Adds `crates/wickra-core/tests/invariants.rs` — a property-based (`proptest`) harness asserting three invariants for **every** indicator and bar-builder in the catalogue (every module entry implementing `Indicator` or `BarBuilder`; the lone non-indicator helper `pattern_swing`/`SwingTracker` is excluded by design) — and fixes the non-finite bugs the harness surfaced.
## Invariants
1. **batch == streaming** — `batch()` must replay `update()` exactly.
2. **reset == fresh** — after `reset()`, re-feeding the same data matches a fresh instance.
3. **non-finite rejected without poisoning** (`f64` / `(f64, f64)` families) — a NaN/inf tick returns `None` and leaves state identical to never having seen it.
## How it works
- A single generic `check_seq<I: Indicator>` covers **all** input families — `f64`, `Candle`, `(f64, f64)`, and the exotic `CrossSection`, `Trade`, `DerivativesTick`, `OrderBook`, `TradeQuote` — via per-family `proptest` generators that produce *valid* inputs (e.g. order books are strictly monotonic and uncrossed).
- `check_bars<B: BarBuilder>` covers the bar-builder trait.
- Outputs are compared by `Debug` string so bit-identical `NaN` outputs count as equal — these properties test **determinism**, not NaN-freeness.
## The 38 non-finite fixes
The harness surfaced **38 more** scalar/pairwise indicators that let a NaN/inf tick poison their state — the same class as the 16 pairwise indicators fixed earlier (#251), but missed by the grep-based audit (`kalman_hedge_ratio` and `spread_bollinger_bands` carried `is_finite` on their **constructor** params, not the update input). All 38 now reject non-finite input via the established first-statement guard; signatures unchanged, so every binding inherits the fix. With these in, the non-finite invariant is enforced for every `f64`/`(f64,f64)` indicator going forward — the permanent regression net that would have caught #251.
Warmup-exactness was evaluated and **left out** (not universal — many multi-component/candlestick indicators emit before `warmup_period` by design).
## Verification
- `cargo test -p wickra-core`: 4225 unit + harness + 464 doc/integration, all green.
- `cargo clippy --workspace --all-targets --all-features -- -D warnings`: clean.
- Coverage runs integration tests, so the new guard branches are exercised by the harness's NaN feed.
Also documents the harness in README's Testing section and adds the pending 0.8.4 entries to CHANGELOG `[Unreleased]`.
* test(keltner): cover periods accessor + name metadata
Codecov flagged 6 lines (file at 95.23%): periods (68-70) + name (106-108).
* test(linreg): cover period accessor + name metadata
Codecov flagged 6 lines (file at 96.10%): period (92-94) + name (142-144).
* test(linreg_slope): cover period accessor + name metadata
Codecov flagged 6 lines (file at 95.91%): period (80-82) + name (125-127).
* test(macd): cover periods/value accessors + name metadata
Codecov flagged 6 lines (file at 95.45%): periods (81-83) + name (135-137).
* test(super_trend): cover params accessor + name metadata
Codecov flagged 6 lines (file at 96.36%): params (99-101) + name (176-178).
`LinearRegression::fit` and `LinRegSlope::update` previously iterated the
full `period`-window on every tick to recompute `Σy` and `Σxy` from
scratch — O(period) per update, in violation of the `Indicator` trait's
O(1) contract. `LinRegAngle` inherits the cost transitively because it
delegates to `LinRegSlope`.
This commit slides the OLS state in closed form. The constant terms
(`Σx`, `Σxx`, the denominator `n·Σxx − (Σx)²`) were already precomputed
in `new`. The new running state is:
- `sum_y: f64` — running sum of the values currently in the window.
- `sum_xy: f64` — running Σ(x · y) where `x` is the position of each
value inside the trailing window (`0` for the oldest, `n−1` for the
newest).
On every push, when the window is already full the front value `y₀` is
popped and the indices of every remaining value shift down by 1; the
identity
new_Σxy = old_Σxy − old_Σy + y₀
closes the slide in O(1). The new value is then pushed at position `k`
(the current length before the push), contributing `k · new_value` to
`sum_xy` and `new_value` to `sum_y`. The output is the same TA-Lib OLS
formula evaluated against the incremental accumulators.
Behaviour is unchanged: same per-tick values, same warmup, same NaN
semantics. Two new tests compare the O(1) result bar-by-bar against a
fresh O(n) refit on a noisy ramp (sliding-phase dominated), a step
function (large pop/push deltas), and constants (tests floating-point
drift) — agreement is within `1e-9`.
`LinRegAngle` benefits automatically through its `LinRegSlope` field.
