CompoundQuant.

The method is the product.

Every claim on this site traces to a documented process. This page is the audit trail — including the parts that make us look worse.

01 · Data

  • Daily prices for ~2,700 NSE symbols from January 2020, pulled independently from two brokers and cross-validated: 94.4% of symbols match exactly.
  • All 151 mismatches (mostly corporate-action adjustments) resolved by documented policy; 20 symbols quarantined; every decision logged.
  • Symbols that stop trading are force-exited at a 15% haircut — never silently frozen.
  • Known limitation: companies fully delisted and removed from instrument masters are absent, leaving residual upward bias in absolute returns. Relative-to-benchmark gaps are the robust claims. Elimination requires point-in-time universe data; procurement is on our roadmap and this line will be updated when done.

02 · Validation

  • All strategy selection is walk-forward: configurations compete on trailing data only; reported curves are stitched out-of-sample segments.
  • Full cost model in every result: 0.40% per round trip covering STT, exchange charges, and slippage stressed to current liquidity.
  • A trial registry logs every configuration ever tested — 238 at last count — so our own results face Deflated-Sharpe scrutiny for multiple testing.
  • Failed engines (an event-drift attempt, a contrarian reversal attempt) are disclosed and their numbers retained in the registry. Survivor-only marketing is the industry's favorite lie; we do not use it.

03 · Risk engineering

  • A structural dual-momentum regime shield de-risks equity sleeves into liquid funds. It is policy — never selectable by trailing performance, never overridden manually. In July 2026 one overlay component was removed after a pre-registered evaluation showed it strictly hurt; the old and new designs now run side by side in live paper trading — the change is being verified in public, not asserted. (Exact parameters are proprietary; the process is not.)
  • Portfolio-level risk budgets are fixed; allocation review is quarterly and never happens inside a drawdown.
  • Live paper-trading of all engines runs daily with frozen specifications — building the verifiable track record that backtests cannot be.
  • Policy v4 (7 Aug 2026), disclosed: a churn governor was added to every stock book — at most roughly ten percent of the book may be replaced per rebalance (worst-ranked names first), and sub-half-percent weight tweaks on held names no longer trade at all. Audit of our own order log showed one book replacing a third of its names in a single event; the governor caps exactly that. This amends running paper specifications — a departure from our freeze rule, made deliberately and dated here rather than silently.
  • Blend concentration (8 Aug 2026), disclosed: the blended portfolios moved from ~80 positions to ~38 (each engine's book concentrated). Our own test showed the tradeoff plainly — higher return, deeper drawdown (-17% vs -12% historically) — and we chose concentration for practicality of execution and lower operational churn. Both curves remain in the registry.

04 · What we will not do

  • No cherry-picked inception dates. Windows of 1, 3 and 5 years are always shown together, including the ones that flatter the benchmark.
  • No mismatched benchmarks. Midcap-risk products are compared to midcap indices.
  • No smoothing, no "since inception" only, no hiding calendar years.
  • No promise of future returns, and no drawdown guarantee — long-only equity can and will fall further than history's worst in a severe enough crash. Our overlays reduce that exposure; nothing eliminates it.

05 · Adding a new strategy to this platform

  • Written economic rationale before any code — who pays us, and why.
  • Backtest on cleaned dual-source data with the full cost model; walk-forward only.
  • Deflated-Sharpe clearance against the trial registry; correlation audit against existing engines — a new strategy must diversify, not duplicate.
  • Four weeks minimum of live paper trading at frozen spec.
  • Quarter-size live deployment before full listing; factsheet auto-generated from the same pipeline as every other product.
  • Live example — Swing Rotator (Jul 2026): after the 2024–26 stretch exposed our stock-picking engines' regime dependence, a faster asset-rotation engine with a structural volatility governor cleared the trial-registry and correlation audits. Its paper verification began 29 Jul 2026; it is listed as Incubation until the four-week gate passes. Several candidate engines from the same research round failed the audits and were rejected — the trial registry records them all.