Turning it up, and what that costs
Set the gate stricter and it holds more. This page shows how much more, on 60 decisions we actually ran — and then asks you for the two prices only you have.
What the dial actually is
One setting: how good the weakest thing a decision rests on has to be, before the decision may go ahead.
Turn it up and more gets held. Turn it down and less does. That is the whole mechanism, and the table on the next panel is what each position did to 60 decisions we ran through it.
Why is the answer the WEAKEST thing, and not an average?
Average four good reasons with one missing one and the missing one disappears. A supplier with nothing to prove comes out level with a supplier who proved everything.
So there is no average here. The answer is the weakest link, and if one required thing has nothing behind it at all the answer is no — however good the rest of it looks. That is the one rule to take away from this page.
What we are paid at each setting
Exactly the same. Not more when it lets something through, not more when it holds something. That is not a promise on a page — no fee in this business may depend on the answer, and a program refuses to build the software if that stops being true.
Why does that matter to me?
Because otherwise you would be taking a strictness setting from somebody who earns more at one end of it.
A vendor who is paid per refusal has a reason to want your gate tight. A vendor paid per release has a reason to want it loose. We are paid per standard per year, so we are indifferent, and the check that keeps us indifferent runs on every build.
What each setting held
60 decisions, run through the real gate once at every setting. Every number below came out of a run.
| Setting | Held | Went through | What it held on |
|---|---|---|---|
| Anything anybody signed a supplier’s own word about its own goods is enough | 30 | 30 | below the floor 10 evidence presented failed 10 nothing submitted 10 |
| Loose a little better than the seller’s own word | 40 | 20 | below the floor 20 evidence presented failed 10 nothing submitted 10 |
| Middling a model’s output will do, a seller’s word will not | 40 | 20 | below the floor 20 evidence presented failed 10 nothing submitted 10 |
| The shipped default an independent record, or a model with a track record behind it | 40 | 20 | below the floor 10 evidence presented failed 20 nothing submitted 10 |
| Strict an independent record, and a good one | 50 | 10 | below the floor 10 evidence presented failed 30 nothing submitted 10 |
| Accredited only nothing below an accredited body’s own report | 50 | 10 | below the floor 10 evidence presented failed 30 nothing submitted 10 |
What do those three reasons mean?
Nothing submitted — nobody ever sent the document. You may simply not have chased it.
Evidence presented failed — somebody DID send something and it did not qualify: it came from the wrong kind of source, or two independent bodies disagreed and neither can be relied on.
Below the floor — a real record from a real source, which is simply not good enough for the setting you chose.
They look the same in a spreadsheet and they are three different conversations in front of an auditor. Only two of them are diligence.
What this corpus is, before you read another number
Invented. 60 decisions built to a stated mixture:
- 10 — an accredited laboratory’s report, recent
- 10 — an accredited laboratory’s report, but eight months old
- 10 — a figure out of a screening tool or a model
- 10 — the supplier’s own certificate about the supplier’s own goods
- 10 — nothing on file at all
- 10 — two independent bodies that disagree with each other
Your mixture is not this mixture, and the right response to this table is "we have far more of the fourth kind than that". The program is what is real: pointed at your own two weeks of records it produces your counts, and those are the ones that decide anything.
Your two prices
We will not tell you what this saves you. We do not know, and anybody who tells you they do is multiplying three numbers nobody has measured.
What we can do is arithmetic on your figures. Two boxes, both empty on purpose:
Fill in both boxes and this works out, for each setting, how often a hold would have to be wrong before that setting stops being worth it.
Why are you asking ME for the numbers?
Because they are yours and they are not ours. What a held load costs depends on your contracts, your dock, and what you were shipping. We went looking for an industry figure and did not find one we could stand behind — the ones in circulation trace back to vendor pages with no method published, and our automotive screen writes up the search and what closed each route.
A business case built on a number we guessed loses the second meeting — which is a thing we would rather say now than find out with you.
What is a break-even here, in plain words?
Take a setting. It holds a number of things. Some of those holds are right and some are wrong, and nobody knows the split — that is the honest position and it is the whole reason for the two-week assessment.
So the arithmetic runs backwards. Instead of guessing the split, it works out the split at which the setting exactly pays for itself: above this share of wrong holds you are losing money, below it you are making it. You then look at that percentage and say whether it sounds like your suppliers.
What this page will not tell you
- What it saves you. Not at any setting, not as a range. Three of the numbers in that sum have never been measured by anybody, and multiplying three guesses produces a figure that looks like a measurement.
- How often it holds something that was fine. Nobody has run this in a plant yet. Any number we gave you would be us marking our own homework, and the two-week assessment on your own records is how that number gets found.
- Which setting to choose. The break-even is arithmetic; the choice is a judgement about your suppliers and your customers, and it is yours.
What it does tell you
That the dial is real, that it moves, that what it holds it holds for a stated reason, and that we are paid the same wherever you set it.
Licence figures on this page — $24,000 for the two weeks, $45,000 a year for the first standard — are read from the pricing register on every build, not typed here.
The counts came from running the shipped gate over an invented corpus of 60 decisions. No saving is computed at any setting and no false-refusal rate is stated, because neither has been measured.
This page loads nothing from another origin, makes no request at run time, and touches no browser storage. It works from a file on disk with no server.
Nothing on this page is legal, financial or tax advice.