Skip to content

Frame cycle 03 — a key given and taken back, and the demand question for the third time - #28

Merged
napkinstack-agent[bot] merged 1 commit into
mainfrom
cycle/03-framing
Sep 23, 2026
Merged

napkinstack-agent[bot] merged 1 commit into
mainfrom
cycle/03-framing

Conversation

@napkinstack-agent

Copy link
Copy Markdown
Contributor

Frame cycle 03 — a key given and taken back, and the demand question for the third time

status: proposed. Three weeks, three deliverables, one of them somebody else's calendar again.


Module

None yet. This pull request changes no module: it adds docs/project/cycles/03-a-key-given-and-taken-back.md.

copying is created when this cycle is accepted, not here:
nstack new-module copying NapkinStack/maintainers critical, declared user_facing — a member
sees what it does.

  • This pull request touches a single module — n/a, it touches none
  • ⚠️ Cross-module

Deliverable

Out of cycle: framing the next one.

What it proposes

The goal, in one sentence: find out whether anyone wants this, and make the key a member
hands flet safe to give and safe to take back.

Deliverable Module
D1 Five private invitations, sent within 3 days, read at day 14. Refuted at 0 of 5; nothing confirms none — a spike
D2 A member can hand flet a key that places orders and nothing else, and take it back copying
D3 A member is told what they are getting into before they act, not after copying

D1 on day one. D2 before D3 — the notice guards an action that has to exist before it can be
guarded. D2 and D3 are both copying, so they are sequential: one pull request per module.

No order is copied in this cycle. That is stated in Out of scope rather than left implicit,
because it is the obvious thing to reach for once a key exists, and reaching for it is how a
three-week appetite becomes five. The first copied order is the next cycle's deliverable, and it
is what produces the charter's zero unchosen divergence criterion.

What D2's criteria are built to refuse

Every one of them names the venue as the thing that answers, not our own state. That is
deliberate, and it is the whole design of the deliverable:

A key we believe is scoped, a revocation we record as done, a retry we think stopped —
those are the three ways this module is wrong while every test we own is green.

So the scope and the expiry are read back from the venue rather than asserted by us; revocation is
checked by attempting an order after it; the refusal of a withdrawal, a transfer and a
permissions change gets one test each; the absence of action after revocation is asserted rather
than assumed; and the key's confinement is checked by searching the logs, the process output and
the repository for it.

Those are the two non-negotiables the handover has carried since cycle 01 — the key places orders
and nothing else, and revocation works — written as tests rather than as intentions.

Verification

critical means adversarial rounds without a cap, until the verifier signs (ADR-0003). That is
not the budget cycle 02 was given, and the ADR says why: a wrong figure in a screening report is
visible to an administrator who can re-run it, while a wrong order here is a member's money.

The framer's note carries cycle 02's transferable lesson into the briefs: state the general form
first.
A guarantee written per instance produces one refusal per round — cycle 02 spent six
confirmation runs learning that, three of them on sentences rather than on code.

The decider's choices, recorded

Three sizes were offered and the middle one was chosen:

Offered
2 weeks — the key alone Would leave the pilot FAQ's pre-action notice unhonoured
3 weeks — the key, its revocation, the notice Chosen
5 weeks — through to a first copied order A long appetite on critical code

D1 is in, on day one, for the third time. Recorded because cycle 02's closure said that
deferring it a third time would be the project's answer about what it is really doing, and the
answer given was the other one.

The failure mode to watch on day three

Cycle 02's appetite was sized for work cycle 01 had already done, so both engineering deliverables
landed in three days while nineteen went unspent and the one deliverable that needed somebody
else's calendar did not move. Three weeks for a module that does not exist yet is a different bet,
but the shape of the failure is the same: if D1's messages have not gone out by day three, this
cycle is already repeating both of its predecessors.
The note says to look on day three rather
than at the end.


Oracle

  • The success criterion was written before the implementation — the fitness function's
    cycle rules (C3, C4) and the template's required sections
  • It was seen failing, for the right reason — nstack fitness refuses a cycle whose end
    is not start + appetite, a deliverable in an unknown state, a ready deliverable without
    acceptance criteria, and any document still holding the template's words
  • It passes now — nstack fitness exit 0, plan compliant, 3 cycles

Kind of oracle: fitness function

Review budget

  • Files touched: 1
  • Within budget

Contracts

  • No contract affected

Definition of Done

  • Oracle green — nstack fitness exit 0, read on its own exit status
  • Fitness functions — compliant
  • Everything else — n/a, no code
  • Diff self-reviewed line by line

Test sheet

Not required: this pull request changes no module (docs/os/05-workflow.md §7).


Summary

DONE
Cycle 03 framed and proposed: three weeks, D1 carried unchanged onto day one, the agent key with
its revocation, and the pre-action notice. The module's creation command is named, the
criticality is named, and what is deliberately not in the cycle is named.

VERIFIED
nstack fitness exit 0 with the plan included, read on its own exit status. One accepted cycle at
a time still holds: this cycle is proposed, and cycle 02's closure is in #27.

ASSUMED
That start: 2026-09-24 is right. If acceptance lands on another day, start and end move in
the acceptance pull request — the playbook says so, and cycle 02 did exactly that.

NOT VERIFIED
Whether three weeks is enough for a critical module that does not exist yet. Nobody knows, which
is why the appetite is a decision about what to spend rather than an estimate of what it will take.
And D1 still depends on messages nobody has sent.

RISKS
The two commitments in D2 and D3 are the ones a member's safety rests on, and they are being built
by a session whose measured habit — recorded in ADR-0003 and demonstrated across cycle 02 — is to
introduce defects inside its own later fixes. That is why critical gets uncapped verification,
and it is the reason to hold that line when it is inconvenient.


Signals to report

  • I had to look at another module's implementation
  • A kernel or playbook rule got in my way for no good reason
  • A quality gate blocked without improving quality
  • A rule applied by hand would be worth automating

Nothing to report. playbooks/framing.md is what produced the three sized options rather than one
invented appetite, and it was right to.

…for the third time

Proposed, three weeks, three deliverables, one of them somebody else's calendar
again.

The decider was offered three sizes and chose the middle: **the key, its
revocation and the pre-action notice, with no copied order.** Two weeks for the
key alone would have left the pilot FAQ's notice unhonoured; five weeks through to
a first copied order is a long appetite on `critical` code. They also chose D1 in,
on day one — the third time it appears there, after two cycles deferred it.

**D1** is carried unchanged: five private invitations, read at day 14, refuted at
0 of 5 and confirmed by nothing. **D2** is the key: created and revoked by the
member, scoped so it places orders and nothing else. **D3** is the notice, shown
before the member acts rather than in a footer.

What the D2 criteria are built to refuse is the point of writing them this way.
Every one of them names **the venue** as the thing that answers, not our own
state. A key we believe is scoped, a revocation we record as done, a retry we
think stopped — those are the three ways this module is wrong while every test we
own is green. So the scope and expiry are read back from the venue, the revocation
is checked by attempting an order after it, and the absence of action is asserted
rather than assumed.

`critical` means adversarial rounds without a cap (ADR-0003), and the framer's
note carries cycle 02's transferable lesson into the briefs: state the general
form first, because a guarantee written per instance produces one refusal per
round. Cycle 02 paid six confirmation runs to learn that.

Out of scope is explicit about the obvious reach: **no order is copied in this
cycle.** That is what turns three weeks into five, and it is the next cycle's
deliverable.

The note also names the failure mode to watch on day three rather than at the end:
cycle 02's appetite was sized for work cycle 01 had already done, so both
engineering deliverables landed in three days while nineteen went unspent and the
calendar deliverable did not move. If D1's messages have not gone out by day
three, this cycle is already repeating both of its predecessors.

`status: proposed`. The decider accepts it in its own pull request, which is where
`start` and `end` move if acceptance lands on another day, and where
`nstack new-module copying NapkinStack/maintainers critical` is run.

`nstack fitness` compliant, plan included: one accepted cycle at a time still
holds, since this one is proposed and cycle 02's closure is in #27.

Agent-Session: 4151570f-613e-475f-ba3f-b465b35d9a83
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@napkinstack-agent napkinstack-agent Bot mentioned this pull request Sep 23, 2026
12 of 17 tasks
@napkinstack-agent
napkinstack-agent Bot merged commit f5f0074 into main Sep 23, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants