test(coppock): lock in warmup_period for every parameter set (refutes R12)

Audit finding R12 claimed `Coppock::warmup_period()` was off by one
because it returns `max(roc_long, roc_short) + wma`, while
`Roc::warmup_period() = period + 1`. After tracing the actual emission
sequence the existing formula is correct: when both ROCs reach `Some`
at 0-based index L (the slower of `roc_long_period` and
`roc_short_period`), the WMA receives its first input there and emits
its `wma_period`-th value at 0-based index `L + wma_period − 1`. The
`warmup_period()` is the 1-based count of inputs needed before the
first `Some`, i.e. `L + wma_period`. R12 was a misread by both Sonnet
audit agents and the Opus verifier — none of them traced the actual
emission timeline.

This commit:

- Expands the doc comment on `warmup_period` with the precise emission
  argument and a worked example for `Coppock::new(6, 4, 3)` (the
  existing test) so a future reader cannot mis-derive the formula.
- Adds `warmup_period_matches_first_some_for_every_parameter_set`,
  which asserts `out[warmup - 1].is_some()` for five parameter
  combinations — including the audit's smoking gun `(4, 2, 3)`. The
  audit's proposed `max + 1 + wma` formula would have predicted index
  7 (the 8th input) for that combination; the real first `Some` lands
  at index 6 (the 7th input), exactly what the current formula
  reports.

No behaviour change — the audit was wrong and the test makes the
contract regression-proof.
This commit is contained in:
kingchenc
2026-05-23 10:38:24 +02:00
parent 2aef8c8db5
commit b340ecd3d6
+42 -2
View File
@@ -104,8 +104,20 @@ impl Indicator for Coppock {
}
fn warmup_period(&self) -> usize {
// Both ROCs must be ready (the longer one is `period + 1`), then the
// WMA needs `wma_period` of their summed values.
// Let `L = max(roc_long_period, roc_short_period)` and `W = wma_period`.
// Both ROCs need `period + 1` inputs to emit; the slower one therefore
// first emits at **0-based index L** (= the `(L + 1)`-th input). From
// that bar onward both ROCs feed the WMA in lock-step, so the WMA
// sees its `W`-th input at 0-based index `L + W 1` — the first bar
// it emits. `warmup_period` is the 1-based count of inputs needed for
// the first `Some` value, which is `(L + W 1) + 1 = L + W`.
//
// Worked example for `Coppock::new(6, 4, 3)`:
// - ROC(6).first_some at index 6 (the 7th input).
// - ROC(4).first_some at index 4 (the 5th input). Both available
// from index 6 onward.
// - WMA(3) consumes 3 inputs at indices 6, 7, 8 → first WMA `Some`
// at index 8 (the 9th input). `warmup_period() == 9`.
self.roc_long_period.max(self.roc_short_period) + self.wma_period
}
@@ -142,6 +154,34 @@ mod tests {
assert!(out[8].is_some());
}
/// `warmup_period() == 1-based index of the first `Some`` for every legal
/// parameter combination — including the parameter set
/// `(roc_long=4, roc_short=2, wma=3)` that an external audit claimed
/// would prove the formula off by one. It does not: the slower ROC first
/// emits at 0-based index 4, the WMA needs 3 such inputs and emits at
/// 0-based index 6 (the 7th input), which is what
/// `roc_long.max(roc_short) + wma = max(4, 2) + 3 = 7` reports.
#[test]
fn warmup_period_matches_first_some_for_every_parameter_set() {
let prices: Vec<f64> = (1..=80).map(|i| 100.0 + f64::from(i)).collect();
for &(long, short, wma) in &[(6, 4, 3), (14, 11, 10), (4, 2, 3), (10, 3, 5), (3, 3, 3)] {
let mut c = Coppock::new(long, short, wma).unwrap();
let warmup = c.warmup_period();
let out = c.batch(&prices);
for (i, v) in out.iter().enumerate().take(warmup - 1) {
assert!(
v.is_none(),
"Coppock({long}, {short}, {wma}): index {i} expected None during warmup, got {v:?}"
);
}
assert!(
out[warmup - 1].is_some(),
"Coppock({long}, {short}, {wma}): warmup_period() = {warmup} but index {} is None",
warmup - 1,
);
}
}
#[test]
fn constant_series_yields_zero() {
// Both ROCs are 0 on a flat series, so the WMA of zeros is 0.