Telegraph Round — A/B Test Report
Proposal: "Turn the telegraph round at the Transition Moment from optional to mandatory" (the boss's first round after a phase transition stays below the full damage band). This report puts the proposal to a controlled A/B test.
Experiment design: 12 parties (6 T2 + 6 T4) × 2 scenarios (Current rule / Telegraph mandatory) × 3.000 repetitions = 72.000 simulated battles. Same parties, same seeds, single variable is the telegraph round. Win rates in percent.
Main Result
| Current rule | Telegraph mandatory | Change | |
|---|---|---|---|
| Overall win rate | %39,6 | %52,9 | +13,3 points |
| T2 average | %53,1 | %63,2 | +10,1 |
| T4 average | %26,1 | %42,7 | +16,6 |
The telegraph round raises the win rate by 13 points overall. But the real story isn't in the average — it's in the distribution: the proposal doesn't help everyone the same — it helps the parties that were supposed to win outright.
Full Results by Party
Bone Puppeteer (T2)
| Party | Archetype | Current | Telegraph | Δ |
|---|---|---|---|---|
| A Standard | balanced | %66,8 | %86,7 | +19,9 |
| B Controller | control | %90,2 | %93,7 | +3,5 |
| C Nova | burst | %8,1 | %22,9 | +14,8 |
| G Balance2 | balanced | %53,7 | %75,9 | +22,1 |
| H Glass-cannon | extremely fragile | %0,0 | %0,0 | +0,0 |
| I Tanky-control | extremely tanky | %99,7 | %99,9 | +0,2 |
Peak Tyrant (T4)
| Party | Archetype | Current | Telegraph | Δ |
|---|---|---|---|---|
| D Key-aware | balance+control | %63,2 | %91,5 | +28,3 |
| E Blaster nova | burst | %0,0 | %0,0 | +0,0 |
| F Melee wall | durability | %27,1 | %54,6 | +27,5 |
| J Full-control | control | %60,0 | %90,0 | +30,0 |
| K Ranged-nova | burst | %0,0 | %0,0 | +0,0 |
| L Balanced-melee | balanced | %6,6 | %20,2 | +13,6 |
Finding 1 — Telegraph saves the right parties
The parties that benefit most from the proposal are the ones in the "critical band": lineups that win %20–70 of the time under the current rule — in other words, those that live or die on the roll of the dice.
Average effect of the telegraph across the 5 parties in the critical band: +25,6 points.
The biggest jumps: Full-control (+30), Key-aware (+28), Melee wall (+28), Balance2 (+22), Standard (+20). These are parties that were "reasonably built but died to an unlucky roll sequence." The telegraph round closes off exactly those unfair deaths — it protects a well-played battle from being sacrificed to the dice.
This was the goal of the proposal, and it's confirmed.
Finding 2 — The two extremes are already independent of the telegraph (expected behavior)
| Extreme | Example | Why the telegraph has no effect |
|---|---|---|
| Extremely tanky | I Tanky-control (%99,7→%99,9) | Already wins almost always; no battle to save |
| Extremely fragile | H Glass-cannon, E/K nova (%0→%0) | Melts so fast that a single soft round isn't enough |
This is a healthy result: the telegraph round is not an "easy-mode trick." It doesn't further empower parties that already win, and it doesn't artificially rescue hopeless cases either. It only fixes the battles in the middle that should be fair.
Finding 3 — The nova-TPK from the first report isn't fully solved, just eased
The striking finding from the first simulation was that nova parties suffered a TPK at the phase transition. The A/B test adds nuance to this:
- Mid-nova (C Nova, T2): %8 → %23. The telegraph tripled it, but it's still low. Nova + no healing is structurally weak in a multi-phase battle.
- Extreme-nova (E Blaster, K Ranged-nova, T4): %0 → %0. These are so fragile that a single soft round isn't enough.
Interpretation: The telegraph round eases the nova-TPK but doesn't solve it. This is actually correct — the nova archetype being weak in a multi-phase boss fight is a balance choice, not a bug. The telegraph turns it into a "fair chance" (C's %8→%23), not an artificial victory.
Finding 4 — Round cost is negligible
Does the telegraph round drag battles out? Average difference: +0,15 rounds for T2, +0,68 rounds for T4. So at most "half a round to one round more." Boss battles already run 8–11 rounds; this difference isn't felt at the table. The proposal doesn't drag the fight out.
Assessment
Is the proposal approved? — Yes, conditionally
Making the telegraph round mandatory:
- ✅ Protects battles that should be fair (+26 points in the critical band)
- ✅ Doesn't disrupt the two extremes (the tanky always wins, the extremely fragile still loses)
- ✅ Round cost is negligible
- ⚠️ Doesn't solve the nova-TPK, only eases it (which is actually the intended effect)
But one caveat: overall win rate is %53
Even after the telegraph, the overall average across the 12 parties is %53. Peak Tyrant at T4 is %43. These bosses are still hard — Peak Tyrant in particular defeats four parties (at a %50+ rate). There are two readings:
- These are campaign-finale bosses (as the framework states). A high death rate may be appropriate for a "set-piece finale."
- But if more than half the tables lose these fights, the lethality may be excessive. It's notable that even reasonable lineups like L Balanced-melee (%20) and F Melee (%55) struggle.
Second suggestion (outside the test, an observation)
The telegraph alone doesn't seem to be enough. Two additional adjustments could be considered:
- Soften the Phase 1 opening Ambush: Peak Tyrant's first round deals 77–84 damage (nearly dropping one character with a single target). This leaves any party that can't solve the key wounded from the start. A cap on the Ambush opener (e.g. "half the target's max HP") could be added.
- Peak Tyrant's Phase 2 DPR (~105) is above the band for T4. Parties without Fire (F, L) get crushed here. Dialing Phase 2 down a notch would give melee/no-Fire lineups a chance.
One-Sentence Decision
Make the telegraph round mandatory — the statistics are clear (+13 points overall, +26 points squarely in the "fair death" band, both extremes undisturbed). But Peak Tyrant's Phase 1 opening burst and Phase 2 DPR should also be reviewed separately; the telegraph is necessary but not sufficient on its own for T4 balance.
Methodology: For each party lineup, damage/round was precomputed and cached with a 600-trial Monte Carlo against the boss's AC; battles were then replayed round-by-round 3.000 times with preserved dice variance. Healing, control, key, and archetype effects were modeled as multipliers. Absolute percentages are sensitive to the model's assumptions; what's actually reliable is the difference between scenarios (same model, single variable telegraph), because systematic errors are equal on both arms. Real tables will deviate based on player decisions; the pattern (critical-band rescue, two-extreme immunity) is structural.