Why Konseki works
the way it does.

Running a full historical context analysis through a model for hundreds of symbols every morning burns tokens on signals nobody needs to see. How the categorical tag fields let a screener triage at scale before a model ever gets involved.

A backtest that returns a table of trades and a Sharpe ratio tells you whether a strategy worked. It doesn't tell you why, or what conditions it depended on. How immutable daily snapshots let a model generate a readable narrative around backtest results, without the lookahead bias that makes most narrative-generation approaches unreliable.

A portfolio risk narrative is not the same problem as a single-symbol historical context summary, repeated several times. What changes when the unit of analysis moves from one symbol to a full set of holdings, and how to structure the aggregation so a model can reason across positions rather than just listing them.

A win rate tells you how often a setup worked. It tells you nothing about how much pain you'd have had to sit through to find out. What MAE and MFE actually measure, and how to use the distributions behind them to reason about stop placement and position size instead of guessing at both.

A trading assistant that answers from training data alone is guessing. Why models hallucinate market statistics, what grounding actually means, and how pre-computed historical evidence closes the gap.

A single similarity score can't tell you why two price setups are considered alike. What the seven distance components behind that score actually measure, and why all seven are needed rather than one.

A match can have a high similarity score and a low match quality rating at the same time. This is not a contradiction in the data — it reflects two genuinely different questions being measured.

Why cross-market pattern matching across hundreds of symbols and a decade of history can't run live — and why pre-computation is an architecture decision, not an optimisation.

The complete technical reference for how Konseki computes historical market context — match scoring, reliability tags, seasonality, and known limitations, explained against the real API schema.

A precise definition of historical market context, how it differs from backtesting and technical analysis, and how Konseki computes it against the real API schema.