System Metrics

CLI Commands

Pipeline Steps

Data Sources

EVM Chains

QF Mechanisms

AI Models

1. Funding History

2. Quantitative Scoring

3. Trust Graph

4. Mechanism Simulation

5. Temporal Anomalies

6. Multi-Layer Scoring

7. Blockchain Scan

8. Code Signals

9. AI Deep Evaluation

10. Adaptive Collection

11. Signal Reliability

11-step evidence pipeline

Live at yeheskieltame-tessera.hf.space

Tessera

AI-Powered Public Goods Evaluation

20 CLI commands. 11-step evidence pipeline. 9 data sources. 9 EVM chains. Trust graph analysis, mechanism simulation, and LLM-driven evaluation for the Octant and Gitcoin ecosystems.

Launch Dashboard GitHub Read Findings

Data Port

Data Port Layer

THE PROBLEM

Public goods evaluators in the Ethereum ecosystem face three core challenges that Tessera solves.

30+ projects

422 donors

Cognitive Overload

Octant distributes millions across 30+ projects per epoch. Each has funding data, donor patterns, on-chain activity, GitHub metrics, and proposals. No human can cross-reference all of this at scale.

Donation pattern chart

Invisible Manipulation

Quadratic funding is vulnerable to whale dominance and coordination. A project can rank #1 while being 90% dependent on a single donor. Secret coordination patterns hide in plain sight.

8 dimensions to evaluate

Qualitative Bottleneck

Proposal quality, team credibility, and community engagement require real-time judgment. Too slow to apply manually across dozens of projects. Cannot be automated with rules alone.

Tessera

Tessera solves these by automating the full evaluation pipeline: collect data from 7 sources, run deterministic analysis, scan 9 blockchains, feed evidence into an LLM for synthesis, then assess signal reliability and adaptively fill data gaps. For concrete evidence, read FINDINGS.md.

11-STEP EVIDENCE PIPELINE

Each step produces structured data that accumulates. Steps 1-8 are deterministic. Step 9 synthesizes via AI. Steps 10-11 assess signal quality and fill gaps.

1Funding History

Cross-epoch allocations, matched funding, donor counts from Octant REST API

ETH

Octant

2Quantitative Scoring

K-means clustering (Lloyd's algorithm). Composite score: 40% allocated + 60% matched

Analysis

3Trust Graph

Shannon entropy, Jaccard similarity, and union-find donor clustering

Analysis

4Mechanism Simulation

Standard QF, Capped QF, Equal-Weight, Trust-Weighted QF with Gini coefficients

Std

Cap

Equal

Trust

Analysis

5Temporal Anomalies

Donor surge/exodus, funding spikes, new whale entries, coordination shifts

Analysis

6Multi-Layer Scoring

5 dimensions: Funding (25%), Efficiency (25%), Diversity (30%), Consistency (20%)

Fund

Eff

Div

Con

ALL

Analysis

7Blockchain Scan

9 EVM chains concurrent scan: balance, transactions, via eth_call

ETH

Base

Arb

Mantle

Scroll

Linea

zkSync

Monad

$ balance33 txs

RPC

8Code Signals

OSO metrics or GitHub API fallback: commits, contributors

OSO/GitHub

Steps 1-8 feed into Step 9 (AI Synthesis)

AI Deep Evaluation

AI Required

Evidence-grounded narrative using ALL data from steps 1-8 via LLM. Produces trajectory analysis, organic vs gaming assessment, counterfactual impact, and confidence-rated recommendation.

ClaudeGeminiOpenAIAI Provider

Tessera Report

Branded PDF

Output: Branded PDF report with funding history, trust profile, multi-layer scores, mechanism simulation, temporal anomalies, blockchain activity, and AI narrative assessment.

Steps 10-11: Signal Quality Framework

10Adaptive CollectionNEW

Detect data gaps, auto-discover OSO, GitHub, Discourse, RetroPGF, Gitcoin

Multi-Source

11Signal ReliabilityNEW

Classify every signal as HIGH/MEDIUM/LOW — cross-verify, assess freshness

Assessment

System Architecture

Go binary (9MB) serves CLI + HTTP API + static frontend. Single process, zero dependencies.

Multi-Chain Scanner (9 EVM Chains)

Ethereum

ETH

USDC, USDT, DAI

Base

ETH

USDC, DAI

Optimism

ETH

USDC, USDT, DAI

Arbitrum

ETH

USDC, USDT, DAI

Mantle

MNT

USDC, USDT

Scroll

ETH

USDC, USDT

Linea

ETH

USDC, USDT

zkSync Era

ETH

USDC, USDT

Monad

MON

Testnet

AI Provider Fallback Chain

Claude CLIprimary

opus-4-6, sonnet-4-6

Claude Code login (no key)

Claude API

opus-4-6, sonnet-4-6, haiku-4-5

ANTHROPIC_API_KEY

Gemini

2.5-pro, 2.5-flash, 3.1-pro, 3-flash, 2.5-flash-lite

GEMINI_API_KEY

OpenAI

gpt-4o, gpt-4o-mini, o3-mini

OPENAI_API_KEY

120s timeout per request. If preferred provider fails, next is tried automatically.

Analysis Algorithms

Deterministic, reproducible analysis using established mathematical methods. Every formula is transparent and auditable.

Composite Score

formula

(normAlloc x 0.4 + normMatch x 0.6) x 100

Matched funding weighted 60% to capture breadth of support via QF amplification.

Shannon Entropy

formula

H = -sum(p_i x log(p_i)) / log(n)

Normalized donor diversity. 0 = single donor, 1 = perfectly uniform distribution.

Jaccard Similarity

|A intersection B| / |A union B|

Donor set overlap between projects. Flagged if > 0.7 (coordinated behavior).

Trust-Weighted QF

score = QF x (0.5 + 0.5 x diversity)

Novel mechanism. Whale-dominated projects (diversity near 0) get 50% penalty. Organic support preserved.

Gini Coefficient

sum((2r_i - n - 1) x a_i) / (n x sum(a_i))

Funding inequality measure. 0 = equal distribution, 1 = all to one project.

Whale Dependency

max(donor_amount) / total_amount

Single-donor concentration. Flagged if > 0.5 (one donor provides majority).

KEY FINDINGS

Real insights from Octant Epoch 5: 30 projects, 1,902 donations, 422 unique donors.

All findings are reproducible by running Tessera CLI commands. For full methodology and interpretation, read FINDINGS.md.

97.9%

Whale Concentration

Top 10% of donors control nearly all funding. Structural, not episodic (92-98% across 4 epochs).

36.6

Rank #1 True Score

Project ranked #1 by composite (89.5) drops to 36.6 under multi-layer scoring. 90% whale-dependent.

Donor Clusters

Jaccard > 0.7 overlap detected. Largest cluster: 39 donors moving in lockstep. Increasing over time.

3,105%

Mechanism Impact

Equal Weight QF would increase smallest project funding by 31x. Trust-Weighted QF balances fairness vs sybil resistance.

Whale-Controlled Projects

Single address 0x2585 controls 90-99% of 5 projects simultaneously. 17% of the ecosystem.

931%

Funding Spike

Epoch 4 to 5 surge with fewer donors. Textbook whale-driven behavior flagged by temporal anomaly detection.

Read Full Analysis in FINDINGS.md

Get Started

From zero to a running dashboard, or try the live demo instantly. Click each section below to expand.

Skip setup and try it now:

Open Live Demo

Try Tessera instantly from your browser.

1. Open the live dashboard

Browser

https://yeheskieltame-tessera.hf.space/dashboard

2. Connect your Claude CLI (optional) Coming Soon

Terminal

npx tessera-bridge

In Development: The tessera-bridge feature (local Claude CLI proxy) is currently being developed. For now, use the Self-Host option below or configure an API key in the dashboard.

Clone the repository and build the Go binary. Requires Go 1.25+ and Node.js 20+.

Terminal

git clone https://github.com/yeheskieltame/Tessera.git
cd Tessera
go build -o tessera ./cmd/analyst/

The binary is ~9MB with zero runtime dependencies. Works on macOS, Linux, and Windows.

Choose one of the following. Claude CLI is auto-detected if installed.

Option A: Claude Max Plan (recommended, no API key)

npm i -g @anthropic-ai/claude-code
claude login

Option B: Gemini API Key

echo 'GEMINI_API_KEY=your-key-here' > .env

Option C: Any other provider

# Set one or more in .env
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
GEMINI_API_KEY=AI...

Fallback order: Claude Local → Claude CLI → Claude API → Gemini → OpenAI. If one fails, the next is tried automatically. Quantitative commands work without any AI provider.

Only needed if you want the web dashboard. CLI works without it.

Terminal

cd frontend && npm install && npm run build && cd ..

Next.js 19 static export. The Go server serves the dashboard from ./frontend/dist/ automatically.

Start the HTTP API and web dashboard with a single command.

Terminal

./tessera serve
# Server starts on http://localhost:3001
# Dashboard: http://localhost:3001
# API: http://localhost:3001/api/status

CLI Commands Reference

The main command. Runs all 9 analysis steps against an Octant project address and generates a branded PDF report.

Usage

./tessera analyze-project 0x9531C059098e3d194fF87FebB587aB07B30B1306 -e 5

# Optional flags:
#   -e <epoch>     Specify epoch (default: latest)
#   -n <oso-name>  OSO project name for code metrics

1. Funding history

2. K-means clustering

3. Trust graph

4. 4 QF simulations

5. Temporal anomalies

6. Multi-layer scores

7. 9-chain blockchain scan

8. Code signals

9. AI deep evaluation

Output: PDF report saved to reports/ directory. Also available via dashboard with real-time SSE streaming.

Evaluate any public goods project across 8 dimensions using AI. Optionally enrich with GitHub data.

Usage

./tessera evaluate "Gitcoin Passport" \
  -d "Decentralized identity verification for sybil resistance" \
  -g "https://github.com/gitcoinco/passport"

# Flags:
#   -d <description>  Project description (required)
#   -g <github-url>   GitHub repo for enrichment (optional)
#   -c <context>      Additional context (optional)

Impact Evidence

Team Credibility

Innovation

Sustainability

Ecosystem Fit

Transparency

Community

Risk Assessment

Analyze an entire Octant epoch. All commands are deterministic and reproducible.

Epoch Commands

# List all projects in an epoch
./tessera list-projects -e 5

# K-means clustering + composite scoring
./tessera analyze-epoch -e 5

# Whale concentration + coordinated donation detection
./tessera detect-anomalies -e 5

# Donor diversity, Jaccard similarity, coordination risk
./tessera trust-graph -e 5

# Compare 4 QF mechanisms side-by-side
./tessera simulate -e 5

# Full epoch intelligence report (4-step)
./tessera report-epoch -e 5

Track a single project across epochs or scan any address across 9 EVM blockchains.

Project Commands

# Cross-epoch timeline + temporal anomalies + multi-layer scores
./tessera track-project 0x9531C059098e3d194fF87FebB587aB07B30B1306

# Scan address across 9 EVM chains (ETH, Base, OP, Arb, Mantle, Scroll, Linea, zkSync, Monad)
# Returns: balance, tx count, contract status, USDC/USDT/DAI balances
./tessera scan-chain 0x9531C059098e3d194fF87FebB587aB07B30B1306

# Gitcoin Grants round analysis
./tessera gitcoin-rounds -r 23 --chain 42161

Commands that use the AI provider chain for qualitative analysis.

AI Commands

# Multi-epoch deep evaluation with trajectory analysis
./tessera deep-eval 0x9531C059098e3d194fF87FebB587aB07B30B1306 -n "octant"

# Proposal verification: checks claims against evidence
./tessera scan-proposal "Project X" -d "We have 10,000 daily active users..."

# Extract structured impact metrics from free text
./tessera extract-metrics "Our protocol processed $2M in Q4, onboarded 500 users..."

# Collect cross-source signals (OSO + GitHub + blockchain)
./tessera collect-signals "gitcoin-passport"

Check connectivity, manage AI providers, and interact on Moltbook.

Utility Commands

# Check all data sources, blockchain RPCs, and AI provider status
./tessera status

# Show configured AI providers and fallback order
./tessera providers

# Moltbook social commands
./tessera moltbook status           # Agent profile, karma, followers
./tessera moltbook post             # Publish a post
./tessera moltbook reply            # Reply to a post
./tessera moltbook follow           # Follow an agent

# Auto-reply to notifications
./tessera heartbeat --loop

Variable	Required	Purpose
ANTHROPIC_API_KEY	1 of 3	Claude API access
GEMINI_API_KEY	1 of 3	Google Gemini access
OPENAI_API_KEY	1 of 3	OpenAI access
OSO_API_KEY	No	Open Source Observer API
MOLTBOOK_API_KEY	No	Moltbook social network
PORT	No	Server port (default: 3001)
CLAUDE_CLI_DISABLED	No	Skip Claude CLI auto-detection

At least one AI provider key is needed for AI commands. Quantitative commands and blockchain scanning work without any key. Claude CLI is auto-detected if the claude binary exists in PATH.

Team

Built for The Synthesis hackathon. Human and Agent collaborating as equals.

Human

Yeheskiel Yunus Tame

@YeheskielTame

Direction, domain context, decision-making, GitHub repo management.

Agent

Synthesis Agent #32417

Claude Opus 4.6 via Claude Code

Architecture, algorithms, implementation, documentation, deployment.

ERC-8004 on BaseView TX

Tracks

$31,308

Prize Pool

69+

Commits

Phases

Ready to evaluate?

Launch the dashboard, enter a project address, and get a full intelligence report in minutes.

Launch Dashboard Live Demo GitHub