From a moment

Point at a market and a minute instead of describing it: the panel reads every versioned feature as-of that bucket and proposes editable chips. Compare up to eight moments for what they honestly share.

Start from a market and timestamp when you have an observation you want to understand. The timestamp anchors the question; it does not establish what caused the later move.

Inspect the recorded state

Open the Research Workbench from an event or replay handoff, or use Investigate. Read the market, UTC time and available lead-up observations. Missing rows or horizons are evidence limits, not quiet markets.

Use the Reading library to check any unfamiliar field, especially its observation window and the distinction between its raw value and a contextual rank.

Turn observations into conditions

Inspect the proposed fields and thresholds before running a query. Keep the conditions you can explain, edit the scope deliberately, and check the resulting exact document. An observed state is a candidate question, not a certified setup.

When comparing selected moments, agreement means they share an expressible condition under the comparison rules. It does not show that the condition is unusual elsewhere or that it explains their outcomes. The API reference owns the exact comparison contract.

Broaden and then evaluate

Review all eligible matches for the proposed condition, including ordinary outcomes. Read the denominator, exclusions and baseline. Event consolidation and eligibility can affect which anchors appear, so a broader scan is not a self-validating proof of your selected moments.

A question derived from moves you already saw is exploratory. Testing it again on the same record is not independent validation. Keep a research log and consult selection bias before interpreting the result.

Save the exact result and inspect contrasting replay examples, subject to their displayed availability and account access.