# Backtest and track record

> There is no historical backtest, on purpose. The validation is a live paper-trading decision log — every input, every rejection, every fill, timestamped and append-only.

![image](https://zmscvxdouuytwoutqtfa.supabase.co/storage/v1/object/public/docs-media/b70167eb-4863-459d-bd8b-2c86daecdec0.jpg)
**There is no historical backtest of this strategy, and that is a deliberate
choice, not an omission.** The parameters in `config/strategy.yaml` are
industry-conventional starting values for a short-vertical-spread strategy, not
numbers fitted to past data. The code, the README and the config all say so in
those words.

## Why not

- **The data to do it credibly does not exist here.** A trustworthy options
  backtest needs historical option chains with quotes and Greeks at each entry
  time. The Basic Alpaca plan does not provide that history. A backtest built on
  synthetic IV, or on the VIX as a stand-in for contract IV, would be *worse*
  than none — it would carry the authority of a backtest with none of the
  validity, and the playbook explicitly forbids using the VIX that way.
- **The edge is documented, not discovered by us.** The volatility risk premium
  is well established in the academic literature. `docs/playbook.md` organises
  every claim behind the strategy by reliability tier — academic research,
  industry convention, or our own choice — with a source on each. The honest
  position is "a known effect, conventional parameters, to be tuned on real
  data", and pretending otherwise with a fitted curve would misrepresent it.
- **The live log is a stronger artifact.** Every cycle writes its inputs, the
  model's reasoning, each risk-check verdict and any fill — append-only,
  timestamped, in Supabase. That is reproducible and auditable in a way a
  backtest we ran ourselves is not.

## What does exist

- **Historical calibration of one gate.** The VIX size taper's shape is set from
  VIX close base rates over 1990–2026: the 1-year percentile sits below 25 on
  about a third of days, below 3 on about 8%, and those deep-tail spells are
  usually short. That informs where the taper starts and where the hard block
  sits — it is calibration of a single rule, not a P&L backtest.
- **A one-day parameter replay.** The entry-slippage settings were tuned by
  replaying one day of real captured candidates to see how many would have been
  tradable versus no-trade. One session, one purpose — not a multi-period
  performance test.

## Reading the track record

The live record is the paper-account equity curve (on the homepage and the
dashboard, pulled straight from Alpaca) and the decision history behind it. Read
it knowing the evaluation window is measured in market days: over a span that
short, P&L is dominated by luck, not skill. What the log demonstrates is
process — that the rules ran, that the rejections happened, and that the losses
were the size they were declared to be in advance.