Define what is being claimed
Pin down the system, the observable, the conditions and what would count as a counterexample.
Independent computational R&D · Bristol / UK
A difficult technical claim deserves a proper test.
Bring a result, model or computational question you cannot quite settle. I work out what evidence would answer it, build the test, and make the result open to examination.
The first answer may be yes, no, or that the evidence is not enough.
TEST DESIGN / 01QUESTION → EVIDENCE
CLAIM
?
INDEPENDENT TEST
assumptions · controls · alternatives
→
RESULTevidencewith limits attached
reproducible where possibleinspectable by design
01 / 05
01Start with the claim
not the tool
not the tool
02Build a test that
could fail
could fail
03Keep the limits
in the result
in the result
04Make the work
reproducible
reproducible
01 / Find your question
What are you trying
to find out?
Start with your field or the kind of uncertainty you face. These are examples of possible investigations, not past client work.
Choose a field to see example questions
↗ Every example shows what would be tested, what inputs the work depends on, and what a useful result could look like.
?
For a consequential technical claim
How strong is the evidence?
A useful investigation can start before the method is clear. Define the claim precisely. Trace the evidence behind it. Reproduce the result where possible. Test assumptions, baselines and failure conditions. Report what survives, what does not, and what remains unknown.
Discuss an independent technical evidence report ↗This is research and engineering work, not regulated certification, investment advice or a guarantee of performance.
- 01Define the claim
- 02Map its evidence
- 03Reproduce and compare
- 04Try to break it
- 05Report the boundary
02 / Selected research
Results with their
boundaries attached.
A small selection of completed investigations with public records. Findings apply to the stated system and conditions; they are not claims about whole fields.
PUBLIC RECORD
Public report or reproduction artifact available. Result status describes the evidence in that investigation only.
03 / The working method
Let the question set
the experiment.
Some questions need a carefully controlled simulation. Others need a fresh calculation from raw outputs, a property checked by a theorem prover, or a comparison against a physical measurement. The method follows the claim.
⌖
Design a test with controls
Choose inputs and an independent reference. Include checks that can reveal a broken instrument or an inconclusive result.
Run, inspect and reproduce
Preserve the relevant code, inputs and environment. Compare alternatives and keep unexpected outcomes visible.
Return evidence and limits
Deliver the result, reproduction materials where appropriate, and a plain account of what it does not establish.
04 / Research infrastructure
Compute is useful
when the question needs it.
Work can begin on local Apple Silicon. Specialist institutional compute is awarded for defined research purposes; its access and evidence status are stated precisely. An allocation, a completed job and a scientific result are different things.
LocalApple Silicon development machineAvailable for development and computational experiments; machine-specific results are labelled.
UKRI AIRRIsambard-AI access awarded20,000 GPU-hours awarded for approved research purposes. No operational compute use is claimed here.AWARDED · NOT USED
EuroHPC QCPVLQ 24-Qubit Star access awarded10 QPU-hours awarded for the approved research purpose. No completed QPU experiment is claimed here.AWARDED · NOT USED
Institutional allocations are restricted to their approved research purposes. They are not general-purpose compute available for client work.
WS
BRISTOL
UK
BRISTOL
UK
05 / About
Warren Smith
Independent Computational R&D
I work across scientific computing, software, AI evaluation and verification. The connecting thread is a practical one: turn a technical claim into a question that can be tested, then make the evidence and its limits clear.
OMEGA is my independent research programme for developing and testing methods for reproducible computational experiments, verification and evidence-backed technical investigation. It supports the work; it is not a prerequisite for understanding it.
Read selected investigations ↗Have a question that matters?
Bring the uncertainty.
Send a short description of the claim, result or model you are unsure about, and what decision depends on it. The first conversation is about whether a useful test can be designed.
Independent Computational R&DBristol · Working across the UK