Operating Record
We publish this brief with our own system, every day. Every number on this page is computed by a program from our own logs — none of it is typed by hand. Each one starts with what it counts; how it is counted, and from which files, sits under "Method and sources".
-
Consecutive days in the site archive
How many days in a row this edition has an archived issue on the site, counting back from the latest one. It counts archive files, not confirmed email deliveries.
107daysMethod and sources
How it is counted: Counted back from the date of the latest archive file: each day with a file adds one, and one missing day ends the run. Each language is counted separately. A day whose archive exists but whose email failed still counts, so this is not proof of delivery every day.
Source:
newsletters/issues/en(the dated issue files behind the site archive, one .md per day) -
Lesson files dated to this month by their text
A lesson file is a "don't do this again" note written after something went wrong and the cause was found. This figure files each one under the earliest date that appears in its text — not the day it was written, and with no deletions subtracted — so it is not "new this month".
348filesMethod and sources
How it is counted: Each lesson file is assigned to the month of the earliest date in its text, and the files falling in this month are counted. A file citing an older date as background lands in that older month; files with no date are left out; deleted files are simply absent, with no deletion record to subtract.
Source:
memory(the lessons directory, one .md per wall) -
Stopped by our own checks, month to date
How many times three automated checks — the pre-ship quality gate, the admission gate and the plain-language gate — have stopped something so far this month, added together. The three use different units (drafts, candidate items, phrases), so this counts interceptions, not articles; a higher or lower count says nothing about quality, only that the checks are catching things.
3,225timesMethod and sources
How it is counted: Three kinds of interception added together (different units — see the table); the total means "interception events so far this month," nothing more. The current month is not finished, so it can't be compared with last month as growth.
Source:
state/pm/prod_events.jsonl(shipping log);state/admission_log.jsonl(admission log);state/term_gate_events.jsonl(outbound plain-language log)
The three kinds of interception
| Which check | What it stops | Month to date | Last month (full) |
|---|---|---|---|
| Pre-ship quality gate | Drafts not good enough to send or publish (pieces) | 10 | 9 |
| Admission gate | Candidate material read, judged not worth adding (items) | 3,071 | 695 |
| Plain-language gate | Internal jargon in outbound text with no plain gloss (spots) | 144 | 253 |
| Total | 3,225 | 957 |
This page was computed at: 2026-10-25 20:04 (US Central CDT)
Latest incident digest: issue 1, published 2026-08-16 →
Limits of these numbers (part of disclosing honestly)
- The ship-gate counter was only installed on 31 July 2026. Earlier months are not comparable with later ones.
- A lesson file's month is the earliest date in its text, not the day it was written. Commit times were all distorted once by a bulk migration, so they are not used; the cost is that a file citing an older date lands in that older month.
- Lesson files with no date in them at all (indexes, roll-ups) are not counted; there are 504 such files out of 2146.
- These three figures are an operating record, not a guarantee of quality or outcome.
What this page deliberately does not show
- No scoring of whether our calls were right. Whether the judgement holds up is a separate matter, not on this page and not implied by these numbers.
- No third parties. Everything counted here is our own event; nobody else's name appears in our failure log.
- Nothing is edited by hand. A wrong number means the counting rule gets fixed, not the page — a hand-edited page is overwritten by the next rebuild.
The above is this site's own operating record. It is not a guarantee of any outcome or quality, and it does not predict future results.