Engine: the DraftFromSpellbook/ManifestDread tracked-set publish belongs in the WaitingFor continuation handler (Kayla's Kindling, Valgavoth's Onslaught, Dazzling Flameweaver)

[lgray]

🤖 AI text below 🤖

Related: #6857

Summary

#6857's root cause is that affected_objects_from_events
(crates/engine/src/game/effects/mod.rs:5590) has no arm for effect heads that affect objects without
emitting a per-object event: they publish an empty tracked set and every downstream
TargetFilter::TrackedSet anaphor binds nothing.

DraftFromSpellbook (2 gated rows) and ManifestDread (1 gated row) are filed together because they
share one mechanism: each parks a WaitingFor mid-resolution, and the objects they produce land on
the continuation's event vector — not on the resolver's. A compensating publish therefore belongs
in the continuation handler, not in affected_objects_from_events.

Both heads are excluded from the #6857 PR because they were dispositioned by code-read only and were
never runtime-measured. Under the maintainer's measured-core ruling that is disqualifying.

The mechanism

head	parks	handler
DraftFromSpellbook	WaitingFor::SpellbookDraft (crates/engine/src/game/effects/spellbook.rs:83)	engine_resolution_choices.rs:6428, GameAction::SubmitSpellbookDraft -> spellbook::complete_draft
ManifestDread	WaitingFor::ManifestDreadChoice (crates/engine/src/game/effects/manifest_dread.rs:66)	engine_resolution_choices.rs:1785, GameAction::SelectCards

Each head has two arms, and only one of them is the continuation case:

• ManifestDread with count == 1 manifests synchronously via morph::manifest_card ->
  zone_pipeline::move_object, which emits ZoneChanged — the existing _ => harvest already sees it.
  count >= 2 parks ManifestDreadChoice.
• DraftFromSpellbook's random arm conjures synchronously and emits
  GameEvent::ObjectConjured { object_id } unconditionally (conjure.rs:221); the ZoneChanged at
  conjure.rs:217 is gated on destination == Battlefield and both spellbook cards conjure to
  Exile, so a ZoneChanged-only harvest sees nothing there. The choice arm parks SpellbookDraft.

Precedent — the same pattern already exists twice

engine_resolution_choices.rs already computes effects::chain_references_tracked_set(...) inside a
continuation handler and publishes when it is true:

// engine_resolution_choices.rs:979
    .is_some_and(|continuation| effects::chain_references_tracked_set(&continuation.chain));
if continuation_consumes_tracked_set {
    effects::publish_fresh_tracked_set(state, chosen.to_vec());
}

// engine_resolution_choices.rs:4325
    .is_some_and(|cont| effects::chain_references_tracked_set(&cont.chain));
if continuation_consumes_tracked_set {
    effects::publish_fresh_tracked_set(state, chosen.clone());
}

chain_references_tracked_set (effects/mod.rs:5118) is the same predicate as
next_sub_needs_tracked_set under a second name, so a continuation publish here would be gated
identically to the resolver publish — no new authority, an existing one applied at the right seam.

Candidate cards

head	card	arm
DraftFromSpellbook	kayla's kindling	choice arm (SpellbookDraft continuation)
DraftFromSpellbook	dazzling flameweaver	random arm (synchronous, ObjectConjured)
ManifestDread	valgavoth's onslaught	count >= 2 parks the continuation; count == 1 is synchronous

What a future PR needs

1. Runtime measurement of all three rows — none exists today. Both arms of each head need coverage
   (continuation and synchronous), because they reach the publish through different code.
2. A control that pins the continuation publish not double-publishing the synchronous path (the
   count == 1 manifest and the random spellbook arm already emit their own events).
3. A rich projection: tracked_object_sets plus per-object zone/counters and the pending WaitingFor.
   A sets-only projection has already produced wrong verdicts elsewhere in this lane.

Evidence provenance

Everything above is code-read and AST classification at BASE 9b7c66e30 — file/line citations are
direct reads. No runtime measurement exists for any of these three rows, and no runtime symptom is
asserted for them.

[cuinhellcat]

Runtime measurement for the ManifestDread row, since it was dispositioned by code-read only.

The compensating publish works, and is measured NOT to be sufficient on its own.

I wired exactly the precedent this issue names — effects::chain_references_tracked_set(&continuation.chain) gating an effects::publish_tracked_set(state, vec![manifest_id]) inside the WaitingFor::ManifestDreadChoice → GameAction::SelectCards handler — and drove Valgavoth's Onslaught end-to-end through GameScenario/GameRunner with X = 2 and a 6-card library.

	before	with the publish	correct
chain tracked set after resolution	{} / chain=None	{TrackedSetId(1): [c1, c2]}	same
+1/+1 counters on the first manifested creature	0	4	2
+1/+1 counters on the second manifested creature	0	2	2

The publish is right; the over-count is a second, independent defect. Manifest dread X times, then put X +1/+1 counters on each of those creatures lowers to one ability with repeat_for: X whose sub_ability is the PutCounterAll { target: TrackedSet(0) }, so the follow-up clause runs once per iteration over a set that grows each time: iteration 1 puts 2 on c1, iteration 2 puts 2 on both. The printed text and the 2024-09-20 ruling ("Players can't take actions in between the time you manifest dread X times and the time you put +1/+1 counters on them") both say the counter clause happens once, after the loop — so repeat_for should not wrap it.

I have therefore left the publish OUT of #7533 rather than shipping a card that goes from "does nothing" to "does the wrong number", and I am reporting this instead. Whoever takes this row will need both halves.

publish_tracked_set (not publish_fresh_tracked_set) is the right call here: it extends the chain's existing set per iteration, which is what makes "each of those creatures" name all X manifested creatures once the repeat scope is corrected.

Adjacent: #7531 (the SINGULAR anaphor after manifest/manifest dread/cloak) is a different slot — TargetFilter::LastCreated — and is fixed in #7533 without touching the tracked set.