How Results Combine
A participant can have many conversations within a single run, and each of those conversations has its own result for a given analysis type. The participant list has room for only one entry per analysis type per person, so the product needs a way to represent several conversation-level results as one participant-level entry.
Both places you see this — the list/grid (the view you see day-to-day) and the participant drawer — render the exact same aggregated chips for a given output type. The only difference is how much of the set is shown: the list/grid truncates to the top 2 values with a +N more chip (hover to see the rest), while the drawer always shows the complete set. The list/grid also names the aggregation strategy once, in the column header — Most Recent for Score, Across All Conversations for Binary and Classification, Most Frequent for List — where the drawer just labels the row with the analysis type's name instead.
Binary (Yes/No)
Each participant's entry shows one chip per true/false value actually present among their conversations for that type, each chip carrying its own count and labelled with the type's own true/false labels (Complaint Detected/No Complaint, or the default Yes/No) — for example Complaint Detected · 3x. Since there are at most two possible values, nothing is ever truncated — the list/grid and the drawer always show identical chips.
Worked example: a participant had 4 conversations analysed for a Binary type, and 3 of them came back Yes. Both the list/grid and the drawer show a Yes · 3x chip alongside a No · 1x chip, using whatever labels that type defines.
Score
The column header reads Most Recent. Each entry shows the score from the participant's most recent conversation as a chip, plus a range bar spanning the minimum, maximum, and mean score across all of that participant's conversations for the type — so unlike the headline chip, the range bar does reflect every conversation, not just the latest one.
In the drawer, the same latest-conversation chip and range bar are shown, prefixed with a Recent Score: label and drawn at a larger size than in the list/grid.
Worked example: a participant has 5 conversations scored for satisfaction, and the newest one scored 8. In the list/grid, their entry shows a chip reading 8, with a range bar underneath drawn from the min, max, and mean of all 5 scores. In the drawer, the same 8 appears next to a Recent Score: label, with a larger range bar.
Classification (Category)
The column header reads Across All Conversations. In the list/grid, each entry shows up to 2 of the participant's distinct categories for that type as chips, sorted by how often each category occurred, plus a +N more chip beyond that (hover to see the rest) when there are more than 2.
In the drawer, every distinct category the participant's conversations produced is shown as its own chip — nothing is truncated, so there's no +N more chip there.
Worked example: a participant's 3 conversations were classified Positive, Positive, and Neutral. Both the list/grid and the drawer show both Positive (2x) and Neutral (1x) as chips — not a single winner (two categories fit within the list/grid's own top-2 limit here, so truncation doesn't kick in either way). If a participant instead had results across 4 distinct categories, the list/grid would show the top 2 plus a +2 more chip, while the drawer would show all 4.
List (Extracted Item)
The column header reads Most Frequent. In the list/grid, each entry shows the participant's 2 most-frequent extracted keys for that type as chips, plus a +N more chip beyond that, each chip carrying its own occurrence count.
In the drawer, every key returned for that participant is shown as its own chip, with no +N more truncation. Both views show a * Showing top X of Y distinct keys note underneath when there are more distinct keys overall than what's shown — that cap is separate from, and can apply on top of, the list/grid's own top-2 display limit.
Worked example: a participant had 6 conversations, and the topic "billing" was extracted in 4 of them alongside a few other topics. In the list/grid, their entry shows chips for the two most-frequent keys (billing (4x) and the next most common), plus a +N more chip if there are others. In the drawer, every returned key for this participant appears as its own chip, with the capped note underneath if the total was limited.
Other Output Types — Shown as "—"
Some output types — Schema is the only one today — have no dedicated combining rule of their own: there's no single value or set of values to fold a participant's conversations into. Both the list/grid and the drawer show a plain — chip for these, with a "This analysis produced no value here." tooltip on hover.