GodzillaBTCCurrent publication policy: v4.8 evidence guard
New EARLY ideas are directional watchlist research only and cannot create a paper position. TACTICAL needs at least 20 non-overlapping observations and 6 chronological holdout outcomes, complete daily/hourly context, a fixed direction/regime cohort, an 82% posterior edge hurdle and positive cost-stressed expectancy. Standard and Confirmed keep their existing sample floors; all tiers require complete context. These are conservative design requirements, not statistically proven accuracy.
The existing 10-bar swing / 1.1–1.8 ATR stop rule, targets and technical direction remain unchanged. A proposed structural-stop/opposition-filter variation failed the limited diagnostic and was not promoted. Legacy positions settle under their original rules. In v4.8, internal data gaps wait for the horizon plus two-hour ingestion grace before conservative ambiguous accounting; partial entry bars and both-level bars remain ambiguous.
Audit every model cohort →Validation & performance evidence
Software correctness is tested. Market profitability remains unproven.
Verified software behavior
The included regression suite checks indicator math, invalid/open bars, stale quotes, timeframe alignment, overlapping labels, stop gaps, ambiguity, publication races, immutable levels, portfolio caps, outages and API behavior. It uses deterministic fixtures and a local SQLite D1 adapter. These tests establish software behavior, not investment returns.
Performance not yet demonstrated
No untouched external market backtest, audited broker execution record, real-money forward evaluation, measured slippage study or live probability-calibration study was supplied. No particular win rate or profit is promised. The revised model needs prospective paper evaluation before financial reliance.
Costs and uncertainty
Historical replay deducts a 0.05R cost floor and tests doubled costs. Forward accounting uses at least that floor or quoted spread/risk plus a 0.02R allowance, whichever is larger. Actual spread, fees, funding and fills can differ. Ambiguous outcomes receive a conservative -1R gross accounting assumption. Beta scores and Wilson intervals do not capture every source of model-selection uncertainty.
Read results correctly
Statistics cover at most the most recent 500 published records. They are not all-time totals. Decisive win rate uses WIN/LOSS only; inspect ambiguity, cancelled trades, timeouts, net R and model version alongside it. High decisive win rate can coexist with negative net R. Legacy outcomes are preserved; current-model results are reported separately.
Data and hosting limits
The model cannot create reliable signals from stale or incomplete history. Biquote documents finite history and a 1,000-bar request limit. Requests also need to fit Cloudflare CPU and D1 limits. Replay caching reduces repeat work, but local tests cannot establish production resource usage. The package does not set up continuous background execution.