CMC Module 3 authoring that checks every limit against your own data before the dossier is assembled
Module 3 drafted in about half the time, every number checked
See one case, screen by screen ↓Software that helps regulatory CMC teams write the quality module of a drug application — the CTD 3.2.S and 3.2.P sections — from development reports, certificates, specifications and stability data. CMC Module 3 Authoring adds six agents that read the sources, draft each section in your house template, check every number and cross-check every limit across sections, for the regulatory CMC lead to decide and sign.
How to choose AI solutions for life sciences →
Updated
Module 3 is written section by section, by different people, from hundreds of reports. The writer retypes values from certificates of analysis and stability tables into Word, and reviewers reconcile them over several rounds. Yet the sections must agree: a limit in 3.2.P.5.1 has to be met by the batches in 3.2.P.5.4, by every stability result in 3.2.P.8.3 up to the shelf life claimed in 3.2.P.8.1, be justified in 3.2.P.5.6 and be quoted the same in Module 2.3. Reviewers read one section at a time, and nobody owns the view across them.
The risk is the late change. A change control narrows a range from release capability, and the stability sections are rarely re-read against the new limit. A process description gives the normal operating range in 3.2.S.2.2 and the proven acceptable range in 3.2.S.2.4 without saying which, so a PPQ batch inside one and outside the other reads as a validation batch run outside its own process. The agency reviewer finds it months later, as an information request or a complete response.
Estimated split for one Module 3 section, by hand and with the solution; the totals are the catalog's design target of about 13.5 h by hand and 5.8 h with the agents.
Agents read the sources, draft each section, check every number and cross-check every limit across sections; the regulatory CMC lead decides.
Northwind Bio is filing a BLA for Veltrimab, a 150 mg/mL solution for injection in a 1.0 mL prefilled syringe. Content lock is Oct 30, 23 days away. On Oct 5 the 24-month interim stability report is released. Here is what happens next, screen by screen, in the working solution.
Maya's screen is the Module 3 cross-check for the Veltrimab BLA: 33 rows — 13 drug substance limits, 14 drug product limits and 6 critical process parameters — against batch analyses, stability, process, justification and Module 2.3. The last run, Oct 6 at 17:40, gave 136 checks: 127 consistent, 8 to look at, 1 contradiction. A banner says stability report STB-DP-031 adds the 18- and 24-month results for three drug product batches.
“The grid below still shows the 12-month data.”
The re-run tests every limit against every stability time point up to the proposed shelf life. Now 125 checks are consistent, 9 to look at and 2 contradict. Specification SPC-VLT-DP-004 version 4, effective Sep 29, 2026, sets protein content at 95.0 – 105.0% of label claim at release and throughout the shelf life. Batches VLT-DP-24S01 and 24S02 read 94.6% at 18 months, while 3.2.P.8.1 proposes 24 months — and 3.2.P.5.6 justifies the limit from release data only.
“Cross-check done — 2 contradictions, 9 to look at. Protein content fails at 18 months.”
The checker proposes two options and decides neither. A: release 95.0 – 105.0%, shelf life 93.0 – 107.0% — the version 3 range, met by every stability result — as ICH Q6B allows when justified, keeping 24 months. B: keep 95.0 – 105.0% and propose 12 months. Maya picks A and sends it to Sam Patel, the specification owner, and Dana Okafor in QA as an amendment to change control CC-2026-118.
Her note: “The 2024 registration batches started at 96% of label claim and fall about 0.1% per month. Every result to 24 months is within the version 3 range.”
The row now reads “Shelf life 93.0 – 107.0% proposed”, waiting for Sam and Dana. The stability chart at 5 °C ± 3 °C draws all three batches from STB-DP-031 against the version 4 limits: 24S01 and 24S02 cross 95.0% between 12 and 18 months; 25A02 stays above 99%. Nothing in the dossier changes until they approve.
“Option A sent to Sam Patel and Dana Okafor at Oct 7, 09:14. The Change tracker drafts the updates when they approve.”
Sam and Dana approve amendment 1 to CC-2026-118, and specification version 5 is drafted. The Change tracker drafts the update in 3.2.P.5.1, 3.2.P.5.6, 3.2.P.8.1 and 2.3 as tracked changes — each still goes through the number check. Protein content reads consistent, “≥ 94.2% · within shelf-life limit”, and the grid is back to 127 consistent, 8 to look at, 1 contradiction.
“Resolved — option A.”
Anion-exchange load conductivity: 3.2.S.2.2 says 6.0 – 8.0 mS/cm, 3.2.S.2.4 says 5.0 – 8.0 mS/cm. Both are right — one is the normal operating range, the other the proven acceptable range in PCR-0412 Table 7-2 — but neither says which. PPQ 2, batch VLT-DS-25012, ran at 5.6 mS/cm. The proposed fix states both ranges in 3.2.S.2.2; Omar Haddad reviews it.
“Read as written, a validation batch ran outside its own process.”
Every cell cites its sources at their pages. A click on PPQ-VLT-DS-01 opens the validation report at page 22, with the passage highlighted: load conductivity 7.1 mS/cm for 25011, 5.6 mS/cm for 25012 and 7.4 mS/cm for 25013 — the 25012 value inside the proven acceptable range of 5.0 – 8.0 mS/cm.
In the batch analyses section, drafted from the approved Corventa text in the Northwind house template v5, the Number checker flags two numbers. The text says monomer ranged from 98.6%, but the lowest result is 98.3% for VLT-DP-26P02 — 98.6% came from the pasted Corventa text. And pH for VLT-DP-26P03 shows the raw 6.15 from the LIMS export where the certificate reports 6.2. Each has a one-click fix that Priya applies; until then the section cannot be signed.
The Number checker proposes the fix; only a person applies it.
The dossier shows the Veltrimab BLA at 76% — 34 of 45 sections approved, content lock in 23 days — beside the MAA at 31% and the Corventa new-site supplement at 8%. The map shows P.5 with one section to fix and one changed, and 2.3 flagged as changed. Maya sees which sections still stand between the filing and Oct 30.
The same filing, seen by the five people who carry it — the lead who decides, the writer, the specification owner, the process lead and QA.
Agents read the sources, draft each section, check every number and cross-check every limit across sections; the regulatory CMC lead decides.
Reads development reports, certificates, specifications and prior filings; maps each to its CTD section and records version and effective date.
Extracts batch results, specification limits and stability results, each value with its page and confidence.
Writes each Module 3 section in the house template, starting from the closest approved prior text and filling tables from the extracted data.
Checks every number in a draft against its source: value, unit, decimals, source version, current specification and the same value elsewhere.
Puts every specification limit and process parameter on one grid and checks it against batch analyses, stability data to the proposed shelf life, the process description, the justification and Module 2.3 — and proposes options for each contradiction.
Watches sources for new versions and new data, finds every place in every filing that uses a changed value and drafts the update as a tracked change.
Chooses the option; signs each section. The agents propose; a named person decides.
Ask in plain words — or say what to cross-check, draft or update. Every answer cites its source.
Which limits does our own data contradict?
Two. Protein content (assay): 3.2.P.5.1 sets 95.0 – 105.0% at release and on stability, but registration batches VLT-DP-24S01 and 24S02 read 94.6% at 18 months — while 3.2.P.8.1 proposes 24 months. And anion-exchange load conductivity: PPQ 2 ran at 5.6 mS/cm, outside the 6.0 – 8.0 mS/cm in 3.2.S.2.2, though inside the proven acceptable range. The options are in the grid; you decide.
Why did the check flag the batch analysis section?
Two numbers in 3.2.P.5.4 do not match their sources. The text says monomer ranged from 98.6%, but the lowest result is 98.3% for VLT-DP-26P02 — 98.6% was pasted from the approved Corventa text. And pH for VLT-DP-26P03 shows the raw 6.15 from the LIMS export, while the certificate reports 6.2. Each finding has a one-click fix.
What changed in the pH specification?
Version 4 of SPC-VLT-DP-004 narrowed pH from 5.8 – 6.4 to 5.9 – 6.3, approved under change control CC-2026-118 because 31 batches ranged 5.95 – 6.18 (Ppk 1.9). Three sections still quote the old limit.
Update every section that still quotes the old pH limit
Done. The Change tracker drafted the update in 3 sections across 2 filings: 3.2.P.5.6 and the overall summary (Veltrimab BLA) and 3.2.P.5.1 (Veltrimab MAA). Each is now Drafted and needs a number check before signing.
13 screens from the working solution, on its sample data. Pick one to see it large.
The template, the checks the agents run, who decides contradictions and who signs are settings.
| Setting | Default | Choose from |
|---|---|---|
| House template | Northwind Module 3 house template v5 | Northwind Module 3 house template v5 · v4 · EU variation template v2 |
| Reuse approved wording | On — start each section from the closest approved prior filing | On · Off |
| Show who wrote each paragraph | Off | On · Off |
| Block approval while numbers are open | On | On · Off |
| Test stability against the proposed shelf life | On — every time point up to the shelf life in 3.2.P.8.1 and 3.2.S.7.1 | On · Off |
| Re-run on every new stability report or specification version | On | On · Off |
| Decides contradictions | Dr. Maya Chen | Dr. Maya Chen · Sam Patel |
| Regulatory approval | Dr. Maya Chen — meaning “Approved for submission” | Dr. Maya Chen · Dana Okafor |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample filing · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Pharma Manufacturing analysis of FDA complete response letters 2020–2024: 150 of 202, 74%, cite quality or manufacturing). Context: ICH M4Q(R1) and the M4Q(R2) draft; ICH Q6B on release and shelf-life acceptance criteria; ICH Q1E on stability data. People, companies, products and batches named on this page are fictional — characters and sample data in the working solution.
Software that helps regulatory CMC teams write the quality module of a drug application — the CTD 3.2.S and 3.2.P sections — from development reports, certificates, specifications and stability data. CMC Module 3 Authoring adds six agents that read the sources, draft each section in your house template, check every number and cross-check every limit across sections, for the regulatory CMC lead to decide and sign.
Every specification limit in 3.2.S.4.1 and 3.2.P.5.1 and every critical process parameter in 3.2.S.2 is a row, tested against the batch analyses, stability up to the proposed shelf life, the process description, the justification and Module 2.3. Each cell is consistent, to look at or a contradiction, and cites both sources at their pages. In the sample BLA that is 136 checks across 33 rows.
The cross-check re-runs and new contradictions go to the person who decides. In the working solution, the 24-month report showed two registration batches at 94.6% at 18 months against a 95.0% lower limit, and the checker proposed separate release and shelf-life criteria or a 12-month shelf life.
No. The Consistency checker never changes a limit, a range or a shelf life. It proposes options with both sources; the regulatory CMC lead chooses, and a change to a limit is approved by the specification owner and QA through change control.
The Number checker compares every number with its source for value, unit, decimals, source version, the current specification and the same value elsewhere — including Module 2.3. It flags a range in the text that does not match the table, or a raw instrument value where the certificate reports another, and proposes the fix for a person to apply.
Yes. The Section drafter starts each section from the closest approved prior filing in your house template and replaces every product fact and number from the new sources, each cited. For the sample batch analyses section it reused 91% of the approved wording.
Yes. Agents draft, check and propose; people decide. Number fixes are applied by a person, updates arrive as tracked changes the author confirms, a section with open findings cannot be signed, and technical review comes before the regulatory approval signature.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your house template, specifications, a sample of development reports and a prior approved filing — and it goes live once every quality gate has passed. We will walk you through it on your own sections first.
Sourced answers from papers, patents and reports in seconds, our own disagreements first.
1-min demoCompetitive Pipeline TrackingEvery rival trial, readout and approval in a sourced brief each Monday
1-min demoPreclinical Study Report AutomationA draft GLP report in hours, every number traced to raw data.
1-min demoWe'll run CMC Module 3 Authoring on a sample of your own development reports and a past filing.