What a Gate Catches, What It Destroys and What It Costs
Last timeWhat Gets Handed Over
A gate between two steps cuts the bad share of handovers from 15 per cent to 5.3, and the same four numbers show what it destroys, why it loses money on compute, and where to set it.
Four steps, each right 95 times in 100, is a job right 81 times in 100. There
are only two ways out of that arithmetic. Use fewer steps, or put a check
between two steps so that bad work is stopped before the next step builds on
it. This is the arithmetic of the second one, and it is less flattering than the
idea sounds.
The service is the same four step one as before: a customer's written question
about an order, answered in four steps. Sort the request, find the
matching order and policy, draft the reply, check the draft against the policy.
Each step sends about 1,500 words of context and instruction, gets back about
300, takes 0.8 seconds and costs 0.009, so the run costs 0.036 and takes 3.2
seconds.
Put a gate on the second handover, where step two passes an order record and a
policy clause to step three. The gate asks one question before any reply is
drafted. Do this record and this clause actually answer what the customer asked?
Three numbers describe that gate, and all three have to be measured rather than
assumed. Write for the share of handovers that are bad, meaning the record
or the clause is wrong. Write for the share of bad handovers the gate flags,
and for the share of good handovers it flags anyway. For this gate,
measured on traffic, is 0.15, is 0.70 and is 0.05.
One warning about before anything is computed from it. It is a property of
the handover and not of a step, and the two are not the same number. A step
that is right 95 times in 100 does not make 5 handovers in 100 bad here: this
handover is bad if either of the two steps above it went wrong, which is already
1 minus 0.95 squared, or 9.75 in 100, and it is bad again whenever the lookup
returns a defensible clause that is nonetheless the wrong one. Measured at the
handover it comes to 15 in 100. Every figure below moves with that number, so
inferring it from a step's accuracy rather than measuring it where the gate
sits is the first way to get a gate wrong.
Of 1,000 handovers
Take a thousand handovers so every figure is a count rather than a decimal.
Fifteen in a hundred are bad, so 150 are bad and 850 are good. The gate flags 70
per cent of the bad ones, which is 105, and misses 45. It flags 5 per cent of
the good ones, which is 42.5, and passes 807.5.
| stopped by the gate | passed on to step 3 | |
|---|---|---|
| bad, 150 of 1,000 | 105.0 | 45.0 |
| good, 850 of 1,000 | 42.5 | 807.5 |
So 852.5 handovers pass on, of which 45 are bad. That is 5.3 per cent, against
the 15 per cent that were bad before the gate.
The lesson stops here
8 more paragraphs to go
You have read the opening. The rest of the argument, the problems that check whether it landed, and the lines worth keeping at the end all come with a plan.
The first lesson of every course in the library reads the whole way through, free, so you can see exactly what the rest of them are.
See the planThe contentsThis is the reading half
Starting the course gives you your own copy of it. Every idea on every page has problems standing under it, marked with a reason rather than a tick, and any sentence you do not believe can be opened and argued with. None of that can happen on a page nobody owns.
The contents