Historical API: Point-in-Time Replay

What's different from the live API

Every endpoint you already know on the live API is mirrored here. Point your existing SDK at a different host, add one query parameter, and you have a backtest-ready data source. No new response shapes to learn - GEX, gamma flip, regime labels, narrative text, VRP percentiles, SVI surfaces and everything else come back in the exact same JSON you already parse.

The analytics run through the same code path as live - so any fix or refinement that lands on api.flashalpha.com shows up here on the next deploy. The only difference is the data store underneath: minute-level quotes and greeks going back to April 2018 instead of the live feed.

The at parameter

Every analytics endpoint takes a required at query parameter - the as-of timestamp. Treated as ET wall-clock (same clock as the stored data). Do not shift by UTC offset.

Intraday-accurate vs EOD-stamped

Not every underlying series is minute-resolution. Here's what moves at the minute you ask for, and what's stamped at the close of that trade day:

Why this is OK: intraday GEX / DEX / VEX / CHEX shifts are driven by greek + spot changes; OI is dealer-positioning context and barely moves intraday anyway.

Coverage: GET /v1/tickers

The only historical-specific endpoint. Lists every symbol with historical coverage, the date range, and any gaps. Regenerated after every backfill run, so whatever comes back is the current ground truth for what you can query.

curl -H "X-Api-Key: YOUR_API_KEY" \
  "https://historical.flashalpha.com/v1/tickers"

Response 200:

{
  "count": 1,
  "tickers": [
    {
      "symbol": "SPY",
      "coverage": {
        "first": "2018-04-16",
        "last":  "2026-04-02",
        "total_days": 2909,
        "healthy_days": 1972
      },
      "tables": {
        "stocks_days":    2009,
        "options_days":   1972,
        "option_eod_days": 2006,
        "svi_days":        1972
      },
      "gaps": {
        "missing_eod": 3,
        "missing_svi": 36,
        "uncovered_calendar": 3
      },
      "updated_at": "2026-04-14T07:07:36.55353"
    }
  ]
}

Available symbols. SPY is fully backfilled from 2018-04-16 through 2026-04-02 and extended on every refresh. Additional symbols are backfilled on demand - ask us to prioritize what you need. Live coverage is always queryable at GET /v1/tickers.

Analytics endpoints

Every live analytics endpoint is mirrored with the addition of a required at. Paths, query params, and response shapes are identical to the live docs - don't re-document them here. Click through for the per-endpoint reference:

VRP date-bounding matters. VRP percentile and z-score only consider history prior to at, so at any historical point the percentile reflects only what was knowable at that moment. No future leakage in backtests.

Example: replay March 16, 2020 (COVID crash, -12% close)

curl -H "X-Api-Key: YOUR_API_KEY" \
  "https://historical.flashalpha.com/v1/exposure/summary/SPY?at=2020-03-16T15:30:00"

{
  "symbol": "SPY",
  "underlying_price": 246.01,
  "as_of": "2020-03-16T15:30:00",
  "gamma_flip": 270.92,
  "regime": "negative_gamma",
  "exposures": {
    "net_gex":  -2633970601,
    "net_dex": -169419489077,
    "net_vex":  152461756844,
    "net_chex":   -13122349
  },
  "interpretation": {
    "gamma":  "Dealers short gamma - moves amplified, trend following likely",
    "vanna":  "Vol up = dealers buy delta - downside dampened if vol spikes",
    "charm":  "Time decay pushing dealers to sell - pressure into close"
  }
}

Real dealer positioning, at 15:30 ET on the day SPY closed -12%. Not a re-derived reconstruction - these are the minute-level greeks as they were at that moment, run through today's analytics.

Field coverage vs live

Most fields are identical to the live API. These are the exceptions:

Everything else - greeks, IV, exposures, flip, walls, regime, narrative, VRP percentiles, SVI fits, variance surface, arbitrage flags - is real, computed from the data as it stood at at.

How coverage grows

New trading days and new symbols are added by a gap-aware backfill that runs on a schedule. The coverage window on /v1/tickers is what you can query right now; it extends forward on each refresh. Transient upstream misses self-heal on subsequent runs, so a gap you see today may fill in later - check /v1/tickers before concluding that a specific date is unavailable.

Latency & caching

Point-in-time reads are fast (~50–300 ms typical). Composite endpoints that rebuild surfaces on demand - /v1/adv_volatility and /v1/stock/{symbol}/summary - can take ~500–1500 ms on a cold request for a new (symbol, at). Cache at your proxy or client for repeat queries on the same tuple - responses are deterministic for a given at, so they're safe to cache indefinitely.