Topics

#ai-agent #bun #claude #consensus #smart-money #solana #typescript #wallet-tracking

Language breakdown

No language data available.

▸ Files click to expand

README

Chorus

Cohort-consensus tracker for smart-money behavior on Solana.

See when smart money is actually aligned instead of just noisy in one pocket.

bun run dev

watches makers, funds, teams, and traders as separate cohorts
ignores one-cohort inventory skew that looks bigger than it is
promotes tokens where multiple cohorts agree on direction at the same time

Consensus Board • Signal Feed • Operating Surfaces • Cohort Separation • Consensus States • Technical Spec • Quick Start

At a Glance

Use case: detect when multiple Solana wallet cohorts actually agree on a name
Primary input: cohort weighting, wallet tiering, timing alignment, agreement ratio
Primary failure mode: mistaking one-cohort inventory skew for broad conviction
Best for: operators who need to know whether smart-money alignment is real or isolated

Live Agreement Wall

Chorus Live Agreement Wall

Live operating view for Chorus: lead token consensus card, four cohort sectors with active and idle smart-money wallets (makers, funds, teams, traders), signal feed, token board, guardrails, and the agent's current ISOLATED, BUILDING, BROAD, or CROWDED state.

Operating Surfaces

Consensus Board: shows whether conviction is broad or isolated to one cohort
Signal Feed: prints the actual names climbing the alignment stack
Cohort Weighting: ranks makers, funds, teams, and traders separately before blending
Agreement Score: converts wallet activity into a usable conviction state

Why Wallet Tracking Usually Breaks Down

Most wallet dashboards collapse all "smart money" into one stream. That looks useful until a single cohort dominates the tape and the board starts pretending there is broad agreement when there is not.

Chorus avoids that mistake on purpose. It keeps funds, makers, teams, and traders separated long enough for the operator to see whether conviction is actually spreading or only echoing within one group.

Why Cohorts Stay Separate

Chorus does not believe that "smart money" is one thing. Funds, makers, teams, and fast traders all behave differently, and collapsing them into one blended wallet list hides that difference.

A maker leaning on inventory is not the same as broad cross-cohort agreement. Chorus exists to keep those cases separate until the market actually lines up.

Consensus States

The board is easier to use when the operator can name the state:

isolated: one cohort is moving, everyone else is absent
building: two cohorts are starting to agree
broad: multiple cohorts are aligned on the same name and direction
crowded: agreement exists, but the move is too late to treat cleanly

How It Works

Chorus turns wallet activity into an interpretable conviction state:

split the tracked wallets into separate cohorts
weight the wallets by tier instead of treating every address equally
measure whether the same token and direction are appearing across cohorts
convert that alignment into a consensus state the operator can understand
print the names where agreement is broad enough to matter

This is why Chorus is more readable than a flat wallet tape. It preserves structure before it blends anything together.

What The Operator Learns From Chorus

The board is trying to answer three practical questions:

who moved first
who confirmed the move after that
whether the agreement is becoming broader or only louder

That is much more useful than a flat list of wallet buys.

Example Output

CHORUS // CONSENSUS SNAPSHOT lead token JUP agreement score 0.79 active cohorts makers, funds, traders state broad late crowding low

operator note: alignment is real across more than one cohort, not just one pocket

Technical Spec

Chorus computes agreement in two layers:

Tier Weight

tier1 = 3

tier2 = 2

tier3 = 1

Cohort Bonus

market_maker = 1.2

fund = 1.1

team = 1.0

trader = 0.9

Consensus Score

consensusScore = agreeingWeight / totalTrackedWeight

Signals strengthen when multiple cohorts align. A single cohort can still be useful, but broad cross-cohort agreement is treated as the higher-conviction state.

When Chorus Should Demote A Signal

only one cohort is active and everyone else is absent
agreement exists, but comes too late after the move is obvious
the same small cluster keeps appearing without broader confirmation
team or maker activity looks more like inventory management than conviction

Risk Controls

cohort separation: prevents one group from impersonating broad agreement
tier weighting: stops low-quality wallet clusters from dominating the score
late crowding check: downgrades names where agreement arrives too late
consensus threshold: requires enough aligned weight before promoting a signal

Why Chorus Works On Busy Days

Wallet tracking becomes unreadable when every feed just dumps buys and sells into one stream. Chorus keeps the interpretation layer intact.

That means the operator can tell the difference between real agreement and one noisy wallet pocket dragging attention around.

Quick Start

git clone https://github.com/ChorusSignals/Chorus
cd Chorus
npm install
cp .env.example .env
npm run dev