Files
stack/docs/plans/2026-09-27_goals-review.md
T

83 lines
3.7 KiB
Markdown

# Goals review, 2026-09-27
Jason asked whether the north star and goals need a review. Sage's answer
is yes. This page gives the evidence, the north star and goals, and
what changes now. Jason ratified the north star and goal order on 2026-09-27.
Everything else on this page is a lead decision (item 27) and is already
in effect.
## Evidence
No file in the repository states a north star. AGENTS.md tells agents to
resolve ambiguity "in favor of the mission, goal, north star", but none is
written down. QUEUE.md says Jason is preparing the target for the next
phase.
The one measured goal is QUEUE row 7: under 10 human messages per closed
issue. The ledger (`packages/ledger`, run 2026-09-27) gives:
| Week (UTC) | Issues closed | Median hours open | Human messages per closed issue |
|---|---|---|---|
| 09-06 to 09-12 (Gate D) | not recorded | not recorded | 18.9 |
| 09-13 to 09-19 | 3 | 7.5 | 47.7 |
| 09-20 to 09-26 | 1 | 283.3 | 33.0 |
The number went the wrong way. Most of the cause is that almost nothing
closes. #1503, #1507, #1508 and #1509 have been open for two weeks.
Pieces are opened, reviewed for several rounds and partly landed, but
not finished.
The review loop costs more than the defects it catches, at least in brief
work. The CHAT-03 brief is over 1,100 lines. It has had three rounds, and
its reviewers found nine blocking issues, all about what pinned Pi can
prove about its own queue and turns, for a piece that is fixture-only.
The findings are real, but the piece has grown past what Gate E needs.
The lead's own time also drifted. On 2026-09-26 a large share of Sage's
session went to a SetSpark service fix in another repository. It was
delegated and worth doing, but it isn't this project's goal.
## North star (ratified 2026-09-27)
Jason states what he wants. The agents take it from request to a
verified, committed result on the new stack, and Jason isn't the
messenger between them.
The measure is the ledger: human messages per closed issue, and the
number of issues closed per week. Both are already computed.
## Goals, in order
1. **Close what is open.** Finish before starting. Each open issue gets
either a closing piece or a written reason it stays open. The target is
#1509 rows 23 to 25 (live checks only), #1512 and #1511 (landed) and
#1503 (MVP ruling).
2. **Queue as data (#1508, Gate G).** Required, and Jason called it
iron-clad. A1 has landed and A2 is under way. This is what lets a seat
start work with no human message, which is the north star's mechanism.
3. **Gate E by the shortest path.** Every seat interactive from the
Console, then Jason's workday ruling. CHAT-03 is rescoped to what Gate E
needs, see below.
4. **Retire the fleet (row 8).** On Jason's word, after Gate E.
## What changes now (lead decision 27)
- **CHAT-03.** Dewey finishes r3, because the findings describe real
limits of pinned Pi that any design has to respect. After that, no
source work starts until Sage rescopes CHAT-03 against Gate E. The
default for anything pinned Pi can't prove is to refuse or report
"unknown". Building more machinery to prove it isn't the default.
- **Review budget.** A brief gets two review rounds. If blocking findings
remain after round two, Sage cuts scope instead of opening a round
three. The CHAT-03 r3 already in progress is the last exception.
- **Other projects.** Work for another repository goes to that
repository's own seats. The SetSpark deploy was handed to the
shared-signals stack operator, not done by Sage.
- **Weekly check.** Sage runs the ledger every Monday and records the
number in QUEUE row 7's log.
## Ratified
Jason ratified this page on 2026-09-27 ("goals-review doc looks good"):
the north star sentence and the goal order stand as written.