Skip to main content

Replay and discuss a recorded session

Session replay presents the messages, tool steps and file versions retained in a session. Use it to follow an analysis, locate its evidence and ask a focused question. Playback does not rerun code or establish that an analysis is reproducible; use reproducibility verification for that separate task.

Open a replay​

  1. Open the session's action menu in the sidebar and choose View replay. For an imported .science session, the Imported research history panel also provides View replay.
  2. Check the session title and branch. When Replay branch is offered, select the branch you intend to inspect.
  3. Use Enter full screen for more space. Exit full screen returns to the workspace; closing the preview does not delete the session.

Imported research remains read-only. You can inspect and discuss it; choose Fork to continue if you need a writable continuation. Follow research-package import before opening its replay.

Follow the recorded steps​

ControlAction
Play replay / Pause replayStart or pause presentation of the saved sequence
Previous step / Next stepMove to an adjacent recorded step
Replay progressSeek to another point in the recording
Playback speedChange presentation speed; this does not accelerate a computation
Browse stepsSelect a message, tool step or file-version event by its label
Notebook / View filesInspect the recorded Notebook material or file list available at that point
Watch againRestart after playback reaches Completed

Expand a tool card to read its retained inputs and outputs. Inspect filenames and version numbers before using a result. The replay position is retained for returning to the same recording. Older sessions may show a timeline reconstructed from archived records; its presentation duration is not a benchmark of the original computation.

Worked example Discuss the evidence in a TP53 pathway analysis

This example opens the recorded Pathway Commons analysis, which selected Reactome Transcriptional Regulation by TP53. Its export contains 3,318 interaction records and 387 nodes. Those are results of this saved analysis, not counts to expect from every future query.

In Browse steps, locate the exported network, original responses, research note and small TP53–MDM2–CDKN1A interaction table. The example has 12 recorded steps. Select the table's file-version step, then read the preceding explanation of the network's scope.

The real TP53 replay with its recorded steps, file versions and playback controls

Ask about a step​

  1. Pause at the relevant step and select Ask about this step. For a broader question, choose Ask about this research.
  2. Check the discussion draft and its Discuss attachment. The selected session, branch and step identify the evidence being discussed. A selected Notebook run includes its recorded context even when some cells are outside the current viewport.
  3. Enter a focused question, choose a connected model and select Send. Opening a discussion or adding context does not send the question automatically.
  4. For imported research, the discussion is a linked child conversation. Return to the parent record to check the original step; the discussion does not make that read-only history writable. Use Fork to continue for a continuation instead.
  5. Inspect any tool approval and compare the response with the recorded evidence. A model explanation can contain mistakes even when the source link is correct.

For the TP53 example above, ask:

From the recorded TP53 pathway analysis, explain why the missing direct
TP53–CDKN1A edge does not show that regulation is absent. Identify the
saved evidence and separate the recorded result from a new biological
claim. Keep the response in English.

The example answer identifies tp53_mdm2_cdkn1a_readable_interactions.tsv and separates the flattened export from a biological conclusion. The screenshot shows that recorded discussion; current linked-discussion controls may differ.

The completed Codex discussion beside the recorded TP53 analysis

To make a narrated presentation of a completed session, follow Research films. Replay, a discussion and a film each reuse recorded evidence; none of them reruns the analysis.

When evidence is unavailable​

A recorded tool call can remain visible even when its original execution environment is unavailable. Read the retained code and output; do not treat the environment warning as a new execution result.

If a file reports Preview unavailable or The recorded evidence is unavailable, check the original session's file card and selected version. If that view also cannot open it, use a separately retained original file or obtain a complete research package from its author. A filename and a completed playback timeline do not prove that the file's contents are available. Discuss only the evidence you can actually inspect.

For a fresh computation, fork the session, supply the required files/environment and run the analysis. To compare a new run with a saved artifact, use reproducibility verification, not the replay's Completed indicator.