Submission readiness as a number, not a meeting
Ask five people whether a study is ready to file and you will often get five answers. The decision usually happens in a meeting, backed by a spreadsheet that someone maintained by hand, and it comes down to judgement. Judgement is valuable, but it does not reproduce. Hand the same package to two senior reviewers and they will disagree on a meaningful share of the items, because readiness has never been written down as something you can measure the same way twice.
The cost of getting that call wrong is concrete. A Refuse-to-File letter, a run of Information Requests, or a review cycle that slips because a gap surfaced late. Each of those is expensive, and each traces back to a readiness decision that no one could check before the package went out the door.
R4SUB, our open source toolkit, exists to make that decision a number you can reproduce. It is MIT licensed, on CRAN, and developed in the open at the R4SUB organisation on GitHub. This post is about the model underneath it, not the code, because the model is the part worth agreeing on first.
From evidence to a single index
The chain has four links. Evidence, indicators, pillars, index.
Evidence is the raw material, gathered from the artefacts you already produce: Pinnacle 21 conformance output, the Define-XML file, the ADaM and SDTM datasets, the Analysis Data Reviewer's Guide. Each finding becomes a row in one machine-readable table with a fixed shape, so a check on labels and a check on traceability sit side by side in the same structure.
Indicators are the questions you ask of that evidence. Does every derived variable carry a method a reviewer could follow. Do the distinct values in a column match the codelist declared for it. Is the reviewer's guide present and complete. Each indicator reads the evidence and returns a score, so a vague worry becomes a specific, named measurement.
Pillars group the indicators into the four things a reviewer actually cares about: Quality, Traceability, Risk, and Usability. A package can be strong on one and weak on another, and the pillar scores say which, rather than averaging the distinction away.
The index is a single composite from 0 to 100, the Submission Confidence Index. It is calibrated to a specific regulatory authority, because readiness for an FDA filing is not the same question as readiness for an EMA one, and a single fixed number would quietly mislead a multi-region programme.
The bands are the part people quote
A number on its own invites arguing about the number. The value is in the decision it drives, so the index maps to four bands that everyone can agree on once, in advance:
85 to 100, ready to file. 70 to 84, minor gaps to close. 50 to 69, conditional, with real work remaining. 0 to 49, high risk, not ready.
Agreeing the bands before you see the score is what stops the conversation from becoming a negotiation about whether 68 is really good enough this time. The band is decided; the work is to move the score into the band you need.
Why a reproducible score beats a good spreadsheet
A spreadsheet can hold the same judgement, and a careful one often does. What it cannot do is run again on the next dry run and tell you what changed, or show a reviewer the exact evidence row behind a low pillar, or produce the same answer when a different analyst opens it. A reproducible score is auditable in a way a spreadsheet is not, because the path from a raw conformance finding to the final number is code that anyone can rerun and inspect.
That auditability is also what makes the score improve behaviour rather than just describe it. When the index drills down to the evidence, a low Traceability pillar is not a complaint, it is a list of variables to fix, and the same run that graded the package tells you where to spend the next day.
Where it sits in the ecosystem
R4SUB does not rebuild what already works. The pharmaverse and Pinnacle 21 are strong at assembling a submission; R4SUB reads their output and scores whether the assembled package is ready. It sits alongside the R Consortium submission pilots, covered in what the R submission pilots mean for sponsors, and it leans on the practices we have written about before: treating Define-XML as the dataset's contract, keeping traceability from CRF to table, and running it all inside a validated computing environment.
None of the four bands is a substitute for reading the study. The point is narrower and more useful than that. It is to make the readiness call the same call every time, backed by evidence anyone can open, so the meeting is about closing the last gaps rather than about whether the gaps exist.
The R4SUB packages
R4SUB is open source and MIT licensed. The Submission Confidence Index is implemented in r4subscore (on CRAN), and the whole ecosystem installs with the r4sub meta-package (CRAN, r-universe).