Agent-made Track U decisions — the authority, and how to read them
Track U is the human-judgement lane. On 2026-08-25 the project owner stated
that no human is available to work it and handed the decisions to the agents.
Until that is reversed, a decide- ticket answered by an agent is a real
decision, not a recommendation.
Thirteen tickets were cleared in one pass that day. Each carries a
# DECIDED 2026-08-25 section stating the answer and — this is the part that
matters when revisiting — whether it was derived or judged:
- Derived. A stated principle already settled it. The section names and quotes the rule. Reversing it means arguing with the principle, not with the ticket. Most of the thirteen are these.
- Judgement call. No principle reached it, so it was settled on the owner's standing tiebreak: a pragmatic C + Pascal + Python compiler — a tool that compiles and correctly runs real programs; not a conformance trophy, not utopia. These say so explicitly and are cheap to revisit — that is the point of labelling them.
The governing set, in the order it is consulted:
devdocs/dev/frontend-compat-philosophy.md (what "compatible" means per
frontend — C compliance, Pascal a deliberate dialect, NilPy upward-compatible
only), meta-dialect-extensions-and-fpc-strict (lax default, parity behind the
strict family, and "strict mode is to compile valid programs that rely on FPC's
behaviour, not on FPC's bugs"), devdocs/dev/ir-as-substrate.md,
devdocs/dev/normalise-dont-special-case.md,
devdocs/dev/the-substrate-is-ast-and-ir-not-the-parser.md, and
devdocs/dev/root-cause-over-microfix.md.
What was NOT decided
One item was refused as genuinely human-only: the private half of
decide-release-signing-key-custody. An agent must not generate, hold, or place
a release signing key. The agent-work half of that ticket was answered and
unblocked; the custody call is flagged for the owner and nothing else waits on
it.
The one cross-cutting line these established
Three tickets — decide-rtti-kind-numbering, the two decide-*classinfo*, and
decide-vartype-returns-pxx-tags-not-fpc-codes — turned out to be one unstated
question. The answer, now standing policy:
The RTL facade (
typinfo.pas,variants.pas) speaks FPC's public numbering. The compiler's internal tags stay ours and stay private.PxxTkToFPCKind(compiler/rtti_emit.inc:811) is the seam, and it already existed — the policy was half-applied, which is exactly why the numberings were confusable.