ARCANAreversion_v1
v2LIVERANKEDby arcana_labs·mean_reversion·us_equities·since 2026-09-09·LIVE
Window analysed
TICKS
18
DECISIONS
18
TRADES · THE AGENT’S
8
protective exits are not counted here
INCLUDING PROTECTIVE
8
0 taken by a level
NAV · FIRST
100,000.00
NAV · LAST
99,575.38
RETURN
−0.4246%
PERIOD
2026-09-10 20:52:53Z
2026-09-13 17:16:34Z
2026-09-13 17:16:34Z
Protective exits
No position was exited by a level; every trade was the agent's own decision.
excluded from: decision_timing, summary.trades
Where the P&L came from
| Symbol | P&L | % of starting NAV | How it was obtained |
|---|---|---|---|
| META | 52.30 | 0.0523 | exact, from held quantity × price change |
| GOOGL | -127.90 | -0.1279 | exact, from held quantity × price change |
| NVDA | -349.01 | -0.3490 | exact, from held quantity × price change |
| cash drag | — | 0.9694 | Counterfactual estimate: idle cash at each tick multiplied by the market index return into the next tick. It assumes the cash could have been deployed at the market rate, which the agent never attempted — an estimate, not a measured loss. |
Per-symbol P&L is exact from held quantity x price change; only cash drag is estimated. 17 ticks could be paired.
Risk
MAX DRAWDOWN
0.6207 %
PEAK → TROUGH
100,121.40 → 99,499.94
2026-09-12 21:11:18Z → 2026-09-13 13:14:31Z
DURATION
4 ticks
RECOVERED?
not yet
| Time (UTC) | Action | Decided by | Symbol | Quantity | Rationale |
|---|---|---|---|---|---|
| 2026-09-12 21:11:18Z | BUY | agent · deterministic | NVDA | 73.2800 | mean reversion: NVDA down -0.37%, buying the dip with 73.28 shares |
| 2026-09-13 05:12:31Z | BUY | agent · deterministic | NVDA | 73.5300 | mean reversion: NVDA down -0.25%, buying the dip with 73.53 shares |
| 2026-09-13 09:13:26Z | BUY | agent · deterministic | GOOGL | 47.0500 | mean reversion: GOOGL down -0.70%, buying the dip with 47.05 shares |
| 2026-09-13 13:14:31Z | BUY | agent · deterministic | META | 24.8100 | mean reversion: META down -0.91%, buying the dip with 24.81 shares |
These trades occurred inside the drawdown window. Co-occurrence is not causation and no causal claim is made here.
Decision timing
{"analysed":true,"trades":8,"window_ticks":5,"buys":{"count":7,"avg_price_percentile":35.9914,"avg_forward_return_pct":-0.3383,"reading":"Percentile 0 = bought at the lowest price in the surrounding window, 100 = the highest."},"sells":{"count":1,"avg_price_percentile":60,"avg_forward_return_pct":-0.2804,"reading":"For a sell, a HIGH percentile means it sold near the top of the surrounding window."},"best_entries":[{"ts":"2026-09-13T13:14:31.768Z","action":"buy","symbol":"META","price":641.4783,"percentile":0,"forward_return_pct":0.3523,"rationale":"mean reversion: META down -0.91%, buying the dip with 24.81 shares"},{"ts":"2026-09-11T21:07:31.511Z","action":"buy","symbol":"GOOGL","price":338.7996,"percentile":30,"forward_return_pct":0.0679,"rationale":"mean reversion: GOOGL down -0.63%, buying the dip with 47.22 shares"},{"ts":"2026-09-13T17:16:34.970Z","action":"buy","symbol":"GOOGL","price":337.624,"percentile":0,"forward_return_pct":0.0046,"rationale":"mean reversion: GOOGL down -0.38%, buying the dip with 23.70 shares"}],"worst_entries":[{"ts":"2026-09-12T21:11:18.964Z","action":"buy","symbol":"NVDA","price":218.6041,"percentile":50,"forward_return_pct":-1.2885,"rationale":"mean reversion: NVDA down -0.37%, buying the dip with 73.28 shares"},{"ts":"2026-09-13T05:12:31.507Z","action":"buy","symbol":"NVDA","price":217.7266,"percentile":44.44,"forward_return_pct":-0.9227,"rationale":"mean reversion: NVDA down -0.25%, buying the dip with 73.53 shares"},{"ts":"2026-09-13T09:13:26.780Z","action":"buy","symbol":"GOOGL","price":339.1625,"percentile":37.5,"forward_return_pct":-0.449,"rationale":"mean reversion: GOOGL down -0.70%, buying the dip with 47.05 shares"}]}
Volatility
agent nav volatility
0.001008
market volatility
0.002944
avg exposure
0.305800
volatility per unit exposure
0.003298
ratio vs market
1.120200
Per unit of exposure the book tracked the market index closely.
The difference may come from trading, from holding a mix that differs from the equal-weighted index, or from both. This measure does not separate them.
What was NOT analysed, and why
sector rotation — The data blocker is gone — the universe is now 50 symbols across 11 GICS sectors and every snapshot quote carries its sector (services/market-data/universe/). What is missing is the analysis itself, which is Autopsy 2.0 work rather than a schema gap. It also needs enough real-market history for a rotation to be distinguishable from a few coincidental trades.
thesis failure — decisions.rationale is populated for every row, but it records the rule that fired ("momentum: AAPL up 0.26% since last tick"), not a forward-looking thesis. There is no claim about the future to test against the outcome, so testing one would mean inventing the thesis first.
Provenance
Prices came from robinhood_pool. This is real market data.
Prices came from robinhood_pool — real market data. Findings describe conduct in the real market, within the limits listed under not_analysed.