The Kinfield open dataset holds the trades of everyone who chose to share them: each decision with the market context it was made in, the action, the intent and the outcome, together with the hash-chained ledger every trade can be checked against. It lives at KinfieldLabs/kinfield-datasets on GitHub.
Whose trades are in it
- Only signed-in traders who chose to share. Kinfield asks once after your first sign-in, and publishes nothing before you answer.
- Guest trades are never published; they never leave your browser.
- Agents appear under their Kinfield names, such as
KIN001. A GitHub username appears only if its owner chose to show it. - Written reasons are private unless their owner shares them; otherwise only a salted commitment is published.
When it is published
Every night at 03:00 UTC, a scheduled job rebuilds the dataset from the ledger and commits it to the repository. Only agents whose files changed are sent, and every file is generated deterministically, so an unchanged record produces byte-identical files and no diff. The commit message lists which agents were updated, renamed or withdrawn.
Turning sharing off, or deleting your account, removes your agent's folder at the next publish. Earlier versions remain in the repository's git history.
What is in it
| Path | Contents |
|---|---|
agents.json | One row per agent, with headline stats, most trades first |
README.md | An overview, the readiness thresholds, and a table of agents |
verify.mjs | The verifier: checks everything below with no dependencies |
agents/<agent>/summary.json | Stats and readiness for each simulated account, with each ledger's length and head hash |
agents/<agent>/trades.jsonl | Every closed trade the person made, one JSON object per line |
agents/<agent>/autonomous.jsonl | Trades the agent made on its own account, if any |
agents/<agent>/ledger/<account>.jsonl | The full hash chain of each account, archived ones included |
The trade record
Each line of trades.jsonl and autonomous.jsonl is one closed trade, ordered by close time:
trades.jsonl, one line, formatted
{
"position_id": "4f1c2a0e-6c1b-4f7e-9a51-0b8d2f6e7c11",
"account_id": "a3d9...",
"order_id": null,
"decided_by": "human",
"market": "BTC",
"type": "market",
"opened_at": "2026-11-17T13:05:41.880Z",
"closed_at": "2026-11-17T14:32:05.112Z",
"context": {
"ret5m": 0.00118, "ret15m": 0.00342, "ret60m": -0.00591,
"vol15m": 0.00071, "rangePos60m": 0.62, "volumeRatio": 1.84
},
"action": {
"side": "long", "size_usd": 10000, "leverage": 5,
"entry_px": 100000, "entry_fee_usd": 4.5,
"tp_px": 104000, "sl_px": 98500
},
"intent": {
"strategy": "Breakout",
"reason_commitment": "5d0c...a7e2"
},
"outcome": {
"exit_px": 102000, "exit_reason": "manual",
"pnl_usd": 190.91, "fees_usd": 9.09,
"return_on_margin": 0.095455, "liquidated": false
}
}| Field | Meaning |
|---|---|
decided_by | human in trades.jsonl, agent in autonomous.jsonl. Agent trades add agent_p_long, its probability of up. |
type | market, or limit with its order_id. |
context | The six market features at the decision, or null if they couldn't be captured. |
action | Side, notional size in USD, leverage, entry price, entry fee, and take profit and stop loss if set. |
intent | The strategy tag and reason_commitment. If reasons are shared, also reason and reason_salt. |
outcome.exit_reason | manual, take_profit, stop_loss, liquidation or time_stop. |
outcome.pnl_usd | Realized P&L net of both fees. |
outcome.return_on_margin | P&L divided by margin (size / leverage). |
summary.json
agents/KIN001/summary.json, shortened
{
"agent": "KIN001",
"github": null,
"shares_reasons": false,
"updated_at": "2026-11-17T14:32:05.230Z",
"lifetime": { "trades": 142, "wins": 79, "win_rate": 0.5563, ... },
"autonomous": { "trades": 36, ... },
"accounts": [
{
"account_id": "a3d9...",
"kind": "human",
"status": "active",
"created_at": "2026-10-12T09:14:00.000Z",
"archived_at": null,
"starting_balance": 100000,
"ledger": { "file": "ledger/a3d9....jsonl", "entries": 311, "head": "e07a...5f19" },
"stats": { ... },
"readiness": {
"as_of": "2026-11-17T14:32:05.230Z",
"checks_met": 5, "checks_total": 7, "meets_all": false,
"checks": { "days": { "met": true, "value": "30 of 30 days", "needs": "30 days" }, ... }
}
}
]
}Stats
lifetime covers the person's trades across every account; autonomous the agent's own; each account has its own stats. All use the same fields:
trades, wins, win_rate, net_pnl_usd, profit_factor, expectancy_usd (average P&L per trade), max_drawdown, best_trade_share, return (net P&L over the starting balance), active_days, avg_hold_minutes, first_trade_at and last_trade_at. See how each is computed.
agents.json
[
{
"agent": "KIN001",
"github": null,
"trades": 142,
"win_rate": 0.5563,
"profit_factor": 1.41,
"net_pnl": 3120.55,
"max_drawdown": 0.0712,
"last_trade_at": "2026-11-17T14:32:05.112Z"
}
]Using it
git clone https://github.com/KinfieldLabs/kinfield-datasets
cd kinfield-datasets
node verify.mjsJSONL loads directly in most tools, for example pandas.read_json(path, lines=True) in Python, or line by line with JSON.parse. Check it first: see Verifying the data.
Licence
The dataset is licensed CC BY 4.0. You may share and adapt it for any purpose, including training models, as long as you credit Kinfield and link to kinfield.si.
All trading in the dataset is simulated, with no real funds. Nothing in it is financial advice, and past simulated results don't predict future performance.