---
type: daily
date: 2026-09-23
day: Wednesday
updated: 2026-09-23
---

# Wednesday 23 September

Session opened 06:30 on Rowan's stated time. **Day nineteen.** The machine clock
still reads 5 September (`hobbs-bkt`, seventeen days at P1); every date here comes
from his sentence or from a file.

Brief delivered first.

## Curation log

| File | Change | Source |
|---|---|---|
| `Calendar/briefs/archive/2026-09-23.md` | The eleventh brief. Spotlit: **`fund-lps` — the capital-call alarm**, the first spotlight this venture has had | Rowan, 2026-09-23 |
| `Efforts/dispositions/2026-09-23-brief.json` | Nine asks, written in the same pass as the brief (ADR-29 clause 2) | the brief |
| `Calendar/daily/2026-09-23.md` | This note | — |
| `Atlas/decisions.md` | **ADR-31**, written before the action; then six records moved out and stubs left at every vacated slot; then the header line claiming ADR-22 bound three ventures corrected | Rowan, 2026-09-23 |
| `Atlas/ventures/ashcroft-tool-die/decisions.md` | **ADR-22, ADR-25, ADR-26** verbatim | ADR-31 |
| `Atlas/ventures/bedrock/decisions.md` | **ADR-18, ADR-24** verbatim | ADR-31 |
| `Atlas/ventures/ashcroft-angels/decisions.md` | **ADR-23** verbatim | ADR-31 |
| `Atlas/ventures/halcyon/decisions.md` | **`HAL-01`** — the runout hold, extracted verbatim from inside a narrative section | ADR-31 clause 8 |
| `Atlas/ventures/halcyon/projects/` | `do-160-certification` · `cascadia-contract` · `mk3-shaft-runout` — the three Rowan named | Rowan, 2026-09-23 |
| `Atlas/ventures/ophelia/projects/` | `refit-and-splash` (**done**) · `tomales-passage` · `race-season` | ADR-31 |
| `Atlas/ventures/rivet-yard/projects/` | `permit-resubmission` · `change-order-84k` | ADR-31 |
| `Atlas/ventures/bedrock/projects/` | `kettle-river-pilot` · `ceo-search` | ADR-31 |
| `Atlas/ventures/ashcroft-tool-die/projects/claire-buy-in.md` | Due 30 September, nine missing inputs | ADR-31 |
| `Atlas/ventures/ashcroft-angels/projects/tessellate-pro-rata.md` | Eighteen days, and Nikhil does not know | ADR-31 |
| `Atlas/ventures/*/someday.md` | Six files. Four carry real parked content; two are empty on purpose | ADR-31 clause 2 |
| Six `VENTURE.md` files | A folder pointer each; halcyon's HOLD section replaced by a pointer to `HAL-01` | ADR-31 clause 9 |
| `Atlas/ventures/INDEX.md` | A shape column, and **the eight refusals with a reason each** | Rowan, 2026-09-23 |
| `Efforts/feedback.md` | `[SELF-FLAGGED]` — three files in one commit, against rail 1 | this session |

**Order of work, because the order is the discipline.** ADR-31 was written and
committed **before** any file moved. Nothing in the table above predates it.

## Notes on the day

- **Both mandated sweeps pass.** P0 (routine step 3a): `hobbs-62a` and `hobbs-bg9`
  have both reached briefs. Doer (step 3b): the five `cos-exec` tasks —
  `hobbs-7q3`, `hobbs-56d`, `hobbs-7vi`, `hobbs-wtt`, `hobbs-z4y` — have all
  reached briefs. Run separately, not merged, per the routine.

- **And the sweeps passing is the finding.** A third query — every open task at any
  priority with any doer — returns **22 handles that have never appeared in a
  brief.** Four were opened yesterday. The other eighteen include `hobbs-77i` at 18
  days, which is the standing fix for the one failure Rowan named on day one as the
  thing he is most determined about. **This is the same shape as the two failures
  the existing sweeps were written for, one scope wider each time:** 3a was scoped
  to a priority and missed a P2; 3b was scoped to a doer and misses everything
  Rowan owns that simply never came up. `hobbs-77i` and `hobbs-707` were pulled
  into today's brief by hand. **A hand-pull is not a mechanism** — a third sweep,
  or one that replaces both, needs an ADR and a rule for what to do when it returns
  eighteen items, because "put them all in the brief" is anti-pattern 1.

- **`brief-prep.sh` produced two negative day counts** — Claire at −9, Dmitri at −8
  — because it subtracts a 2026-09-15 last-contact from a machine date of
  2026-09-06. **The clock bug reached the arithmetic, not just the header.** Every
  Quiet figure in today's brief was re-derived from the frontmatter date against
  23 September and none was decremented from yesterday's brief (SL-01). A script
  that returns a negative age should say so rather than print it; noted against
  `hobbs-bkt`, which is now the second thing that bug has corrupted.

- **The cap was missed again: 4,802 characters against ~4,000.** Eighth brief
  running over, and the trend is upward. The routine calls the cap the editorial
  policy, which means eight consecutive overruns is a policy nobody is enforcing —
  either the cap is wrong for a thirteen-venture day or the brief is carrying lines
  that belong in the weekly. **Not fixed today, and not quietly dropped either.**

- **The spotlight went to `fund-lps` and the choice is arguable.** Eleven briefs and
  this venture has never had one, while `home` has had two. The rotation is by
  least-recently-modified venture note, which rewards ventures that get edited —
  and `fund-lps` was edited on the 20th precisely because something went wrong in
  it. **A staleness heuristic that reads a post-mortem as freshness is backwards.**
  Chosen by hand against the archive's `spotlit:` lines instead, which is what
  ADR-19 actually asks for.

- **Nineteen asks were open and unanswered when this brief was written** — eight
  from the 22nd, eleven from Sunday's review — so no line above assumes an outcome
  for any of them. `hobbs-som` in particular was asked yesterday and expires
  tomorrow; today is the last brief that can carry it as a decision rather than as
  a lapse.


## The restructure — what was decided and what it cost

- **Six ventures took the folder, eight did not**, and the test is in ADR-31
  clause 1 so the fifteenth is decided rather than argued. **The refusal list is
  the part Rowan asked for** — *"that list is more interesting to me than the list
  you did"* — and it is in `Atlas/ventures/INDEX.md` as well as in the ADR,
  because the ADR log is read every session and the index is read when someone is
  looking at ventures.

- **Four of the eight refusals rest on an ADR that already exists**, which is the
  finding worth keeping: `maya` on ADR-06, `family` on ADR-14, `health` on ADR-11,
  `norton` on its own 20 September entry. **The guide's tree has empty slots in it
  and this vault has four rules that say those slots stay empty.** A structure
  taken off a shelf does not outrank a decision made here.

- **`money` is the one that could go either way and it is recorded as such**, with
  the tipping condition named — a second workstream beside the will, or the ledger
  being built. **Named so it is not re-argued from scratch in a month.**

- **ADR-22 contradicted itself and the contradiction was load-bearing.** Its header
  says it sat in the cross-venture log *"because it binds three ventures."* ADR-25
  reversed decision 1 and ADR-26 superseded decision 3, which leaves decision 2 —
  the Halcyon supply question — as the only cross-venture element, **and it is not
  decided.** A pending cross-venture question does not make a decided
  venture-local record cross-venture. **When it is answered it comes back up as a
  new ADR**, and that instruction is written in three places so it cannot be lost.

- **Numbers did not move and stubs stayed behind.** ADR numbers are identifiers
  cited across venture notes, people notes, briefs and tasks; renumbering six of
  them would have been SL-01's failure with a blast radius. Every vacated slot in
  `Atlas/decisions.md` names the file its record moved to.

- **What this cost, stated rather than left to be discovered.** Twenty-one new
  files in one afternoon, on a system with nineteen days of habit, the day after
  eighteen charters. **Anti-pattern 8 is the live risk and ADR-31's first Reverse
  if is aimed at exactly it:** four weeks with no project note edited since the day
  it was created means the projects were paragraphs after all and the folders
  collapse back.

- **Two decision logs now exist for six ventures.** A reader who greps only
  `Atlas/decisions.md` will miss six records. The stubs are the mitigation and
  they are only as good as the habit of reading them — which is why the second
  Reverse if fires on a record being cited and not found.