TSI-100R

paper only 6-trial frontier — selection is dev-in-sample generated —
prospective window from 2026-11-01 — model replay, no real orders panel through — family: TSI-90 · TSI-TILT-90 (live) · both shadows

The return-maximizing point under a stated 30% drawdown limit: the same family with a lighter trend ballast and the cap + funding brake, but without the throttle — more growth per year, less tail protection.

Paper account value (prospective, from €1,000)

Model replay of the frozen spec — no fills, no broker. Bars before the prospective window would be backtest, so this chart starts only where observation begins.

Prospective gate scoreboard (fixed in advance)

Current model weights (target row, net of guardrails)

Historical development (backtest, log scale) — this book vs the live family book

Selected/characterized on 2019 → 2026-09-30 — part of this strength is selection luck by construction. The prospective chart above is the only out-of-sample evidence this page is building.

Backtest results on the audited frozen panel

Dev CAGR (2022-04 → 2026-09)179.2%
Dev Sharpe2.18
Dev max drawdown-30.0%
Dev CAGR @ 2x costs157.5%
Holdout ≤2022-0395.0%
Full window (2019-09 → panel end)141.8%
P(≥30% DD in random 1-yr)17.3%
CAGR if best day of every month missed25.4%
CAGR under +25bp/day adverse funding-4.0%
Correlation with live book1.000
Neighbourhood (ballast weight 0.15→0.35 dev CAGR)0.15:184.0% · 0.20:179.2% · 0.25:171.1% · 0.30:162.5% · 0.35:153.8%
Selection: best of a 6-trial pre-registered grid (ballast weights 0.15–0.35 + one re-lever trial) chosen on the dev window; w=0.15 scored 184% but breached the 30% drawdown limit at −31.5%. The +8 pp/yr edge over the live book is inside selection noise until 90 prospective days judge it. Cohort stress not re-run for this variant (cap+brake hardening expected to help; unverified).

Retire / review lines at day 90 (fixed before the window opened)

RETIRE if: 90-day cumulative ≤ −25.8% · prospective max drawdown deeper than −40% · ≥10 funding-brake trigger days · any reconciliation break. Bands (from the frozen flagship dev stream): p5 −17.6% · p50 +25.4% · p95 +209.1%. ACCEPT requires clearing these lines; nothing else re-opens.

What is believed to break this book

· A funding-cost regime flip: the book pays rather than collects >15%/yr for a sustained month (the brake halves it; +25bp/day stress shown above).
· Momentum continuation dying: the returns live in a handful of surge days — the drop-best-day row above is the honest sensitivity.
· Delisting-generation markets (2025+ listings): pump-into-short squeezes; stress case above shows the hardening cuts, not removes, that risk.
· Execution decay: one-day-lag fills cost ~45% of CAGR (family-wide weak point).
1. Rank
Identical ranking engine as the flagship family: TSI(25,13) momentum over the eligible, liquid perp universe.
2. Blend
80% main momentum book at 1.5x gross, dollar-neutral + 20% BTC-gated trend sleeve — one notch more attack than the live book's 75/25.
3. Harden
Same 25% per-name cap and same funding brake (halve while 30-day funding cost > 15%/yr). No Bitcoin throttle: that protection was traded for return, per the stated drawdown budget.

Exact frozen specification

Glossary

Standing limitations

· Every curve except the prospective chart is backtest on data used for design/selection; guardrails in these specs were chosen on the same window.
· No real orders exist: fills modeled at 00:00 open, 7 bps/side costs, no slippage; funding reconciled settlement-level against Binance archives (0 mismatch over 107k cells).
· The 2019–2022 holdout is the only pre-design era for the core; regime concentration in 2026 applies to the whole family (~45% of dev growth).
· Money is simulated. No real capital is authorized for these books under any circumstances without a separate explicit decision.
· Lineage: specs frozen in DECLARATION_TSI100.md / DECLARATION_tsi100R.md (2026-11-04/05), trial-registered (N=1,759 / 1,765), audited panel + backups per house protocol.