Skip to content

T27 closes for the libraries; T28 opens where the sweep found a gap - #253

Merged
omercelikdev merged 1 commit into
mainfrom
thread-sweep
Sep 8, 2026
Merged

T27 closes for the libraries; T28 opens where the sweep found a gap#253
omercelikdev merged 1 commit into
mainfrom
thread-sweep

Conversation

@omercelikdev

Copy link
Copy Markdown
Collaborator

T27 — the delivery cycle reached every named repository

repo how
mockifyr #417 — only the three steps its own loop lacked
qorpe/ui #49 — npm shape
mediant #166 — thin hook, because the build IS the gate
api-portal #74
factoring-poc omercelikdev/factoring-poc#7 — kept implement/test-writer as the authority for story work, added only the defect path
specdrift #13
specanchor #1

Remaining on the thread: coexist's regeneration, then D6's spec chain.

What the new gates found

The row records this, because a gate that has never caught anything is a gate nobody trusts.

specdrift — the README pinned 0.4.1 for its Docker image and Action ref while the csproj shipped 0.4.3, so the run-it-anywhere table was two releases stale. And --help printed specdrift mcp below the paragraph explaining the other two verbs' flags, so the third verb read as a footnote.

specanchor — the ledger called the shipped licence gate "still owed"; the README's Next: named the gates, the CLI and MCP, all three built and running in CI; specanchor mcp appeared in no document at all; and one section was written in Turkish, in the repository whose entire product is an auditable trail read by client-side reviewers.

T28 — nothing keeps the copies in step

The sweep closed T27 and opened the gap underneath it. cycle.md now lives in nine places. scripts/skills-parity.sh compares the three copies inside this repository and nothing compares them across repositories.

That matters because the cycle has already been revised once — #252 made §6 stack-neutral — and every adopter that had already taken a copy would have kept the old text silently. goldpath init does not ship the skills family: template-completion RFC D5 assigns that to the transformation pack, which is not built. So today's propagation mechanism is a person remembering, which is exactly the failure mode the ledgers exist to prevent.

The thread carries the no-extra-spend constraint in its proof column, because that is what rules out the obvious answers (a scheduled cross-repo job, a hosted checker). The proof it demands: a planted drift in one adopter's copy must fail something, without anyone remembering to look.

Gates

ledger-check     every issue referenced by a ledger has the state the ledger claims
docs-freshness   links resolve · no retired names · 23 packages, 13 CLI verbs in sync
skills-parity    16 files identical across 3 copies, 5 deliberate divergences accounted for
schema-honesty   every roadmap-only value is annotated as such
template-pins    templates and adopter docs all on 0.1.0-preview.8

…ft a gap

Every repository named in the delivery-cycle RFC now carries the layer:
mockifyr #417, qorpe/ui #49, mediant #166, api-portal #74, factoring-poc #7,
specdrift #13, specanchor #1. Only coexist's regeneration and D6's spec
chain are left.

The two repositories that carried NO agent layer also carried no docs gate,
and writing one for each found live drift in both — recorded in the row
because a gate that never caught anything is a gate nobody trusts:

  specdrift   README pinned 0.4.1 while the csproj shipped 0.4.3, so the
              Docker tag and Action ref in the run-it-anywhere table were
              two releases stale. `--help` printed `specdrift mcp` below
              the prose about the other two verbs.
  specanchor  the ledger called the shipped licence gate "still owed"; the
              README's "Next:" named the gates, the CLI and MCP, all three
              built and exercised in CI; `specanchor mcp` was documented
              nowhere at all; one section was written in Turkish, in a
              repository whose whole product is an auditable trail read by
              client-side reviewers.

T28 records what the sweep exposed. cycle.md now lives in nine places and
nothing compares the copies ACROSS repositories — skills-parity.sh only
compares the three inside this one. The cycle has already been revised once
(#252 made §6 stack-neutral); the next revision reaches adopters only if a
person remembers, and `goldpath init` does not ship the skills family
because template-completion D5 assigns that to the unbuilt transformation
pack. The thread carries the owner's no-extra-spend constraint in its proof
column, because that is what rules out the obvious answers.

All five gates green.
@omercelikdev
omercelikdev merged commit 40b09ee into main Sep 8, 2026
4 checks passed
@omercelikdev
omercelikdev deleted the thread-sweep branch September 8, 2026 23:31
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.

1 participant