Harbor
Skip to content
AgentsDecision ChainSentiment
Agent 02 · Sentiment

Measure the crowd.
Before you join it.

The Sentiment agent reads the positioning layer of the market so Harbor can tell whether consensus is a tailwind or a trap. It matters because a technically strong idea can still be a bad decision when the crowd is already too far on one side of the trade.

Agent status
Active
Crowd posture
LEANING BULLISH
Aggregate · live universe read
Contrarian score
0.84
Elevated crowding risk
Last evaluation
Mar 16, 2026
Triggered by Macro
Chain position
2 of 6
After Macro · Before Signal
Decision Chain Position

Where Sentiment sits in the pipeline.

Sentiment runs immediately after Macro. It inherits regime posture, adds crowd and positioning context, then hands a more honest universe to Signal and Valuation.

Sentiment Spectrum

Five zones. One read.

Sentiment keeps Harbor from confusing excitement with edge. It tells the whole universe whether participation is healthy, stretched, or dangerously one-sided, and that changes how much trust the rest of the chain should carry forward.

Extreme Fear
0.00 – 0.20

Contrarian support. Harbor can trust upside recovery more if Signal and Valuation confirm it.

Bearish
0.20 – 0.40

Cautious posture. Reduce trust in weak setups and keep governance more alert.

Neutral
0.40 – 0.60

No sentiment override. Downstream agents lean mainly on their own evidence.

Bullish
0.60 – 0.80

Crowding risk rises. Good setups can still work, but Harbor should watch for overheating.

Euphoria
0.80 – 1.00

Late crowd behavior. Tighten trust, sizing, and review posture on new longs.

Capabilities

What Sentiment reads.

It combines crowd behavior, positioning, narrative flow, and volatility context into one readable layer for the rest of Harbor.

NF
News flow scoring

Reads tone and velocity of narrative flow so Harbor can tell whether a move is supported or merely sensationalized.

narrative
OF
Options flow analysis

Uses unusual flow and positioning cues to detect whether conviction is informed, speculative, or too one-sided.

flow
CP
Crowd positioning

Measures how crowded a setup is becoming so Harbor can separate healthy support from late participation.

crowd
VS
Vol surface context

Adds ticker-level volatility posture so fear pricing and crowd enthusiasm are read together instead of separately.

vol
CI
Contrarian index

Synthesizes all modules into a crowding score so Harbor can tell when agreement is useful and when it becomes dangerous.

meta
EC
Event context

Reads earnings build-up and expectation pressure to detect when a name is priced for perfection before the next catalyst.

event
Evaluation Pipeline

From regime to crowd posture.

Every cycle follows the same five-step path before Sentiment publishes a crowd-aware context object the rest of Harbor can trust.

STEP 01
Receive

Read Macro context first so Sentiment starts from the regime Harbor is already operating inside.

STEP 02
Scan

Pull narrative flow, crowd participation, and options posture across the active universe.

STEP 03
Score

Resolve crowd posture into a composite read and a contrarian risk score the rest of Harbor can use.

STEP 04
Flag

Mark consensus extremes that should slow trust, reduce sizing freedom, or trigger governance review.

STEP 05
Emit

Publish Sentiment context into the shared decision surface for Signal, Valuation, Risk, and Execution.

Data Sources

What feeds the agent.

Sentiment pulls from flow, narrative, and positioning sources so crowd posture can update faster than a daily-only process.

SourceDataFrequencyProvider
Options flowUnusual activity, sweep posture, crowd positioning cuesStreamUnusual Whales
Narrative toneHeadline and reaction posture across the active universeLiveContext overlay
Sleeve leadershipSector sponsorship and crowd follow-throughLiveSignal fabric
Benchmark internalsBreadth, participation, and market toneLiveHarbor runtime
Analyst revision driftExpectation changes and revision velocityDailyEstimate overlay
Event pressurePre-earnings and catalyst crowdingScheduledEvent layer
Output Schema

What Sentiment emits.

Sentiment publishes a context object the chain can inspect instead of a vague mood label. Signal and Valuation can then reason with the same crowd-aware frame.

SentimentContextharbor_context.py
{
"crowd_posture": str // EXTREME_FEAR | BEARISH | NEUTRAL | BULLISH | EUPHORIA,
"composite_score": float // aggregate sentiment posture,
"contrarian_score": float // crowding risk,
"news_flow": { "tone": float, "velocity": int },
"options_flow": { "pressure": str, "unusual_activity": list },
"crowding_flags": list[str],
"event_context": { "pressure": str, "summary": str },
"evaluated_at": datetime
}
Configure & Interact

Tune the read.

Sentiment should stay explorable and configurable. Buyers should understand what it does, and operators should be able to shape how Harbor treats crowd posture.

01
Set contrarian thresholds

Control when Harbor upgrades crowding from useful sponsorship to a review-worthy warning.

02
Weight the modules

Raise the importance of options flow, reduce narrative influence, or bias toward event posture depending on your style.

03
Filter flow inputs

Tighten what qualifies as unusual activity so Harbor focuses on informative flow, not noisy participation.

04
Query the read

Ask what changed in crowd posture and why, then trace the explanation back to the modules that moved it.

Try it out

See how conviction builds.

Sentiment shapes how much trust Harbor should carry into Signal, Valuation, and Risk. Test it directly or keep walking the chain.