configure(config_s)Set caps
reset()Clear accumulator
setBase(baseSize, referencePrice)Validated base + price for notional
setParticipationVolume(volume)Volume for participation cap
multiply(multiplier)Chain regime/confidence/win-rate factors
finalize(instrumentPtr)Apply caps + venue grid snap
product()Raw accumulated product

sizingGuards.h

Header-only fail-closed helpers.

FunctionPurpose
contractSizeOrOne(contractSize)Sentinel-safe multiplier; venues without contract size behave per-unit
makeSizeOrZero(instrument, rawSize)Grid snap or reject; sizes below venue minimum return 0, never inflated

Pipeline Flow

CalculatePositionSize() → base size
→ sizer.setBase(base, price)
→ sizer.multiply(regimeFactor)
→ sizer.multiply(confidenceFactor)
→ sizer.finalize(instrument)
  ├─ product cap clamp
  ├─ contract cap clamp
  ├─ notional cap clamp
  ├─ participation cap clamp
  └─ venue grid floor snap
→ result ≤ 0 ? → no order sent

Protection Policy

Pure-state TP/SL geometry with breakeven ratchet, trailing stop, and profit lock. Owns no order-manager coupling — computes prices only. Geometry expressed in ATR multiples and R units. Stop is monotone ratchet-only: tightens, never loosens. Replaced twelve-plus hand-rolled copies across algos.

Configuration (config_s)

FieldDefaultPurpose
decStopAtrMult1.6Initial stop distance in ATRs
decStopMinDistance0Absolute floor (0 = off)
decStopMaxBps0Cap in bps of entry (0 = uncapped)
decTpAtrMult2.2Target distance in ATRs
decBreakevenTriggerR0R-multiple to arm breakeven (0 = never)
decBreakevenOffsetR0Offset past entry in R
decTrailFrac0Trail distance in live ATRs (0 = off)
decTrailMinDistance0Absolute trail floor
decProfitLockTriggerR0Peak R to arm lock (0 = off)
decProfitLockFrac0.5Locked fraction of peak R

API Reference

MethodPurpose
configure(config_s)Set geometry parameters
reset(entry, isLong, initialAtr)Arm for one position
stopPrice(mark, liveAtr)Monotone ratchet stop
targetPrice()Entry-latched take-profit
initialRisk()|entry - stop|
profitLockFloor(mark)Peak-R lock floor price
breached(mark)True once mark crosses lock floor
isArmed()Armed state
peakR()Best favorable excursion in R

Venue Protection

Resting venue-side stop as a safety net behind the algo's internal stop. Only fires when the process is not acting (crash, frozen feed). State machine: FLAT → PROTECTING → PROTECTED → AMENDING → CANCELING. Never widens after placement.

Configuration (config_s)

FieldDefaultPurpose
fEnabledtrueMaster switch
decBufferBps5.0Distance behind internal stop
decAmendMinBps2.0Minimum amend delta
ui32AmendMinIntervalS2Amend throttle interval
ui32PlaceTimeoutS3Ack timeout → retry
fRequiredtrueNo protected stop → flatten

API Reference

MethodPurpose
configure(config_s)Set parameters
reconcile(...)Keep venue stop in step; call on every fill and heartbeat
beginClose(...)Cancel before internal close
onFlat(...)Tear down when flat
venueTrigger(internalStop, isLong)Compute behind-internal trigger price

Session Guard

Deterministic daily/continuous session guard consolidating eight near-identical private copies. Covers trade caps, net + gross daily-loss latches, loss-streak freeze, pre-flatten/dead-zone/open-quiet windows. All time inputs are explicit for testability.

Configuration (algoSessionGuardConfig_s)

FieldDefaultPurpose
eModeDAILYCONTINUOUS skips time windows
ui32FlattenMinuteUtc1320 (22:00)Flatten deadline (0 = disabled)
ui32NoEntryMinutes30Pre-flatten entry block
ui32ReopenMinuteUtc0 (midnight)Venue reopen
ui32OpenQuietMinutes0Quiet window after reopen
ui32DailyMaxTrades0 (unlimited)Trade cap
decDailyMaxLossFraction0 (disabled)Loss latch fraction
ui32LossStreakTrades0 (disabled)Consecutive losses to freeze
ui32LossStreakFreezeMs300000Freeze duration

API Reference

MethodPurpose
configure(config_s)Set parameters
reset()Clear counters
onEquitySnapshot(equity, nowMs)Anchor day-start equity
onTradeClosed(netResult, nowMs, equityFallback)Update counters/latches
canEnter(minuteUtc, nowMs)False when any gate blocks
shouldFlatten(minuteUtc)True inside dead zone
parseJsonConfig(params)Read standard keys from JSON
toJson() / fromJson()State persistence
sessionIndex(nowMs)Advances on rollover

Staleness Guard

Wall-clock-anchored quote-staleness watchdog. Returns PROCEED, LATCHED, or SUSPEND_ARMED. Includes per-instrument kill resolution and observed quote-cadence histogram.

Core Functions

FunctionPurpose
algoEvaluateStaleness(lastBboMs, suspendUntilMs, wallMs, killMs, suspendMs)Core verdict decision
algoResolveStalenessKillMs(fallback, instrumentOverride)Per-instrument clamped kill timeout
algoEvaluateCadenceConfig(cadence, killMs)Warn when kill < p99 cadence

Quote Cadence (algoQuoteCadence_s)

MethodPurpose
observe(arrivalMs)Record quote arrival
ui32GapP99Ms()P99 inter-arrival gap

Cost Model

Round-trip execution cost model and strategy admission gate. Computes fees (notional-rate or per-contract), spread, latency drift, variance break-even hold time, and full admission check. All static methods — no instance state, no allocation.

Venue Latency Constants

ConstantValueVenue
ce_i32LatencyLighterMs300msLighter REST + signing path
ce_i32LatencyIbkrMs150msIB TWS order round trip
ce_i32LatencyDefaultMs50msCo-located crypto venues

Input Struct (algoCostInput_s)

FieldPurpose
decFeeMakerPerSideMaker fee fraction
decFeeTakerPerSideTaker fee fraction
decFeePerContractUsdPer-contract USD fee (IB)
decHalfSpreadBpsHalf quoted spread in bps
decLatencyDriftBpsVol-scaled latency drift
decMidPriceFor per-contract conversion
decContractSizeFor per-contract conversion

API Reference

MethodPurpose
venueLatencyMs(exchange)Default RTT by venue name
latencyDriftBps(latencyMs, volBpsPerSec)Expected movement over latency
feePerSideBps(input, isTaker)Fee in bps per side
roundTripBps(input, isTaker)2×fee + 2×spread + drift
minHoldMs(rtBps, sigmaPerGridBps, gridMs)Variance break-even hold time
targetBps(rtBps, kEdge)Required move to clear cost
admit(input, sigma, gridMs, maxHoldMs, isTaker)Full admission gate

Value Feedback

Realised-edge learner nudging kEdge, captureFrac, and pyramidGate toward observed outcomes. Bounded per-cycle step, clamped to configured bounds. Self-tunes strategy admission thresholds from closed-trade P&L without manual intervention.

Bounds (algoFeedbackBounds_s)

FieldDefaultPurpose
decKEdgeMin / Max1 / 5kEdge clamp range
decCaptureMin / Max0.2 / 1Capture fraction range
decPyramidMin / Max0.1 / 1Pyramid gate range
decMaxStep0.1Max adjustment per cycle
decReferenceEdgeBps5Reference edge for normalization

API Reference

MethodPurpose
init(kEdge, captureFrac, pyramidGate, bounds, minTrades)Configure
recordTrade(netEdgeBps, isWin)Feed one closed trade
endSession()Apply adjustment cycle
kEdge() / captureFrac() / pyramidGate()Current gate values
hitRate() / meanEdgeBps()Session statistics

Order Brackets & Helpers

Thin orchestration layers over algoOrderManager_c for placing protection brackets, managing per-leg user values, and preventing double-cancel bugs.

algoBracketOrders

Places stop-market + limit-target with per-leg user values, re-syncs sizes after partial fills, ratchets stops only when strictly tightening.

FunctionPurpose
algoPlaceProtection(mgr, pos, closeId, stopPrice, targetPrice)Place both legs
algoPlaceProtectionTicks(mgr, pos, closeId, stopPrice, targetTicks)Target as tick distance
algoSyncProtectionSizes(mgr, pos, closeId)Re-sync after partial fills
algoRatchetStopTrigger(mgr, order, newTrigger)Tighten-only modify

algoOrderLegIds

Per-leg user-value scheme so findOrder can distinguish close, stop, target, and TP2 legs without collisions. Stride is 8.

ValueOffsetPurpose
CLOSE+0Flatten leg
STOP+8Resting native stop
TARGET+16Take-profit / graceful exit
TP2+24Second take-profit

algoCancelOrder

Cancels an order through the manager and nulls the caller's pointer. Prevents double-cancel and dangling-handle bugs.

FunctionPurpose
algoCancelOrder(algoOrderManager_c&, orderInfo_s*&)Cancel + null the handle
Lifecycle & Warmup
The support library also includes algoWarmupGate_s — a bitmask gate preventing an algo from acting until dependent subsystems (HMM steps, history replay, book kinetics) are warm. Conditions: BOOK_KINETICS, HMM_STEPS, HISTORY, CUSTOM. One-shot consumeOpen fires exactly once on the closed-to-open transition. Combined with algoHistoryReplay which walks the newest contiguous run of closed candles oldest-first so bar-driven algos rebuild state before the live stream resumes.