For the first present tag, return {period_end: (filed, fy, fp, val)} keeping only ~quarterly (80–100 day) periods, earliest-filed per end (point-i

(ticker N-day return) − (benchmark N-day return), in percentage points.

specs = list of (open, close); high/low bracket them. Prev close → this open is the gap.

RS ratio aligned to the ticker's bars. None until a benchmark close exists.

Return (mid, upper, lower, pctb, bandwidth_pct), aligned to `closes`. mid = SMA(period); band = k·population-σ over the trailing window. %b =

Fuse the two daily in-market series into a daily target exposure in [0,1]. All branches are pure functions of the two leg signals (+ a static

Per-bar (z_i, beta_i), each using the window of bars strictly before i.

Most recent days-to-cover STRICTLY BEFORE d (publish-lagged short interest).

Merge revenue / gross profit / net income quarterly series into per-quarter records keyed by period end, sorted ascending.

Discretionary insider flow over (as_of - lookback_days, as_of], using only filings filed on/before as_of (lookahead-free).

四餘 mean longitudes (deg). Node/apogee are standard mean elements; 紫炁 is a defined fictitious slow point (period ≈ 28 yr from a fixed epoch).

Reconstruct a missing/truncated core item from its trustworthy canonical neighbours. Returns number of items recovered. Mutates `items`. Only

陽遁: 冬至→夏至 (順布); 陰遁: 夏至→冬至 (逆布). Approx by day-of-year.

Tiny mustache-style replacement: {{key}} → value. No logic, no escaping.

Fractional, daily-rebalanced backtest of a target-exposure series. `exposure_by_date` is a *holding* series: its value for day i is the fraction

Return (system_prompt, user_template) parsed out of the prompt file.

Disclosure dates of buys and sells (the only thing the signal needs).