How each feature works — inputs, limits, and what it does not do.
One page, five features. For each: what it inspects, what it returns, its known limitations, and how we handle your data. Where a policy is still being finalized, we say so explicitly instead of inventing one.
RUBRIC v2.1 · TWO TRACKS · EACH SUMS TO 100
Feature-by-feature governance
For each feature: what goes in, what comes out, what it does not do, and how we handle your data. Retention and deletion follow the commitments on the Trust page: analyses are auto-deleted 30 days after they are created unless you save them to your Document Vault, and manual deletion requests are confirmed within 72 hours. The per-feature data-flow table on our privacy page states exactly what is stored.
Submission Check
- Engine type
- Deterministic PDF parser and rules engine. This feature calls no AI model.
- What it inspects
- Font family and point size, margins, page count, page geometry, and file/PDF metadata — checked against a SQL-backed table of NIH format rules with a verified-on date.
- Outputs
- Pass / warn / fail findings per attachment, with the exact rule cited and the excerpt or measurement that triggered it.
- Model exposure
- The uploaded PDF's contents are not sent to any language model. Files are used in-memory during the check and are not persisted.
- Limitations
- It verifies mechanical compliance only — it does not judge the science, and passing this check does not mean an application is competitive.
Submission Match (Quick and Deep Score)
- Inputs — Quick Score
- Self-reported intake fields: mechanism, project stage, team, preliminary data, and (for SBIR/STTR) commercial framing. No documents required.
- Inputs — Deep Score
- Everything in Quick Score, plus uploaded Specific Aims, Research Strategy, and other application attachments.
- Rubric depth
- Two mechanism-specific tracks, published in full below. R-series uses 4 pillars (importance, rigor & feasibility, expertise, submission readiness). SBIR/STTR uses 6 pillars — the same four plus commercialization and an eligibility / work-split check. Weighted sub-scores sum to 100% within each track.
- Outputs
- A 0–100 Merit Readiness score, a confidence rating, per-pillar sub-scores, a ranked Best Destination table by agency fit, and prioritized gaps.
- Typical turnaround
- Quick Score: about five minutes end-to-end. Deep Score: a few minutes of scoring after documents are uploaded.
- Limitations
- The score is an advisory readiness signal to help prioritize revisions. It is not a prediction of NIH's funding decision and not a numeric reproduction of NIH peer review. Under NIH's Simplified Review Framework (due dates on or after January 25, 2025), Factors 1 and 2 receive 1–9 criterion scores and Factor 3 is rated sufficient or not sufficient; reviewers then assign a single holistic Overall Impact score (1–9) informed by all three factors.
Winner Intelligence
- Data source
- NIH RePORTER — the official federal award database — covering NIH projects (including SBIR/STTR under R41/R42/R43/R44).
- Update cadence
- Weekly refresh via a scheduled backend job.
- Outputs
- Award counts, medians, IC and activity-code slices, disease-area filters, state and organization filters, and concept-similarity matches against funded abstracts.
- Roadmap (not yet live)
- PubMed, ClinicalTrials.gov, and USPTO expansions are on the roadmap and are not part of the current dataset.
- Limitations
- RePORTER only reflects awards NIH has published. Very recent awards and unpublished pending decisions are not represented.
Funding Directory
- Scope
- NIH R-series and SBIR/STTR funding opportunities relevant to for-profit biotech applicants.
- "Verified" definition
- Source URL, last-checked date, eligibility fields, deadline status, and update cadence per listing.
- Update cadence
- Every listed program carries a per-row verification date and source link; rows past their recheck window are flagged stale.
- Limitations
- Program officers change program rules faster than any directory can mirror. Always confirm current eligibility and deadlines with the NOFO before submitting.
Document Vault
- Purpose
- Keep the structured analyses you generate — scores, mock summary statements, resubmission plans, compliance checks — so you can reopen and re-export them. The Vault stores these records, not your original files; we operate no file storage bucket.
- Model-training policy
- Your content is used only to produce your analysis. It is not used to train any model.
- Encryption in transit
- TLS on every request.
- Encryption at rest
- Provided by our managed storage backend (Supabase). We have not implemented custom encryption of our own.
- Access
- Only the authenticated account owner can read a document. Row-level security enforces this at the database layer.
- Retention window
- Analysis results are auto-deleted 30 days after they are created, unless you save them to your Document Vault. A daily job performs the deletion; saved items are kept until you delete them.
- Deletion
- Every result page carries a Save to Vault control; leave it unsaved and a daily job hard-deletes the record on day 30. Manual deletion requests are confirmed within 72 hours.
Resubmission Lab
- Inputs
- The reviewer summary statement returned by NIH (PDF or pasted text) plus, optionally, the prior application's Specific Aims and Research Strategy.
- Engine type
- LLM analysis over the summary statement, prompted to distinguish reviewer-identified problems from applicant-side interpretation and to preserve reviewer language verbatim in quotations.
- Outputs
- Severity-tagged critique cards (FATAL, MAJOR, MINOR, plus reviewer CONSENSUS callouts), a suggested Introduction-to-Resubmission outline, and a typeset PDF export that keeps quotes and severity labels intact for print or grayscale review.
- Rubric version
- v2.1 — matches the two-track scoring rubric described below; verdict language differs for R-series vs. SBIR/STTR because NIH does not percentile SBIR/STTR.
- Model exposure
- The summary statement text is sent to the AI gateway only to produce your analysis; nothing is used to train any model. Your PDF is parsed in your browser and never reaches us. The resulting analysis follows the 30-day auto-delete policy unless you save it to your Vault.
- Limitations
- The Lab reports what reviewers wrote; it does not predict the resubmission's next score. Treat the FATAL/MAJOR/MINOR labels as prioritization, not adjudication.
The scoring rubric that Submission Match uses is published in full below, unchanged.
For due dates on or after January 25, 2025, NIH scores three factors: Factor 1 (Importance of the Research = Significance + Innovation, scored 1–9), Factor 2 (Rigor and Feasibility = Approach, scored 1–9), and Factor 3 (Expertise and Resources = Investigators + Environment, rated sufficient or not sufficient — not numerically scored). Reviewers then assign a single holistic Overall Impact score (1–9) informed by all three factors; Factor 1 effectively sets the ceiling for the best possible Overall Impact, and Factors 2 and 3 reinforce or worsen it. Our rubric reflects this shape without claiming to reproduce a reviewer's holistic judgment.
R-series rubric
For academic PIs applying to R01, R21, R03, or parent-R mechanisms. There is no commercialization pillar — reviewers do not score it.
Importance of research
Model-scored. Maps to NIH Factor 1 (Significance + Innovation).
| Sub-score | Detail | Weight |
|---|---|---|
| Unmet need | Is the disease burden / gap real and clearly established? | 10 |
| Innovation | Meaningfully novel vs. already-funded work. | 10 |
| Significance framing | The 'so what' is stated in reviewer terms, not lab terms. | 15 |
Rigor & feasibility
Model-scored. Maps to NIH Factor 2 (Approach).
| Sub-score | Detail | Weight |
|---|---|---|
| Approach | Aims are testable with named methods and pitfalls. | 10 |
| Preliminary data | Missing prelim data for a key aim is fatal for an R01. | 15 |
| Rigor | SABV, power, authentication of key resources. | 10 |
Expertise & resources
Model-scored, sufficiency-oriented. Maps to NIH Factor 3, which NIH rates 'sufficient / not sufficient' rather than scoring 1–9.
| Sub-score | Detail | Weight |
|---|---|---|
| PI track record | Prior awards, publications in-topic, ESI status handled honestly. | 8 |
| Team & environment | Every critical role named; facilities matched to specific aims. | 8 |
| Prior NIH | Prior funded NIH awards where relevant. | 4 |
Submission readiness
Deterministic — computed from user-confirmed booleans, not model-scored. Includes the DMS-plan gate: if the plan is not confirmed on an R-series application, overall is capped at 74.
| Sub-score | Detail | Weight |
|---|---|---|
| DMS plan present | Data Management & Sharing Plan confirmed complete. | 5 |
| Registrations | SAM.gov / UEI and eRA Commons live. | 3 |
| Docs ready | Biosketch, facilities, budget templates in hand. | 2 |
SBIR / STTR rubric
For small businesses applying to R41, R43, R44, Fast-Track, or Direct-to-Phase-II. Reviewers explicitly score commercialization, and eligibility/work-split is a compliance gate.
Scientific merit
Model-scored. Significance + Innovation for a small-business context.
| Sub-score | Detail | Weight |
|---|---|---|
| Unmet need | Real, established burden. | 7 |
| Innovation | Meaningfully differentiated approach. | 8 |
| Significance | Reviewer-facing 'so what'. | 10 |
Approach · rigor & feasibility
Model-scored. Can this be de-risked within the phase's scope?
| Sub-score | Detail | Weight |
|---|---|---|
| Approach | Testable aims, pitfalls named. | 8 |
| Preliminary data | Supports the core hypothesis. | 8 |
| Feasibility | Realistic for the phase and budget. | 4 |
Team & environment
Model-scored. STTR requires a qualifying research-institution partner; SBIR rewards credible advisors in lieu of one.
| Sub-score | Detail | Weight |
|---|---|---|
| PI track record | Record relevant to this mechanism. | 6 |
| Disciplines | All required disciplines covered on the team. | 5 |
| STTR partner / advisors | STTR: qualifying partner. SBIR: SAB credibility. | 4 |
Commercialization & impact
Model-scored. Hand-wavy commercialization is the #1 avoidable reason SBIRs miss the payline. If this pillar scores very low, the SBIR cap fires and overall is capped at 64.
| Sub-score | Detail | Weight |
|---|---|---|
| Market size | Credibly defined bottom-up TAM. | 8 |
| IP / freedom to operate | Filed IP and named FTO analysis. | 7 |
| Letters of interest | Payers, partners, or investors. | 5 |
| Regulatory | Named pathway (510(k), De Novo, 505(b), orphan). | 5 |
Eligibility & work-split
Deterministic — computed from user-confirmed booleans. SBIR: ≥ 2/3 of Phase I work at the small business; PI primary employment with the small business. STTR: 40% small business / 30% research institution.
| Sub-score | Detail | Weight |
|---|---|---|
| Work-split confirmed | User has confirmed the SBIR/STTR split rule for this mechanism. | 3 |
| PI primary employment | SBIR PI primary employment with the small business. | 2 |
Submission readiness
Deterministic — SAM.gov/UEI, eRA Commons, doc-package readiness, prior-submission experience.
| Sub-score | Detail | Weight |
|---|---|---|
| SAM.gov / UEI | Registration live and current. | 3 |
| eRA Commons | Account live. | 2 |
| Docs ready | Biosketch, budget, letters in hand. | 3 |
| Prior experience | Prior NIH submissions. | 2 |
Cross-cutting rules
- Mechanism-fit flag. The model returns a fit flag alongside the pillar scores. A clear scope/mechanism mismatch (for example, an R01-scale discovery program submitted as an SBIR Phase I) caps overall at 25 (WRONG_FIT). A marginal fit caps overall at 55.
- R-series DMS-plan gate. On the R-series track, if the Data Management & Sharing Plan is not confirmed complete, overall is capped at 74 (GO_WITH_FIXES).
- SBIR commercialization cap. On the SBIR/STTR track, if the Commercialization & Impact pillar scores below 10 (out of 25), overall is capped at 64 (CONDITIONAL).
- Advisory, not a prediction. Under NIH's Simplified Review Framework, reviewers assign one holistic Overall Impact (1–9) informed by all three factors. Our 0–100 composite is a readiness signal for prioritizing revisions — it does not reproduce that judgment.
What the score doesn't do
- It doesn't predict a specific overall impact number. Study sections vary; we publish a band, not a point estimate.
- It doesn't read your science for novelty. Factor 1 is inherently judgment. We tell you whether reviewers will be able to tell what the novelty is — not whether they'll agree it's enough.
- It isn't re-weighted per applicant. Every applicant runs against the same published rubric.
Change log
- v2.1 · 09 JUL 2026 — Weights rebalanced after the Simplified Review Framework took effect; current weights are as published above.
- v2.0 · 15 JAN 2026 — Split Submission Readiness out of Team pillar into its own pillar.
- v1.3 · 02 OCT 2025 — Commercial Pathway added; SBIR/STTR-specific.