See for yourself · 3 agents · running non-stop

The discount climbs in plain sight.
And everything stays green.

Approved policy is an 8 % discount. The tool allows up to 15 %, because some cases justify it. Three agents work Monday to Friday, forty quotes a day — and not one step breaks a single rule all week.

Watch the discount Pricing is applying, and the counter on the right. Then switch DiaCroma on and see why nobody was stopping it.

DiaCroma off

discount Pricing is applying

8,00 %

approved policy: 8 % · the tool allows 15 %

given away above policy · this week

0 €

adding up the contracts signed so far

Pricingok

Keeps the discount table inside approved policy.

> set_discount(8,00 %)
✓ tool is authorised
✓ within the allowed range (≤ 15 %)
✓ no personal data
All clear

what DiaCroma sees

deviation per cycle
spent of its own
arrived from others
allowed
Quotesok

Reads the discount table and prepares customer quotes.

> read_discount_table
> quote #1 ·
✓ uses the current table
  discount given:
All clear

what DiaCroma sees

deviation (persistence)
spent of its own
arrived from others
allowed
Contractsok

Reads approved quotes, signs them and sends them to the customer.

> sign_contract #1
✓ tool is authorised
✓ quote is approved
  above policy:
All clear

what DiaCroma sees

deviation (persistence)
spent of its own
arrived from others
allowed

The whole week, in money

what the system doescontracts
signed
given away
above policy
No trajectory governance0 0 € every step was legal
Yes/no mark, strict — “there is a mark, do not sign” 00 € stops the clean days too
DiaCroma — the mark carries a quantity 00 € cuts when the number says so

No step of the week ever crosses a threshold. The largest discount the tool allows —15 %— produces a per-cycle deviation of 0.35, and the warning sits at 0.50. Not even the worst possible step trips anything on its own. That is why a control looking step by step cannot see this: it isn't that it misses it, it's that there is no step to look at.

And the obvious version doesn't work either. Marking the data as “someone else touched it” is a yes/no: either it stops the clean days too —and the business switches it off within a week— or it stops nothing. What changes things is the mark carrying how far the one who wrote it had already strayed.

The five working days this page strings together are produced by demo/semana.py against the same governance kernel that runs in the service, with the seed 20260728 written into the file. The page shuffles them; it does not invent them. An automated check fails if the page and the kernel stop matching.