Book a call

The barWhat we assess, and what we hold a system to

The bar is set for your system

Before we change anything, we assess the team and the product in ten areas. From what we find, we set the bar that fits: what your reality shows is needed, what is reasonable to expect of a system like yours, and what the law requires. Nothing is added for show.

It is the same bar whether we build the system for you or your own developers build it with the method. Three ways to work with us

Describe a system

19 of 37 apply

  1. 01

    Code

    Can a newcomer follow it?

    7 of 8 · not in this bar
    • Your conventions are kept. A pattern changes only where the new one is clearly better, and we say why

      every system
    • One responsibility per file and per function, judged by what reads clearly, not by a line count

      every system
    • Names carry the meaning, so the code reads from top to bottom. Comments say why, never what

      every system
    • Pure logic at the core, with reading and writing kept at the edges

      every system
    • Types that make a wrong state hard to write, and errors as typed values

      every system
    • One definition of each rule. Nothing built before it is needed, and nothing left behind: no stubs, no dead paths

      every system
    • Immutable by default, explicit dependencies, a small public surface. Operations are safe to run twice, and we measure before we optimise

      every system
    • Deterministic logic: the same input always gives the same result

      when errors cost money
  2. 02

    Tests

    Can the tests fail?

    2 of 3 · not in this bar
    • Every change ships with tests that are able to fail

      every system
    • Each property that must never break has a test of its own

      every system
    • Property tests, fuzzing and a model-checked core rule

      when errors cost money
  3. 03

    Gates

    What stops a bad change?

    2 of 3 · not in this bar
    • Nothing merges on a red gate

      every system
    • Dependencies are pinned and audited

      every system
    • A person reviews every merge

      with real users
  4. 04

    Infrastructure and delivery

    Is there a way back?

    1 of 3 · not in this bar
    • Infrastructure exists only as code, and only the pipeline deploys

      every system
    • A rehearsed way back for every release

      with real users
    • Backups that have been proven by a restore

      with real users
  5. 05

    Security

    What can an agent reach?

    2 of 4 · not in this bar
    • Agents work where production data and credentials cannot be reached

      every system
    • One register of findings, each closed with the test that proves the fix

      every system
    • A full security and quality audit before going live

      with real users
    • An attack by reviewers who did not build it

      when errors cost money
  6. 06

    Data protection

    What does the law ask?

    0 of 3 · not in this bar
    • An impact assessment and processing agreements

      with personal datarequired by law
    • Retention periods and a register of everyone who receives the data

      with personal datarequired by law
    • Only anonymised data ever reaches an agent

      with personal data
  7. 07

    Operations

    Who notices first?

    0 of 4 · not in this bar
    • Every alarm has a written reason and a runbook

      with real users
    • What production shows comes back as proposals a person approves

      with real users
    • Service targets on the paths users feel

      with real users
    • When an outside service fails or slows, the product degrades instead of breaking

      with real users
  8. 08

    How the team works

    Where does the work wait?

    3 of 3 · not in this bar
    • A written spec before any code

      every system
    • Research before building. Every piece of work gets two fresh review passes by agents before a person sees it

      every system
    • Decisions that are yours stay yours: spend, credentials, deploys

      every system
  9. 09

    Evidence

    Can a stranger check it?

    2 of 2 · not in this bar
    • Every number carries its source and says whether it was measured

      every system
    • Decisions are logged with their reasons

      every system
  10. 10

    Agents in the product

    Would you know if it was quietly wrong?

    0 of 4 · not in this bar
    • Every run leaves a trace

      with agents inside
    • Deterministic checks score every run for failures that raise no error

      with agents inside
    • Actions that always need a person are held by construction

      with agents inside
    • Each agent has a limit on steps, cost and time

      with agents inside

APrinciple

How the bar is set

  1. Needed

    What your reality shows

    A defect that reached users, a bottleneck we measured, a risk we found in the code.

  2. Reasonable

    What a careful team would expect

    Of a system like yours, at the stage it is in. A prototype is not held to the rules of a payment system.

  3. Required

    What the law and your contracts ask

    Where personal data is processed, that part of the bar is not a choice.

If nobody can give the reason for a rule, it is not in your bar.

We put in place what the situation needs, and add the next rule when reality asks for it. A bar that fits gets followed. A bar built to impress gets worked around.

Every item on this page comes from systems we have built and run. The systems behind the method

BAssessment

What you receive within a week

  1. A scorecard

    Each area, what we found in it, and the evidence by file and line.

  2. What would hurt first

    The three findings that matter most, in plain language, with what each costs if it is left.

  3. Your bar

    Which items apply to your system and why, and which we would leave out.

  4. A sequence

    What to put in place first, and the first real feature to prove it on.

The assessment is yours to keep whether or not we continue.

CExample

What one looks like

Taken from an assessment we delivered in November 2026. The client's name and everything that could identify them are removed, and so are the file names. The wording and the figures are the report's own.

Readiness assessment · excerptClient details removed
Where the team is

Level 1 of 3

Developers use code completion and occasional chat. The target is level 3, orchestrated agentic development. It can be reached without a rewrite. The path starts with building the safety net.

What is strong
  • The application framework is current. No large rewrite is needed.
  • Money is handled correctly. On synthetic data, the core of the calculation gave the same results as an independent reference calculation.
  • Most of the code is small: the median function is 13 lines long.
The three largest risks
  1. High

    There is no safety net. No working tests, no pipeline and no locked list of dependencies, so an agent cannot tell whether it broke something. On synthetic data we found three calculation errors that stay invisible without tests.

  2. High

    Three serious security risks: one in sign-on, one in permission checks, one in an outdated library that processes what users upload. Each fix is small.

  3. Medium to high

    Customer environments are not code and not under version control, and they are not isolated from each other. This has to be fixed before any AI feature reaches customers.

The cost of leaving it

The gaps stay open and the calculation errors go unnoticed. Agents would copy the patterns already in the code, such as large functions and hand-written permission checks, far faster than a person would.

One finding, as it is written up
Finding
A permission check that never blocks. The check is a method, but it is used without being called, so the condition is always true. Accounts that the user interface restricts can read a register through the interface for integrations.
Severity
Medium. It needs an account, so an outsider cannot use it.
Evidence
Four call sites and the method's definition, each by file and line.
Fix
Four missing pairs of brackets, and a regression test that proves the fix.
The path, first three phases
PhaseWhat it holdsThe usual wayperson-daysWith the methodworking days
1Immediate fixes: eight small corrections before any other development10.5 to 18.53.1 to 5.5
2The safety net: locked dependencies, tests, a pipeline, instructions for agents21 to 404.6 to 8.7
3A new interface where permissions are decided in one place, with an audit log24 to 435.0 to 8.9
Phases 1 to 356 to 10213 to 23

Both columns are estimates. Person-days describe an experienced team working the usual way. The second column is building and testing with the method; it leaves out the client's own time. In the first two phases the safety net is still being built, so the gain is smallest there.

Why this order

Agents are only as good as the safety net of the code base. Without tests and a pipeline, an agent does not notice that it broke a calculation, and a person notices only when a customer calls.

DConfidentiality

Your code, while we work on it

  1. Isolation

    A machine of its own

    Your project lives in a dedicated virtual machine, used for your work and nothing else.

  2. Separation

    Nothing crosses over

    Your code and your secrets never meet another client's project. What we carry forward is what we learn about the method, never your code.

  3. Access

    Only what the work needs

    For the assessment, that is read access to the repository and the pipeline. We do not ask for access to production.

  4. The end

    Destroyed when we finish

    When the collaboration ends, the machine is destroyed with everything on it. Your code stays where it always was: in your repository.

The models

Where your code goes when an agent reads it

By default
Agents read your code in sessions with our model provider. Those sessions are not open to anyone outside. The one exception: if the provider's automatic checks flag a session as abuse, its dedicated security team may review that session.
On request
The work can run on dedicated private models instead, for example on AWS Bedrock in an account you control. We agree this case by case, and it changes the time and the cost of the assessment.

Contact

Start with 20 minutes

Tell us what you are building, where your team is stuck, or what your production keeps telling you. You leave the call with a clear next step.

hello@sterlon.eu
Book a 20-minute call