Separate outputs from outcomes, with the application quoted

Each measurement element, quoted or flagged

Define the measurement elements your rubric asks for: outcome statement, baseline, data collection method, timeline, and responsible party. For each one, the agent shows the quoted passage from the proposal, or flags that the section is silent on it. You see where an applicant named an outcome but never said how it would be measured. You also see where the plan stops at counting outputs.

See where outputs are passed off as outcomes

A proposal that promises to train 200 teachers has stated an output. One that measures whether student reading scores moved has stated an outcome. The agent checks whether the proposal addresses the outcome your criterion asks about, then quotes the text it found. When the only supporting passage describes activity counts, that sits on the page next to the criterion. The conflation is visible instead of buried in the narrative.

A person judges whether the design is strong

Whether a measurement plan is rigorous is a judgment call, so the agent never scores it alone. It confirms the element is present and quotes it. Then it routes the rigor question to a reviewer with the criterion and the quoted passage attached. You decide whether a pre-post comparison with no baseline is good enough, or whether the timeline is realistic. The agent does the locating. You do the judging.

Silence is flagged, not guessed

If a proposal never states a baseline, the agent reports the gap against that criterion rather than inferring one. The same holds for a missing data source or an unnamed responsible party. A declined applicant who asks why gets a specific answer: the rubric asked for a baseline, and the application did not address it. Every finding traces to the passage it came from, or to the absence of one.

How a review runs

1. Step 1

   Define your measurement criteria

   List the elements your rubric asks for in the measurement section: outcome statement, baseline, method, timeline, responsible party. The agent checks against that, not a generic template.

2. Step 2

   Bring the proposals

   Upload the submissions or export them from your grants platform. Each review is scoped to the set you provide and stays in your account.

3. Step 3

   Deterministic checks locate each element

   For every proposal and every criterion, the agent quotes the supporting passage or flags that the section is silent. It runs the same way for each applicant.

4. Step 4

   A reviewer judges the design

   The rigor of each plan is routed to a person with the criterion and quoted text attached. The agent proposes. The reviewer decides, and the decision is recorded with its evidence.

Common questions

How does it tell an output from an outcome?

It does not decide that on its own. Your rubric defines the outcome each criterion is asking about. The agent checks whether the proposal addresses that outcome and quotes the supporting passage, or flags that the only relevant text describes outputs. A reviewer reads the quote and makes the call. The agent makes the conflation visible. It does not rule on it.

What measurement elements can it check for?

Whatever your rubric lists. Common ones are an outcome statement, a baseline, a data collection method, a measurement timeline, and a responsible party. For each element you define, the agent reports the quoted passage or flags the gap. Change the rubric and the elements it checks change with it.

Does it score the quality of the evaluation design?

No. Confirming an element is present is a deterministic check. Judging whether the design is rigorous is a judgment call, and that always goes to a person. The agent attaches the criterion and the quoted text so the reviewer can decide quickly. The rigor verdict stays the reviewer's.

What if a strong plan uses unconventional language?

That is exactly why a person makes the judgment call. The agent quotes the passages it found against each criterion, including ones that do not match the usual phrasing, and routes the assessment to a reviewer. You are never relying on the model to recognize good evaluation design. You are reading the applicant's own words against your criterion.

Does it connect to your grants management system?

Not as a self-serve feature today. Most teams export submissions and upload them. Ask about direct connectors to systems like Fluxx, Submittable, and Foundant as part of an Enterprise engagement. Write to [email protected].

How do I get started?

Sign in and upload a few recent proposals with the measurement criteria from your rubric. Or schedule a 30-minute walkthrough on your own evaluation review.