Every application scored to your rubric. Every score survives the audit.
EvalLens is a structured first-read layer for your existing review process: it reads every proposal in full, scores it against your rubric anchor by anchor — a quote and a page reference behind every score — and hands your panel a pre-read brief. Never a decision.
Your reviewer pool is heroic. The math isn't
- ICC 0.26measured agreement between independent reviewers across 23,414 ratings at a national science fund. Your consistency problem is documented, not hypotheticalPLOS One
- 2×application volume growth at a major national funder since 2017 — while award rates fell from 36% to 19%UKRI, via LSE Impact Blog
- 18+ moend-to-end decision cycles at major schemes. Applicants call. Boards noticefunder publications
- Midnightyour confidentiality policy is only as strong as your most tired reviewer at midnight — with a free chatbot one tab awaythe reason NIH wrote the rule
“Doesn’t NIH prohibit this?”
Two different rules, and we answer both plainly.
NIH NOT-OD-23-149 bars reviewers from uploading confidential proposals to consumer AI tools. So do we: that rule exists because individual reviewers paste applications into public chatbots, unlogged and retained by whoever runs the tool. An organizer-governed perimeter is a different object, and we will walk your counsel through the distinction.
The ERC guidance of March 2026 goes further: reviewers may not delegate the assessment of scientific merit to AI at all. That is a boundary no perimeter fixes, so we do not argue with it. In programs under rules like these, EvalLens runs the administrative half of the round, intake, completeness, eligibility, comparability checks and the record, while merit stays entirely with your reviewers. Prize programs, foundation calls and competitions that set their own rules can use the scored first read as well. If you are not sure which side your call sits on, that is the first question we ask, before any document moves.
Three years later, someone asks “why 4.2?”
Today the answer lives in scanned scoresheets and a departed reviewer’s inbox. With EvalLens, staff opens the record for that application in one click: the AI read and the panel score both preserved side by side, each finding tied to the quote and page it came from, and the human sign-off attached. Field-level change history and rubric versioning are on the roadmap, and we will tell you plainly where that line sits before you buy.
When an appeal lands, you open the record. You don’t reconstruct it. Independent reads — reviewers never see each other’s scores; disagreement surfaces to the panel, never averaged away; bias made inspectable with score distributions by geography & org size, on request.
Your process, pre-read. Seven steps
Your reviewers remain the reviewers of record. EvalLens prepares the consistent first read underneath.
- 01
Your rubric, locked before the call opens
Criteria, anchor descriptions, weights and eligibility rules configured in one working session — then applied identically to every application that will ever arrive. Procedural fairness, by construction. You get: a documented methodology you can publish.
- 02
Applications flow from your existing intake
Submittable, SurveyMonkey Apply, Fluxx, SmartSimple, OpenWater or your own forms — we ingest the batch. Applicants change nothing. You get: no migration, no applicant-facing change.
- 03
The administrative screen runs itself
Completeness and eligibility checked against your rules, gaps flagged as info / warning / critical. Staff handles exceptions, not the pile. You get: staff weeks back before review even starts.
- 04
Independent AI reviewers — named honestly
Not people: independent AI reviewer roles, each reading the full proposal through its own lens, composed per program — including domain-matched technical reads. Each scores blind to the others, evidence before score, and every model's read is logged — so “why is this reviewer qualified” has an answer too. Your human panel's COI and recusal workflow stays exactly where it is. You get: 3–5 reviews' worth of reading on every application.
- 05
Panels read briefs, not piles
Every proposal arrives pre-read: comparable scores, laid-out evidence, ranked open questions. Your judges read 15 briefs, not 900 PDFs — expertise goes to judgment on the borderline. You get: panel meetings that start at the finish line.
- 06
The committee decides. The memo comes from the record.
Finalists and awards are built from your panel's scores; AI reads stay advisory. The selection memo is generated from the live review record — not reconstructed for the board. You get: a board-ready memo with page references.
- 07
Feedback for every applicant — approved before it leaves
Rubric-grounded feedback is drafted from the evidence and reviewed by your staff before anything is sent. Every applicant gets a real answer — and the decision record stands behind every word of it. You get: goodwill at zero marginal staff time.
Day 0 vs award day.
Your governance, as it stands
Your rubric with anchors and weights · your COI and confidentiality policy · your existing intake system and this round's batch. No rubric written down? The setup session turns how your panel already decides into one — yours to keep either way.
A defensible round, documented
Every application read in full with coverage logged · evidence-linked reviews with quotes and page references · a comparable ranked pool with disagreements surfaced · awards decided by your panel, logged · approved feedback for every applicant · an exportable review record that outlives staff turnover.
The block your counsel reads first.
Never used for training
Applicant documents are processed solely for your evaluation, under a GDPR-aligned DPA. Never used to train models — in writing.
The security pack, up front
Architecture and data-flow diagram, sub-processor and model-provider list, DPA template, retention terms, EU processing and data-residency options, and our EU AI Act position. Bring your questionnaire to the first call — review completes before any document moves.
Deployment on your terms
Closed managed perimeter today; dedicated processing, your approved models, your environment and your retention policy on the Enterprise track.
One round, in parallel. Then decide.Your reviewers work one round as usual; EvalLens reads the same applications. You compare reading coverage, panel-hours, shortlist overlap against a documented human baseline, and the completeness of the decision record. Fixed-fee parallel pilot, 4–6 weeks, security review first; then a program licence sized on the pilot’s own numbers. See pricing or book a call.
Asked by every program officer
- Does the AI pick the winners?
- No — by design. AI reads are advisory input; finalists and awards are built exclusively from your panel's scores, with every confirmation and override logged. There is no mode where the machine decides.
- Our donors contractually require human review.
- And they keep it. Your reviewers and panel remain the reviewers of record — they confirm, override and decide, with their COI and recusal workflow untouched. What changes is what they review: a complete, consistent pre-read with documented evidence instead of a raw pile.
- How does this square with NIH-style AI prohibitions?
- Those rules bar reviewers from uploading confidential applications to consumer tools — a rule we agree with. An organizer-governed closed perimeter with logged processing, no training on applicant data, and template applicant disclosure is a different legal object. We'll walk your counsel through it and leave the memo.
- What do applicants get told?
- Whatever your policy requires — and we make it easy: template disclosure language for the call documents, plus a plain-language description of how AI assists (and never replaces) the human panel.
- What about bias?
- The panel applies the same anchored rules to every application — no fatigue curve, no order effects, no Friday-afternoon reviewer. Every score's evidence is logged, disagreement is surfaced, and on request we produce score-distribution breakdowns by geography, organization size and budget band — the chart your DEI committee will ask for.
- Who has actually run this?
- We're onboarding a founding cohort of programs — which is why the engagement starts with a measured parallel pilot on your own round rather than someone else's logo. You see agreement, misses and record quality on your data before you commit.
Pick one round. Measure us against your panel.
30 minutes with you: we map the rubric, agree the pilot metrics, and schedule your security review first. Bring the questionnaire. The first run is free through August 31, for batches up to 10 decks.