Tests & results
Expected outcomes, what actually happened and supporting evidence.
File format
JSON + CSV
← All services
NERACIA AI SYSTEM TESTING / Testing in a separate test environment
AI System Testing
Check AI errors and permissions before wider use.
We test one AI process that can carry out tasks, check incorrect or unauthorized actions, and provide evidence-based fixes.
Fees & payment
Pricing follows the AI process, test setup, run counts, evidence and retesting. Model/software costs, additional development and system repairs are separate.
WHAT WE DO
The work we do
for you.
Example: AI can prepare orders, but important changes must not happen without an authorized person’s approval.
- 01Agree permitted actions and those requiring approval.
- 02Run normal, error and failure situations in the test environment.
- 03Report issues, repair priorities and agreed retest results.
WHAT YOU RECEIVE
When the work is done,
this is what you receive.
For teams whose AI can take actions but have not established what happens on errors, failures or missing permissions.
Issues & repair priorities
Issues ordered by impact so your team knows what to fix first.
File format
CSV + HTML
Results & retest summary
Retested issues, unresolved items and limits on the conclusions.
File format
HTML + Markdown
Confirm formats in the written scope. Sample packs use fictional data and do not show client results.
OUTPUT PREVIEW
See the shape
of the work.
A fictional example of findings and next actions. Codes in the first column identify sample items only. Figures in the download files can differ.
| Reference | Finding | Next action |
|---|---|---|
| SYN-CASE-01 | Tool permission too broad | Restrict permission; retain transcript |
| SYN-CASE-02 | Grader not calibrated | Review rubric and failure example |
| SYN-RETEST-03 | Fix needs a retest | Repeat scenario and record version |
CHECKING THE OUTPUTS
We check the work
to support your decisions.
Which failures need fixing before broader workflow use?
We check outputs against the agreed inputs and work. Outstanding items stay recorded with reasons and next actions.
See detailed checks (for the relevant team)
Starting records Non-production workflow/harness, configuration version, permission and 75–100 agreed scenarios.
Execution coverage
Executed cases ÷ in-scope cases; show failed, blocked and not-run separately.
Finding closure
Priority findings with passing retest evidence ÷ priority findings in retest scope. Untested findings remain open.
The reviewer checks evidence and residual risk; a passing result does not guarantee production security.
Re-evaluate after material workflow changes; agree extra retests in writing.
Agree starting records, targets and period before starting. Examples and targets are not client results.
WHAT IS INCLUDED
Work limits
One process in a test environment, one configuration version, 75–100 test situations, one results discussion and one retest of agreed findings. Access and a repeatable test setup are required.
What you provide
- AI process, permitted actions and approvers.
- System version/configuration and a repeatable test setup.
- Fictional data and permitted access to action evidence.
Not included
Not certification, a security guarantee, penetration testing, compliance audit or repairs to the live customer system. Results are limited to the versions and situations tested.
SUBSEQUENT WORK
AI System Testing
For an existing AI system: test one process in a separate environment before deciding on wider use.
Test AI behavior
Agree system version, permissions and 75–100 situations. Record actual actions and evidence.
Fix and retest
The system owner makes repairs unless separately agreed. One retest follows agreed findings and run limits.
Review again when the system changes
New AI versions, software or rules may need a new review with separate fees.
Who is responsible?
You provide permission, the test environment, rules and a decision reviewer. The report explains evidence and testing limits; it is not certification or a security guarantee.
Each engagement starts after agreeing scope, people, access, timing and payment. Subsequent stages are excluded unless written in the quote.
FROM CONVERSATION TO HANDOVER
You know
what happens next.
Arif is your discussion contact. Required delivery and review roles are designated before engagement.
Tell us the problem
You do not need a package name. Explain the task causing difficulty and the result you need.
Review the work and price
We explain tasks, outputs, limits, fees and timing in writing before you decide.
We do the work and check it
We check inputs, carry out the work and discuss unresolved questions with you.
Receive the work and explanation
Receive output files, an explanation and next actions. Discuss corrections under the agreement.
Fees & payment
Quote after reviewing test requirements
Pricing follows the AI process, test setup, run counts, evidence and retesting. Model/software costs, additional development and system repairs are separate.
Timing & corrections
*Estimated 7–10 working days. Final timing follows complete inputs and agreed scope, access, roles and payment terms. Discuss changed inputs or blockers before changing the schedule.
Includes one correction round on the same inputs and scope, requested within seven days of handover. New periods, data or additional work require agreement on scope and costs.
BEFORE STARTING
Common questions.
Is subsequent work included?
The system owner implements repairs unless separately agreed. Discuss further tests after system changes with new fees and limits.
What if data is missing or there are differences?
We record gaps, their impact and required information. Conclusions requiring evidence remain on hold; discuss extra work and schedule changes before proceeding.
How is business data shared?
Use general context in the initial brief. Before sharing real data, agree minimum access, transfer channels, storage, recipients and retention. Do not share main account passwords.
I am unsure which service fits. What should I do?
Send a brief through WhatsApp or email. We discuss fit and availability, then prepare work and a quote. The website form is not an order or payment.
NEXT STEP
Tell us about one task.
Let’s define the first step.
Tell us about the business problem. We help define work, outputs, fees and timing before you decide.