Rally Comps × Joiner Skills × the Damage Formula (Gen 1~12)

🔑 And there is one more axis — the investment threshold

Those four layers give you the theoretical ceiling. Who actually shows up in the rally has one more factor multiplied on top: only the top four by skill level are taken, so what gets picked is not "the hero with the best effect" but "the hero who is realistically at Lv.5" → ⑦.

Without this axis you mistake "an empty slot nobody uses" for the optimum. A slot being empty ≠ opening it is worth it.

🔑 Exactly two rules

1. Several heroes in the same slot → they add inside one bracket
2. Different slots → the brackets multiply

The back end decides the slot via the skill's internal effect_op id. The practical test: same skill description text = same slot.

🔑 Two ways to read this

1. Leader diagnosis — sum the slots the three leaders occupy. Mia and Alonso each carry an extra slot-A skill (⑪); Logan alone fills C, D and H (⑫).
2. Joiner judgement — as a joiner, only S1 counts. Flint and Natalia have good S2/S3 skills, but as joiners they only ever deliver S1 (Infantry-locked / a proc D).

🔑 How to read row ④

Jessie2+Seo-yoon2 loses there because the leader team already filled slot B twice. Adding Seo-yoon to slot B is just addition, so the value dies — the four-way split that shifts weight toward slot A wins instead.

Practical rule: ask the rally captain for their three leader heroes first, cross out the slots those three occupy, then fill what is left.

How to read the two-number ratios

Three-number ratios are always infantry/lancer/marksman. Two-number shorthand means the missing class is 0, so it points at a different slot every time.

60/40 = 60 inf · 40 lancer (0 marksman) · 70/30 = 70 inf · 30 lancer (0 marksman) · 50/50 = 50 inf · 50 marksman (0 lancer) · 49/49 = 49 inf · 49 marksman (2 lancer) · 40/60 = 40 inf · 60 marksman (0 lancer).

There is exactly one exception — the "50/50 that breaks a 40/60 ultra-defense" is 50 inf · 50 lancer. The whole point there is sniping their 60% marksmen with lancers, so a marksman-flavored 50/50 defeats the purpose.

The same 60/40 vs 50/20/30 matchup, opposite verdicts

Attacking with 60/40 → 🚫 banned. Zero marksmen means their 50% infantry never melts, and your 40% lancers only have 30% marksmen to hunt, so the surplus is wasted. Fail to break through in time and you lose.

Defending with 60/40 → ✅ actually favorable. You do not need to melt anything, so zero marksmen costs nothing, and 60% infantry stalls all day. Their 20% lancers have no target at all, because you brought no marksmen.

Defensive ban rule (empirical, NOT derived from mechanics — read the correction below): there is exactly one axis — if their rally is lancer-heavy (≥30%) and your marksmen are heavy (≥30%), it is a ban. Marksmen have the lowest defense and health while being the lancer counter target (+10%, plus T7 Ambusher's 20% chance to bypass the front line), so losses are heavy. Conversely, if their rally is marksman-heavy (≥40%), pushing your lancers to 40% is correct — and that is why 60/40 is the standard defense.

The ban table is a labelled dataset

7 counters and 5 bans — 12 points in total. Rather than hand-picking thresholds, we can solve for them.

The three-role model allows exactly three failure channels, and each ban in the table is caused by a different one.

Defense drops P — which is why 60/40 flips

A garrison is under no obligation to kill anything inside a timer, so "cannot melt them" is not a loss condition. Only T and B remain. That is why the same 60/40 is a ban on offense (P=50) and the standard on defense (T=0, B=0).

The hand-picked defensive rule from earlier — "their lancers ≥30% && my marksmen ≥30%" — is exactly min(m, L) ≥ 30, i.e. the upper edge of channel B's feasible range. The derivation validated the guess; the data-centred value is 26.

What changes

P = max(0, I − N·m) weakens from "cannot clear their infantry" to "forfeits the marksman's +10% counter bonus". The claim that 60/40 cannot kill enemy infantry was simply wrong.

B = min(m, L/N) assumed every enemy lancer reaches my marksmen, but the actual bypass is 20%. That is a 5× overestimate.

The guide's own phrasing — "their 40% lancers delete your marksmen instantly" — is likewise not supported by a +10% counter and a 20% Ambusher. The source itself overstates it.

🔑 Only the top four by skill level are taken

So what gets picked is not "the hero with the best effect" but "the hero who is realistically at Lv.5".

🔑 The general principle

Slot scarcity and effect scarcity are different things. An empty slot is usually empty because it is not worth opening, not because everyone overlooked it. Conversely Hendrik (slot F) is a permanent pick precisely because nothing cheap can replace what that slot does.

🔑 Why stacking three Norahs is the exception

One Norah fills damage taken −15% (D-like) and damage dealt +15% (A-like) at the same time — one hero, two slots. And her effect is Infantry/Marksman only, which in a 48/4/48 covers 96% of the army, so the restriction costs nothing. It is the only defensible exception to "never stack the same hero".

🔑 Hendrik owns slot F (enemy Defense −25%) alone

Nobody else uses that slot, so he multiplies cleanly with any leader combination. That is the real reason he never drops out of the tables from Gen 8 onward.

⚠️ If Sonya (slot A, 20%) is on the leader team, she adds with the Jessie* family — send joiners to other slots.

④ Filter by the investment threshold ⭐new

• Check the candidate is actually at Lv.5. A half-levelled legendary loses to a maxed epic (⑦).
• Check the hero is not locked into someone's leader team — if they are, they cannot come as a joiner.
• If a slot looks great in theory but nobody uses it, first suspect a slot left shut because substitutes are cheaper.

🔑 Logan solves all three survival slots by himself

One hero fills C, D and H. The survival axis a defense needs is already thick at the leader stage, so all four joiner slots can go to damage. That is why the sheet spends three of them on Mia, Jessie and Seo-yoon.

⑪ (Jeronimo+Mia+Alonso) was the opposite — slot A was saturated at 1.90, so the joiners got pushed toward defensive picks like Philly and Zinman. Same generation, and one change of leaders flips the whole character of the joiner set.

🔑 In one line

Put the hero whose kit loses the fewest skills in your formation.

Generation tables change; this rule does not. There are two axes.

🔑 The marksman slot splits almost mechanically by widget direction

Offense rows = Greg, Alonso, Hendrik, Blanchette, Rufus / Defense rows = Zinman, Lynn, Bradley, Xura.

The decisive case: Gen 7~10 defense at 60/40/0 runs Bradley with 0% marksmen. All three of his skills are troop-wide so he works without marksmen, and his defensive widget fires. Put Hendrik there instead and his S3 dies and his widget stays dark — a double loss.

Picking joiners is not taste — it is filling empty slots. Cross out the slots the leader's nine skills already occupy, then fill the rest with four different heroes. Four in one slot is +100%; four across four slots is +144% — that 44% is what rally skill actually is. On top of that, check whether your damage is normal attack or skill damage (Reina / Wu Ming), whether your ratio matches the matchup (avoid the banned ones), and finally whether the hero is actually at Lv.5 (the investment threshold). However good the theoretical value, an unmaxed hero never even makes the top four.