the entry the clock condemned hardest, and the board it was never measured on

Issue #200, 2026-09-13. Constructed, Duel (485). Rules version 2026-07-16. Card text verbatim from data/corpus_flat.txt; rules pasted from data/Riftbound-Core-Rules-2026-07-16.txt. Read the unopposed clock first — this play is about what that document's baseline is and is not for.

Subject: svellsongur-copy-hold — a BURST, T6, and a member of the worst-scoring population in the whole catalogue on the day's headline numbers, and three separate counts say so.

• in BURST, the class where 20 of 23 rows pay later than the do-nothing Hold curve;
• in the --stalled pass's HOLD bucket, labelled "DIES WITH THE CURVE … the stall that makes this line necessary is the same stall that switches it off";
• one of the 21 entries that are both, which is the sharpest negative any instrument in this project produces.

Those three counts read 20, the same bucket, and 8 when this play was first written. They are re-measured here on 2026-09-13 at catalogue 766 — and then re-measured a second time the same afternoon, because the clock learned to price an engine's own output and the unopposed headline fell from 45 to 38 of 80 while this paragraph was being written. Re-derive them; do not quote them. Nothing in the argument below turns on any of the three, and the subject entry's own T6 is untouched by both changes because it folds no fuel.

And every one of those verdicts is measured against a board this line was never for. On the board it is for it wins three turns early. This play is the correction, and it is a correction to instruments I wrote or changed myself today.

1. What it does, in one Hold

OGN-066 Ahri, Alluring is four words long — "When I hold, you score 1 point." SFD-059 Svellsongur copies its carrier's text, and the copies compose rather than add: 434.1.c gives the Top-Most Card all the Effect Text of all attached cards, so the second Svellsongur copies what the first already appended. Three of them make 2³ = 8 instances of Ahri's sentence on one body. UNL-087 Blue Sentinel then reads "Your hold effects for holding here trigger an additional time" — K = 1, so 8 × (1 + 1) = 16, plus the Hold's own Score, is 17 points in one Beginning Phase at one battlefield.

Nothing in that needs a second battlefield, an enemy garrison, a combat, or a turn after it.

2. The verdict the instruments give, and why it is the wrong board

npm run adversarial -- --turns puts it at T6 against a T5 baseline — slower than doing nothing. That baseline is the unopposed Hold curve: both battlefields yours, 315.2.b.2 Holding both, 2 points a turn, 8 by turn 5. Calm reaches it on 5 because VEN-043 Steel Paws costs one Energy.

But a player holding both battlefields unopposed does not need a finisher, and this project already knew that — the unopposed-clock play says it outright: "Unopposed, the Hold curve has already won." Measuring a finisher against that board asks "is this redundant when I was winning anyway?", and the answer is yes for almost everything, which is why the number is 41 of 80.

The board a finisher is for is a contested one. There are three board states and both instruments model two:

The middle row is T9 in every identity, because one body on turn 1 costs two Energy and two runes pay for it — the Energy-1 units that make the unopposed curve T5 are not needed to move one body. 471.1.a.1 is why the eighth point lands unconditionally: the Final Point restriction at 471.1.b.1 is scoped to a Conquer, and these are Holds.

Against that curve this line is T6 against T9 — three turns early. Measured over the whole catalogue the re-baselining moves the class from 20 of 23 slower to 1 of 23, and the whole table from 38 of 80 to 4 of 80 (2026-09-13, catalogue 766, with the engine-output discount on). Both numbers are true; they answer different questions, and the script now prints both.

3. The turns, on the board it is for

Calm/Mind, going first (485.7 gives the extra rune to the player going second). R = runes after the Channel Phase. The opponent has taken the other battlefield and garrisoned it; you hold yours.

Turn 6 — and the two earlier versions of this table both said T7, for two different wrong reasons. That is worth the space, because the second wrong reason became a line in CLAUDE.md.

The first version said the clock's T6 "does not include" the Equip costs. It did not: costsOfSet read a card's printed Energy and Power while 818.1 makes [Equip] a separate Activated Ability with its own cost. That was fixed and priced, and the clock still said T6, so the missing turn was never the money.

The second version said it is ORDERING — that deployTurn would happily pay a Svellsongur's [Equip] before the Svellsongur. The premise is true and the conclusion is false. Ordering is now modelled exactly (--selftest proves the allocator can see a turn where a hand-derived case owes one), and it moves none of the 80 rows, including this one. 33 of the 80 carry a linked [Equip] cost and not one of them is constrained by it.

What actually cost this table a turn was the table. It spent one card a turn and left Energy idle on every one of them — E5 of 5 on T3 wastes nothing, but E4 of 5, E5 of 6 and E6 of 7 each leave one behind, and T6 spent E3 of 7. Packed against the same rune curve it fits a turn earlier. Verified twice and by two different algorithms: the shipped allocator's exact search, and an independent breadth-first enumeration written from the rules rather than from the allocator, which prints exactly the schedule above.

The specific belief that cost the turn is that a gear played this turn cannot be attached this turn. 359.2.d refutes it — "If it is a Non-Unit Gear, it enters the Board Ready at the player's Base" — so the gear is a Game Object on the board the moment it resolves, 380 is satisfied, and no [Equip] cost in this pool contains an exhaust. Run the clock with --strict-ordering, which forbids the same turn and is therefore wrong as rules, and this row reads T7: the old table was the optimum of a game one paragraph stricter than the real one. Six rows move under that probe and the unopposed headline goes from 38 to 41 of 80, so the whole gear half of the catalogue sits exactly on that boundary.

Two assumptions this table inherits from the clock and does not hide. It ignores which domain a rune is when paying Power, and that is a draw assumption rather than a choice: 114 shuffles the Rune Deck at setup, 108.5.d makes its order Secret Information, and 430.1 channels from the top. And it assumes the cards are in hand when wanted. Both are the same optimism as "perfect draws", applied to the other deck.

T6 against a contested curve of T9 is three turns early, and you only ever needed one battlefield. Section 2 said three turns and section 5 said two; three is right.

4. Breaks to

• OGN-133 Flurry of Blades does nothing here, and that is unusual enough to say: Ahri is Might 4 and the Sentinel is Might 4 with [Shield 2]. This line has no small body at all, which is exactly why it is slow — see the walk document's gradient.
• Gear removal, and this is the real answer. 718.5.b: "Attached cards still can be chosen or targeted by game effects while Attached." Three Svellsongur on one carrier is 2³ = 8 instances; killing one takes it to 2² = 4 and halves the burst. Sixteen printings in the pool kill a gear and none carries [Reaction], so it must be cast a turn early — which, on this line, is the turn the attaches happen.
• Killing Ahri. Everything is appended to her Rules Text, so she is the single point of failure, and 719.5 detaches all three Svellsongur when she leaves the board. She is Might 4 with no protection.
• Contesting your battlefield. 469.2 defines the Hold on a battlefield you control, so losing it turns 17 points into none. This is the real cost of a one-battlefield plan and it is why the third board state exists.

5. Verdict

On the unopposed board it is redundant and the clock is right. On the contested board it is three turns faster than the alternative and the clock's verdict was an artifact of the comparison. The same entry, the same number, two opposite readings — and the only thing that changed is which board the baseline describes.

The general form is the one this play exists for, and it applies to the eight entries the old comparison condemned hardest:

A Hold-gated finisher is not switched off by "a stall" — it is switched off by holding NOTHING, and it is at its best holding ONE. The --stalled pass's HOLD label says "the stall that makes this line necessary is the same stall that switches it off", which is true of a total stall and false of the ordinary contested game. Three board states, and both instruments were modelling two.