Understanding Findings
When a review finishes, Judicio reads through every check against every page and surfaces what it found as a list of findings in the rail on the right of the results screen. Each finding is one issue, answer, or observation — with everything you need to judge it in a single card: how serious it is, how sure the AI is, where it lives in the document, and a one-line explanation. This page teaches you to read that card so you can triage a long list in seconds and spend your time where it matters.
The document renders in the centre; the findings rail sits on the right. Click any finding to open it — the rail switches to a detail view with the original clause and the resolution actions (covered in Resolving Findings). First, learn what every part of the card is telling you.
Anatomy of a finding card
Reading top to bottom, a card carries: the title, then a row of badges (risk, an optional typed-value pill, and confidence), then a one-line description, then a footer with the category, the page number, and the section. The right edge shows the finding's status — Review pending, or Reviewed once you've acted on it.
The risk badge — what to fix first
Every finding opens with a coloured risk pill. Risk is Judicio's single best signal for triage order — it answers "if I only had time for a few of these, which ones?"
| Badge | Colour | What it means | What to do |
|---|---|---|---|
| High Risk | Red | A serious problem — a clause that could expose your client to real legal or financial harm, or a required term that is missing or one-sided. | Review first, every time. Fix it with an AI suggestion or your own edit, or flag it for a colleague. Don't leave a high-risk finding untouched without a deliberate decision. |
| Medium Risk | Amber | A genuine concern worth a look, but not necessarily a deal-breaker — often a clause that deviates from your preferred position into a workable-but-not-ideal range. | Read it carefully and decide in context: fix, negotiate, or accept. |
| Low Risk | Green | A minor observation or an easy improvement — a small deviation, or a clause that's fine but could be tightened. | Handle when time permits. Often safe to accept once you've confirmed it reads sensibly. |
A finished review can return dozens of findings. The colour is there so you never have to read them in order — sort your attention by risk and clear the High Risk items before anything else. Use the status tabs (below) together with risk to carve the list into a manageable queue.
The confidence score — how much to trust it
Next to the risk pill, most findings show a confidence badge — the AI's own estimate of how certain it is, as a band plus a percentage. Hover it to see the exact reading ("Model confidence: 84%").
| Badge | Reading | How to treat it |
|---|---|---|
| High conf. | 80% and above | The finding is well-grounded in the text. Still worth a glance, but you can move quickly. |
| Med. conf. | 50–79% | Plausible, but read the cited clause before you act — the wording may be ambiguous or the match imperfect. |
| Low conf. | Below 50% | Treat as a prompt to look, not a conclusion. Open the citation and verify against the source before you fix or accept. A low score often means a poor scan, unusual phrasing, or a borderline call. |
A low-confidence finding isn't wrong — it's the AI flagging that it's unsure. That's exactly the kind of item a human should look at. Click through to the clause and make the call yourself.
Older reviews run before confidence scoring was added simply omit this badge — nothing is wrong; there's just no score to show.
The typed-value pill — a structured answer at a glance
When a check is configured to return a structured answer rather than free prose — a Yes/No, a date, a currency amount, a percentage, a number, or a tag from a fixed list — the card shows that answer as a typed-value pill right after the risk badge.
- A Yes / No check gets a strong signal: a green pill for Yes, a red pill for No, so a compliant-or-not answer reads instantly across a whole list.
- Every other structured type — a date like 1 January 2024, an amount like $50,000, a percentage like 12.5%, a count like 3 items, or a tag — uses a neutral pill. The value is data, not a verdict, so it stays understated.
- If the AI declared a structured format but couldn't produce a clean value, you'll see a muted Format mismatch chip instead. The original answer is still in the description — the chip is just a heads-up that the output didn't conform, so give that finding a closer read.
Free-text checks (the default) have no pill — their answer is the description line.
The description
A plain-language, one-line summary of what the check found and why it matters. It's written to be skimmable; the full context — the actual clause — is one click away via the citation.
The category tag
In the footer, a subtle category tag groups the finding by the kind of check that produced it (for example Liability, Confidentiality, Termination). Categories let you read the document by theme — scan all the liability findings together, then all the indemnity findings — instead of jumping around.
The page and section citation — your jump to the source
Every finding is anchored to where it came from. The footer shows:
- Page number — the page in the source document the clause sits on.
- § Section — the clause or section label when Judicio could identify one (hover to read a long label in full).
This is the most important habit in Document Review: click into a finding and use View in document to jump straight to the cited clause, highlighted in place, so you read the AI's finding against the real text — never in isolation. The detail view's View in document button (see Resolving Findings) scrolls the centre panel to the exact passage.
Occasionally a finding can't be tied to a precise location — a missing-clause finding, for instance, has no clause to point at. Those show a small "no location" marker, and View in document will scroll to the page rather than highlight a span. Everything else about the finding is unaffected.
Judicio's analysis is strong, but it is an aid, not a substitute for your judgement. Every finding is cited precisely so a single click confirms it — make that click your default before you accept, fix, or rely on any finding, especially anything marked high-risk or low-confidence.
The status indicator
The right edge of each card shows where the finding stands:
- Review pending — you haven't acted on it yet.
- Reviewed — you've resolved it, with a small badge naming how (for example AI fix, Edited, Accepted).
Pending, Flagged, and Resolved
Above the list, three tabs split your findings by status, each with a live count so you can see your progress at a glance:
| Tab | What's in it |
|---|---|
| Pending | Findings you haven't touched yet — your working queue. Start here. |
| Flagged | Findings you've set aside for someone else to weigh in on — a partner, a colleague, or the client. Each carries the note you left. See Resolving Findings. |
| Resolved | Findings you've finished with — fixed, edited, or accepted. Your audit trail of decisions. |
The counts update the moment you act on a finding, so a review is "done" when Pending reaches zero (with anything you couldn't decide sitting in Flagged for follow-up).
- Open Pending and let the risk colours order your attention — clear every High Risk finding first.
- For each one, click the card, read the description, and View in document to check the clause in context — especially anything low-confidence.
- Resolve it (fix, edit, accept) or flag it with a note for whoever needs to decide.
- Work down to medium and low risk. When Pending hits zero, your review is complete.
Next steps
- Resolving Findings — act on each finding: AI Fix, a custom edit, Accept, Flag, Undo, and refining the wording.
- Review Modes — how concise and deep modes change the depth of analysis behind these findings.
- Exporting Results — turn your reviewed findings into a report or a redlined document.