engine: plan and implement the next high-leverage compatibility tranche

[ninthworld]

Parent: #713.

Problem

SAGE has 134 supported functional definitions and a generated, freshness-checked list of 36 excluded mechanic families. The compatibility report is honest, but there is no active issue turning the largest blockers into a prioritized next engine tranche.

The goal is not to create 36 speculative mechanic issues. It is to choose a small group of primitives by measurable card payoff, implement them through real cards, and regenerate the report after each lands.

Candidate families

Use crates/sage-engine/data/exclusions.json and the M19 card pool to measure payoff. Obvious candidates include:

• General replacement/prevention effects
• Equipment and non-Aura attachments
• Modal spells and X costs
• Cost modification and richer activation/additional costs
• Control-changing and untap/skip-untap effects
• Broader trigger events/selectors
• Multi-face/copy mechanics
• Optional or conditional targeted effects

Scope

1. Produce a small table mapping candidate primitive → blocked cards/families unlocked → architectural dependency/risk.
2. Select the next tranche; open implementation issues only for selected primitives.
3. Implement each primitive with representative real catalog cards and rules-text generation.
4. Update schema documentation, protocol only where necessary, exclusions, and generated compatibility report.
5. Preserve the pure-engine/server-policy/client boundary.

Acceptance criteria

• Prioritization is based on cards/mechanics unlocked, not intuition alone.
• Each selected primitive has at least one representative catalog card and focused engine tests.
• Client-visible behavior is fully server-projected; no client rules interpretation is added.
• make compat removes or narrows the corresponding exclusions.
• Generated rules text and executed behavior remain one source.
• General replacement effects are not claimed until an actual ordering/choice-capable replacement layer exists.
• Unselected families remain honestly excluded.
• Work is split into focused implementation issues rather than a single mega-PR.

Non-goals

• Claiming complete M19 or complete Magic coverage.
• Hand-authoring cards whose mechanics the IR cannot faithfully express.
• Creating one issue per exclusion before prioritization.

[ninthworld]

The prioritization table this issue asks for

Measured rather than intuited: the M19 printed set has 280 base-set cards, the catalog defines 176 of them, and each of the 104 unimplemented cards was mapped onto the primitive that blocks it. Cards blocked by several primitives are counted under each.

Primitive	M19 cards blocked	Complete on this alone	Protocol change?	Issue
Costs paid by sacrificing a chosen permanent, or by discarding	13	5	yes	#721
Amounts derived from other than a permanent count	11	2	no	#722
Casting/activating from zones other than the hand	10	2	no	#723
Optional/conditional effects that choose a target	7	2	no	#725
Selectors that filter by power	6	3	no	#724
Keyword removal and loses-all-abilities	5	1	no	#726
Conditions beyond a permanent count, mill, or discard	5	2	no	#727
Announcement-time choices — modes and X	5	0	yes	#733
Tokens created attacking, and copy effects	5	1	no	#734
Equipment	4	1	no	#728
Trigger conditions: draw, another attacker, activation, keyword filters	4	1	no	#732
Gaining control of a permanent	3	1 (after #730)	no	#729
Replacement effects	3	0	no	#731
Naming a colour, type, or card	3	1	no	#738
Untap, and stopping a permanent untapping	2	1	no	#730
Cost reduction and increase	2	0	no	#735
Damage prevention	2	1	no	#736
Attack/block requirements, extra blocks	2	1	no	#739
Evasion naming a subtype, ignoring hexproof	2	1	no	#742
Statics reaching beyond the controller's creatures	2	0	no	#743
Optional costs paid with anything but mana	2	1 (after #721)	no	#744
Fight / two differently-specified targets	1	1 (with #722)	no	#737
Auras that enchant something other than a creature	1	1	no	#740
Combat damage assigned by other than power	1	0	no	#741
Maximum hand size	1	1	no	#745
Choosing the order of cards put back on a library	1	1	yes	#746
Multi-face cards	1	1	no (large)	#747
Mechanics no exclusion names yet	22	—	no	#748

Nine exclusion entries block no M19 card at all and are left alone: emblems with an activated ability, mana filtering, kicker, the legend-rule choice, graveyard returns to a third zone, and the rest.

Recommended tranche

1. engine: selectors that filter by power #724 selectors that filter by power — purely additive, no protocol change, three complete cards. In progress.
2. engine: maximum hand size modification #745 maximum hand size and engine: untap effects, and stopping a permanent untapping #730 untap — the two smallest complete cards left, and engine: untap effects, and stopping a permanent untapping #730 unblocks engine: gaining control of a permanent #729.
3. engine: costs paid by sacrificing a chosen permanent, or by discarding #721 chosen-sacrifice and discard costs — the largest payoff on the list, and the first protocol change; worth doing once the cheap ones have shaken out the authoring loop.
4. engine: trigger conditions for drawing, attacking, and activating #732 trigger conditions and engine: optional and conditional effects that choose a target #725 optional targeted effects — broad, cheap, no new layers.

Deferred deliberately: #731 replacement effects (no shortcut, per this issue's acceptance criteria), #747 multi-face (one card, largest change on the list), #741 combat damage by toughness (needs three other pieces first).

#748 is the finding worth acting on first. 22 M19 cards are blocked by mechanics the exclusion list does not name, so the compatibility report currently claims a completeness it does not have. That is a correctness bug in the artifact, not a feature request.

_{The card census was built from the printed M19 set list; no Oracle text entered the repository.}

---

Generated by Claude Code

[ninthworld]

Dependency graph for the #706 tranche

Edges are hard must-land-first relations taken from the issue bodies, not guesses.
✅ = merged. ▣ = wave 1 (running now). ▢ = ready, not started. ⏸ = blocked or deferred.

graph TD
  subgraph landed[" Landed "]
    I724["✅ #724 power selectors"]
    I730["✅ #730 untap / skip-untap"]
    I745["✅ #745 maximum hand size"]
    I721a["✅ #721a sacrifice as a cast cost"]
  end

  I721b["▢ #721b discard + activation costs"]
  I744["⏸ #744 optional non-mana costs"]
  I722["▢ #722 amounts from other sources"]
  I737["▢ #737 fight / two target specs"]
  I729["▣ #729 control change (layer 2)"]
  I728["▣ #728 Equipment + attachment model"]
  I740["⏸ #740 Auras on non-creatures"]
  I731["⏸ #731 replacement layer"]
  I736["⏸ #736 damage prevention"]
  I734a["▣ #734a tokens created attacking"]
  I734b["⏸ #734b copy effects (CR 707, layer 1)"]
  I732["▣ #732 draw / attack / activate triggers"]
  I725["▣ #725 optional effects that target"]
  I727["▣ #727 life-gained conditions"]
  I748["▢ #748 unnamed mechanics (exclusion truth)"]
  I741["⏸ #741 damage assigned by toughness"]
  I747["⏸ #747 multi-face cards"]
  I726["▢ #726 keyword removal / loses-all"]
  I723["▢ #723 casting from other zones"]
  I733["▢ #733 modes and X (protocol)"]
  I735["▢ #735 cost modification"]
  I738["▢ #738 naming a colour/type/card"]
  I739["▢ #739 block an additional creature"]
  I742["▢ #742 subtype evasion + ignore hexproof"]
  I743["⏸ #743 statics reaching further"]
  I746["▢ #746 ordering cards onto a library"]
  I750["▢ reflexive triggers (CR 603.11)"]
  I751["▢ as-though-no-defender"]

  I721a --> I721b --> I744
  I730 --> I729
  I724 --> I722
  I722 --> I737
  I728 --> I740
  I731 --> I736
  I731 --> I734b
  I729 --> I734b
  I725 --> I750
  I732 --> I741
  I751 --> I741
  I748 --> I751
  I748 --> I750
  I743 -.card incomplete alone.-> I735

Loading

Roots — everything with no unmet prerequisite

#722 #723 #725 #726 #727 #728 #729 #732 #733 #734a #735 #738 #739 #742 #746 #748 #721b

Seventeen of the twenty-four open issues are unblocked. The graph is wide and shallow: the
real constraint on parallelism is not dependency, it is shared-file contention and the
fixed e2e ports.

The two things that actually serialize this work

1. Five files every PR touches: data/exclusions.json, data/sets/M19.json,
   docs/generated/compatibility.md (generated), docs/card-schema.md, and
   crates/sage-server/src/rules_text/. All additive, all rebaseable, none mergeable
   without a rebase queue.
2. make verify binds ports 4173 and 9000 with --strictPort. Concurrent agents
   cannot run it. Wave 1 was chosen to be pure-engine so make check is a sufficient
   agent-side gate and the browser tier runs once, serially, before merge.

Wave plan

Wave	Issues	Chosen because
1	#732, #725, #728, #729, #727, #734a	Six near-disjoint seams: trigger events, targeting, attachments, layer 2, conditions, combat/tokens. All pure-engine, no protocol change. 25 M19 cards behind them.
2	#722, #726, #738, #742, #739, #723	Ready now, but overlap wave 1 seams (layer 6, target legality, create_token). Start as wave 1 lands.
3	#733, #746, #721b	All three are protocol changes — client + TS mirror + docs/protocol.md. Serialize them; they cannot share a rebase queue cheaply.
4	#731 → #736, #734b; #728 → #740; #748 → #750, #751 → #741	Gated by wave 1/3. #731 is the one the issue forbids shortcutting.
—	#747 multi-face, #743	Deferred on the issue's own recommendation: one card each, largest change on the list.