Account value over time
The two ranges have different evidence quality and are never merged ▾
“Paper account” is the simulated account from its modeled open on 2026-06-01 (June–September overlap the period used to choose the strategy; October was a frozen-holdout month — see Evidence & risks). “Model history” is a pure backtest of what the model would have done since 2019 — not an account that existed.
TSI-90 compares the recent price strength of eligible crypto markets and each day holds positions that benefit from continued rises in stronger markets and falls in weaker ones. The style is momentum continuation — deliberately not an oversold-reversal trade. All money here is simulated. Fills are modeled at the next day's opening price with a 0.07% fee per side, with funding and EUR/USD effects included.
Where the change came from
Model components from the start of the paper account to the latest recorded value. Published components are rounded separately; a labeled rounding bridge connects their sum to the authoritative current value. This is a calculation check, not proof the strategy works.
How the accounting checks work — and what they don't prove
Current paper positions
Long rises · Short falls · perpetual-futures exposure — both sides can lose money together in a sharp move.
What "long" and "short" mean here ▾
Long: benefits if this market rises. Short: benefits if this market falls. These are perpetual-futures exposures — the account does not own the underlying coins, and long and short positions can lose money together in a sharp market move.
Recorded positions
Exposure is valued at the recorded market close (basis below). “Intended” is the target allocation from the latest signal — an intention, not an executed trade. Open a market for its chart and the model's signal-boundary estimate.
Planned changes — not executed
Not executed. These are what the strategy intends at the next modeled daily open. “Recorded → intended” compares the current position with the intended allocation; the difference is the direction of change, not a promised trade size (exact traded quantity/value is not modeled). Raw BUY/SELL actions and weights stay available below and in exports.
details column keeps the raw action (BUY/SELL) and weight change from the engine.
Signal watchlist — markets near a boundary
Potential entry: outside the selection zone and close to entering it. Possible removal from main signal: currently held and close to leaving the zone. “Gap” is a distance in signal units — it is not a probability, and not a price distance. Nothing happens until a completed daily close crosses the boundary; fills then occur at the NEXT open, at a price nobody knows yet.
Activity & account history
Daily account history
Return-interval-ending NAV in EUR: return label t carries the outcome from open(t) to open(t+1), translated at FX(t). It is not the opening balance on t. Interval boundaries and FX dates are shown separately; the latest close-marked estimate uses its recorded close and that day's FX. The return column is the USD-book net return, not the EUR NAV percentage change.
Trade journal — newest first
Displayed newest settlement first; exports keep the full filtered set in source order.
Every row is price P&L only — trading costs and funding are accounted separately, so a winning row is not necessarily a profitable episode, and closing rows exclude the open part of a position. REDUCED = a partial close (some P&L realized; position stays open). Missing-price settlements appear tagged with their settlement penalty (journal basis) — none has occurred yet.
Signal weight changes — most recent 200
Not the complete fill history — this is a signal-weight-change stream, not a complete quantity-fill ledger. It records changes in signal target weights with a reference daily-open price; it omits quantity rebalances at unchanged signal weights. The latest-200 cap is a separate limitation. A weight delta is not an executed quantity or traded value; accounting costs and journal quantities include rebalances absent from this list. The fee assumption is 7 bps per side, not an actual fee charged per displayed row.
Evidence & risks
What is proven, what was chosen with the same data it is measured on, and what remains untested — in the order the evidence gets stronger.
The headline historical numbers are large and should be treated with suspicion. This section exists so you can see exactly what is known, what was chosen in-sample, and what remains unproven.
What kind of evidence is each number?
Prospective observations are the only evidence the strategy cannot have influenced: daily results after the research freeze, computed as each day completes. The count shown on the Overview (recorded execution days strictly after the documented freeze cut) is the authority — it updates every nightly run. Nothing here is real-money trading.
Historical development results
Exact development window: . This window was used to choose TSI-90 among 674 registry books — the reported strength and the selection process share the same data, which is precisely why prospective tracking exists.
By year — full recorded daily history (display only; the exact development window is stated in the table)
Stress tests — what breaks the result
Reliability of the evidence
Drawdown simulations (conditional, not forecasts)
Standing limitations — read before trusting anything above
Strategy guide
In three steps
The full seven-step method, the blend, costs and glossary follow below.
The full seven-step method
The blend and the multiplier
TSI-90's daily target = 90% main TSI momentum sleeve + 5% volume-confirmed momentum sleeve (only markets whose recent 5-day traded value exceeds the prior 30-day average qualify) + 5% a small signed-momentum tilt, then the whole book is scaled by a 1.5× gross-exposure multiplier. The book targets dollar-neutral (equal intended long and short exposure), so its 1.5× multiplier means about 75% long and 75% short of account value at target. Eligibility and allocations change daily; a target is never a guarantee of diversification or of loss protection. Direction is momentum continuation (buy recently strong, sell recently weak) — verified against the implementation, not an oversold-reversal strategy.
Glossary
Advanced — data provenance, field dictionary, build & review chronology
Technical process detail, kept out of the everyday reading surfaces. This page is generated nightly from a gated pipeline (publication gate → deploy content gate) and published at best-tsi.pages.dev/tsi90-dashboard. The full evidence trail, every critic-review round and the handoff live in the repository (D:\\Best, backed up per D:\\TSI90Backup). Data sources: Binance daily perpetual bars via the frozen research panel (binance vision monthly zips); FX via ECB (Frankfurter); all daily boundaries UTC.
Where every number comes from
UI release — — an identifier for this page's presentation build only; it is separate from (and never changes) the financial snapshot stamps above.
The page never computes account figures itself — it reads them from the gated daily data file produced by the strategy's accounting system. Presentation-only arithmetic (deltas, percentages, drawdown dates over published daily values) is labeled below.