Pulse.

The pulse is the org’s report on itself, produced each morning by a script with no AI model involved. It reads 8 areas, called axes, and prints a verdict for each: the founder’s open directives; whether a run that reported success wrote its output; whether data freshness monitors received data; which automated workflow runs failed in the last day; whether a scheduled job is failing or stuck; whether, within the bounded run window it reads, console commands sent through the API lack matching runs or records, or runs or records lack matching commands; whether the kill-switch file that halts the org was set or cleared without a recorded command; and which open pull requests have had failing checks for three days or more.

Each axis gets one verdict. Observed and clean means the observed portion had no findings; partial reads and stated limits still apply. Observed with a warning held means the axis was read and a check chose not to mark a finding red. Red means the axis was read and something is wrong; items are listed by public name. NOT OBSERVED means the axis could not be read. It is never clean: seeing nothing is not a pass. When the kill-switch is set, checks do not run and the checked axes read Halted. The directives axis is still read from its file.

This is the pulse of September 17, 2026, read from the org’s own record each time this site is built, never typed by hand. The line the pulse put at the top of its own report that morning:

6 failed run(s) in the last day.

1 of 8 axes was red that morning. A red axis is not a failure of the pulse; it is the pulse doing its job. The failure it exists to prevent is a red nobody saw.

The ladder is the org’s staged rollout: five stages, and the ladder decides how much of the org runs unattended. Each stage has promotion criteria, and a script, the evaluator, scores them every day by fixed rules with no AI model. A seat is one agent job in the org, with a charter that says what it does and a lane, the files it is told to write. The evaluator wrote its last verdict on September 17, 2026, and the active stage is Stage 2.

Stage 2 is holding on 3 unmet criteria, and the evaluator says why in its own words below. The Inspector General is the org's audit seat, the one that checks the other seats' claims against code and data; the letters ig in a criterion name are that seat.

ig_runs needs an IG smoke-eval scenario; ig_coverage needs the coverage manifest; no_org_fabrication needs at least one org-targeted IG module that measured (org_modules_observed in the report frontmatter), and the battery has none yet. All three are recorded Stage 2 build debt; stage2 holds honestly until they exist.

Every week the Chief of Staff seat writes a briefing to the CEO, rolled up from what each department reported. The briefing stays private, because it names security findings before they are patched, legal questions before they are answered and finances before they are closed. Its header names none of that, so it is published here: who wrote it and for whom, the period and the date, how many decisions were waiting on the CEO, which departments proposed a lesson, and how current each department’s last report was.

This is the header of the briefing of September 17, 2026, from the Chief of Staff to the CEO, read from the committed file each time this site is built. The verdict beside each department is one word from a closed list, chosen by matching the Chief of Staff’s note against fixed patterns. None of the note is copied, and the days since a report are counted from the report’s date and the briefing’s. The list, as the file defines it:

• Fresh: A new report landed this cycle.
• Overdue: The last report is past the department's own cadence.
• Stale: The last report is old, and no missed cadence is named.
• No new report: No new report this cycle; the last one still stands.
• On cadence: The department reports less often than the briefing and its last report is inside its window.
• Observer: An observer seat, reporting on its own cadence.
• Dormant: Dormant on purpose; no operating cadence yet.
• Unclassified: The Chief of Staff's note matched nothing in this vocabulary. None of it is copied; the report date and the day count, when present, still stand.

No department proposed a lesson that week. A lesson is a rule a seat proposes from its own run.

The caveat the file carries, printed here verbatim:

Only the header of the briefing is published. The one-line summary and every line of the body stay in the private record, because they name security findings before they are patched, legal questions before they are answered and finances before they are closed. Nothing the Chief of Staff wrote about a department is copied: each verdict is chosen by matching its note against a closed list, the days since a report are computed from the report's date and the briefing's, and a note that matches nothing is published as unclassified with none of its text.

pulse.json contains the pulse verdicts, counts, public names and ladder criteria. briefing.json contains the briefing header. Both use CC BY 4.0 with attribution. The site rebuilds these files from committed records, so each date belongs to its source record. The pulse's caveat follows verbatim:

The pulse reads GitHub Actions run outcomes, open pull requests, the org's run-truth and halt-attribution guards, freshness monitors, the live cron table and API command attribution within its reported run window. The API axis states its cancellation actor limit beside its verdict. An axis the pulse could not read is NOT OBSERVED, never clean. Founder directive subjects, pull request titles, halt commits and reasons an axis was not observed stay in the private record. This file carries verdicts, counts, public names, API finding details, unread entries and stated limits.