Toolbox · Sortable verification table

Claim & entity matrix

A reading of a draft is not a review. This turns the draft into a working table: every figure, named organisation and link is extracted as a row, each row carries a status and a source URL, and the whole matrix sorts so you can work one claim type at a time and watch the unresolved count fall.

@tanstack/react-tableRuns in your browserNothing uploaded

Twenty-five instruments on one bench.

Live tool

Offline · nothing leaves this tab

Source URL
Statistic

2025

Model Context Protocol adoption tripled in 2025, with 42 percent of platform teams reporting a production d…

Statistic

42 percent

Model Context Protocol adoption tripled in 2025, with 42 percent of platform teams reporting a production deployment. Accord…

Statistic

1,200

…ints after the switch. The Whitespace Institute benchmarked 1,200 workflows and found latency fell to 240 milliseconds per re…

Statistic

240

…itute benchmarked 1,200 workflows and found latency fell to 240 milliseconds per request. Full methodology is published at…

Statistic

63 percent

…-benchmark. Teams at Northwind Logistics cut review time by 63 percent.

Proper noun

Model Context Protocol

Model Context Protocol adoption tripled in 2025, with 42 percent of platform teams…

Proper noun

Gartner

…tform teams reporting a production deployment. According to Gartner, retrieval accuracy improved by 18 points after the switch.…

Proper noun

Whitespace Institute

…retrieval accuracy improved by 18 points after the switch. The Whitespace Institute benchmarked 1,200 workflows and found latency fell to 2…

Proper noun

Northwind Logistics

…is published at https://example.com/mcp-benchmark. Teams at Northwind Logistics cut review time by 63 percent.

URL

https://example.com/mcp-benchmark

…milliseconds per request. Full methodology is published at https://example.com/mcp-benchmark. Teams at Northwind Logistics cut review time by 63 percent…

0 verified · 0 flagged · 10 unverified · 0 sourced

0% resolved

See what the certificate still needs

A list of claims is only useful if it can be worked

Reading a draft for accuracy fails for a boring reason: the work is unstructured. You notice a figure on page two, decide to check it later, and by page four you cannot remember which numbers you already confirmed. The matrix removes that bookkeeping. Every claim becomes a row with an identity, a state and a place to put the evidence, so the next decision is always about one row rather than the whole document.

Sorting is what makes it usable on a long piece. Sort by type to do all the statistics in one pass while the calculator is open, then by status to clear whatever is still unverified. Sort by claim text when a client asks whether a specific figure was checked — it is one lookup instead of a re-read.

What the three states commit you to

Verified means you found evidence and it supports the sentence as written: the row should carry the URL where you found it. Unverified is the honest default and stays that way until you act — a matrix full of unverified rows is a true picture of the work left. Flagged is for the claims that fail: a number nobody can source, a company that does not exist, a link that 404s or points somewhere unrelated.

The source URL is not decoration. A verified status without a source is an opinion, and the certificate will not accept it. Paste the page you checked, even when the same document supplies several rows — future you, or the client's compliance team, needs to re-open the evidence without repeating the search.

Saved between sessions, not to a server

Long-form work rarely finishes in one sitting. The draft, the statuses and the source URLs are kept in this browser's local storage as you type, so closing the tab and coming back tomorrow returns you to the same table. Nothing is uploaded, which is also why the tool cannot recover your work on another machine: the record lives on this device until you clear it.

Questions about this tool

Where is my progress stored?
In this browser's local storage, under the toolbox keys. It survives a reload or a new session on the same device and is cleared by the reset button or by clearing site data.
Does it check the sources for me?
No. No network requests are made. The matrix records what you verified and where; opening the source and judging it is your job.
Can I paste a very long draft?
Yes, though the matrix caps at 80 extracted rows so the table stays workable. Split a long report into its sections and run each one.

Toolbox

Fix the prose here, then check what the prose cannot show.

The console runs structural, claim and formatting checks on the same draft, and a named reviewer takes on anything with consequences.