The Replay Is Not a Copy Anymore

When a reviewer watches a recorded session, the thing they are watching used to be a reconstruction: a separate interface built to look like the one the candidate used, maintained on its own, and free to quietly drift out of sync with it as the real interface kept changing underneath it.

What changed

Piece by piece, the replay was rebuilt to use the real, live interface components instead of a lookalike. The chat transcript is now the actual chat bubble the candidate saw, not a separately styled approximation of one. The way a multi-step agent's work folds and expands, its plan, its tool calls, its reasoning, now folds the same way in playback as it did live, instead of showing a flattened summary. The problem panel and its tabs are the real problem panel, including a "Rules" tab that used to sit there doing nothing in the replay and now behaves exactly as it does for a live candidate.

Each of these landed as its own small change rather than one large rewrite: the chat bubble first, then the problem panel, then the turn-folding behavior, so each piece could be verified working before the next one started. A reconstruction built in one large pass is harder to trust than several small swaps, each checked against the real thing before moving on.

Why it matters

A reconstruction is a second product that has to be kept honest against the first one, forever, by hand. Every time the live interface changes, somebody has to remember to update the replay's copy of it too, and that kind of manual synchronization is exactly the sort of thing that quietly falls out of date. It is also a second product a candidate never had a reason to trust, since a reviewer relying on it is one small drift away from seeing something the candidate did not actually see, or missing something they did.

Reusing the real interface removes the second product entirely. There is one implementation, used live and used in playback, and the replay is a recording of it rather than a description of it. When the live interface improves, the replay improves for free, on the same day, with no separate effort required to keep it current.

The multi-step agent view is a good example of why this is worth the effort rather than a nice-to-have. Watching an agent's plan, tool calls, and reasoning fold and expand the same way in a recording as it did live is genuinely a different reviewing experience than reading a flattened text summary of the same turn. Detail that only exists in the live product is detail a recruiter loses the moment they switch to reviewing a recording, unless the recording is built from the same pieces.

This work happened in the same slice-by-slice way across the whole replay, not only in the agent view. The chat transcript was swapped first and checked on its own. The problem panel and its tabs came next, again checked on their own before anything else changed. Each piece earned its place by matching the live product exactly, rather than the whole replay being rewritten at once and trusted to be right.

Where it points

The same principle applies to every other surface that shows a candidate's work back to a reviewer: prefer showing the real thing over building something that resembles it, even when the resemblance is very close. Close is exactly the gap that erodes trust the moment someone notices it.