Skip to content

RFC-0011 — Partial-sangha resolutions and participant-scoped binding

  • Status: Accepted
  • Track: G4 (governance)
  • Relates to: RFC-0010 (explicit baselines), RFC-0009 (participants = tips read), CONSTITUTION.md Articles II / VI, sangha/PROTOCOL.md, docs/sangha-resolution-flow.md
  • Versioning layers touched: template (protocol commentary); reuses RFC-0010's baseline declaration
  • Downstream reference case: one derived repository, 29 resolutions — measured, see below.

Decided 2026-09-04: the declaration binds, not the merge. Proposal 1 is settled in the form below and the rule now lives in sangha/PROTOCOL.md, enforced at Info by elector-holds-unadopted. Every RFC in this repository carries Status: Draft and there is no Accepted state to move to, so settlement is recorded here rather than in the field — inventing a lifecycle for one RFC would say more about process than about this decision. The three questions the amendment added are answered under What this decides, below.

Amended 2026-09-04. The premise holds and is stronger than the original claimed: partial resolutions are not merely permitted, they are 9 of 29 in the one repository running a sangha, and the repository's owner is the abstainer in four of them. What does not hold is the closing claim that scoped binding "forecloses the failure mode where a subset silently rebases the collective." It forecloses nothing. All nine partial resolutions already bind their abstainers — absorbed through merge main, with no adoption act and nothing recording that it happened. The proposal below is therefore a change to what is true, not a guard against something that might become true, and it has to reckon with a workflow already going the other way.

Summary

A subset of electors can resolve while others abstain, and the constitution never says whether the resulting rigpa binds the abstainers' baseline. This RFC proposes: a partial resolution is baseline-of-record for its participants only; abstainers keep their prior baseline until they adopt, and adoption is an explicit act — RFC-0010's baseline declaration supplies it. Two RFCs, one field.

The measurement changes what that proposal costs. It is not a rule that formalises current practice; it contradicts it. Nine of nine abstainers are bound today, silently, by the ordinary merge main the protocol prescribes for a different purpose.

What was measured

One derived repository: 29 resolutions, 3 registered electors, 126 elector commits, read-only.

1 — Subsets are the real case, not a hypothetical. Of 29 resolutions, 9 read fewer than the seats registered on that date. Discounting the 12 that predate the third elector's registration, this is a live pattern rather than a startup artifact — and it runs both ways: ma/advocate abstained from 5, ma/goedelsoup — the owner — abstained from 4.

2 — Every abstainer is already bound. For all 9 of 9, the abstaining elector's branch contains the settlement. Six of them also contain the rigpa/* tip; the other three contain the settlement through main while the branch was never merged. There is no case of an elector holding a baseline that excludes a resolution they sat out.

3 — It happened without an adoption act. Across 126 elector commits: 1 adopt: commit, 72 merge main commits. The mechanism that bound the abstainers is the same merge the protocol prescribes in step 3 for reading what the others filed mid-loop.

4 — And the gate cannot see it. Those merge subjects are authored (merge main — the baseline through property-claims), so they are not the git-generated form the commit check exempts. They are outside the closed vocabulary — no-verb — and main..ma/auditor reports 30 such findings. The default range never walks an elector branch, so a workflow used 72 times is invisible to CI.

Problem

Subsets are normal. Resolution triggers on a question "sufficiently explored across ≥2 ma/* branches"; nothing requires the whole sangha, and the "participating ma/* branches" phrasing throughout presumes non-participants exist.

Binding is unaddressed, and the documents cut both ways.

  • Against global binding: Article VI — the sangha "exercises the minimum authority needed", "positions that do not conflict are inherited", resolution "focuses on genuine tensions, not on imposing uniformity" — and an elector's branch "may diverge freely from rigpa/*... Divergence is normal and expected; it is not a violation."
  • Toward global binding: "The new rigpa/<evolution> is the active baseline", read as one global fact.

And practice has already decided, without deciding. The reading that won is global binding, by default, through a merge nobody performs as an adoption. That is the worst of the three outcomes: it is not the scoped rule Article VI argues for, it is not a declared quorum rule either, and it leaves no record of having happened.

Proposal

1 — Binding is participant-scoped. A rigpa is baseline-of-record only for the electors whose tips it read. Abstainers are not bound; their declared baseline (RFC-0010) is unchanged by a resolution they were not part of. This is Article VI applied literally, and Article II denying a subset the power to impose a baseline on the whole.

2 — Adoption is an explicit act. An abstainer adopts by re-declaring their baseline to it. Until then they keep their prior baseline.

3 — Not bound ≠ invisible. The rigpa is a permanent, legible evolution for the whole sangha: anyone may read, cite, adopt or build on it. "Not your baseline yet" is a statement about measurement, not visibility.

4 — What to do about the nine. They stand. A resolution absorbed a year ago is not undone by a rule written today, and rewriting elector history to un-adopt them would breach Article III far more seriously than the silent adoption did. What the rule changes is what happens next, and the honest migration is that the declaration starts from where each branch is.

What this decides

Merging main binds nobody. It is how the corpus stays shared and step 3 requires it mid-loop, so it cannot be the act that binds without making the ordinary workflow illegal. The Baseline: trailer from RFC-0010 is the adoption act, and it is the only thing that binds.

That is what makes scoped binding enforceable in a repository with a shared main, and it costs the nine nothing: they hold what they hold, no history is rewritten, and the rule changes only what being measured against something means.

An abstainer's baseline is what they last declared. A branch that holds a settlement it has not declared is measured against an older evolution while its tree carries a newer corpus. That is an expected state, not a defect — it is what every branch is in today, and saying so plainly is cheaper than leaving a reader to wonder.

Holding without adopting is reported, at Info. elector-holds-unadopted names, per elector, the resolutions its branch holds, took no part in, and its declaration does not reach. Info because none of them is wrong: an elector who has absorbed a resolution and not declared it has done nothing the protocol forbids. It is a prompt to decide, and it clears by declaring.

And a resolution older than the seat is inherited, not abstained from. An elector cannot have sat out a settlement that predates their registration, so the report is bounded by the commit that first put their branch in electors.md. Measured, this is not a rounding detail: it takes one elector from 13 reported resolutions to 4.

One case the boundary decides that neither RFC anticipated. PROTOCOL allows registering an elector in the first resolution they participate in, and the measured repository did exactly that once — ma/advocate's registration and the electoral-purpose settlement are the same commit. That resolution's tips: do not name the new seat, so a naive reading calls it an abstention. It is not: you cannot have sat out the resolution that seated you.

What this proposal had to answer, and did not

The original treated scoped binding as the obviously-correct reading with no cost. Measurement says it has one, and the RFC is not ratifiable until it is answered:

  • merge main binds you to everything on main. An elector merging the baseline picks up every settlement, including resolutions they abstained from. Under scoped binding, is that merge now wrong? It cannot be — it is how the corpus stays shared, and step 3 requires it mid-loop. So either the declaration is what binds (and the merge does not), or scoped binding is unenforceable in a repository with a shared main. Lean: the declaration binds. Holding a commit is not the same as being measured against it, and separating those two is the whole content of RFC-0010.
  • Then what is an abstainer's baseline, concretely? If they hold the settlement but have not declared it, they are measured against an older evolution while their working tree contains a newer corpus. That is a coherent state and an odd one, and the protocol should say plainly that it is expected rather than leaving a reader to wonder whether it is a defect.
  • Is a non-participant who holds the settlement worth reporting? It is decidable — 9 of 9 today — and reporting it is not the same as calling it wrong. Lean: report it once the declaration exists, so that "bound without adopting" is visible; before then there is nothing to compare against and the finding would be every abstainer, every time.

Migration & compatibility

Template-layer protocol commentary; no new schema — it reuses RFC-0010's baseline declaration. Reword PROTOCOL's "the active baseline" to name participants, and add an adoption note. 29 resolutions and 9 already-bound abstainers exist; the original's "existing resolutions are unaffected (none exist)" was written before any did.

Alternatives considered

  • Global binding by default. Rejected as a rule, and it is what practice does. A subset imposing a baseline on non-participants contradicts minimal authority (VI) and epistemic equality (II). If a domain wants it, it should say so — see quorum below.
  • Rigpa is purely advisory. Rejected: the participants who synthesized want a shared baseline; producing one is the point of resolving.
  • Quorum makes binding global. Offered as an opt-in: a domain may add a quorum article so a resolution meeting quorum binds the whole sangha. Given that 9 of 9 partial resolutions already bound everyone, a domain adopting quorum would be ratifying its own practice rather than changing it — which is a legitimate thing for a domain to do and an argument for making the extension easy to reach for.
  • Leave it undecided. Rejected, and this is the alternative measurement kills. Undecided did not produce caution; it produced global binding with no record.

Open questions

  • Later conflict. If an abstainer's un-adopted position conflicts with the participant baseline at a subsequent resolution, is that an ordinary tension? Lean: yes — it becomes an open-question node or a new synthesis like any other.
  • Stale baselines. Does a long-abstaining elector's baseline expire? Lean: no forced expiry — divergence is normal (Article VI), and staleness is visible from the declaration.
  • Who counts as a participant? Decided: tips-read. Measured, it is also the only available definition — nothing records who was notified, so "notified and declined" is indistinguishable from "never asked". The distinction cannot be drawn from the record as it stands, and inventing a notified: field to draw it would be recording an act nobody performs.
  • The vocabulary question this turned up. Decided: neither. merge main — … is used 72 times and is outside the closed vocabulary, and the reason it does not need a verb is the decision above — that merge is not an adoption, so nothing about it wants naming. A baseline merge is written with git's own subject, which the commit check already exempts, and carries the Baseline: trailer that does the binding. The 72 existing commits stay as they are; no history is rewritten to satisfy a rule written after them.