Startup Regulatory Operating Blueprint
What a startup has to stand up, in what order, and — the part most roadmaps omit — what it should deliberately not build yet. One configuration at a time, because the differences between them are the whole point.
The same obligations land differently depending on what you make and what you physically operate. A device startup’s regulated object is its design; a diagnostics laboratory’s is the laboratory as an operation; a virtual sponsor’s is its decision record; a CDMO-dependent sponsor’s is the release decision it takes over work it did not perform. Across the four configurations the blueprint sequences 40 capabilities and deliberate deferrals, and every one of them names both its minimum viable form and the over-build it invites — because startups fail this in both directions. Nothing here is an applicability determination or regulatory advice.
Six questions, then the closest blueprint
The regulated object follows from this more than from anything else.
Accountability follows the role, not the equipment.
Determines which obligations have already attached.
The single biggest driver of how much facility work applies.
The blueprint sequences toward this, not toward completeness.
Sets the honest ceiling on how much process can be operated.
Answer any question to jump to the closest configuration. Nothing here is an applicability determination.
Class II device startup entering first clinical use
Twenty-two people, one product, one leased assembly and test space. The design is nearly frozen, first clinical use is two quarters out, and the quality system exists as four procedures written for the last funding round. Nothing about the facility is qualified, and the design history file is the real deliverable.
Building a facility quality system before a design quality system. The gating evidence for first clinical use is design control and risk management, not room classification — but the facility is visible and the design file is not, so the facility gets the attention.
Applicability — what actually applies, and when
Each row is an applicability decision with a scope and a reason, not a checklist item. Scoping something out is a decision that gets recorded, dated and revisited — not a silent omission.
| Scope | Verdict | Why, for this configuration | Revisit when |
|---|---|---|---|
| Design controls and design history file | Applies now | The binding obligation for a device at this stage, and the one auditors open first. | Already in force |
| Risk management to ISO 14971 | Applies now | Feeds design controls, labelling and the clinical protocol simultaneously. | Already in force |
| Quality system for manufacture | Partly | Needed for assembly and inspection, not for the whole plant. Scope it to what is built in-house. | At design transfer |
| Sterilisation process validation | Scoped out | Performed by the contract steriliser under its own validated cycle; the startup qualifies the supplier, not the cycle. | If sterilisation is brought in-house |
| Cleanroom classification of assembly space | Applies later | Product is terminally sterilised and packaged externally; classification is not the control that protects it. | If the process becomes aseptic |
| Post-market surveillance and complaint handling | Applies later | No market product yet, but the procedure must exist before first commercial sale. | At 510(k) submission |
Build order
Sequence matters more than completeness. Each step names its minimum viable form and the over-build it invites, because startups fail this in both directions.
Establishes the record that everything else references, and forces the design freeze conversation early.
A design plan, an indexed design history file, and one review per phase with real attendance.
A documented gate for every engineering change, which stalls the last two quarters of iteration.
Connects hazard to control to verification, which is the argument a reviewer follows.
One hazard analysis maintained as a live document, not a submission artefact.
A separate FMEA per subsystem with no link back to the hazard analysis.
Sterilisation, moulding and packaging carry more product risk than anything done in-house.
Three qualified suppliers with agreements naming who owns which specification.
Auditing every supplier including the office stationery vendor.
Makes the units built for clinical use defensible as having been built to the frozen design.
Controlled work instructions, trained operators, recorded inspection at defined points.
Full MES deployment for a line building forty units a month.
The point where design becomes something repeatable rather than something remembered.
A device master record that matches what the floor actually does.
A validated electronic DMR system before the paper one is stable.
Labelling, traceability and complaint capture for units placed with investigators.
Unit-level traceability and a route for investigator feedback to reach the design file.
A full post-market surveillance system for a product not yet on the market.
Evidence spine
Every category names the system that holds it. SPEQ is not the system of record for any of them.
| Category | System of record | State |
|---|---|---|
| Design history file Indexed and current, not assembled the week before an audit. | Document management system | In review |
| Risk management file Live document; every design change reopens it. | Document management system | Approved |
| Verification and validation reports Bench and biocompatibility data against the frozen design. | Document management system | Draft |
| Supplier agreements and audit reports Sterilisation, moulding and packaging only. | Contract and vendor system | Approved |
| Assembly and inspection records Unit-level, retained per unit placed with an investigator. | Paper batch records, scanned | Approved |
| Training records Operators on the current work instruction revision only. | Learning management system | Approved |
Minimum role set
Roles, not people. What can be fractional, what can be a vendor, and what cannot be either.
Owns design control, risk file and the submission narrative. Cannot be a consultant.
Accountable for the frozen design and every change to it.
Owns work instructions, operator training and inspection points.
Submission strategy and formatting are episodic and specialised.
Held by the contract steriliser under its own validated process.
Independence matters more than availability at this size.
Milestone pathway
Five milestones, each a decision rather than a date. The last column is what newly attaches once the milestone passes, which is the part most plans omit.
Is the design stable enough to transfer?
Design history file current, risk file live, verification complete against the frozen design.
Change control over the design. Every later change now costs assessment.
Can someone else build this to the same specification?
Device master record matching the floor, trained operators, inspection points defined.
Manufacturing controls, operator qualification, and unit traceability.
May units be placed with investigators?
Unit-level traceability, labelling, investigator feedback route into the design file.
Complaint capture and the obligation to feed use back to the design.
Is the evidence set coherent and complete?
Submission dossier, supplier agreements, sterilisation validation from the contract steriliser.
Commitments made to the authority, which constrain operations afterwards.
Can supply be sustained under control?
Post-market surveillance procedure, distribution controls, complaint handling in operation.
Post-market obligations, reporting clocks, and field action capability.
Risk of deferring
Operational consequence of postponing a capability past the point it was needed. No financial modelling, no invented return — just what breaks and whether it can be fixed afterwards.
| Capability | If deferred past | What actually happens | Afterwards |
|---|---|---|---|
| Design control spine | Design freeze | Decisions exist only in engineering tools. Reconstructing the rationale later is guesswork presented as record. | Unrecoverable |
| Risk management file | Design freeze | Hazards are identified after the design is fixed, so controls become inspection rather than design. | Expensive |
| Supplier qualification | First clinical build | Units are built with materials from unassessed suppliers, and the build may not be defensible. | Expensive |
| Assembly controls | First clinical build | No contemporaneous record that units matched the frozen design. | Unrecoverable |
| Post-market procedure | Commercial launch | Complaints arrive with no route to assess or report them, against a reporting clock. | Recoverable |
Do not build this yet
Deliberate deferrals, each with the condition that reopens it. A deferral is a decision with a trigger, not an oversight.
Eight procedures do not need a platform, and configuring one consumes the quarter that design transfer needs.
The product is terminally sterilised and packaged externally; classification is not the control that protects it.
Nothing is on the market. The procedure must exist; the system need not.
Auditing all suppliers equally dilutes attention from the three that carry product risk.
Illustrative reference blueprint for interface design. Sequencing and applicability judgements are generic teaching examples and are not regulatory advice for any specific product, site or route.