Guide
Reporting
BrandKind reports in five places, each answering a different question. This page says what every number counts, what it deliberately excludes, and which screen to open when you want a particular answer.
Which screen answers which question
| Screen | Answers | Who can see it |
|---|---|---|
Reports | What has this workspace produced, and is it on brand? | Anyone in the workspace |
History | What did the system actually do, when, for whom, and did it work? | Anyone in the workspace |
Audit log | Who changed what, and when? | Managers and admins |
Generation runs | What is happening across every tenant? | RoodtSquared support staff only |
Platform health | Is the service itself healthy? | RoodtSquared support staff only |
Everything is scoped to one workspace
Reports
The at-a-glance view of what the workspace has made. Everything on it comes from a single query, so the whole page is one consistent snapshot rather than four figures taken at four moments.
The headline figures
| Figure | What it counts |
|---|---|
| Voices | Voice presets that exist in this workspace right now. Not how many have been used. |
| Pieces | Every piece ever created in this workspace, in any format, at any quality. |
| Images | Every image generated in this workspace, including ones never chosen. |
| Average brand fit | The mean brand-fit score across scored pieces only. Unscored pieces are excluded rather than counted as zero — otherwise a workspace would appear to get worse simply by generating more. |
Pieces by format
The split of your output across the six formats. Its total matches the Pieces figure above, and it is the quickest way to notice that a team believes it is producing a mix when it is in fact producing blogs.
Brand-fit distribution
How many pieces landed at each brand-fit score from 1 to 10. This is worth more than the average: an average of 7 made of consistent sevens is a healthy workspace, and the same average made of nines and threes is a voice that works for some topics and not others. The average hides that; the distribution shows it.
Scores that never occurred simply do not appear. A gap is an absence of data, not a zero.
Activity
Pieces created per day over the last 30 days, by creation date. Days with no activity are genuinely empty rather than missing. It counts pieces, not runs — a failed run produced no piece and so contributes nothing here, which is why this chart and History disagree.
History
Where Reports counts what survived, History records what the system did — including the runs that failed, the ones still going, and the voice previews that never produce a piece at all. This is the screen to open when something did not appear.
What a run is
Three kinds of work are recorded: generation (writing a piece), images (producing a set of images) and voice_capture (learning a voice from examples). Each moves through queued, running, then done or failed.
| Column | Meaning |
|---|---|
| Type | Which of the three kinds of work this was. |
| Label | The brief for a generation run, or the source it learned from for a brand capture. |
| Format | The piece format for generation runs; blank for captures. |
| Voice | The brand voice used. Kept as a snapshot if it has since been deleted, so old rows stay readable. |
| Requested by | The person who triggered it. Blank for older system-created rows. |
| Trigger | manual if a person started it, scheduled if a content job did. |
| Status | Where the run got to. |
| Error | Why it failed, in plain English. Empty for successful runs. |
| Duration | Wall-clock time from start to finish. Empty while a run is in flight. |
Sample runs
Previewing a voice while editing it starts a real run, and that run appears here badged as a sample. Samples never create a piece — that is the intended behaviour, not a failure. If your run count looks higher than your piece count, this is usually why.
The statistics strip
Above the list, per run type: how many ran in total, how many finished, how many failed, how many are in flight, and the average and median duration.
Durations cover successful runs only
done. A failure that hung for two minutes before giving up does not appear in either. Read the failure count alongside the timings — the timings alone will look healthy during an outage.Prefer the median to the average. One pathological run drags an average noticeably in a workspace with modest volume; the median tells you what a normal run feels like.
There is also a per-voice breakdown — total, finished, failed and average duration for each voice. A voice with a markedly worse failure rate than its siblings usually has a problem in its definition rather than bad luck.
Filtering
You can narrow by run type, by outcome, by voice, and by time window. The window defaults to the last 30 days and can be set to all time. Outcome offers everything, finished, failed, or still in flight — the last of these is the fastest way to answer “is it stuck, or is it just slow?”
Audit log
A record of consequential actions in the workspace — invitations, role changes, deletions, configuration changes — with who did each and when. It is read-only for everyone, including admins, and nothing in the product can edit or remove an entry.
It answers governance questions, not usage questions. “Who removed this person?” belongs here; “how much have we produced?” belongs in Reports. Managers and admins can see it; members and viewers cannot. More in Team and access.
The two platform screens
Generation runs and Platform health belong to RoodtSquared support staff and are not visible to customers at any workspace role. They exist so that when you report a problem, someone can see it across tenants rather than asking you to reproduce it. They are described in Platform admin.
Reading the numbers honestly
- Reports counts outputs; History counts attempts. They are supposed to differ. Failed runs and voice samples explain nearly every gap.
- Average brand fit rises as you delete weak work. It is a measure of what you kept, not of what you produced. Judge a voice by the distribution instead.
- A high Human score is not a claim about quality. It measures how likely text is to be flagged as machine-written, and nothing more. See Human score.
- Nothing here is billing data. Runs are not credits — some actions cost more than others, and failures may not cost at all. Use Billing and credits for spend.
Getting the data out
There is no scheduled report and no reporting API. Workspace admins can export the workspace's data, which is the supported route into a spreadsheet or your own warehouse — see Workspaces and branding.