· 3 min · updated Sep 27
how i think
these are not values. they are four mistakes i made expensively enough to remember.
every one has a specific system attached. the example is the point: the slogan is the part that is easy to agree with and useless.
In this note04 sections
1 · put the invariant where it cannot be forgotten
a rule enforced by convention holds exactly as long as nobody writes the one line that forgets it.
- the usual multi-tenant answer is a tenant filter applied in application code on every read and write. it works until one query is written without it, and that query looks completely normal in review.
- so the boundary went into the database instead, forced, and the services check it against the live catalogue at boot and refuse to start if a table is unprotected. the invariant became something a machine asserts rather than something a reviewer remembers.
if a rule matters, make it impossible to violate rather than embarrassing to violate.
2 · disagreement is a signal, not noise
a system that produces one confident answer has thrown away the information you most needed.
- clinical pdfs defeat a single extractor in ways that fail silently — downstream, nothing can tell a sentence that was missed from one that was never written.
- running six engines and reconciling them by vote costs six times the compute per page. what it buys is that where they diverge, the pipeline knows the page is hard instead of quietly emitting one engine's guess.
the same shape shows up in agents arguing before they answer. cheap certainty is the expensive kind.
3 · a number without a source is decoration
if you cannot say where a figure came from, it is not evidence.
- evidane holds every finding to that rule. eighty-seven agents can argue all they like, but a finding that cannot name its clause, page and line never reaches the lawyer.
- brekfuz does the same with answers: each one carries the sources it was built from, and only the sources the asker is cleared to read.
a claim that cannot be checked is a claim nobody should act on.
4 · prove it before you say it
a bug report without a reproduction is an opinion.
- breken does not comment on a pull request until a runner has proved the finding, and it does not open a fix until the patch passes the repository's own typecheck, tests and build in a sandbox.
- it is the rule i follow upstream too. a pull request that arrives with the failing case and the test that pins it is one a maintainer can merge. 100+ have been, mine and breken's.
proof first. the argument gets shorter every time.
show your work — cite specific text, admit uncertainty.
— evidane's house style, written into the product
where this shows up in practice is building; where it gets argued in public is writing.