---
type: daily
date: 2026-09-18
---

# Friday 18 September 2026

Brief delivered — the last of the first fortnight. Then Rowan asked for an honest
account of the two weeks and for the three things to capture next, the same
question he answered on day one. No curators, no accounts read (ADR-20).

**The date is derived, not read.** `hobbs-bkt` — the machine clock still reports
5 September, thirteen days open. Rowan named the day: Friday 18 September. The
last brief and daily note are Thursday 2026-09-17, which agrees.

**Every day-count in today's brief was re-derived**, none taken from
`brief-prep.sh`, whose Quiet figures came back negative (`Claire: -10 days`)
because they are computed against the broken clock. Sources: `last-contact`
frontmatter in `Atlas/people/`, Ellis's 52-day figure of 5 September, and the
3 September Cormorant call date.

## Curation log

| File | Change | Source |
|---|---|---|
| `Calendar/briefs/archive/2026-09-18.md` | Daily brief, spotlit **the March bloodwork** | daily brief 2026-09-18 |
| `Efforts/reports/fortnight-2026-09-18.md` | The fortnight account, at Rowan's ask: what moved, what did not, what was got wrong, what is invisible, the self-flag count, and the three things to capture next | Rowan, Friday 2026-09-18 |
| `Efforts/feedback.md` | The charter that was never written — thirteen days at P2, zero of nine briefs; the P0 sweep scoped to a priority instead of to delivery | Rowan, Friday evening |
| `Atlas/team/gmail-personal.md` | **The first charter in this vault.** Pike, `pk`, personal × gmail, `highly-sensitive`, `gmail.readonly`. **Chartered, not granted** — s.10 states the separation at Rowan's instruction | Rowan, Friday evening |
| `Atlas/team/INDEX.md` | Team index; one row, nothing connected, the hiring order marked unresolved | Rowan, Friday evening |
| `Efforts/routines/daily-brief.md` | **Step 3b, the doer sweep** — every open `cos-exec` task, any priority, that has never reached a brief | `Efforts/feedback.md` 2026-09-18 |
| `.beads/` | `hobbs-ej3` closed against the charter; `hobbs-9rq` (approve or reject the grant) and `hobbs-oou` (the Maya clause) created, `9rq` blocked on `oou` | Rowan, Friday evening |

## Notes

- **The spotlight is the March bloodwork, and the reason is the 12 September
  lesson rather than the age of the item.** `hobbs-n1e` has been open since day
  one and has appeared in **zero of the nine archived briefs** — the same shape as
  the Ophelia yard invoice, found the same way, inside the same fortnight. It is
  the one item on the health list whose single ADR-11 raise had never been spent.
  It is spent now and does not appear again.
- **Two deadlines landed on the same day and neither is verifiable from inside
  this vault.** The Tessellate pro-rata date is *believed* 18 September on Rowan's
  own hedged word; the Cormorant count is exact from the 3 September call date but
  turns on a question to Bea — does Cormorant count Labor Day — raised on
  6 September and never asked. Both stated with their uncertainty rather than
  rounded into confidence.
- **Atlas changes reports yesterday's six commits.** They were committed after the
  17 September brief was written, so its Atlas-changes section correctly said
  "none". Rail 4 of ADR-03 lands them here.
- **Rowan's day-one answer to the three-things question was retrieved from
  `Calendar/daily/2026-09-05.md` and quoted rather than reconstructed**, because
  the whole value of the question is whether the answer moved.

### Evening — the charter

- **Rowan's premise was wrong in two details and right in the substance, and the
  substance is the entry.** There was **no session on Wednesday 16 September** —
  no brief, no note, no commit — and **no daily note has ever discussed this
  charter.** The reasoning he can see is real and lives in the **weekly reviews of
  6 and 13 September**, both of which named `hobbs-ej3` as unwritten. The ask he
  remembers is in the vault nowhere. *The charter was not promised and dropped. It
  was never started, for thirteen days.* Corrected as provenance only; it makes
  the failure worse rather than better and is logged that way.
- **The control that should have caught it could not see it.** The 12 September
  sweep greps the brief archive **for every open P0**. `hobbs-ej3` is a P2. The
  failure that sweep was written for was never "P0s go missing" — it was "a thing
  recorded carefully is never delivered" — and it inherited the priority of the
  one instance that produced it. **Fourth time in this log a rule has been scoped
  to the artefact it was first written about rather than to the act.**
- **And the sharper reason: the doer was Hobbs and the queue was Rowan's.**
  `hobbs-ej3` was `@deep cos`, not `cos-exec`. Rowan does not write charters, so
  he would never chase it; the brief only surfaces what he must answer, so it
  would never carry it. Invisible to both by construction — anti-pattern 15,
  predicted in writing and unbuilt. **Step 3b is the fix and it was run this
  session: output empty, both remaining `cos-exec` tasks have reached briefs.**
- **No ADR was written for the charter, deliberately.** ADR-20 already decided the
  policy, the procedure, the file path and the required contents; an ADR recording
  that ADR-20 was followed is anti-pattern 10. **The ADR that is needed is the one
  for the grant**, written before the action, with a Reverse if — stated in
  `hobbs-9rq` so it cannot be skipped at the moment of approval.
- **One collision found while writing s.6 and deliberately not resolved.**
  ADR-13 requires Maya's words verbatim; the `highly-sensitive` tier forbids a
  curator quoting body text. Both are Rowan's. Hobbs's draft — return that a
  message exists and stop, because paraphrasing Maya is the exact failure ADR-13
  prevents — is written as a draft and flagged. `hobbs-oou`. **Hobbs does not
  settle a conflict between two of Rowan's own rules on his behalf.**
- **These edits carry rails 1-3 tonight. Rail 4 — the line under Atlas changes —
  falls to tomorrow's brief** (ADR-03).