SlopChecker

Check whether a grant proposal holds up, before you spend an hour reading it.

Upload a grant application. SlopChecker verifies its citations against the published record, runs AI-detection signals over the text, compares it against submissions you've already received, and tags it by topic and type — then hands back one page of evidence with every finding tied to the sentence it came from. A score is a signal, not a verdict: it doesn't rate the proposal and it never rejects anything. That stays with you.

Check a proposal or look at a finished report first.
What a finding looks like

This proposal cites a study that does not exist, quotes a paper that never said it, and scores 0.96 on an AI detector.

Why it exists

Proposals increasingly cite sources that don't exist. The reference looks correct, but the paper isn't there — or it is, and never said what the proposal claims.

Catching that means looking sources up one at a time, which nobody has time for across a submission window. SlopChecker does it mechanically, and shows you its working: every finding is attached to the sentence it came from, so you can confirm any one of them yourself in about thirty seconds.

More on what it does and doesn't do
  • It does not decide anything. There is no overall score, no ranking, no flag that removes a proposal from consideration. Findings go to a human reviewer, who decides what they mean. A passage the tool never flags can still be a weak passage.
  • Two kinds of check, kept apart. Some are lookups anyone can repeat by hand: does the DOI resolve, do the publication details match, does the quoted sentence appear in the source. Others are judgements made with a language model: whether a reference actually supports the claim it's cited for, how closely a submission overlaps another. The report labels which is which, and both kinds are pinned to a quote you can read for yourself.
  • An AI-detection score is a signal, not a verdict. Passages are scored by Pangram, a third-party detector. The score sits in its own lane, marked as an estimate. An applicant drafting with AI is not what this tool objects to — claims that can't be verified are, regardless of who or what wrote them.
  • It reports its own blind spots. A paywalled source, a scanned PDF with no readable text, a check that could not run — each is listed as a gap in coverage rather than passed over quietly. You always know what was not checked.
DOI
The permanent ID attached to a published paper, like a catalogue number. If a DOI does not resolve to anything, the cited paper does not exist at that address.
Detector score
A third-party estimate, from 0 to 1, of how likely a passage is machine-written. An estimate, not a measurement, and never grounds for rejection on its own.
Evidence report
What you get back: the proposal with findings marked in place, plus a full list of every check that ran, passed, failed, or was skipped.

How it works

  1. STEP 1

    You upload

    A grant application or blog post — PDF, Word, Markdown, or HTML. The text comes out with its page positions kept. Attach your own solicitation alongside it and the submission is checked against your rules too.

  2. STEP 2

    Checks run

    Every citation and quotation is looked up. Claims are extracted and compared against your other submissions. Each check records its own result.

  3. STEP 3

    You get evidence

    One page, on screen or printed: each finding marked where it occurs, its evidence beside it, and a list of everything that ran.

  4. STEP 4

    You decide

    SlopChecker recommends passages for a human to look at. It never accepts or rejects a proposal, and it doesn't rank them.

What it checks

Citation integrity

References are pulled out of the document and checked four ways — the last one matters even when a source is perfectly real:

  • Is the citation validly formed, and does its DOI or link resolve?
  • Do the author, title, and year match the actual source?
  • If a passage is quoted, does that sentence appear in the source?
  • Does the source actually support the claim it's cited for?

AI-generation signals

Passages are scored by Pangram, a third-party detector, through its API. The score is reported as a score — context for a reviewer, in its own lane, never a pass or a fail.

Overlap with other submissions

The core claims of a submission are extracted and compared against proposals you've already received — and against your reviewer pool where you have one — so recycled and near-duplicate applications surface.

Compliance and budget

Where you supply the solicitation, submissions are checked against it: required sections, eligibility, scope. Budget figures are reviewed for internal consistency and plausibility against what's being proposed.

Background on the applicant

A short report on the submitting team and their topic, assembled from public sources and cited throughout, so every statement in it can be traced back.

Topics and routing

Each submission is tagged with its topics, document type, and the kind of organisation that sent it — so a large batch can be sorted and routed to the right reviewer.

Every run reports which of these checks ran, which were skipped, and why. A missing API key, a paywalled source, or an unreachable server is recorded as a gap in coverage — never as a quiet pass.

Open source · read the code Grant applications and blog posts PDF · Word · Markdown · HTML

How to read a result

NO

Failed a check you can repeat

The citation didn't resolve, or the quote isn't in the paper. You can confirm it yourself without taking this tool's word for it.

YES

Passed a check you can repeat

The source exists and its publication details match. Passing isn't an endorsement of the argument, only of the citation.

0.96

A score, not a verdict

AI-detector and similarity scores sit in their own lane. They are context for a human reviewer, never grounds for rejection.

Check a proposal

Upload a proposal Drop a PDF here, or choose a file. You get a report, not a decision — nothing is accepted or rejected automatically.

Nothing attached — compliance is reported as unchecked, not passed.

Attach the RFP or scoring criteria and the proposal is checked against your own rules — budget ceilings, required sections — with the funder's line quoted beside the passage that breaks it.

Rather not upload anything yet? Open a sample evidence report for a proposal we wrote ourselves.