Reviews

​

What are Reviews?

The Reviews tab surfaces contracts and statements that scored low on extraction quality and need a closer look. When an item’s review score falls below the quality threshold, it is flagged here. The tab is a visibility surface: each flagged row shows the item’s score and the exact scoring rules that triggered, so you can open the contract or statement and correct it. Contracts and statements appear together in the same list.

​

Accessing Reviews

Navigate to Dashboard → Reviews tab. The tab lists every contract and statement currently flagged for review in the project, with the most actionable items shown first.

​

Overview banner

At the top of the Reviews tab, a banner summarizes the health of everything you’ve ingested into the project across two panels.

​

Ingestion

The Ingestion panel shows how much staged content needs attention, split into two cards: Each card shows a Rejected count and a Failed count — the items that need attention — alongside a progress bar tracking how much of the staged total is not rejected or failed.

​

Scores

The Scores panel summarizes extraction quality across your Contracts and Statements with a card each. Every card breaks the total into:

​

Review Rows

Each flagged item appears as a collapsible row: Click anywhere on a row (or press Enter/Space) to expand it. The expanded row is split into two tabs:

​

Status tab

The Status tab shows the triggered rules under a Score heading — the scoring checks the item failed or warned on, grouped by section so you can see exactly why it was flagged.

• Contracts group their rules by extraction section (parties, dates, royalties, and so on), plus document-level cross checks.
• Statements group their rules into Document Cross Checks (whole-statement reconciliation, shown first), then Sheet Cross Checks (per-sheet totals reconciliation), and then the pipeline phases — Scan, Parse, and Map — matching the groups on the statement’s score breakdown.

For a contract that has already been through a repair review, the Status tab also shows the full verdict — so you can see the outcome without opening the contract (see Verdict and root causes) — and the collapsed row shows the diagnosed root causes inline. Every row links to the item’s file name so you can open it directly.

​

Agent tab

The Agent tab shows the repair agent trace for contracts — the agent’s tool calls and reasoning as it investigates the flag. When a repair review is running, the trace streams in live; once it settles, the recorded trace stays available here. If no repair agent has run — including every statement row — the tab shows a No Repair Agent Trace placeholder.

​

Status

​

Verdict and root causes

When a contract’s repair review reaches Ready, expand the row and open the Status tab to see the verdict card. It records the proposed outcome — Repair (corrections were found) or Ignore (no change recommended) — along with the reasoning behind it and, for a repair, a summary of the corrected fields. A Ready row also shows a verdict badge on the collapsed row.

​

Root causes

Alongside the verdict, the card shows one or more root cause chips — the diagnosis of why the contract was flagged. A single review often spans several causes at once (for example, a missing-context miss and a genuine extraction miss in the same flag), so every cause found is listed rather than forcing one. Chips are ordered with the most significant cause first. The same causes appear inline on the collapsed row, so you can scan them before expanding. Root causes help you spot patterns across your reviews — for example, whether misses come from the extraction itself or from missing context.

​

Filters

Click Filter in the actions bar to narrow the list. A count badge shows how many filter groups are active. Use Clear within a group to reset just that group, or Clear at the top of the popover to remove all filters at once.

​

Sorts

Click Sort in the actions bar to change the order. Pick a field and a direction: Choose Ascending or Descending, and use Clear to return to the default order. By default, reviews are ordered by actionability — items with a verdict awaiting review come first, then flagged and reviewing items, with the newest first within each group.

​

Pagination

Long lists are paged. At the bottom of the list you can:

• Choose how many reviews to show per page (20, 50, or 100)
• See the total number of flagged items
• Page through results with the previous, next, and page-number controls

Your per-page choice and sort selection are remembered between visits.

​

Next Steps