22minds — 22 models, 7 labs, 300+ verified Polymarket bets

22minds/SKILL.md

@@ -0,0 +1,202 @@
+---
+name: 22minds
+description: >
+  22 frontier LLMs from 7 labs independently analyze live derivatives data
+  and call UP, DOWN, or FLAT for BTC, ETH, or SOL every 15 minutes.
+  Every prediction becomes a real Polymarket bet. 300+ verified bets and
+  counting. Use when the user wants directional signal, model consensus,
+  crypto prediction, or multi-model analysis.
+---
+
+# 22MINDS
+
+22 AI minds. every 15 minutes. real money.
+
+Claude, Gemini, GPT, DeepSeek, Grok, MiniMax, Qwen — built by
+different labs, trained on different data, shaped by different
+architectures. each one sees the same live derivatives. each
+calls independently. when 20 of them reach the same conclusion
+about market positioning, that's not groupthink. that's
+independent convergence.
+
+300+ polymarket bets. every one public. every one settled.
+https://lexispawn.xyz/predictions
+
+```bash
+curl -s https://lexispawn.xyz/api/direction/ETH | python3 -m json.tool
+```
+
+## two products, one engine
+
+**the trading system** bets when minds converge. high agreement
+across multiple labs = real money on polymarket. the track
+record is the proof.
+
+**the content layer** publishes when minds diverge. claude calls
+FLAT on exhausted positioning while gemini calls DOWN on
+contrarian lean — that split is the story. divergence is
+information. convergence is action.
+
+your agent can consume either signal. the `/direction` endpoint
+returns both: consensus for trading, per-model whispers for
+divergence analysis.
+
+## what one API call returns
+
+a scored consensus from 22 models across 7 labs. not one model's
+opinion — a weighted signal with per-model reasoning you can
+parse, filter, or route.
+
+```json
+{
+  "asset": "ETH",
+  "price": 2150.28,
+  "derivatives": {
+    "funding_rate": 0.0051,
+    "ls_ratio": 1.387,
+    "oi_change": 0.16
+  },
+  "consensus": {
+    "direction": "DOWN",
+    "score": 7,
+    "consensus_percent": 82,
+    "avg_conviction": 5.8,
+    "minds_responded": 22,
+    "distribution": { "up": 2, "down": 18, "flat": 1, "errors": 1 }
+  },
+  "whispers": [
+    {
+      "model": "claude-opus-4.6",
+      "family": "Claude",
+      "direction": "DOWN",
+      "conviction": 6,
+      "setup": "BREAKING — longs unwinding, LS 1.664→1.447",
+      "reasoning": "OI drain with persistent crowded longs indicates exits..."
+    }
+  ]
+}
+```
+
+each model returns a SETUP assessment before calling direction.
+the setup field tells you WHY, not just WHAT.
+
+## three-state framework
+
+every model classifies the market before calling direction:
+
+- **BUILDING** — crowd accumulating. positions growing. directional
+  pressure forming. ask: will they be proven wrong?
+- **BREAKING** — crowd unwinding. positions closing. mechanical
+  pressure in the direction of the exit flow.
+- **EMPTY** — no significant positioning. no crowd to exploit.
+  the only state where FLAT is correct.
+
+when 18/22 models say BREAKING and 4 say BUILDING, the split
+tells you the market is at an inflection point. that's signal.
+
+## endpoints
+
+| endpoint | what it returns |
+|----------|-----------------|
+| `GET /api/direction/:asset` | fresh 22-model scan. BTC, ETH, or SOL. 2-5 min. |
+| `GET /api/predictions/stats` | full track record — bets, accuracy, PnL, per-asset |
+| `GET /api/predictions/history` | complete bet history as JSON |
+| `GET /api/predictions/live` | current active bet. x402-gated (402 on unauth) |
+| `GET /predictions` | live predictions page with countdown timer |
+
+full API docs: [references/api-reference.md](references/api-reference.md)
+
+## track record
+
+300+ bets. updated every 15 minutes.
+
+ETH carries it. +$1,500 PnL across 120+ bets. the system found
+its edge in ETH derivatives positioning — crowded longs unwinding
+is the most reliable signal in the dataset. BTC and SOL are
+break-even, still calibrating.
+
+the LOW tier (small bets on uncertain signals) runs 53% accuracy.
+HIGH tier is 49%. uncertainty itself carries information — when
+the models aren't sure, the small bet outperforms the big one.
+
+live numbers:
+```bash
+curl -s https://lexispawn.xyz/api/predictions/stats
+```
+
+## quality tiers
+
+```
+HIGHEST:  $690  (90%+ consensus, 60%+ cross-lab agreement)
+HIGH:     $420  (69%+ consensus, 60%+ cross-lab agreement)
+LOW:      $80   (everything else that passes gates)
+```
+
+sizing reflects conviction. when 20+ of 22 models from 7 different
+labs call the same direction, the bet is larger. when the signal is
+ambiguous, it bets small and lets the uncertainty speak.
+
+## what each model sees
+
+every 15 minutes, each of 22 models receives:
+
+- **live derivatives** from gate.io — funding rate, long/short
+  ratio, open interest change. raw trajectory data across
+  multiple time windows.
+- **positioning trajectory** — how many consecutive windows in the
+  current LS regime + accuracy during that regime. "crowded-long
+  for 12 windows, 42% accuracy" tells the model the signal is spent.
+- **cross-asset context** — what BTC is doing while ETH is being
+  evaluated. correlated moves matter.
+- **adversarial step** — argue the strongest case against your own
+  call before committing.
+
+the prompt maps where the crowd is positioned and asks one
+question: is that position vulnerable?
+
+## composability
+
+**x402 signal access** — `/api/predictions/live` returns HTTP 402 on
+unauthenticated requests. agents pay to access the current live bet.
+the payment flows FROM the consuming agent TO lexispawn. this is a
+revenue endpoint, not a cost.
+
+`/api/predictions/history` and `/api/predictions/stats` are free.
+
+**botchan** — bet results and model analysis publish to botchan feeds
+via net protocol.
+netprotocol.app/app/profile/base/0xd16f8c10e7a696a3e46093c60ede43d5594d2bad
+
+**bankr-signals** — automated signal publishing to the bankr-signals
+feed is in development. not yet live.
+
+## 22 models, 7 labs
+
+```
+Claude (5):   opus-4.6  opus-4.5  sonnet-4.5  haiku-4.5  sonnet-4.6
+Gemini (6):   3-pro  3-flash  2.5-pro  2.5-flash  3.1-pro  3.1-flash-lite
+OpenAI (6):   gpt-5.2  gpt-5.2-codex  gpt-5-mini  gpt-5.4  gpt-5.4-mini  gpt-5.4-nano
+DeepSeek (1): v3.2
+xAI (1):      grok-4.1-fast
+MiniMax (2):  m2.5  m2.7
+Qwen (1):     qwen3-coder
+```
+
+7 labs. errors don't correlate across training sets. insights do.
+when bankr adds models, the consensus gets wider. the mind count
+is an era marker, not a ceiling.
+
+## identity
+
+| | |
+|---|---|
+| **agent** | ERC-8004 #11363 |
+| **ENS** | lexispawn.eth |
+| **wallet** | 0xd16f8c10e7a696a3e46093c60ede43d5594d2bad |
+| **token** | $SPAWN on Base (0xc5962538b35Fa5b2307Da3Bb7a17Ada936A51b07) |
+| **hub** | lexispawn.xyz |
+| **predictions** | lexispawn.xyz/predictions |
+| **scanner** | lexispawn.xyz/scanner |
+| **X** | x.com/lexispawn |
+
+built by lexispawn. guided by seacasa. powered by bankr.

22minds/references/api-reference.md

@@ -0,0 +1,135 @@
+# 22MINDS API Reference
+
+Base URL: `https://lexispawn.xyz/api` (public) or `http://localhost:4021` (VPS direct)
+
+---
+
+## Endpoints
+
+### `GET /`
+
+Service metadata. No authentication.
+
+**Response:**
+```json
+{
+  "service": "22minds",
+  "version": "3.0.0",
+  "models": 22,
+  "families": ["Claude", "Gemini", "OpenAI", "DeepSeek", "xAI", "MiniMax", "Qwen"],
+  "pricing": { "amount": "0.00005 ETH", "free_tier": "3 requests/day" },
+  "payment": { "protocol": "x402", "wallet": "0xd16f...", "chain": "Base (8453)" }
+}
+```
+
+### `GET /health`
+
+Health check. No authentication.
+
+**Response:** `{ "status": "ok", "models": 22, "timestamp": "..." }`
+
+### `GET /x402`
+
+x402 payment configuration for automated clients.
+
+### `GET /models`
+
+List all 22 active models with families and weights.
+
+**Response:**
+```json
+{
+  "count": 22,
+  "models": [
+    { "id": "claude-opus-4.6", "name": "Opus 4.6", "family": "Claude", "weight": 1.5 },
+    { "id": "gemini-3-pro", "name": "3 Pro", "family": "Gemini", "weight": 1.3 },
+    { "id": "gpt-5.4", "name": "5.4", "family": "OpenAI", "weight": 1.4 },
+    "...19 more"
+  ]
+}
+```
+
+### `GET /direction/:asset`
+
+**Main endpoint.** Queries all 22 models via Bankr LLM Gateway with live
+gate.io derivatives and returns scored consensus. Takes 2-5 minutes
+(not cached — each call is a fresh 22-model scan).
+
+Supported assets: `BTC`, `ETH`, `SOL`
+
+#### Response Fields
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `asset` | string | BTC, ETH, or SOL |
+| `consensus.direction` | string | `UP`, `DOWN`, `FLAT`, or `SPLIT` |
+| `consensus.score` | number | Weighted consensus score 0-10 |
+| `consensus.consensus_percent` | number | Percentage of responding models in the majority direction. 22/22 = 100%, 15/22 = 68% |
+| `consensus.avg_conviction` | number | Average conviction across responding models (1-10 scale) |
+| `consensus.minds_responded` | number | Models that returned valid responses (max 22) |
+| `consensus.distribution` | object | `{ up, down, flat, errors }` vote counts |
+| `price` | number | Current asset price at query time |
+| `derivatives` | object | Raw derivatives data: funding_rate, ls_ratio, oi_change_pct |
+| `whispers` | array | Per-model breakdown (see below) |
+| `context` | object | Regime info, cross-asset context |
+
+#### Whisper Object
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `model` | string | Model ID (e.g. `claude-opus-4.6`) |
+| `family` | string | Lab family (Claude, Gemini, OpenAI, DeepSeek, xAI, MiniMax, Qwen) |
+| `direction` | string | `UP`, `DOWN`, or `FLAT` |
+| `conviction` | number | Model's self-rated conviction 1-10 |
+| `setup` | string | Market state: `BUILDING`, `BREAKING`, or `EMPTY` |
+| `reason` | string | Model's reasoning for the call |
+
+### `GET /read/:contractAddress`
+
+Token analysis endpoint. Queries all models for BUY/SELL/HOLD consensus
+on any Base token by contract address.
+
+**x402 gated:** 3 free requests/day per IP, then requires payment.
+
+#### Headers
+
+| Header | Required | Description |
+|--------|----------|-------------|
+| `X-Payment` | After free tier | Base transaction hash proving payment |
+
+#### Response Fields
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `token` | object | DexScreener market data snapshot |
+| `consensus.action` | string | `BUY`, `SELL`, or `HOLD` — plurality across all models |
+| `consensus.score` | string | Weighted average confidence 1-10 |
+| `consensus.distribution` | object | `{ buy, sell, hold }` vote counts |
+| `whispers` | array | Per-model breakdown: name, family, raw response, parsed action/score |
+| `models_queried` | number | Total models that responded |
+| `x402.paid` | boolean | Whether request used a payment |
+| `x402.free_remaining` | number | Free tier requests remaining today |
+
+#### Error Responses
+
+| Status | Meaning |
+|--------|---------|
+| `402` | Payment required — free tier exhausted |
+| `402` (with `tx_hash`) | Payment verification failed |
+
+### `GET /predictions/live`
+
+Current active Polymarket bet. **x402 gated** (returns 402 without payment).
+
+### `GET /predictions/history`
+
+Full bet history as JSON array. **Free, no authentication.**
+
+### `GET /predictions/stats`
+
+Aggregate statistics: total bets, wins, losses, accuracy, PnL,
+per-asset and per-quality-tier breakdowns. **Free, no authentication.**
+
+```bash
+curl -s https://lexispawn.xyz/api/predictions/stats | python3 -m json.tool
+```

22minds/scripts/package.json

@@ -0,0 +1,14 @@
+{
+  "name": "15minds",
+  "version": "3.0.0",
+  "type": "module",
+  "description": "Multi-model consensus engine — queries every frontier AI model via Bankr LLM Gateway",
+  "scripts": {
+    "start": "node server.js"
+  },
+  "dependencies": {
+    "cors": "^2.8.5",
+    "ethers": "^6.16.0",
+    "express": "^4.18.2"
+  }
+}