Harbor
Skip to content
OUTCOMESDISTRIBUTIONSPROBABILITYCASCADEINPUTS
Probability Engine

Engine

Conditional probability surfaces across every indicator, regime, and timeframe.

Scroll to explore ↓

Get the highlights.

Conditional Probability
Every indicator. Every combination. One surface.

Joint probability distributions across RSI, StochRSI, MACD, Bollinger, and 98 more indicator states — computed simultaneously, not sequentially.

Built for conviction.
From the data up.

The Probability Engine transforms raw indicator states into conditional probability surfaces. Every agent, every recommendation, every risk call is built on this. So you can act on probability, not intuition.

Probability-first decisions. The engine replaces qualitative calls with conditional probability surfaces. When Macro says "cautious stance," the engine quantifies exactly how cautious — regime-specific win rates, drawdown distributions, time-to-reset.

engine.query(
  given=[("VIX", "> 25", "daily")],
  what="forward_distribution",
  horizon=[10, 21, 63]
)
→ P(+5%|10d) = 34.2%
→ P(-5%|10d) = 12.8%
→ win_rate_63d = 68%

Cross-agent intelligence. One surface feeds six agents simultaneously. Signal reads win rates, Risk reads tail probabilities, Options reads vol forecasts, Valuation reads distributional fair value. Different questions, same truth.

surface = engine.query("CRDO")

signal  → PASS  (73% win rate)
risk    → SIZE_DOWN (Kelly: 3.15%)
options → NO_EDGE (IV near fair)
val     → ENRICH (dual confirm)
exit    → HOLD (no reset signal)
macro   → TIGHTEN (elevated vol)
The Gap

Before the engine.
After.

Every agent produces sharper, deeper, more actionable output when the Probability Engine is underneath. The transformation isn’t incremental — it’s structural.

BEFORE ENGINE
RSI oversold on daily.
Entry signal.

Binary. No probability.

ENGINE
AFTER ENGINE
PASS
Win Rate
73%
+73%
Fwd Return
+4.2%
+4.2%
Max Adverse
-6.8%
Quantified

Dual timeframe confirmation found in 847 historical matches — high-confidence setup with regime-adjusted sizing.

Take a closer look

Explore the Engine.

102 conditions.
RSI, StochRSI, MACD, ADX, Bollinger, Volume, Fibonacci, Stochastic, ATR, Moving Averages.

Every indicator is computed daily and weekly across 346 tickers. The engine doesn't check one at a time — it queries combinations. RSI > 70 AND StochRSI pegged AND MACD contracting. That's one query. There are 330,000+ valid combinations.

102
indicator columns
346
tickers tracked
5yr
history depth
Wilder RSI on daily + weekly timeframes
StochRSI K/D with reset detection
MACD with histogram expansion state
ADX + directional movement for trend strength
Bollinger %B + bandwidth compression state
Volume regime classification
10,000 paths.
Indicator-aware simulation with RSI, StochRSI, and MACD tracked on every path.

Not a random walk. Each simulated path computes Wilder RSI, StochRSI K/D, and MACD at every step using the exact same math as real data. The engine tracks when each indicator resets — on which day, at what price, through what drawdown.

10K
paths per scenario
4
drift scenarios
63d
simulation horizon
GBM with GARCH-informed vol clustering
Indicator computed on every simulated step
Reset-day distribution by percentile
MAE (max adverse excursion) at 5th / 50th / 95th
Forward return distribution at 5d / 10d / 21d / 63d
Regime-conditional drift scenarios
Q1 through Q8.
Each question narrows the probability surface. The chain runs automatically.

Q1: How long to reset? Q2: What if price continues? Q3: What if it crashes? Q4: Can time alone reset it? Q5: How does VIX change this? Q6: MACD expanding or contracting? Q7: Full combination with confidence. Q8: Vol diagnosis.

8
cascade questions
auto
fallback logic
5
confidence levels
Q1 — reset time distribution
Q2 — continuation probability
Q3 — crash asymmetry
Q4 — time-only reset path
Q5 — VIX regime conditioning
Q6 — MACD expansion state join
Q7 — full combination w/ confidence
Q8 — IV vs engine forward vol
IV meets probability.
The engine computes forward vol from its own distributions — then compares to market IV.

IV Rank, VRP, skew, term structure. The engine knows what volatility should be based on the current indicator state. When market IV diverges from engine vol, that's either a trade or a warning.

34
vol regime states
VRP
edge detection
IV×HV
spread analysis
Front-month IV vs engine 21d forward vol
VRP in basis points, by strike
Skew: 25Δ put vs 25Δ call spread
Term structure: front / back ratio
IV rank on rolling 252d window
Ranked across universe for edge
6 agents consume it.
Signal, Risk, Exit, Macro, Size, Reconcile. Every agent reads the same probability surface.

The ProbabilitySurface is the one data object that flows through the entire system. Signal Agent reads forward returns. Risk Agent reads max drawdown distributions. Exit Agent reads reset timing. No agent guesses — they all compute from the same surface.

6
governed agents
1
shared surface
3
governance states
Signal — forward return scoring
Risk — tail + MAE quantiles
Exit — reset-time distribution
Macro — regime conditioning layer
Size — Kelly with regime adjust
Reconcile — cross-agent consistency
PASS. BLOCK. REVIEW.
Every decision is governed, traceable, and auditable at the constraint level.

Concentration limits, drawdown circuit breakers, regime-aware exposure policy. The governance layer doesn't just flag — it blocks. A signal with 70% win rate still gets blocked if it would exceed sleeve concentration or breach a regime guardrail.

3
decision states
100%
audit trail
real-time
constraint checks
Concentration cap — sleeve + ticker
Drawdown circuit breakers
Regime-aware exposure policy
Correlation ceiling per sleeve
Event-proximity blocker (earnings / CPI)
Full constraint audit trail with IDs
Live indicator matrix
RSI_14
30.0
STOCH_K
41.0
STOCH_D
52.0
MACD_H
63.0
ADX
74.0
BB_%B
30.0
ATR
41.0
VOL_Z
52.0
INDICATORSENGINE.INDICATORS
↕ scroll to explore · click any feature to expand ↕
The Calibration Loop

The engine recalibrates
every night.

On every trade. In every regime. Six stages cycle nightly at 03:00 UTC.

01CAPTUREEvery scan's full context loggedVIX · news · indicators · volume02FINGERPRINTProprietary hash maps each setupto its combination bucket03AGGREGATEMaterialized view groups outcomesBayesian posterior computed04BLENDHistorical prior ⊕ Harbor realizedConfidence-weighted average05SURFACEHarbor-adjusted predictionon every scan, with confidence06VERSIONEngine v2.4 → v2.5Changelog publishedNIGHTLY03:00 UTCREFRESH
Tonight’s refresh — forecast
Trades ingesting
once Supabase + Polygon are wired, this goes live
Current version
v1.0.0
foundation release · every scan stamped
Bucket coverage
climbing as users + outcomes accumulate

The engine you use tomorrow
is measurably different from
the one you used this morning.

Performance

Benchmark the Engine.

Measured accuracy, latency, and coverage across every query configuration. Expand to explore.

74%
Peak accuracy
with vol regime
330K+
Query combos
statistically valid
3.8s
Full cascade
end-to-end
285
Ticker universe
5yr daily + weekly
The Engine

Built to answer
every question at once.

10,000Monte Carlo paths per query
102indicator conditions
8cascade questions
4regime states
756day lookback default
Tech Specs

Compare across tiers.

Engine capabilities scale with your tier. Every tier includes the dashboard, signals, and journal.

Probability Engine
Conditional probability queries
Basic
Monte Carlo simulation
1K paths
10K paths
Cascade (Q1–Q8)
Q1–Q3
Q1–Q8
Vol surface integration
Probability surface output
Summary
Full
Research mode (25K paths)
Coverage
Ticker universe
25
100
300
285+
History depth
1yr
3yr
5yr
5yr
Indicator conditions
12
48
102
102
Custom indicator combos
Limited
Unlimited
Options Specialist
Vol scanner
Basic
Trade plan generation
VRP scorecard
Structure optimization
7 structures
Kelly sizing
Agent Integration
Signal Agent (probability-backed)
Read-only
Risk Agent (regime-aware sizing)
Exit Agent (conditional stops)
Agent Console
Limited
Infrastructure
API access
REST
Prediction logging
Broker connection
Paper
Live
All tiers include the operator dashboard, trade journal, signal layer, and watchlist management. Engine features are additive — each tier includes everything below it. Prices reflect founding access rates.
Pipeline

Ticker in. Surface out.

01
Ingest
Ticker + State
02
History
756-Day Lookback
03
Cascade
Q1 → Q8 Chain
04
Monte Carlo
10K Regime Paths
05
Surface
P(x) + Confidence
The complete Harbor system.
Inputs → Engine → Products → Agents → Governance → Execution
INPUTSENGINESYNTHESISPRODUCTS + DISTRIBUTIONAGENTSCUSTOMERSGOVERNANCEEXECUTIONfeeds back into engineTicker + Market StateSupply Chain DataSPLC · earnings · capexHistorical DB5yr × 346 tickersMonte Carlo10K indicator-aware pathsConditional CascadeQ1→Q8 chainVol IntegrationIV · skew · termProbability Surfacereturns · resets · confidenceunified data object → all agentsNexusinternal agent busOptions SpecialistSaaS productSupply Chain IntelSaaS productSignal AgentRisk AgentExit AgentMacro AgentSaaS CustomersConsulting / APIGovernancePASS / BLOCK / REVIEWRetail ExecutionIBKR · Schwab · TDAInstitutional ExecutionFIX · Bloomberg EMSXINPUTENGINESURFACENEXUS / AGENTSPRODUCTSSUPPLY CHAINGOVERNANCE

Access the engine.

Probability surfaces for every ticker in the Harbor universe.