Spot a signal that knows the future

Export or retain the source availability time, decision time, order time and eligible fill time for an actual signal. Use the simple check below on those timestamps. The sample is deliberately invalid: the final close arrives after the proposed decision. It contains no price or performance observation.

Run the check once with the sample, then replace the timestamps with one actual order trace. A failed assertion identifies a violated information boundary. A passed row only checks that row; review operands such as indicators and filings as well as the final signal.

For filing-based research, list the actual filing metadata with sec_edgar using the JSON below. Keep the returned filing/acceptance time separate from the fiscal period. Do not call ingest_filings merely to demonstrate a read.

from datetime import datetime

def utc(value):
    return datetime.fromisoformat(value.replace("Z", "+00:00"))

available = utc("2026-01-05T21:00:00Z")
decision = utc("2026-01-05T20:59:00Z")
order = utc("2026-01-05T20:59:30Z")
fill = utc("2026-01-06T14:30:00Z")
assert available <= decision <= order <= fill, "Information was unavailable at decision time"

Submit this read request through your connected NexusTrade client. Check the returned filing dates before joining fundamentals into historical screening. It does not run a backtest or establish a point-in-time fundamentals series by itself.

{
  "ticker": "GOOG",
  "action": "list_filings",
  "forms": [
    "10-K",
    "10-Q"
  ],
  "limit": 5
}

Lookahead bias occurs when a historical decision uses information that was not available when the decision could have been made. The date attached to a daily close, financial quarter or revised series may differ from the time that value became usable.

For a moving-average strategy, decide when the close and indicator become known and which subsequent fill is eligible. An engine can preload future rows for efficient evaluation; correctness still requires every condition and fill to respect its evaluation timestamp.

This is a timing example, not a measured fill-price series. Buying at the same final close used to decide the order needs an explicit executable order protocol; knowing the closing value after the fact does not establish one.

Scroll sideways to compare all columns. Select a heading to sort.

No matching rows. Clear the filter to see all records.

Write the timestamps into your audit output. Shifting a column by one row is sufficient only if its actual availability and your fill protocol match that shift.

BAD: decision at t uses close(t) finalized after the order cutoff
CHECK: signal_available_at <= decision_at <= order_at <= eligible_fill_at
CHECK: every operand.available_at <= decision_at
CHECK: forward return belongs to the label, never the signal

A company quarter ending June 30 does not make the filed revenue available on June 30. A historical stock screen must use the filing release or acceptance time and the version available at that time. Joining all June-quarter values onto June 30 decisions leaks reports that arrived later.

Store period end, filed/accepted timestamp and any later amendment separately. At decision time, filter available_at <= decision_at before selecting the latest eligible record. Today’s revised fundamentals are not automatically the values a trader could have observed then.

Scroll sideways to compare all columns. Select a heading to sort.

No matching rows. Clear the filter to see all records.

Dates come from the actual plan. UTC boundaries reflect the generated market calendar; preserve them instead of hand-rounding an adjacent day into both windows.

The actual first GOOG fold has a 14-calendar-day gap between training and validation. A gap separates adjacent samples; it cannot make a future filing available earlier or turn a post-close signal into an executable pre-close order. Indicator warmup, label overlap, universe membership and revisions need their own checks.

Audit a few actual signals near date boundaries, then trace all operands and the selected fill. Compare the recorded rules with the engine’s evaluation interval. Daily indicators and daily backtest evaluation do not establish a broker’s deployment cadence.

Retain signal timestamps, source availability timestamps, eligibility rules and fill assumptions with the result. If the source lacks historical availability or revisions, state that limitation. A chronological train/test split can still contain lookahead leakage inside either window.