---
type: daily
date: 2026-09-13
---

# Sunday 13 September 2026

No brief. The **second weekly review**, 17:05, two hours before Ruth's call
(ADR-15). Rowan brought one thing: **he has started the Bedrock CEO search.**
That was filed before anything else was done with it (constitution rule 1), and
the ADR that governs it was written before the acts it authorises (s.5).

**Step 0 did not run.** Bea's Sunday email was not opened. Recorded as a gap, not
as an empty list.

No curators, no accounts read (ADR-20).

## Curation log

| File | Change | Source |
|---|---|---|
| `Atlas/decisions.md` | **ADR-24** — ADR-18's Reverse if has fired; the search has started; "The real reason" is kept as written, not struck | Rowan, Sunday |
| `Atlas/ventures/bedrock/VENTURE.md` | The Marcus call filed **verbatim and without comment**, as asked | Rowan, Sunday |
| `Atlas/ventures/bedrock/VENTURE.md` | **Correction:** the task handle in the commit above was invented; re-derived as `hobbs-p9u`. "No CEO search" retired in Open threads | this session |
| `Atlas/people/Marcus Feld.md` | Asked twice over two years, answered by being asked for something; the ball is now his | Rowan, Sunday |
| `Atlas/people/June Park.md` | "There is no search" is no longer true; **whether she has been told is not recorded and is not assumed** | Rowan, Sunday |
| `Atlas/ventures/norton/VENTURE.md` | **Correction:** `hobbs-3v9` does not exist and never did; the 19-20 September ask is `hobbs-1dk` | this session |
| `Efforts/feedback.md` | **[SELF-FLAGGED]** fourth and fifth invented handles; SL-01 read this session and did not fire | this session |
| `Efforts/feedback.md` | **SL-01 clause 5** — the write-time clauses cannot see what is already on disk | this session |
| `Efforts/routines/weekly-review.md` | **Step 1.7**, the handle sweep against the task graph; reports, never auto-fixes | this session |
| `Efforts/cos-guides-seen.json` | Third consecutive failed fetch, and **not for the assumed reason**; read from the on-disk artifact instead; seen-map deliberately not written back | this session |
| `Atlas/ventures/fund-lps/VENTURE.md` | Cormorant re-derived from the 3 September call date: **four or five**, not nine | this session |
| `Atlas/ventures/money/VENTURE.md` | Bea's email not opened this week, recorded as a gap; yard bill at 60 by count | Rowan, Sunday |
| `Atlas/ventures/maya/VENTURE.md` | The 26-27 September booking list delivered, `hobbs-0si` closed; splash dependency cleared; "where north" spun out as the gating decision | Rowan, Sunday |
| `Efforts/reports/weekly-2026-09-13.md` | The review | this session |
| `.beads/` | `hobbs-nsj` and `hobbs-0si` closed; `hobbs-p9u` (waiting on Marcus, follow-up 20 Sep) and `hobbs-2g6` (where north) created | Rowan, Sunday |

## Notes

- **The record he asked for, and the two words that shaped it.** *"Don't
  editorialise, just record that the thing the note says I didn't want to do, I
  did."* The Bedrock section is his three sentences, a bullet naming what was
  called and when, and a bullet naming **what is not known and is not assumed** —
  what he asked Marcus for beyond introductions, what Marcus said, whether a date
  was set, whether June knows. No comment on the two years. He named the gap
  himself in his own sentence; adding to it would be the editorialising.
- **"The real reason" was not touched, and that is the whole of ADR-24's first
  clause.** ADR-18's Reverse if said *superseded, not deleted*. The reason a named
  avoidance is worth keeping is that it is now the thing the act can be measured
  against. Struck-through prose would have made the file tidier and the record
  worse.
- **What "properly, not as a task" was read to mean, stated because it was a
  judgment call.** Read as: *he did the thing, rather than adding it to a list* —
  not as an instruction that no task may exist. ADR-18's own Reverse if says the
  opposite in Rowan's pre-approved words: *"then it becomes an ordinary
  concrete-action task."* One task was created, `hobbs-p9u`, flagged `waiting` on
  Marcus with a follow-up date, because the ball is genuinely his. **The reading
  is flagged in the review rather than buried; it is one word to reverse.**
- **June's note is the edit with the most in it and the least said in it.** It
  asserted "there is no search" as a fact and that fact is now false. What
  replaced it is the correction plus one sentence: she has not been told, and that
  is not a recommendation. She is on the Quiet list, she is carrying Kettle River
  and its counterparty alone, and the post being filled is the one above her.
- **The handle sweep found something on its first run, which is the argument for
  it.** `hobbs-3v9` had been in the Norton file since 6 September — through a
  weekly review, five daily briefs, and two later edits to that same file — while
  the correct handle `hobbs-1dk` sat four sections below it. No amount of care had
  found it; one `comm -23` did.
- **And the sweep's own session produced a sixth instance.** `hobbs-mkl` was
  written into the Bedrock note today, in the same commit as Rowan's verbatim,
  for a task that did not exist. **SL-01 was read in full at the top of this
  session.** The lesson is not "be more careful" — it is that SL-01 is a rule that
  fires only when consulted, which is the exact critique SL-01 makes of the two
  lessons it replaced. The 12 September session reached the same place about a
  different failure and answered it with a command. This one is now step 1.7.
- **The guides feed, and the thing worth correcting.** Rowan's read was that the
  two failures last week were the server. **The server is up** — `workerd` on
  127.0.0.1:8788 — and returns **404 on every path probed, including `/`.** That
  is a different fault from a placeholder URL, and `hobbs-21n` is correct but no
  longer the whole story. The repo it serves is not the vault and is not Hobbs's
  to change (authority row 13), so it was diagnosed and reported, not fixed. The
  32-guide build artifact was read from disk and diffed by hand, and **Section 12
  says in its first line that it is not built from a fetch.**
- **Eighteen guides surfaced and none marked seen.** Writing the seen-map back
  would bury eighteen unread guides the moment they were surfaced once — "routed
  is not done," anti-pattern 15. They surface again next week if he does not
  dispose of them, which is correct behaviour and not a bug.
- **The staleness rule fired for the first time and caught the right two.** Not
  the four eight-day ventures — `rivet-yard`, `home`, `cos`, `bedrock` — but the
  two at the seven-day boundary: `fund-lps` and `money` hold the only hard money
  deadline this week, went a whole week without an edit, and `fund-lps` was still
  carrying the sentence *"nine business days remaining"* on the page today. **The
  rule keys off note edits and never task counts, which is why it saw them.**
- **Cormorant is the sharpest number in the review.** Nine either way on
  6 September; **four or five** today, depending on Cormorant's Labor Day reading.
  A week ago that ambiguity decided which day was last. It now decides the number.
  Re-derived from the 3 September call date, never decremented from last week's
  figure (SL-01).
- **Twelve "disposition needed" lines, and that is named in the document as being
  too many.** Three were marked as the ones to answer if he answers only three:
  Bea's email, the Cormorant day, and where north.
- **One thing said once, per s.8, and not repeated:** the search started. Two
  years. It is in the review's opening and nowhere else in the document.