Procurement
The DDoS Buyer's Checklist, and the Claims That Need a Proof of Concept
Last updated: August 2026 · Three piles, and nothing belongs in none of them · Reading time ~14 min

Every claim about a DDoS product belongs in exactly one of three places: verifiable from a document, measurable in a test, or written into the contract. Sorting them that way removes most procurement arguments, because the disagreement is usually about which pile a claim belongs in. Anything that fits none of the three is a statement of intent and should be recorded as one.
Procurement disputes in this category are rarely about whether a claim is true. They are about what kind of claim it is — and once that is settled, most of them resolve themselves.
So the checklist below is sorted by how a thing can be established rather than by subject. Three piles: documents, tests, contracts. Anything that fits none of the three is a statement of intent, which is a legitimate thing for a supplier to make and an illegitimate thing for a buyer to score.
At a glance
| Appliance | Claim | Why it cannot stand alone | The measurement that replaces it |
|---|---|---|---|
| Zero false positives | Unfalsifiable without a denominator | Transaction completion against a baseline, with sample size | |
| Automatic mitigation | Every product automates something | Operator interventions counted during a live scenario | |
| AI-powered detection | Describes an implementation, not a result | Detection time on a vector the product has not seen | |
| Unlimited protection | No device exceeds the circuit in front of it | Sustained packet rate at 64-byte frames, model-specific | |
| Line-rate performance | Line rate at which frame size, in which direction | The RFC 2544 frame-size matrix | |
| Multi-layer coverage | Coverage is not efficacy under load | Concurrent capacity with two countermeasures active |
None of these claims is dishonest. Each is unfalsifiable as stated, and each has a straightforward measurable form that a confident supplier will accept.
Pile one — settled by a document
These need no test. They need a current document, with a date, naming the model quoted.
- Capacity of the specific model, in packets per second at 64-byte frames, mitigation active, direction stated.
- New sessions per second, separate from concurrent sessions.
- The RFC 2544 frame-size matrix, or an explicit statement that it was not measured.
- Which ceilings are licensed rather than physical, with the cost and lead time of each step.
- Deployment modes supported, and which are in scope for this quotation.
- Bypass mechanism: hardware or software, trigger conditions, stated failover time.
- Which functions require reachability to a supplier-operated service, answered per function.
- Export formats for telemetry and incident records.
- Platform generation the figures describe, and the document’s revision date.
- Licence expiry behaviour and grace period, as a number of days.
Item 9 catches the trap described in reading a datasheet: a public specification page can describe a superseded platform for years, and two suppliers compared across different generations are not being compared at all.
Pile two — settled only by a test
No document can establish these, because they are properties of the product and your traffic together.
- Legitimate transaction completion under attack, against a measured baseline.
- False-positive behaviour against the awkward sources: shared addresses, unusual clients, genuine peaks.
- Sustained packet rate at 64-byte frames with mitigation active, on the quoted model.
- Concurrent capacity with two countermeasures running, which is always lower than either alone.
- Behaviour when the supplier path is blocked for the full evaluation window rather than an hour.
- Time to mitigate, measured from the first attack packet rather than from the alert.
- Operator interventions required, counted.
- Recovery behaviour after the attack stops, including late-clearing state.
- Evidence export performed by your staff, in your format, without supplier assistance.
- Failover triggered deliberately, with the time measured.
Item 15 has a specific failure mode worth naming: grace periods run on timers, so a short disconnection test passes regardless of the period’s length. The duration is the test.
The full method is in the proof-of-concept methodology and the environment in the test lab guide.
Pile three — settled only in the contract
Things no document proves and no test reveals, because they are about the future.
- Support renewal percentage, its basis, and a cap for the expected life.
- Behaviour at end of support term, and at licence expiry, with a grace period in days.
- Capacity step pricing, fixed in advance.
- Emergency capacity uplift: whether it exists, who authorises it out of hours, how long it takes.
- Response commitments by severity, in your time zone and language, with the escalation named.
- Exit terms: what you keep, what stops, what is returned, over what period.
- Acceptance criteria from pile two, with the thresholds you set, as a condition of acceptance.
- Notification obligations if a component’s dependency model changes during the term.
Item 27 is what connects the three piles. A test result that is not a contractual condition is information; the same result written into acceptance is leverage.
The claims that need converting
Six recurring phrases, each unfalsifiable as stated and each with an obvious measurable form. The conversion is not adversarial — a supplier confident in the product will usually propose the measurement themselves.
“Zero false positives.” Over what sample, against what baseline? Replace with transaction completion under attack, with a denominator.
“Automatic mitigation.” Every product automates something. Replace with a count of operator interventions during a live scenario, and the escalation point where automation stops.
“AI-powered.” Describes how something was built, not what it achieves. Replace with detection time against a vector the product has not previously seen, and the false-positive rate that accompanies it.
“Unlimited protection.” No device exceeds the circuit in front of it. Replace with sustained packet rate for the quoted model, and a written division of responsibility with the upstream tier.
“Line-rate performance.” At which frame size, in which direction, with mitigation active? Replace with the frame-size matrix.
“Multi-layer coverage.” Coverage is a feature list. Replace with concurrent capacity — the figure with two countermeasures active, which is the one multi-vector campaigns attack and which appears on no datasheet.
Before signing
Four questions that a good evaluation still frequently misses:
- Which capabilities demonstrated in the evaluation are separately licensed? Ask before the demonstration, not after.
- What happens on the day support lapses? The answers vary across the category by more than most technical differentiators.
- Who answers at 03:00, from where, in which language? With the escalation named rather than described.
- What would make you tell us this product is the wrong fit? A supplier with an honest answer is describing a boundary; one with none is describing a sales position.
The last one is not a trick. Every product in this category is wrong for some estates, and a supplier who can say which ones has told you something no datasheet contains.
Technical specification checklist
A clause-by-clause checklist for a DDoS tender. It covers what to require on hardware bypass, packet-rate capacity, application-layer coverage, logging and support, and each line is written so it can be pasted into a specification and answered yes or no.
Sent by hand, usually within a working day. The address is used for that and for nothing else.
Frequently asked questions
- What if a supplier refuses a proof of concept?
- Then everything in the second pile moves to the third, and the contract carries the risk instead of the test. That is a legitimate outcome, and it should be priced: a supplier unwilling to demonstrate should be willing to commit, and one unwilling to do either has answered the question.
- Is a reference customer as good as a test?
- For different things. References are the best available source on what the supplier is like to work with in year three, which no test reveals. They are a poor source on performance figures, because their traffic is not yours. Use them for the relationship and the test for the numbers.
- How long should the evaluation take?
- Long enough to include one genuine business peak, which for most estates means weeks. The commonly proposed short evaluation is long enough to confirm that the product functions and too short to find the two things that matter: false positives at a real peak, and behaviour when a dependency lapses on a timer.
- What is the single most skipped check?
- What happens when the support term ends. It has a factual answer, it varies across the category from "nothing changes but updates" to "protection ceases", and it is almost never asked because it sits between the technical and commercial evaluations and belongs to neither team.
Sources
- RFC 2544 — Benchmarking Methodology for Network Interconnect Devices
IETF · 1999-03 · standard · accessed 2026-08-17
- Regulation (EU) 2022/2554 (DORA)
EUR-Lex · 2022-12-14 · regulator · accessed 2026-08-17
Contractual content requirements for ICT third-party arrangements, which is why the third pile is not optional for regulated entities.
- Directive (EU) 2022/2555 (NIS2)
EUR-Lex · 2022-12-14 · regulator · accessed 2026-08-17
Published: August 2026 · Last reviewed: August 2026
Reviewed means the sources above were re-read on that date; the text is only reissued when something material changed.
This guide is updated as vendors release new models and pricing. How we compare vendors