Skip to content
Trazum

1.51.0

The record, and the minor

Added#

trazum report --year — the year, from what was already written down. Chapter ten of docs/plan-1.51.md, the last of the arc.

No new data. Everything comes from the store and the plans a team already keeps, and nothing is computed that cannot be checked against a document that already exists. That constraint is the design: an annual report is the document most likely to be quoted out of the room it was written in, and the one nobody goes back to verify.

Three outcomes, never two. "Eleven of fourteen arrived" reads better than "eleven arrived, one did not, and two could not be judged" — and the second sentence is the one that tells somebody their measurement has a hole in it. A year is where the temptation to collapse them is strongest.

Missing months are named, never filled. A year report that quietly covers nine months and prints an annual total is wrong by a quarter and says nothing about it.

There is deliberately no arrivedUsd. A verification says whether each action landed; it has never carried a per-action dollar figure for the saving that arrived. Summing one out of the observations would mean deciding which of several numbers per action is "the saving" — a judgement the verification refused to make and this module has no standing to make on its behalf. So the year says what was promised and how many promises were kept, and says plainly that it cannot put a figure on the kept ones. The alternative is precisely the annual-report arithmetic this document exists to replace.

It lists its own blind spots in a cannotSay array, which is the only reason the rest of it is worth acting on.

It reports the record, not the team. No per-person anything — an annual document is exactly where a cost tool starts being used for performance review, and the way not to be is to hold no data that could be. A test asserts the document carries no field that could name somebody.

conform grows an outcome chapter and an annual chapter, so another tool emitting this format has to handle a missing numerator the same way. The standard is only worth something if its refusals travel with it.

Cross-field rules in conform. The refusals worth carrying turned out to be relational: a per-field contract can say "a number or null", and it cannot say "null when nothing was recorded, and a number otherwise" — which is the whole refusal. A rate of 0 is valid when calls were recorded and none succeeded, and a lie when nothing was recorded at all, and the difference lives in a different field.

docs/our-own-medicine.md — this project's own record, kept the way it asks users to keep theirs: what each arc refused to ship, what it got wrong and for how long (a version claim wrong for seventeen releases; a roadmap running in two directions for twenty-four), and what it cannot say about itself. It ends without a score, because every miss on it was found by the same process that made it.

The doctrine, second edition — two rules added from this arc: cheaper per call is not cheaper per outcome and an unallocated share is never spread.

Fixed#

The annual record attached an outcomes object whenever calls were parsed, rather than whenever an outcome was recorded — so a year with no outcomes at all reported "0 of 120 recorded" and the honest sentence was unreachable. A zero dressed as a measurement, which is the exact failure that field's own contract forbids. Caught by a test written for the sentence that could not print.