Signal → direction

The signal-to-direction lifecycle is how Kraite collapses 36 indicator readouts per symbol into a single tradeable direction. It runs hourly at :30 via kraite:cron-conclude-symbols-direction, computes LONG / SHORT / none per symbol, propagates the Binance-derived conclusion to the other exchanges, and exposes the result for the token-selection layer to rank against.

This is the lifecycle lens view of the indicator pipeline. For the indicator inputs themselves, see Indicators.

End-to-end flow

   :30 tick
      │
      ▼
   kraite:cron-conclude-symbols-direction
      │
      ▼
   for each tradeable Binance symbol:
      │
      ▼ (rate-limited via TAAPI throttler)
   fetch 12 indicators × 3 timeframes (1h / 4h / 12h) = 36 readouts
      │
      ▼
   per-timeframe collapse → tf_direction[1h], tf_direction[4h], tf_direction[12h]
      │
      ▼
   cross-timeframe agreement → symbol_direction = LONG | SHORT | none
      │
      ▼
   persist to exchange_symbols.direction + indicators_synced_at
      │
      ▼
   CopyDirectionToOtherExchangesJob
      │
      ▼
   Bybit / KuCoin / Bitget exchange_symbols rows updated to match Binance

Per-timeframe collapse

Twelve indicators per timeframe, each contributing a vote (LONG / SHORT / neutral) based on its threshold rules. The timeframe's direction is the majority vote, with neutrals discarded. A timeframe with no clear majority resolves to none.

The 12 indicators are not all equal weight — directional indicators (RSI, MACD signal cross, EMA stack) carry more weight than amplitude indicators (ATR, Bollinger band width). The exact weighting lives in kraitebot/core and is treated as tunable, not API-stable.

Cross-timeframe agreement

   1h:  LONG    │
   4h:  LONG    │ ─►  symbol_direction = LONG   (full agreement)
   12h: LONG    │

   1h:  LONG    │
   4h:  none    │ ─►  symbol_direction = LONG   (4h neutral, the others agree)
   12h: LONG    │

   1h:  LONG    │
   4h:  SHORT   │ ─►  symbol_direction = none   (conflict on a primary anchor)
   12h: LONG    │

   1h:  SHORT   │
   4h:  none    │ ─►  symbol_direction = none   (no anchor majority)
   12h: none    │

The 4h and 12h timeframes are treated as primary anchors; conflict between them is fatal to the direction. The 1h timeframe is allowed to disagree with the anchors only if both anchors agree with each other.

Cross-exchange propagation

Architectural decision

Direction is concluded once on Binance and copied to every other exchange's matching exchange_symbols row. The reasoning: directionality is BTC-anchored, BTC moves the same way across exchanges to <0.1 % drift on liquid tokens, and TAAPI's rate-limit budget is finite. Duplicating the indicator query per exchange would burn the budget without changing the answer. CopyDirectionToOtherExchangesJob does the propagation as the last step of the conclude pipeline. Symbol-name divergences (BTC→XBT on KuCoin, 1000SATS→10000SATS on KuCoin / Bybit) are handled by the token_mappers table.

Freshness

indicators_synced_at stamps every successful conclude. The token-selection layer treats a stale stamp (older than the configured freshness budget) as "skip this symbol" — better to leave a slot empty than to open against rotten signal. A symbol that fails to conclude (TAAPI failure, all neutrals, genuine none) does NOT get its indicators_synced_at advanced — that would mask the failure on the next tick.

What this lifecycle does NOT decide

  • Which symbol to trade — that's token selection's job, downstream.
  • Position size, leverage, entry price — the position lifecycle's open phase decides these.
  • Whether to open at all — gated by the market regime's shouldBlockOpens().

This pipeline answers "if this symbol were to be opened right now, which way?". Nothing more.