Skip to main content

Limits. Reports. Deviations. Changes. CMC knowledge. Why every number is what it is, and where the evidence stops.

In Seal a limit keeps the report it rests on, at its version. Ask every limit one question, and Neil, Seal's AI, says where one runs past it.

A saved view of ten recipe steps for mAb-24 in Seal, each with its limit and the study report it cites at its version, and a Neil column asking "Does the report support the whole limit?" Step 3.1's answer is open, Neil's reading of v6: Partly. The report supports 26 h, not the full 36 h stated limit. It quotes EXP-332's report, p. 8, the highest bioburden count, 5 CFU/100 mL at 26 h, and p. 11, the study supports a hold of up to 26 h. Of the other nine, four say Yes, three Partly, one No and one Cannot tell.

The limit says 36 h. The study stops at 26.

Step 3.1 took its 36 h from a client's process description. EXP-332 studied 6 to 26 h. Past 26 h, filtration waits for QA.

MB-24 step 3.1 in Seal at v6, hold the formulated bulk. Limit: 36 h at 2 to 8 °C, which came from mAb-24 process description v2, pinned at v1. Studied: 6 to 26 h, in EXP-332, pinned at v5. Past the study: past 26 h, filtration waits for QA, a rule from DEV-041, bulk held 29 h 40 min, past the 26 h studied.

Neil quotes the page it read.

EXP-332 concluded that a hold past 26 h needs further data. The step went on allowing 36 h. This is the page, at its version.

EXP-332's report, protocol v1, open in Seal at its conclusion: the study supports holding formulated mAb-24 bulk at 2 to 8 °C for up to 26 h before sterile filtration. It does not support a longer hold; a hold past 26 h, such as the 36 h in the process description, needs further data.

What the plant learns goes back on the step.

DEV-041 put QA on any hold past 26 h. CC-118 then cut the limit to 26 h, and Neil's answer shows the version it read.

The same column after CC-118 changed step 3.1's limit to 26 h at 2 to 8 °C, to 36 h with QA approval: its answer is marked Changed, and opened it reads Neil's reading of v6, which has changed since, quoting the limit as it was, 36 h at 2 to 8 °C.
  1. ChangedNeil read v6. CC-118 made the step v7, at 26 h.

Ten limits. Four rest fully on their report.

Four are Partly: step 2.1 allows pH 5.8 to 6.2, and its study chose pH 6.0. Step 4.6 rests on a protocol, so Neil says No.

The same column with step 2.1's answer open: Partly. The report supports the selected pH 6.0, but not the entire 5.8 to 6.2 range. It quotes EXP-287's report, p. 4: 20 mM histidine, pH 6.0, with 8% sucrose and 0.02% polysorbate 80, is selected for mAb-24 drug product.

Where Neil stops.

What it reads, what it can't, and what stays with your people.

What does Neil read?
What you can read. It reads with the access of the person asking, so two people can get different answers.
Where are the exact words?
In a Neil column. Open an answer: the file and page, in its own words. A chat links records and versions. Seal does not check those links.
Can it read a scanned report?
Not in a column. A column reads text, so a scan comes back Cannot tell. What is on paper, Neil hasn't read.
Will it answer the same way twice?
Not always. Neil's answers can differ between asks. A column shows its rule and the words it read, so a person can check each one.
Can our client or CDMO ask it?
Not across companies. Each company's records are its own. The other side gets what your people sign and send.
What does it decide?
Nothing in GMP work. Neil drafts, finds and proposes. People decide, sign and send.

Questions about CMC knowledge.

Do our documents have to move into Seal?

No. A file can stay in SharePoint, Box, Google Drive or Veeva Vault and be captured into Seal at its version, so a record links the version it rests on. Reports from before Seal can also come in attached to the records they belong to.

Does Neil read our old development reports?

Yes, where they have text: PDF, Word, Excel and CSV files. A Neil column reads text, so a scanned page without text comes back Cannot tell. In a chat Neil reviews a scan's pages as images, so check what it reports against the page. What is only on paper, Neil hasn't read.

Is this a knowledge graph?

No. It is your records and the references between them: a limit to its study, a result to its method version, a lot to the recipe version it was made to. Neil reads those records and follows their references.

How do we check an answer?

Open what it links. Neil is instructed to link each record and version it read; Seal does not check a chat's links, so open them. A Neil column shows the field, or the file and page, with the exact words, and marks an answer Changed when its record has changed since. A person confirms anything that matters.

Does Neil write our Module 3?

It drafts sections, tables and responses from the records, with their sources, for your people to edit, sign and send as Word or PDF. Dossier assembly, publishing and submission stay in your regulatory tools. See the guide.

We outsource manufacturing. Does this help us?

Yes, for what you hold: your studies, specifications, methods and decisions as records, and your CDMO's reports captured at their versions from where they arrive. Neil answers from what you hold; it can't read your CDMO's own systems.

How does this fit ICH Q10?

ICH Q10 names knowledge management as one of the quality system's two enablers, and says: “Product and process knowledge should be managed from development through the commercial life of the product up to and including product discontinuation.” In Seal that knowledge is the record the work is done on, kept with its versions.

Is Neil trained on our records?

No. Under Seal's agreements your data isn't used to train the underlying models. Notes Neil keeps for the organisation are saved only after an admin reviews them.

In depth

The CMC knowledge guide7 sections, with figures and references

All 8 capabilities
Limits that reference their evidence
A recipe step's limit references the study and records it rests on, pinned to the version it was set on. Anyone with access opens the same fields, and sees what references them.
Answers that link their records
Neil answers CMC questions from the records the person asking can read, and links the records and versions it read, for you to open. What they don't show, it says.
One question, asked of every row
A Neil column asks one question of every study, lot or deviation in a view, and of the text of its files, by a rule anyone can read. Each answer opens to the field, or the file and page, in its own words.
Files from before Seal
Reports come in attached to the records they belong to, or are captured from SharePoint, Box, Google Drive or Veeva Vault at their version. Neil reads their text; a Neil column can't read a scan.
Module 3 sections, drafted
Neil drafts a section, a table or a response from the records, with their sources, for your people to edit, sign and send. Publishing stays in your own tools.
Versions and their history
Recipes, specifications, methods and studies are versioned. A change goes through a changeset, a change to signed content keeps its reason, and every lot keeps the versions it was made to.
Stability studies
The protocol, each physical pull and the original result, together for the life of a stability study. Neil prepares the next timepoint’s work.
Technology transfer
The receiving site starts from a copy of the configured process, not a document package. Neil prepares the gap assessment and the local changes.

See your process in Seal.

Working 48 hours after we agree its scope: tested, with evidence your QA approves.