Final authority before capital
Risk is the last place Harbor can still say no. It converts chain intelligence into a capital decision instead of letting conviction slip straight into execution.
Agent 05 · Risk
Gate the capital.
Enforce the limits.
The Risk agent is Harbor’s enforcement layer. It receives the full chain, asks whether the live portfolio can actually absorb the idea, and only then allows capital to move toward execution. This is the page that shows a buyer Harbor is governed, not just intelligent.
Governance engine
PASS · BLOCK · REVIEW
Nothing reaches Execution until Risk decides the portfolio can carry it.
Portfolio drawdown
—
Live from peak equity
Largest live position
41.9%
ASX currently binds size
Breakers tripped
0
2 healthy · 4 live positions
Chain position
5 of 6
After Valuation · Before Execution
Decision Chain Position
Where Risk sits.
Risk is the last gate before Execution. It reads every layer that came before it, then decides whether the portfolio can absorb the proposal without breaking Harbor’s policy envelope.
Protects the book, not just the idea
A good trade in isolation can still be a bad portfolio decision. Risk checks the proposed position against the book Harbor already carries.
Makes governance legible to a buyer
This is where Harbor proves it is not just a ranking engine. It can enforce policy, document its reasons, and create auditable boundaries around capital.
Governance Verdicts
Three states. No ambiguity.
Risk does not whisper. Every proposal receives one governed verdict the rest of Harbor can act on immediately.
✓
PASS
0
All constraints are satisfied. The proposal can route into Execution with a computed size and a fully logged governance trail.
✕
BLOCK
0
A hard limit is violated or the proposed trade fails policy. Block is the system saying no, not a suggestion to think about later.
⟳
REVIEW
0
The trade is not rejected, but Harbor wants human eyes on it because the posture is close to a limit or governance pressure is elevated.
Current pressure
The latest scan shows where governance is currently doing the most work inside the live universe.
Below minimum
0
Good setups that size below Harbor’s starter floor.
Manual only
0
Names requiring human routing rather than automatic promotion.
Wide stop
0
Technically interesting ideas that still fail stop-distance discipline.
Latest flow
Blocked signals0
Executed signals0
Pending requests0
Circuit Breakers & Policy Gates
Hard stops. Live headroom.
These are the live gates Risk is evaluating right now. The page reads like an explainer, but the numbers come from Harbor’s actual portfolio and governance surfaces.
Armed
Portfolio drawdown gate
Threshold: -15% from peak
Harbor keeps a hard drawdown buffer in front of new capital. If the book loses too much from peak, new entries stop until the portfolio is reassessed.
Current: —
Tripped
Single-position cap
Threshold: 40% of portfolio
No single holding should dominate the book. Risk watches the biggest live position first because that is where concentration risk becomes real.
Current: ASX at 41.9% · Headroom: 0.0%
Armed
Stop-distance ceiling
Threshold: 15% max stop
Signal quality is not enough if the stop has to live too far away. Risk uses this ceiling to block ideas that require too much distance to survive normal noise.
Current: GSIT at 12.9% · Headroom: 2.1%
Armed
Stress drawdown buffer
Threshold: -15% post-stress
Risk does not only watch the current book. It asks what the book would look like after a bad tape and whether that would break Harbor’s tolerance envelope.
Current: —
Armed
Starter minimum policy
Threshold: $100 minimum starter
Harbor will not treat a microscopic ticket as a real position. The policy floor keeps tiny computed sizes from masquerading as conviction.
Current: no live request
Armed
Open-position breach monitor
Threshold: 0 active breaches
Every live position stays under continuous watch. Risk tracks whether any open holding is already outside the published constraint envelope.
Current: 0 breached · 2 healthy
Sizing Governors
From conviction to permitted size.
Harbor’s live sizing surface is governed, not decorative. These are the current multipliers compressing or preserving proposal size before Execution ever sees it.
Live proposal
—
Current candidate size
Score weight
0.89x – 1.00x
Signal conviction weighting
VIX governor
0.43x – 0.86x
Regime drag from Macro
GARCH governor
0.25x – 1.00x
Volatility conditioning
Final band
0.17x – 0.68x
Range Risk will actually permit
Average correlation
—
Max observed pair
—
Days in drawdown
—
Analytical Modules
What Risk evaluates.
Risk is not one rule. It is a stack of policy, portfolio, and sizing checks that all contribute to the final governance verdict.
CL
Concentration limits
Checks whether the proposed trade would make one name, one sleeve, or one thesis too dominant inside the live book.
exposure
CR
Correlation analysis
Reads pairwise correlation so Harbor can catch hidden concentration that would otherwise look diversified on a holdings list.
correlation
DD
Drawdown monitoring
Tracks live and post-stress drawdown posture so Harbor knows when the portfolio has lost the right to keep adding risk.
drawdown
SG
Sizing governors
Turns signal strength into capital through regime-aware and volatility-aware governors, not by trusting raw conviction alone.
sizing
PG
Policy gates
Enforces starter minimums, stop discipline, and routing boundaries so tiny or structurally weak ideas do not slip through on technicality.
policy
GV
Governance verdicts
Publishes the final PASS, BLOCK, or REVIEW decision that Execution and the operator surfaces can both understand immediately.
verdict
Evaluation Pipeline
From proposal to verdict.
Harbor’s governance pass is sequential and auditable. A proposal does not jump around the risk stack or skip unpleasant checks.
STEP 01
Receive
Read the proposed trade with the full chain context already attached: regime, sentiment, signal quality, and price discipline.
STEP 02
Check
Challenge the idea against concentration, stop, drawdown, and live book constraints instead of only evaluating the new trade in isolation.
STEP 03
Size
Apply Harbor’s live sizing governors so capital reflects regime and volatility, not just enthusiasm about the candidate.
STEP 04
Verdict
Collapse the constraint stack into one governance state the operator can act on without ambiguity.
STEP 05
Emit
Publish RiskContext so Execution, the dashboard, and later audits can all see the same decision and the same reasons.
Output Schema
What Risk emits.
Risk publishes a reusable verdict object so Execution, the dashboard, and later audits all see the same governance decision and the same reasons.
RiskContextharbor_context.py
{
"verdict": str // PASS | BLOCK | REVIEW,
"ticker": str,
"proposed_size": float,
"proposed_pct": float,
"sizing_governors": {
"score_weight": float,
"vix_gov": float,
"garch_gov": float,
"final_multiplier": float
},
"constraints": { "stop": bool, "drawdown": bool, "concentration": bool },
"violations": list[str],
"reasoning": str,
"evaluated_at": datetime
}
Configure & Interact
Set the boundaries.
Governance is the most sensitive layer in Harbor. Changes here alter what the system is allowed to risk, so the page makes that sensitivity obvious.
01
Adjust circuit and policy gates
Change starter minimums, stop ceilings, or concentration caps when Harbor’s governance posture needs to evolve with the operating mandate.
02
Tune regime-aware governors
Modify how aggressively Risk compresses size when volatility rises or macro posture deteriorates.
03
Review constraint history
Inspect why good-looking candidates were blocked, then decide whether the policy is working or too restrictive for the mandate.
04
Query the reasoning trail
Ask Harbor why a trade was blocked or sent to review and get the exact governing factors, not a vague confidence score.
Try it out
See governance in context.
Open the live operator surfaces, inspect the execution handoff, or keep walking the chain into the final agent.