KRW market prices, turnover, breadth and activity are updated continuously on the Analysis Server.
METHODOLOGY
How market state becomes a real trading decision.
ByteCosmos makes the decision in three steps: identify the current market regime, check how each strategy performed in that same regime, then decide whether actual trading is allowed. One overall win rate does not decide everything.
X = capital flow, Y = breadth, Z = established trend persistence. The current L1 cell is refreshed through Sliding 15-vote.
Every strategy keeps a separate Forward SIM + Actual record for every XYZ cell it has experienced.
The current XYZ cell alone decides whether the strategy may open a new actual position.
Qualified strategies can trade. Everyone else continues Forward SIM and accumulates regime-specific evidence.
1. L1 Market Regime
The execution regime is a 4 × 4 × 5 categorical space: 80 distinct market states.
XCapital Inflow
Rolling 15-minute capital-flow pressure derived from cross-market return and activity behavior.
YMarket Breadth
The share of the market participating in the recent move. This distinguishes broad moves from narrow leadership.
ZTrend Persistence
MA5 / MA20 slope structure on synthetic 15-minute bars. It measures established trend direction, not just the latest minute.
Primary decision
Sliding 15-vote. Every minute, the latest 15 completed Rolling-15M classifications vote for the current L1 XYZ state.
Shadow decision
Fixed 15M Majority. Recorded in parallel for research and comparison only. It never controls live execution.
Rolling preview
The newest single Rolling-15M observation. Useful for seeing change early, but it is not the live gate by itself.
2. Strategy evidence
Each strategy has up to 80 independent evidence buckets. Good performance in one regime is not cancelled by poor performance somewhere else.
Live evidence
Forward SIM + Actual combined. Both sources are counted inside the same Strategy × XYZ cell.
Historical backtest
Context and research only. Historical performance is not the live promotion gate when Forward/Actual evidence exists.
Strategy-wide Win Rate
Context only. It can explain overall behavior, but it cannot block a current XYZ cell that independently qualifies.
Continuous learning
Forward SIM keeps running whether a strategy is Potential, Live or Exit Only, so every XYZ cell can continue accumulating evidence.
3. Live thresholds by execution group
Canonical Strategy Families define the Strategy Grammar for new Adaptive af_* lineages. Existing strategies may be mapped to a Family for comparison, but their logic stays fixed and they are never mutated or versioned by the Adaptive engine. Live thresholds are applied separately by execution group.
| Execution group | Minimum evidence | Enter Live | Maintain Live |
|---|---|---|---|
| Scalp | 10 combined closed | Combined Win Rate ≥ 80% | Combined Win Rate ≥ 70% |
| Lead-Lag | 10 combined closed | Combined Win Rate ≥ 80% | Combined Win Rate ≥ 70% |
| Core | 10 combined closed | Combined Win Rate ≥ 60% | Combined Win Rate ≥ 50% |
Profitability check
The current cell must also have positive cumulative return, Profit Factor ≥ 1 and positive capital-weighted return for both Live entry and Live maintenance. Failing any maintain gate removes new-entry permission.
Funding
Actual funding is ₩30,000 per trade when the current XYZ cell is Live-qualified and execution capacity is available.
4. Unified lifecycle
Core, Scalp, Lead-Lag and all promoted adaptive Strategy Versions share the same XYZ-based lifecycle.
POTENTIAL
Default state. The strategy can become Live in any XYZ cell that later qualifies.
LIVE
The current XYZ cell qualifies on Forward SIM + Actual combined evidence.
EXIT ONLY
The cell no longer qualifies, but an actual position is still open. No new entries; the strategy manages its own exit.
ARCHIVED
Reserved for strategies with statistically severe failure confirmed across the entire 80-cell XYZ space.
Archive evidence
ARCHIVED can be produced only from post-reset Forward SIM + Actual evidence under the primary L1 XYZ decision method. Historical Backtest, registry labels, aliases and research-only objects can never archive a strategy.
Coverage requirement
Every one of the 80 XYZ cells must contain at least 20 closed trades. A single unobserved or under-sampled cell is enough to keep the strategy POTENTIAL.
Severe failure
All 80 cells must independently satisfy all four conditions: the 95% Wilson upper confidence bound of Win Rate is below the family's Maintain-Live threshold, cumulative return is negative, Profit Factor is below 1.0, and capital-weighted return is negative.
No aggregate shortcut
Low strategy-wide Win Rate, poor Historical Backtest, replacement by another branch, or lack of recent signals is never sufficient for ARCHIVED. Actual or Forward evidence is mandatory.
ARCHIVED therefore means 80 / 80 sufficiently observed XYZ cells and 80 / 80 statistically severe failures. Anything less remains POTENTIAL somewhere in the market map and continues validation.
5. What happens when the market changes?
The Market Regime can move every minute because the Sliding 15-vote window is updated every minute.
New L1 cell
All strategies are immediately evaluated against their evidence for the new XYZ cell.
Qualified strategy
It can move to LIVE for that cell and open new actual trades when capacity and funding rules allow.
Unqualified strategy
It stays POTENTIAL and continues Forward SIM. No actual entry is opened.
Open actual position
If a strategy loses qualification while holding a position, it becomes EXIT ONLY until that position is closed.
6. How to read the dashboard
The dashboard surfaces the same logic used by execution.
Strategy Coverage
Shows all 80 XYZ cells, how many strategies have evidence there, and which strategies are Live-ready or near the threshold.
Live / Near Live / All
Live shows actual entry-eligible strategies for the current XYZ. Near Live shows current-cell proximity. All shows every strategy and its current-cell evidence.
Primary / Shadow / Preview
Primary controls execution. Shadow is comparison research. Preview is the newest rolling observation before voting.
Forward / Actual / Combined
Forward and Actual remain visible separately, while Combined is the evidence basis used for live qualification.