Do announcements improve profit after costs?

LONG: 5-minute return ≥ 0.05%, 15-minute return between 0.55% and 2.5%, 60-minute return ≥ 0.05%, and taker buy share ≥ 52%. SHORT: symmetric negative return thresholds, and buy share ≤ 48%. Otherwise, valid inputs produce WAIT. These are initial research thresholds, not parameters shown to be profitable.

Returns use the same last completed one-minute close and the closes 5, 15 and 60 minutes earlier. Buy share is total taker buy base volume divided by total base volume across the last five completed candles. The original price strings are used for decimal arithmetic. No open-interest level or change is used as a signal.

Symbols: BTC, ETH, SOL, BNB, XRP, DOGE, ADA, AVAX, LINK, LTC on Binance USD-M. New legacy-v2 live-collector triggers are read regardless of their LONG/SHORT label or PASS/WATCH/SKIP decision. Their old scores, returns and flow values do not drive the new rule. However, the source already selects by price impulse, volume, pool rank and a four-hour deduplication per direction. This is a conditional sample, not the whole market. Overlapping observations are dependent.

Each new decision uses a fresh, bounded request for completed one-minute candles. All 61 required minutes must be consecutive, with valid prices, volumes and exchange close times; the latest close must be within 90 seconds of receipt and entry. Invalid inputs are excluded as data failures, not recorded as WAIT. Inputs and the rule decision are stored before execution quotes are requested.

At most three common observation slots, one per symbol. WAIT reserves its slot too. Both directional references must pass the same quote-time, spread (≤ 3bp) and quantity checks. No historical entry is recreated after a failure. Four-hour exits allow at most three minutes of delay; an unresolved exit keeps its slot. At most 40 feature attempts per UTC day. Cash earns zero interest and has no trading costs.

Funding uses official settlement marks and rates in (entry receipt, exit receipt], with history bracketing both boundaries. Missing funding leaves directional net unresolved. Waiting for the following settlement can delay the result. Seven days ends new enrollment automatically; existing observations and funding checks finish afterward. Results are visible as observations close, with no automatic extension or live-trading activation.

The market collector, original Jev protocol and original paper ledger are preserved. This experiment has its own versioned ledger. Protocol SHA-256: f530cb257302c48cef9b5d58536bae7853fe8cb283c04767ca9772b2997b9949.

Symbols: BTC, ETH, SOL, BNB, XRP, DOGE, ADA, AVAX, LINK, LTC. New oi_flow_momentum_v2 PASS candidates from the live collector. The source currently permits SHORT only; its directional advantage has not been independently validated. This is a separate market-entry simulation, not a reproduction of the legacy strategy.

Jev sees code-generated trend alignment, impulse, funding-burden and liquidity categories. No announcement text is used for this first A/B experiment. Only a skip result with confidence at least 0.70 filters a candidate. Errors, insufficient information and budget limits make B follow A, with failure costs retained. A deterministic filter's answer is logged for comparison without a third account.

Up to three shared slots, one per symbol. B's skipped slot stays reserved. Both accounts enter after the model decision and use the same fresh quote. Long: ask entry / bid exit. Short: bid entry / ask exit. A 0.05% fee applies to each side's notional. Adverse 2bp slippage per side is included in fill prices once; 5bp is a cost sensitivity case. Quantity is fixed from the assumed entry fill. No compounding.

Candidate maximum age: 10 minutes. Quote maximum age: 10 seconds. Two bounded public quote attempts. Four-hour holding period; exit observed within three minutes of its due time. Missing an exit window remains unresolved. These observations do not guarantee an exchange fill or lot-size validity.

Funding uses official settlement marks and rates in (entry, exit]. A bounded history response must bracket both boundaries before the simulated net is final under these assumptions. An empty unverified response is not zero funding. Waiting for the following settlement may delay the final number. Actual exchange charging near a boundary may differ.

AI cost uses returned input tokens at the pinned public price, with USD treated as USDT for this simulation. It is an estimate, not an account invoice. Missing charges keep B net unresolved. Reported closed balances and drawdowns omit unrealized movements and are not mark-to-market equity.

Seven days is a first review point, not proof of profitability. Enrollment stops automatically; open positions finish and funding reconciliation continues. No automatic live trading or extension. Protocol SHA-256: 4d64750ff0ade663e98ebf25674de00c5bc018bb9254050ac8aadb6635221625.