Skip to content

docs(governance): a human may ratify a bounded class of change in advance - #502

Merged
transitrix merged 1 commit into
mainfrom
docs/governance-standing-grant
Aug 18, 2026
Merged

docs(governance): a human may ratify a bounded class of change in advance#502
transitrix merged 1 commit into
mainfrom
docs/governance-standing-grant

Conversation

@vkgeorgia

Copy link
Copy Markdown
Collaborator

What changed

New §2.1 in method/08-governance.mdRatification may precede the change — the standing grant.

§2's asymmetry ("an agent may propose; only a human may commit the repository to a decision") is about who decides. Read as a rule about when, it forces a false choice: either every instance of a repeating mechanical change waits for a person, or a change is in force with nothing having ratified it. §2.1 states the third case — a repeating mechanical change is usually not a decision each time, but the Nth execution of one decision, so a human ratifies the class ahead of time in a standing grant that names the exact change permitted, the conditions that must hold on every use, and an end condition. An agent acting inside a ratified class executes a decision rather than taking one, and the record of each instance is the run rather than a fresh decision record.

The guarantee is unchanged, and is restated at every point a skimming reader lands so nobody comes away with a weaker one:

  • §2's closing rule and §6's summary test now say it directly — every change in force was ratified by a human, the instance or the class, and an agent never supplies that ratification itself.
  • §2.1's own bullets: an agent may not author, widen, or ratify a grant (including its own); a change outside every grant falls back to the default of propose-and-wait, because "no applicable grant" is the ordinary case rather than permission; a grant does not travel to another repository, so an adopter inherits the mechanism and never anyone else's grants.
  • §4 carries the load-bearing condition: a grant's conditions are in force only where a guard evaluates them on every use, so a grant whose guard is missing has not moved a judgement earlier — it has removed one.

§2's table row for methodology/catalog version upgrades is amended to match.

Two files restated the doctrine in a form §2.1 would contradict, and now point at it instead:

  • method/09-releases-and-propagation.md §5 — splits into the default per-instance path (unchanged, including the check-adl.mjs A3 gate) and landing inside a ratified class, with a MAJOR bump or an unmet condition taking the default path.
  • method/07-decisions.md §4 — a ratified class needs no per-instance record; the author: agent gate in the table above is untouched, and the grant itself is an ordinary human-authored, human-accepted, immutable record.

How it is verified

  • node scripts/check-notations.mjs — clean (20 view notations; examples, internal links, version pins and notation counts consistent). Two pre-existing SIZE1 soft-ceiling warnings on 06-team-operations.md and 07-decisions.md are unchanged: this PR adds one ### subsection and no ## sections.
  • node scripts/check-adl.mjs — clean.
  • Documentation only; no script, schema, or notation behaviour changes.

Checklist

  • Based on main, one concern per PR.
  • Commits are signed off (git commit -s) — DCO.
  • No hub/work-item references in committed files, this description, or commit messages.

…ance

§2 draws the authorship asymmetry as a rule about who decides. Read as a rule
about when, it forces a false choice: either every instance of a repeating
mechanical change waits for a person, or a change is in force with nothing having
ratified it. A repeating mechanical change is usually not a decision each time —
it is the Nth execution of one decision.

New §2.1 states the third case. A human ratifies the class ahead of time in a
standing grant naming the exact change permitted, the conditions that must hold
on every use, and an end condition. An agent acting inside a ratified class
executes a decision rather than takes one, so the record of each instance is the
run rather than a fresh decision record.

The guarantee is unchanged and is restated where a skimming reader lands: every
change in force was ratified by a human — the instance, or the class — and an
agent may never author, widen, or ratify the grant it acts under, nor read the
absence of an applicable grant as permission. §4 carries the condition the whole
construct rests on: a grant's conditions hold only where a guard evaluates them
on every use, so a grant whose guard is missing is not in force. §2's table row
and §6's summary test are amended in the same pass.

07 §4 and 09 §5 restated the doctrine in a form the new section would contradict,
and now point at it: a ratified class needs no per-instance record, while the
`author: agent` gate and the default per-instance path are untouched.

Signed-off-by: Valerii Korobeinikov <valerii@korobeinikov.consulting>
@transitrix
transitrix merged commit fd32395 into main Aug 18, 2026
9 checks passed
@transitrix
transitrix deleted the docs/governance-standing-grant branch August 18, 2026 11:42
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