Why this exists as a ticket at all

The wasm branch's whole strategy is move the conflict, don't manage it (feature-a-wasm32-target-registration-skeleton): registration went to master first so the branch would add only new files. PLAN.md named the expected escapes as Phase 4 (VMT fixups) and Phase 5 (exceptions), and set a standing rule — a shared-file edit outside that list is "a surprise worth stopping for".

Phase 1's milestone is "pxx emits a .wasm". Reaching it needs compiler.pas. That is a shared file, at Phase 1. So the lane stopped, built the encoder to be provable without the wiring — a standalone self-test that includes the two new files directly (test/wasm/check_phase1.sh) — and put the routing to the coordinator rather than arguing itself across the boundary on an "it's only three lines" basis. That argument is exactly how the strategy erodes.

The registration commit had in fact anticipated this moment: "that branch adds only new files and never merges a shared one until it emits". It emits now.

What was granted, and the edges

Granted by the coordinator after measuring rather than assuming — git log on both files showed only the landed registration commit 290ee8ca4, with frankA in pasparser_*/defs.inc and frank-optimize in ir_codegen.inc. Both files uncontended. Filed as Track A and self-resolved: the combined-track pattern CLAUDE.md describes, whose only requirement is a coordinator confirming no concurrent A holder.

compiler/compiler.pas — {$include wasmenc.inc} beside the encoders, {$include asmtext_wasm.inc} beside the text emitters, and the TARGET_WASM32 output arm calling writeWasm instead of erroring.
compiler/exception_emit.inc:437 — the message text only, never the mechanism.

Anything beyond those two is a new conversation.

The guard, and why it is not a placeholder

writeWasm errors when WasmFuncCount = 0. With no codegen the module model is empty, so writing an empty-but-valid .wasm for a program that plainly has code would be a silently wrong output — the failure class this project treats as the expensive one. Every real program takes the error branch until Phase 2, and takes it for an accurate reason. The guard stays correct once codegen exists.

The exception_emit.inc correction

The wasm arm said the module "needs the exception-handling proposal or a trampoline". It needs neither. Pending-flag threading uses no engine proposal, no version dependency and no trampoline — which is precisely why PLAN.md chose it over the EH proposal, whose legacy-vs-final opcode split is live fragmentation. It was prototyped and diffed against a native build before the replacement text was written (devdocs/dev/wasm/phase5-exceptions.md, test/wasm/proto/check.sh).

Erroring there is still correct — exceptions genuinely are not implemented. Only the stated reason was wrong. A wrong reason attached to a correct behaviour is the most durable kind of wrong, because nothing ever fails to make anyone check it.

Gate — seen, not assumed

This is the first wasm change that could reach the self-host loop at all: the branch converged at 325b4479070a up to now only because nothing compiler.pas includes had moved. Two includes ends that.

make compiler/pascal26 → converged after 1 round(s), cb55d648a657, differing from pinned 325b4479070a. The convergence line was watched for specifically; its absence is the only tell and there is no error to wait for.
No existing target moved: 4 programs x 6 targets = 20 output hashes, identical before and after, with the "before" compiler rebuilt from the reverted files and reproducing 325b4479070a exactly.
test/wasm/check_phase1.sh and test/wasm/proto/check.sh green under the new binary.

Consequence the next session needs

The wasm branch now touches shared files. It was true until today that a master merge could not conflict in compiler/**; it is not any more. CHARTER.md's escapes table records this. Merge master in more often, and treat a conflict in compiler.pas or exception_emit.inc as a coordination event rather than a merge to resolve locally.