Home/Methodology
The methodology
How AgentCrush ranks the agent economy
AgentCrush is the evidence-ranked index of the agent economy. We don't pick winners — we publish multi-signal evidence with transparent weights. Different agent categories leave different evidence trails, so we run four category-specific methodologies, each with its own signal sources, weights, and evidence-ready rule.
Principles
Multi-signal corroboration. No agent is evidence-ranked on a single signal. Every category requires at least 3 of N signals available, AND at least one of those signals must be a capability signal — not just popularity. Downloads and stars are vanity metrics on their own.
Per-category methodology. A model family leaves HuggingFace downloads and LMArena scores; a tokenized agent leaves on-chain liquidity and holder distribution; a service agent leaves GitHub forks and Agentverse interactions. Running one universal scoring function across all of them would average away the truth.
Methodology travels with data. Every category page publishes its full signal set, weights, formulas, evidence-ready rule, and scope notes. The same methodology is exposed via our MCP server so LLMs querying AgentCrush can correctly explain HOW a ranking was computed — not just what it is.
Honest gaps. Where a signal isn't yet populated for an agent (no LMArena coverage, no citations indexed, etc.), the methodology returns NULL — not 0. That distinction matters: NULL means "unmeasured," 0 means "measured at zero." The composite weights unmeasured signals as missing rather than failing.
Live coverage
179 total evidence-ranked agents across 5 categories.
Open source
Scoring views — read the SQL
Every ranking comes from a Postgres view. Each one is published verbatim with weights, evidence-ready rules, and a GitHub link.
Proof of Index — on-chain data integrity
Every night, AgentCrush computes a SHA-256 digest over that day's full snapshot export and notarizes it on Base. Once a digest is on-chain, the historical archive behind every ranking and every Ghost Index reading is tamper-evident — you don't have to trust that we didn't rewrite history, you can check. Oracle attestations from /api/oracle/attest reference the latest digest.
latest digest bb5630512f25efff436f14f1e64f7d4a67448067a2ed40e96737cc3714622ed5
covers 1,394 snapshot rows · 2026-06-29
notarized 0xb1b544dfd3fafc579d05b7f3f5dec814105e0acb15cb9bd33a58fab48e7acb1d
Verify: recompute the SHA-256 over the canonical JSON of the day's snapshot rows (recursively key-sorted, rows ordered by id) and compare with the tx calldata. Full history: /api/proof-of-index/v1
Ghost Index ingestion coverage
The Ghost Index counts an agent as “alive” when activity_status = ‘active’ or last_event_at is within 30 days. Whether that signal exists depends on whether a worker writes to it for that category. v1.0 wires activity ingestion for some categories, not others — and we surface that gap honestly instead of reporting 0% for categories we're not actually measuring.
For pending categories the index correctly shows the agents as indexed-but-no-activity-signal — surfaced on /ghost-index as pending rather than 0%. The aggregate headline liveness number includes all categories: it's a lower bound, and will rise as ingestion coverage grows.
Model Families
v1.4-with-deploymentScores model families (Hermes, Llama, Mistral, Qwen, DeepSeek, etc.) on adoption, capability, downstream usage, research impact, and cross-protocol agent-economy deployment.
Signals
Weighted basket of 5 sub-scoresLEAST(100, ROUND((MAX(arena_score) − 700) / 8))LEAST(100, ROUND(LOG10(SUM(derivatives_count)) × 25))LEAST(100, ROUND(LOG10(SUM(citation_count)) × 16))LEAST(100, ROUND(LOG10(SUM(deployment_count)) × 30))Evidence-ready rule
3 of 5 signals AND ≥1 capability signal (derivatives, LMArena, citations, or deployment).
Scope & coverage — what this version measures, and what's next
- Covers the big-10 model families (OpenAI, Claude, Gemini, Llama, Qwen, DeepSeek, Mistral, Grok, Cohere, Hermes). New families enter the view as soon as they leave public signals.
- Paper citations follow Semantic Scholar indexing — recently published papers can take weeks to register.
- Deployment counts measure family-level adoption breadth, not deployment of one specific variant. Read them as reach, not precision.
Changelog
Tokenized Agents
v1.1-tokenized-tvlScores tokenized AI agents (Virtuals Protocol, etc.) economics-first: market cap, on-chain liquidity, holder distribution, capital locked, plus social visibility.
Signals
LEAST(100, ROUND(LOG10(market_cap_usd) × 12))liquidity_score × 0.65 + volume_score × 0.35holders_count_score × 0.55 + (100 − top10_pct) × 0.45GREATEST(0, LEAST(100, 50 + price_change_pct))LEAST(100, ROUND(LOG10(tvl_usd) × 14))socially_visible ? 100 : 0Evidence-ready rule
3 of 6 signals AND ≥1 economic signal (mc, liquidity, holders, or TVL > 0).
Scope & coverage — what this version measures, and what's next
- Coverage today: Virtuals Protocol (16 evidence-ranked). Additional tokenized ecosystems are on the integration roadmap.
- Social signal is a curated flag in v1.1; v1.2 integrates X follower volume + Farcaster engagement.
- Cross-protocol presence is tracked but unweighted until the signal has enough ecosystem coverage to be meaningful.
Changelog
Service Agents
v1.1-service-forksScores service agents (A2A protocol, Agentverse, x402, ERC-8004) on adoption, source quality, activity recency, protocol breadth, fork engagement.
Signals
GREATEST(stars_log×18, interactions_log×22)GREATEST(a2a_signal_strength, ROUND(av_rating × 20))Time-bucketed: 7d→100, 30d→80, 90d→60, 180d→40, 365d→20LEAST(100, COUNT(protocols) × 25)LEAST(100, ROUND(LOG10(forks) × 22))currently NULL (placeholder)Evidence-ready rule
3 of 6 signals AND ≥1 adoption signal (stars > 0, interactions > 0, or forks > 0).
Scope & coverage — what this version measures, and what's next
- Sources today: A2A protocol crawl (28 agents). Agentverse ingestion is wired and awaiting fresh crawl data.
- Roadmap v1.2: ERC-8004 registry (29K agents) and Bazaar x402 endpoints (46K) join as additional service surfaces.
- Cross-protocol presence is tracked but unweighted in the v1.1 composite.
Changelog
Developer Agents
v2.c-publicScores developer-tool agents (frameworks, runtimes, dev tools) on GitHub activity, package usage, dependency adoption, ecosystem links, docs, discourse, and trust signals. The universal ranking surface.
Signals
weighted by active_weight_totallog-scaled per ecosystemlog-scaled countcomposite heuristic 0-100graph-distance scorelog-scaledcomposite 0-100Evidence-ready rule
Multi-signal coverage threshold OR top-100 ranked OR single signal ≥ 90 with ≥ 2 corroborating signals > 50.
Scope & coverage — what this version measures, and what's next
- Weights are computed per agent from available signal coverage (active_weight_total) rather than fixed percentages — agents are scored on what can actually be measured about them.
- The public ranking lists the evidence-ranked subset; the universal ranking scores the full index behind it.
Changelog
Agent Payments Stack
v1.0-apsA 6-layer map of the agent payments infrastructure — settlement, wallets, routing, protocol, governance, application. Scores projects (companies, protocols, standards) by stack depth. Inspired by Keyrock "Who Pays the Agent?" (May 2026), kept live and methodology-disclosed.
Signals
presence × 4presence × 3presence × 2presence × 4presence × 5presence × 3Evidence-ready rule
All tracked projects are displayed. No evidence-ready gate — layer coverage is the qualification.
Scope & coverage — what this version measures, and what's next
- This is a project taxonomy, not an agent ranking — entries are companies, protocols, and standards.
- Layer coverage is determined by documented public evidence only; self-reporting is not accepted.
- Scores here are not comparable across categories — the maximum is 21 (all 6 layers covered).
- Source: Keyrock "Who Pays the Agent?" (May 2026), extended and kept live by AgentCrush.
Changelog
For machine consumers
The same methodology is exposed via our MCP server. LLMs (Claude Desktop, Cursor, custom agents) can query AgentCrush as a live data layer and explain ranking decisions accurately.
Endpoint
POST https://agentcrush.xyz/api/mcp/v1Discovery
GET https://agentcrush.xyz/.well-known/mcp.jsonVersion history
Each version bump changes signal weights, adds new signals, or adjusts the evidence-ready rule. Agents that were borderline evidence-ranked may move when a methodology version changes.