Interpret replay coverage and corrected history

Read the timeline policy before judging outcomes

Retained replay reconstructs final-corrected canonical event time. It does not reproduce every provisional value originally seen by every exchange participant. A correction can change an event's final data or timestamp, and canonical resolution occurs before source-time filtering. This distinction is important when comparing a replay with an older screenshot that captured a provisional live publication before later corrections.

Audit one replay's evidence

Coverage and quality describe the available source, while selected start/end describe the request.

1. Open the replay and record its mode and quality label.
2. Read requested and observed bounds, gaps, completeness and limits.
3. Inspect event/seed counts and available resolution where shown.
4. Pause at a known position and inspect the analytics frame.
5. Seek earlier and verify later model inputs remain unavailable.
6. Compare only identical exact instruments and corrected selection ranges.

No lookahead means inputs can be unavailable

A chain or reference seed can initialize a replay only when eligible at or before its start. Later inputs become usable according to their source/model availability time. Gamma frames end at current position rather than drawing later history as known. An unavailable model frame can therefore be correct even while executions are already present. A later complete bucket cannot be used to manufacture an earlier valuation that lacked its required inputs.

Worked example: a corrected event moves outside the range

Suppose an original event appeared inside a selected interval, then its accepted final correction moved its source timestamp outside that interval. Canonical replay should use the final placement, not count both versions or keep the obsolete in-range copy. Your older live screenshot can legitimately include the provisional observation. Record the timeline policy and correction context before treating the differing total as missing replay data. The tape is a corrected reconstruction, not an original-arrival forensic archive.

Troubleshoot partial profiles and book expectations

A frozen execution profile describes eligible frozen executions, not current full-session totals. Replay cannot infer resting depth from trades or a present DOM snapshot. Unknown direction and missing inputs must remain explicit. When asking support about an outcome, include replay ID, exact position, requested range, coverage and expected field. Explain whether the comparison source was live provisional data, final retained data or a synthetic fixture, because those are different evidence contracts.

What the isolated timeline can establish

Retained replay uses a final-corrected canonical event timeline. Each retained identity's highest accepted revision is placed at its final event timestamp. This reconstructs corrected event time, not the exact information an exchange participant saw at every original arrival. Dated chain/reference seeds at or before the start can initialize the isolated engine. Later model inputs remain unavailable until their own source/model availability timestamp, and historical gamma frames stop at the current replay position.

Replay does not open a live provider feed, write live analytics or fire live alert callbacks. It cannot reconstruct a synchronized DOM or resting-liquidity book from an execution tape. A partial frozen tape is not expanded with today's full-session profile. Explicit synthetic fixture replay has a different mode and quality from retained replay. Read those labels and coverage together before interpreting a result. A saved replay selection demonstrates the requested analysis, while its actual eligible inputs determine which frames can be produced at each clock position.

Compare replay evidence with an old live screenshot

An old live image records one displayed publication. It may predate a late correction, reflect a partial session prefix or use a different model selection. Replay uses eligible retained canonical final revisions under its stated timeline policy. To reconcile them, match exact instrument and range, then compare capture scope, source/model times and correction status. A different total is a question to investigate, not proof that replay should reproduce every obsolete intermediate display. Preserve the older screenshot as historical evidence without forcing the final tape to count both versions.

No-lookahead is a separate issue from final correction. A corrected event can have its final event time while a later model input still remains unavailable until its own eligibility time. Check both constraints when reviewing an earlier frame. A later complete gamma bucket must not leak into that frame merely because it exists in the frozen range. Similarly, current synchronized depth and today's session profile cannot repair a partial frozen execution tape. State exactly which evidence contract you expect: final-corrected chronology, original-arrival behavior or current live state. The existing replay provides the first with explicit limits, not all three simultaneously.