# Drybench Pipeline: methods and checks

How Drybench Pipeline reaches its results: every calculator, analysis or view with its methods, the references behind them, how far each has been checked, and the app's limits.

[Open Drybench Pipeline](https://www.drybench.com/pipeline/app) · [What the levels mean](https://www.drybench.com/methods/levels) · [Every app's methods and checks](https://www.drybench.com/methods)

## Landscape

The registered trials for a condition, counted by phase and status as ClinicalTrials.gov records them, by lead sponsor merged by company, and by a modality label the app derives from intervention names.

No method record is kept for this view, so it carries no level.

### How it is made and tested

- **Trials per search:** 1,000 trials per page, newest start date first, at most 10 pages; the page always says how many of the matching trials it analysed.
- **Study scope:** Interventional studies with a drug, biologic or genetic intervention.
- **Modality naming stems:** 151 International Nonproprietary Name stems, checked against the WHO stem book; names that match nothing are shown as unclassified.
- **Modality labels:** Modality labels come from International Nonproprietary Name stems. How often a label is right has not been measured.
- The modality classifier: antibodies under old and new naming schemes, antibody-drug conjugates, small molecules by stem, peptides, nucleic-acid drugs, cell and gene therapies, vaccines, placebos and diagnostic agents, names that stems misread, and a well-formed rule table.
- Build-time check: all 151 stems checked against the WHO stem book.

## Trials

Each registered trial with its phase, status, dates, sponsor and why it stopped, as its sponsor entered them.

No method record is kept for this view, so it carries no level.

### How it is made and tested

- The ClinicalTrials.gov query: filters combined into one request, paging tokens, cleaned condition text, partial dates, phase buckets and the parsing of a page of studies.
- Stop reasons grouped by the first matching rule, with mentions that are not the reason and denials left out.
- Filter checks with the failing input named, and filters that round-trip through the address with bad values dropped.

## Charts

Trial counts by phase and start period; durations from completed trials with actual start and primary completion dates only.

No method record is kept for this view, so it carries no level.

### How it is made and tested

- Quantiles, status groups, durations from completed trials with actual dates only, the breakdowns by phase, period, sponsor and modality, and tracer-only trials set apart.

## US approvals

US FDA approvals whose labels mention the condition, matched on label wording.

No method record is kept for this view, so it carries no level.

### How it is made and tested

- openFDA label searches (phrases, possessives, accents and Greek letters), application counts, Drugs@FDA and NDC fallbacks, label wording that rules a condition out, and parent names without salt words.

## FDA precedent

US products whose label names the condition, with how FDA approved each one's original application: pathway, designations, review time and any published complete response letter. Each value is cited to the FDA record.

Its result goes through 1 method. The lowest level among them: Tested, not reference-checked.

Below the top level: Review time (Tested, not reference-checked).

### Methods

#### Review time

- **Level:** Tested, not reference-checked
- **Reference:** FDA, Compilation of CDER New Molecular Entity (NME) Drug and New Biologic Approvals, Data Dictionary: FDA Receipt Date, the date the first and complete marketing application was received by FDA; FDA Approval Date, the date FDA originally approved the application.
- **How it was checked:** Against 50 products transcribed from Drugs@FDA approval letters in the build session; not independently checked (Q-19).

### How it is made and tested

- **Join:** By application number only (method appno-exact 1.0.0): a product joins FDA's Compilation of CDER NME and New Biologic Approvals when one of its applications equals one of a row's, or one the row's own note names (products moved from NDA to BLA on 23 March 2020 keep their digits; some older biologics were given a new licence number). Nothing is joined by name. FDA's breakthrough therapy lists carry application and submission numbers, so they join exactly too.
- **Review time:** Method review-time-receipt-approval 1.0.0: FDA Approval Date minus FDA Receipt Date, in days, both from the compilation, whose data dictionary defines receipt as the date the first and complete marketing application was received. It includes any complete-response cycle and the sponsor's time between cycles. Months are days divided by 30.4375. Where the compilation holds no receipt date the view says “Receipt date not in a structured public source” and why; no review time is ever estimated, from a PDUFA goal or otherwise.
- **Designations:** As the compilation records them at the original approval, qualifiers kept (“Yes (indication \[A\] only)”). “N/A” reads “Programme did not exist” only for approvals before the programme began: accelerated approval 15 April 1992, fast track 21 November 1997, breakthrough therapy 9 July 2012, priority review vouchers 27 September 2007 (the data dictionary). For products the compilation does not cover, breakthrough therapy comes from FDA's lists and priority review from Drugs@FDA; a product absent from the lists is “not on CDER's list as of” the lists' date, never “No”. Where the compilation and the lists disagree the cell says so.
- **Fixture sets:** Status: draft. Expected values transcribed from FDA documents in the build session, 5 October 2026; not independently checked (Q-19). FDA precedent's release gate is not met until a reviewer with regulatory-affairs experience signs these fixtures.
- **Review time and designations: products checked against their approval letters:** The join by application number, the receipt and approval dates, the review time in days and every designation flag's reading, for a seeded sample of FDA's compilation stratified by oncology, NDA or BLA and approval before or after 9 July 2012. Expected values from: Each product's Drugs@FDA approval letter (its receipt sentence quoted with the page and the PDF's SHA-256), Drugs@FDA's approval date and review priority, and FDA's breakthrough lists. Tolerance: Dates and days exact. Where an approval letter and the compilation differ, the difference is recorded and the app keeps the compilation's date, which the data dictionary defines; an unrecorded difference fails the test.
- **Complete response letters joined to the later approval:** Whether each letter joins to the later approval of the same application, and the interval in days; at least five negative cases (a supplement, another letter type, an application never approved). Expected values from: The application number printed on the letter's first page (or FDA's file name for a scanned letter), the date in FDA's file name, and the original approval in Drugs@FDA. Tolerance: Exact.
- **Trials named in a label's Clinical Studies section:** The NCT numbers read from openFDA's section 14 text, in order, against those in the label's PDF. Expected values from: DailyMed's PDF of the same label set id and version as openFDA's record. Tolerance: Exact, including labels that name none.
- **The seven example searches:** For every US product of each example search: the compilation row it joins to, its pathway, and its review time or the reason it has none. Expected values from: A second implementation of the rules, in Python, reading FDA's compilation spreadsheet rather than the app's snapshot. Tolerance: Exact.
- Every FDA precedent snapshot carries its source, licence id, read date, retrieval date and each file's SHA-256, and none is within 14 days of turning stale; the example snapshots are checked the same way.
- Weekly once its workflow is merged: the openFDA fields FDA precedent reads, the fda.gov table headers and the request count of the seven examples' US approvals.

## What else is checked

- **Saved snapshots:** Dated snapshots of the seven example searches, shown only while they are 60 days old or less.
- Snapshots older than the limit are not shown, and shipped snapshots carry their terms.
- Fails on any mention of the removed industry success-rate tables outside docs/.

## Data sources

- [ClinicalTrials.gov API v2](https://clinicaltrials.gov/data-api/api): ClinicalTrials.gov Terms and Conditions: no charge; attribute the source, keep the data current, show the processing date and state modifications (the data carry copyright outside the US)
- [openFDA drug label, Drugs@FDA and NDC endpoints](https://open.fda.gov/license/): CC0 1.0
- [WHO: Use of stems in the selection of INN for pharmaceutical substances, 2024](https://www.who.int/publications/i/item/9789240099388): CC BY-NC-SA 3.0 IGO (reference only)
- [FDA: Compilation of CDER NME and New Biologic Approvals 1985-2025, with its data dictionary](https://www.fda.gov/drugs/drug-approvals-and-databases/compilation-cder-new-molecular-entity-nme-drug-and-new-biologic-approvals): US government work, public domain (FDA website policies)
- [FDA: CDER Breakthrough Therapy Approvals (current calendar year and cumulative earlier years)](https://www.fda.gov/drugs/nda-and-bla-approvals/breakthrough-therapy-approvals): US government work, public domain (FDA website policies)
- [openFDA Complete Response Letters (transparency/crl)](https://open.fda.gov/apis/transparency/completeresponseletters/): CC0 1.0
- [FDA Oncology Center of Excellence: Project Confirm](https://www.fda.gov/about-fda/oncology-center-excellence/project-confirm): US government work, public domain (FDA website policies)
- [FDA: Table of Surrogate Endpoints That Were the Basis of Drug Approval or Licensure](https://www.fda.gov/drugs/development-resources/table-surrogate-endpoints-were-basis-drug-approval-or-licensure): US government work, public domain (FDA website policies)

## Limits

> Not medical or investment advice. Nothing here is advice about any medicine, treatment or condition, and nothing here is an assessment of any person's variant, sample or molecule. Decisions made with it are yours.

> Not regulatory or legal advice. This describes FDA records as they stand. It is not advice on any regulatory strategy, submission, designation or interaction with any agency, and it does not forecast any regulatory outcome.

> Not a forecast. Stages, scores, rates, rankings and similarities shown here describe records as they stand, or historical averages; they are not a forecast for any molecule, trial, drug, company or target.

- Not medical or investment advice, and not regulatory or legal advice. The FDA precedent describes FDA records as they stand and is not a forecast for any trial, drug or company.
- Registered trials are not a complete commercial asset inventory: unregistered and preclinical programmes, confidential assets, some international programmes and commercial context are absent. Trial counts are not unique drug or programme counts.
- The patient-group check recognises two forms of wording (efficacy 'established in a population with' a condition, and cardiovascular risk reduction 'in adults with' it). The cardiovascular form is applied to type 2 diabetes only. Other wording is not recognised, and those products are still listed.
- FDA precedent covers CDER new molecular entities and new biologics only; CBER products, such as cell and gene therapies and vaccines, are not in FDA's compilation. It describes each product's original application at approval, which may have been for another condition.
- Review time is receipt of the first complete application to approval and includes any complete-response cycle and the sponsor's time between cycles. It is computed only where FDA's annual compilation holds both dates (approvals to the end of 2025); for later approvals the receipt date is not in a structured public source, and no review time is estimated.
- Designations are shown as FDA recorded them at the original approval. “Programme did not exist” marks an approval before the programme began; orphan designation is a designation, not orphan approval or exclusivity. Breakthrough therapy lists are CDER's own.
- Complete response letters are those FDA has published; absence is not evidence none was issued. Only letters on the original application that came before its approval are shown, each beside that approval.
- Complete response letters are shown as FDA published them, by application, company, date and type, with a link to the letter. We do not summarise, classify or explain them.
- The trials a label names in its Clinical Studies section are not necessarily the trials that supported approval, and many labels name none.
- The precedent figures describe the listed products, with how many each covers; a median review time is a historical figure, not a forecast. Modality tags are automatic, and the share left unclassified is shown.
- Trial records are entered by sponsors. Statuses can lag, dates can be estimates, and the condition search uses ClinicalTrials.gov's own matching, which can include trials where the condition is not the main focus.
- Searches with more than 10,000 matching trials analyse the 10,000 with the latest start dates, and the tool says so.
- Modalities come from drug names. Many interventions have no stem (company codes, older drugs), so they stay unclassified, and a stem describes how a name was built, not how every product is made. ClinicalTrials.gov registers radiotracers and imaging agents as drugs; the tool sets them apart as diagnostics but can miss tracers named only by a company code.
- Durations use completed trials with actual start and primary completion dates only, so trials that stopped early are not in them. They measure single trials, not the years a drug spends in development.
- Approvals are US FDA only and are matched on label wording, so they can include passing mentions and miss differently worded labels. openFDA links some current labels to no application: the example searches trace those through the NDC codes on the label, other searches only count them. A label is flagged when every mention of the condition in it rules the condition out; every label is checked for the examples, and up to 12 applications for other searches. A discontinued product's label can stay in openFDA. Each date is labelled by its source: the original approval of the product's application where Drugs@FDA records one, which can predate the approval for this condition, otherwise the NDC marketing start, which is not an approval date.
- The saved snapshots cover the seven example searches with default filters. They were saved on 28 September 2026 (ClinicalTrials.gov data of 25 September 2026), keep only the 100 most recently started trials for the list, and are not shown once they are more than 60 days old.
- FDA's files behind the precedent were saved on 5 October 2026 and are marked as stale once they are more than 90 days old; FDA updates the compilation once a year and openFDA's letters infrequently.
- Dates are labelled by source: original approval (Drugs@FDA), NDC marketing start (not an approval date) or type not recorded. An original approval can predate approval for this condition.