What gets better by itself, and what does not
Two things could get better on their own here, and only one of them does. Which one is which is the most important thing on this page.
1. What you are told to chase gets shorter
Three things are required and nothing has been filed. What to ask for, in order:
- plot_geolocation — ANY admissible evidence
closes a gap - ochratoxin_screen — ANY admissible evidence
closes a gap - moisture_content — ANY admissible evidence
closes a gap
One accredited record arrives. The list is now:
- ochratoxin_screen — ANY admissible evidence
closes a gap - moisture_content — ANY admissible evidence
closes a gap
Why is the order not just alphabetical, or the order I wrote them in?
Because a gap holds the whole decision at nothing, so closing one outranks everything. After that, only a claim that could still change the answer is listed.
Which means the evenings go on documents that could still change the answer, rather than on the ones that could not. A gate that only says no leaves you to work out which is which.
2. Who you believe moves with what actually happened
Three outcomes were attested against this laboratory's record — two where it turned out right, one where it did not. Its track record moved:
| Before | After | |
|---|---|---|
| Track record | {"n":1,"successes":1} | {"n":4,"successes":3} |
| The most its word can ever be worth | VERIFIED | VERIFIED |
The first row moved. The second row did not, and cannot.
What is the difference between those two rows?
The first is how well this particular laboratory has done — earned, and it goes up and down with what actually happened.
The second is the most a record from that kind of source may ever be worth, which comes from your signed register of who is accredited to do what. Doing well cannot promote anybody into a class they were not accredited into. A good run of results is not an accreditation.
3. Your requirements do not move at all
Nothing on this page changed what the standard requires. No outcome, no volume, no pattern the system noticed. Requirements change when a person signs a new version, and every decision already made goes on re-checking under the version it was made under.
Why would I not want it to tune my standard for me?
Because the moment the arithmetic that produced a verdict has improved itself, the record stops fixing the computation — and a counterparty re-running it two years later is answering a different question from the one you answered.
And from inside a loop, a rule that tightened because outcomes justified it and a rule that loosened because refusals were expensive look exactly the same. You are the one accountable for the standard. It should not move without your signature on it.
4. And here it is being refused
Saying a thing cannot happen is worth nothing on its own, so the attempt is made here, in this run. The supplier has filed its own moisture reading about its own coffee, which lands — correctly — as SELLER-ASSERTED. The loop then tries to promote that record into the class an accredited laboratory occupies. What comes back is not a warning and not a log line —
OUTCOME_CLASS_CEILING_VIOLATION: an outcome attestation moved the class of record mem_afd378dea45a from SELLER-ASSERTED to VERIFIED (promotions undefined -> 0). An outcome may move a track record and may never move a class ceiling: class derives from who observed a thing, never from having been right about it.
The check runs on every outcome that is ever applied, and it throws rather than returning false, so there is no path on which a ceiling moves and something carries on regardless.
Every line on this page came out of a real run — the worklist from whatWouldHelp(), the track record from applyOutcomeTrial(), and the refusal from assertClassCeilingUnmoved(). The laboratory and the lot are invented; the workings are not.
This page loads nothing from another origin and makes no request at run time.
Nothing here is legal advice.