A report is what an interview run adds up to. It reads every transcript, finds what held across your synthetic participants, and puts it in front of you as a decision rather than a pile of quotes. This article walks the report top to bottom and tells you what to trust, what to question, and where the raw evidence sits.
You don’t launch synthesis yourself. When every session in an auto-interview run reaches a terminal state, Candor starts synthesis automatically and the progress bar on the run page shifts to the synthesis phase. When it finishes, a banner appears with a View report link.
One run produces one report. Running the same guide twice gives you two reports, and both are kept.
All four interview types lay out the same way, in four sections. Only the contents of Findings really change.
Everything except What Matters Most starts collapsed. That is deliberate: the report is meant to be skimmed to a decision first and audited second. Cross-reference links will open a collapsed section for you when you follow one.
Inside What Matters Most, below the headline and alongside the sample details, is a learnings block. Its learnings are numbered 1, 2, 3 so you can refer to them. It is the part you should always read, and it’s worth knowing that it appears in a slightly different form depending on the interview type.
In problem discovery and problem validation reports, each learning also carries clickable links down to the findings it was drawn from. Use those rather than scrolling: they take you straight to the evidence for that specific claim.
Reports do not label individual findings with a confidence level. A single word invites you to skim past everything underneath it, and it flattens the thing worth reading, which is how the evidence behind a finding is actually shaped.
What tells you instead is on the page, and it differs by type. Problem discovery and problem validation put a participant count on each finding. On a newer validation report, a tested hypothesis shows how many participants agreed out of those who were asked about it, how many agreed only in part, how many disagreed, and who they were, and it quotes the pushback participants gave when the interview closed by asking what didn’t fit. Pushback on how the problems were described as a whole gets its own short list at the bottom of the Tested hypotheses group. An older one counts everyone who spoke to the topic, including people who rejected the hypothesis. Newer discovery reports count more strictly, counting only people who said more than one distinct thing about a finding, and they say so in the sentence above the Findings list. Older ones use the looser count and that sentence reads “A ranked list ordered by how broadly each finding showed up” instead, so check which one you have before comparing numbers across two reports. Concept testing gives each concept an explicit verdict instead; on a card-sort test (five or more concepts) each concept also says how many participants were probed in depth on it. Price testing leads with a recommended price and the reads behind it.
Three things are on every report whatever its type. The yellow banner at the top, when the run was thin or partial in a way that should make you cautious throughout. The paragraph the report opens with, which says in plain words how settled its findings are. And, in problem discovery only, a Thin, verify first mark on a finding the evidence barely supports.
None of it changes the basic rule: every report is directional. It tells you where to look and what to ask real people about next. It does not settle the question on its own.
One thing worth knowing about validation reports. The problems that surfaced on their own carry a strength score worked out from the transcripts. You never see it, on the page or in the download. It shows up only as ordering: within the same severity, better-evidenced problems rank higher in the at-a-glance table. So if two problems of the same severity sit in an order that surprises you, that is why.
Some report types return an explicit verdict as a coloured pill. Each type uses its own vocabulary, so the words differ by report.
The words differ by type, so your type’s article lists its own set. The colours do not, because every report draws them from the same palette: green means go, yellow means change something, red means stop, grey means the run couldn’t tell you, and blue means the verdict is about how something was described rather than about the thing itself.
A yellow banner at the top of a report means the run was partial or thin in a way that affects how far you should push the conclusions. It lists each issue on its own line, with the participants involved where that applies. The usual causes:
Those last three all mean the same thing for the numbers: that interview is subtracted from the participant count at the top of the report, and you are not billed for it. So the count you see always matches what the report was built from. If you ran 24 interviews and one could not be read, the report says 23.
The report is still worth reading when this banner appears. Treat it as a narrower base, and if the decision is expensive, run again with more participants before committing.
Everything above holds for any report you open. What each section actually contains does not, and neither does the verdict vocabulary or what lands in the appendix. Each type has its own article covering its sections and the judgement calls specific to reading it.
The export menu in the report header offers three scopes.
Reach for Markdown or CSV when you want to do your own analysis elsewhere. Neither truncates, so you can hand the whole run to another tool. PDF is the one to send to someone who just needs to read it.
Every finished report ends with one question, after the last section: how useful was the information in it, on a five-point scale, with an optional box for what worked or what was missing. That answer goes to the Candor team and nowhere near the research. Each person with access to the report is asked once for that report; answering or closing the question retires it for you only, so a colleague opening the same report is still asked.
A report is built in two stages: first every interview transcript is read, then the report is written from what was read. The page always names which stage it stopped in, and the button always says what pressing it will do. Most failures are transient and a single press clears them.
If the write-up stage failed, you get Retry from failed step. It picks up at the step that failed rather than starting over, and it never re-reads a transcript.
If the reading stage stopped partway, so most transcripts were read and a few were missed, you get Re-read the transcripts. It reads only the ones that were missed. Everything already read is kept, and you are not charged twice for it.
If every transcript was read but the report never moved on to writing itself up, you get Finish this report. Nothing is read again; it starts the write-up from what is already there.
Two failures offer no button at all, and in both cases your transcripts are saved and you are not charged for the report. If none of the interviews could be read, there is nothing to build from, and retrying cannot make a transcript readable. If a report never started reading at all, there is nothing to pick up. Contact support for either and we can look at why.
Start with the learnings block inside What Matters Most, which is the only section open when the report loads. In problem validation, concept testing, and price testing it's titled Top Learnings and Recommended Next Steps and pairs each learning with the action it implies. When that action is a step in Next Steps, the learning links to the step instead of repeating it. In problem discovery it's titled Key Learnings and carries no per-learning action, because discovery is for understanding: the actions live in Findings and Next Steps, Either way, read it, then drop into Findings only for the learnings you want the evidence behind. If you only have five minutes with a report, spend them there.
Reports do not put a confidence label on individual findings. A single label invites you to read past everything under it, and it hides the thing that actually matters, which is how the evidence is shaped. What to read instead depends on the type. Problem discovery and problem validation put a participant count on each finding, so a finding two people raised tells you something different from one eight people raised. On a newer validation report, a tested hypothesis shows how many participants agreed out of those who were asked about it, how many agreed only in part, how many disagreed, and who they were (in a card sort, everyone who sorted the hypothesis counts as asked, and anyone whose answers gave no position is placed by their sort), and it quotes the pushback participants gave when the interview closed by asking what didn't fit; an older one counts everyone who spoke to the topic, including people who rejected the hypothesis. Newer discovery reports count more strictly, counting only people who said more than one distinct thing about a finding, and they say so in the sentence above the Findings list. Older ones use the looser count and that sentence reads "A ranked list ordered by how broadly each finding showed up" instead, so check which one you are looking at before comparing numbers across two reports. Concept testing gives each concept an explicit verdict instead; on a card-sort test (five or more concepts) each concept also says how many participants were probed in depth on it. Price testing leads with a recommended price and the reads behind it. Three things apply to every type: the yellow banner at the top, which names anything about the run that should make you cautious; the paragraph the report opens with, saying in plain words how settled its findings are; and, in problem discovery only, a Thin, verify first mark on a finding the evidence barely supports. One thing worth knowing about validation reports: the problems that surfaced on their own carry a strength score computed from the transcripts, which is never shown to you but does decide their order within each severity band in the at-a-glance table.
That banner means the run was thin or partial in a way you should know about before you trust the numbers. The common causes are two or fewer synthetic participants, one or more sessions that ended before eight exchanges, a study with no explicit learning goals so they had to be inferred, sessions filtered out before synthesis ran, and individual interviews that contributed nothing. There are three distinct versions of that last one and the banner names the participant for each: the transcript could not be read, the interview was readable but produced no findings, or the session left no usable transcript at all. Each row names the problem and, where it applies, the participants affected. An interview that contributed nothing is subtracted from the participant count at the top and is not billed, so that number always matches what the report was built from: 24 run with one unreadable shows 23. The report is still valid, but read it knowing the base is narrower than usual.
Usually yes, and the button on the page tells you which case you are in. If most transcripts were read and a few were missed, you get Re-read the transcripts: it reads only the missed ones, keeps everything already read, and does not charge you twice for it. If every transcript was read but the report never moved on to writing itself up, you get Finish this report: nothing is read again, it just starts the write-up. If the report never started reading at all, there is no button, because there is nothing to pick up; your transcripts are saved, nothing is charged, and support can look at why.
Then no report is produced. Candor needs at least one readable interview to build from, and a report assembled from nothing would look authoritative while resting on no evidence at all. The run fails, the page tells you plainly that none of the interviews could be read, and there is no retry button because retrying cannot make a transcript readable. Your transcripts are saved and you are not charged for the report. Contact support and we can look at why it happened. This is different from two other things. One or two interviews being unreadable is common and simply shrinks the report: those are named in the yellow banner, subtracted from the participant count, and not billed. And a report that stopped partway through reading is recoverable: it offers a Re-read the transcripts button that picks up only the ones that were missed.
They share a spine and diverge in the middle. Every report has four sections: What Matters Most, Findings, Next Steps, and an evidence appendix, with the learnings block inside the first of those and only that section expanded when the report opens. What sits inside Findings changes completely by type, along with the verdict vocabulary and what lands in the appendix, so each type has its own article rather than a paragraph here. One naming quirk worth knowing up front: price testing titles its third section What's Next rather than Next Steps.
Yes, from the export menu in the report header. Report gives you the written report as Markdown or PDF. Transcripts gives you the raw interviews as CSV, Markdown, or PDF: CSV arrives as one file containing every session, while Markdown and PDF arrive as a zip with one file per participant. Everything gives you a zip with the report plus one transcript file per participant. Markdown and CSV are the ones to reach for if you want to run your own analysis in another tool, since neither truncates the content.
A yellow banner above it says the project is archived and read-only. The report reads exactly as before and every export still works. What an archived project can't do is run new research: the buttons that would start it are greyed out on the project's other pages, and if this report failed to build, its retry button is greyed out too. Unarchive the project to run more interviews or retry the report.
Start with a free project, or walk through Candor with us first.