docs: record structural-key registry reopen outcome

Author: lesnik512

planning/decisions/2026-07-12-reject-structural-key-registry.md

@@ -55,6 +55,35 @@ the key-*name* list from each owning module.
   snapshot + disjoint tests already guard loudly (no silent-bug risk). Not worth
   it at this size and stage.
 
+## Reopened 2026-07-15 — reaffirmed
+
+Reopened at the maintainer's request during an architecture review. The registry
+question was handed to **two independent reviewers, each blind to this decision**
+— one mandated to build the strongest case *for* a structural-key registry, one
+*against*. They converged on this decision's holding: a universal registry is a
+false seam, and `secrets`/`configs` (`stores.py`), `dns`/`sysctls`/`extra_hosts`
+(`pod.py`), `deploy` (`resources.py`), `depends_on`/`networks`/`hostname`/
+`container_name` (`graph.py`), and the positional `command`/`entrypoint`/`image`
+argv keys all belong in their owning module. The clinching disproof of the
+universal case: a single `entrypoint` value emits tokens on *both sides* of the
+`image` token and a string `entrypoint` silently cancels `command`
+(`emit.py:199-207`) — no `emit(value) -> list[Token]` slot can express that.
+
+The reopen surfaced exactly one crack, and it does **not** reach this rejection.
+Four per-service keys have no owning noun — `environment`, `env_file`, `volumes`,
+`tmpfs` — and of those, `environment` is provably `_map("-e")` (same
+`validate_map`, same `["-e", Expand(str(pair))]` emit loop as `labels`/
+`annotations`, same `pairs_to_mapping` map-merge), fitting the *existing*
+`emit(value)` signature at zero interface cost. `tmpfs` is nearly `_list("--tmpfs")`.
+But `env_file`/`volumes` need `project_dir`, which would force widening
+`KeySpec.emit` to `emit(value, ctx)` across all ~29 current keys to house two —
+a real cost this decision still declines. So the strongest "reopen" collapses
+from *"structural-key registry"* all the way down to *"optionally move
+`environment` (± `tmpfs`) into `SERVICE_KEYS`"*, which is this decision's own
+narrow revisit-trigger #2, not an overturn of it. That narrow move was examined
+and left unscheduled (same category as the option-2 single-sourcing declined
+above); it is recorded here so a third review does not re-derive it from scratch.
+
 ## Revisit trigger
 
 Reopen — and when reopening, reach for the **narrow name single-sourcing**, not