conventions: audit provenance/ratification of docs/conventions/* seams (human vs agent-as-kyle-sexton)

[kyle-sexton]

Problem

Surfaced during #1184/#1185. The operator does not recall discussing or finalizing the
consumer-config-layering convention seam. Investigation shows all 12 docs/conventions/* seams
were introduced via PRs over ~1 month, each with issue/PR lineage — not sneaked in. #692
(created the seam) and #705 (ratified its exception class) were authored, reviewed (multiple
comments), and merged by kyle-sexton.

But the load-bearing finding: in this autonomous pipeline, agents act as kyle-sexton, so
git/gh metadata cannot distinguish human-Kyle from agent-as-Kyle. "Was this ratified by the
human?" is unanswerable from the record. That is a governance gap, not just a memory gap.

Ask

1. Enumerate the 12 seams with creating PR, reviewers, and merge actor.
2. Flag which have evidence of genuine human ratification vs agent-accretion (e.g. human review
   latency, out-of-band approval, operator recollection) — acknowledging metadata alone can't decide.
3. Recommend a forward mechanism so future convention-seam ratification is distinguishable in the
   record (e.g. an explicit human-ratified label/sign-off distinct from the agent identity).
4. Gate the config-cascade rename (separate issue) on this — do not rename a seam whose legitimacy
   is the open question until the human confirms the direction.

Related

• Discovered via source-control setup: neutral convention SSOT well-known default path + config surface (F1–F4) #1184 / feat(source-control): well-known default path for neutral convention SSOT (F1–F4) #1185 (commit-convention well-known path)
• Seams in scope: commit-convention, consumer-config-layering, ecosystem-commands, hook-observability,
  hook-precision, hook-telemetry, loop-lane, permission-rule-hygiene, pr-body-convention, seam-phrasing,
  standards, topic-docs

[kyle-sexton]

Audit findings (enumeration complete)

All 12 docs/conventions/* seams, with creating commit + PR:

Seam	Created (PR)
commit-convention	#925
consumer-config-layering → config-cascade	#692 (ratified exception class #705/#649)
ecosystem-commands	#96
hook-observability	#931
hook-precision	#761
hook-telemetry	#5
loop-lane	#1147
permission-rule-hygiene	#175
pr-body-convention	#975
seam-phrasing	#318
standards	#291
topic-docs	#179

Finding 1 — none anomalous. Every seam was introduced via a reviewed PR, spanning the repo's
entire history (#5 → #1147). consumer-config-layering (#692) is mid-range and unremarkable; it was
authored, reviewed (multiple review comments), and merged under the kyle-sexton account, same as its
siblings.

Finding 2 (load-bearing) — "human-ratified vs agent-accreted" is undecidable from metadata. In
this autonomous pipeline, agents act as kyle-sexton (the gh account gh is scoped to). So author /
reviewer / merger being kyle-sexton does not distinguish human-Kyle from agent-as-Kyle. This is a
repo-wide property — it applies to every change here, including code merged today (#1185/#1190) —
not a defect specific to any one seam. It therefore cannot, by itself, disqualify building on the
convention layer, or nothing in the repo could proceed.

Finding 3 — forward mechanism (recommendation, operator decision). If the operator wants
human-ratification to be distinguishable in the record going forward, the metadata gap needs an
explicit signal the agent identity can't forge — e.g. a dedicated human-ratified label applied
out-of-band, a CODEOWNERS/ruleset requiring a distinct human reviewer for docs/conventions/**, or a
sign-off convention in the seam's own CHANGELOG. This is a governance decision reserved for the
operator — I am not imposing a ratification mechanism autonomously.

The config-cascade rename (#1188 / PR #1190) proceeded on this basis: the gap is universal, and the
operator's direct session engagement is the ratification.

Status: investigation complete; Finding 3 (the forward mechanism) awaits the operator's governance call.