Progression and pacing

How to keep a player moving toward the next tier, and how to prove with a simulation that they can't get stuck.

Contents

• Make the next goal explicit
• Gate goods vs auto-sell
• Time milestones with a sim
• Sims are optimistic
• Upgrades must beat a new copy
• Serial gates and lulls
• Prestige worth doing
• Daily and tier-up rewards
• Soft-lock detection by simulation

Make the next goal explicit

Players asked "how do I get to Tier 3?" because the requirements lived on three screens (rings, research, goods). Fix it with two pieces:

• A next-goal pill on the HUD that names the current blocker only, and walks through them in order:

  1. "Clear Ring 2: conquer 2 more (5/7)"
  2. "Take the capital to clear Ring 2"
  3. "Get 700 Iron for Tier 3 (420/700)"
  4. "Save $24,000 for Tier 3"
  5. "Lab busy: Drill finishes first (2:39)"
  6. green READY: "Research Tier 3 in the Lab"
  7. "Researching Tier 3 (2:32)" with a bar

  Tapping it opens the right screen, focused on the right item.

• A checklist under the pill (tap a tier chip): every requirement with a tick or cross, progress, the current blocker outlined, and "Unlocks: …" so the player knows why they want it.

Build both from the same config the server checks (tier list + research costs). Never hand-write the text, or it drifts. Hide the pill while a building panel or the tutorial card is up, so it doesn't show through.

Gate goods vs auto-sell

Bug: an Auto Sell feature kept a flat 150 iron and sold the rest every 10 s. The Tier 3 gate needed 700 iron. With Auto Sell on, iron never got past ~150, so the tier could never be paid, and the message just said "Need 550 iron". It hit paying players first, because Auto Sell was a paid pass.

Rules:

• One keep function decides how much of each resource is safe to sell. Every sell path uses it: auto sell, "Sell All" and per-row sell buttons. (A per-row button that skipped it let tutorial players sell the wheat they needed for the next step.)
• Priority: the player's own keep setting > the next tier gate's goods (once its other requirements are met) > the next build's cost > the tutorial reserve.
• When a sell keeps everything, say why in a toast.
• The "Need N iron" message adds "(Auto Sell is on)" when it is.
• Better still: remove goods from gates. One redesign went cash-only for this reason.

Same family: storage caps must never be the only thing blocking a purchase. If a cost exceeds the warehouse cap, the player can never afford it. Raise the cap floor to the largest cost the player can face next (next research level, not level 1).

Time milestones with a sim

Write a plain-Luau (or Python) balance sim that plays the game with a greedy bot using the real config and formulas. Use it to time milestones across many seeds:

Also measure lulls: the share of each 5-minute block with no purchase or win for 20 s or more. One sim showed 9 to 28% waiting at minutes 15 to 25 because the next zone's weakest enemy was 3× the last boss. Adding two weaker "gateway" enemies on the road out fixed it. Later lulls had 35 to 64% waiting; four gateways, one tier weaker, fixed those.

Re-run the sim after every balance change, on 4 to 9 seeds, and report min / median / max.

Sims are optimistic

A sim bot is not a kid. Known ways sims flattered one game:

• It bought missing goods with cash, which the real game can't do, so 700 iron cost it a trivial $9.5K.
• It reacted instantly (attacked the moment it was ready); a real player paced 3 to 5× slower on the first conquest.
• It picked units it couldn't actually train, and attacked at ratios real fights didn't support.
• It didn't model the single research slot being busy with side research, which blocked the tier gate for an hour.

So:

• Mirror real constraints in the sim (no buying goods, slot limits, caps).
• Feed the sim measured numbers. One game ran 102 real battles to measure win rates by power ratio, then used those in the sim.
• Treat sim times as a lower bound. Check the important ones with one real paced playthrough.
• When the sim stalls, first ask: bot bug or game bug? Replay the bot's exact state in the real game.

Upgrades must beat a new copy

Counter-example: upgrade cost was 2.2 × level^1.45 per step, so reaching level 10 cost 223× base in total for 5.5× output (linear +0.5 per level). A new copy cost 1 to 7.6× base. Copies always won, the sim built 138 copies vs 32 upgrades, and the map filled with clutter to walk over.

Rule: the marginal cost of output from upgrading should be close to, or better than, building another copy. Check it with snippets/upgrade_vs_copy.py. Fixes: lower the cost exponent (≈1.1), or raise the output step (≈+1.0 per level). Also:

• show the upgrade's effect before buying ("+0.9/s → +1.35/s Wood"), computed from the server's own formula;
• make every level change something visible on the building;
• say "Max for Tier N" with what unlocks more, not just "Max level".

Serial gates and lulls

If every tier requires "clear the zone, then pay, then wait 10 / 40 / 90 minutes", the player sits idle against enemies a tier above them.

• Let the tier research start before the zone is cleared; the tier unlocks the moment both are done.
• Keep gate timers short (5 / 15 / 30 min).
• Give the gate research its own slot, or let it pre-empt side research.
• When the research slot is idle, suggest a one-tap node (gate first, then production).

Prestige worth doing

A reset that pays 2 stars (≈+10% income) for 2 to 3 hours of buildings, land and research is a trap, and repeating "New Campaign unlocked!" on every zone clear nudges kids into it.

• Show "you'll lose X, you'll gain Y" before confirming.
• Pay enough that the second run is visibly faster (e.g. 3 stars per zone + 1 per 5 wins, instead of floor(sqrt(wins/5))).
• Announce it once, when first available.
• Make the dialog text true. One said "you keep research" and then reset every node to level 1 and dropped running jobs.
• List what survives a reset explicitly. An idle game rebuilt state from newState() and silently re-armed the daily reward, the wheel, quests, redeemed codes and the tutorial on every prestige. Keep an Account list of fields to carry; anything new and permanent must be added there.
• Leaderboards rank a monotonic total, never "earned this run".

Daily and tier-up rewards

• Daily login streak: a 7-day calendar, cash scaled to current income, premium currency on days 3 and 7, a free item on day 7. Store {day, lastClaim} in the save and use server time.
• Tier-up reward: a "TIER n UNLOCKED!" card with fanfare and particles, a gift (e.g. 10 / 20 / 30 / 40 gems), and the new buildings and units it unlocked. A tier-up that only flips a number feels like nothing happened.
• Offline: a welcome-back screen that says "Collect" and gives $0 feels broken. Give some offline income (e.g. 25% for up to 2 h) or label it honestly.
• Count free premium currency on the main path against what rushing costs. ~28 free gems vs 1 gem per minute of rush meant kids never got to feel a rush.

Soft-lock detection by simulation

Bug: on an endless map of rings, countries left over in an old ring became unattackable once the player's current ring moved on (ring == currentRing instead of ring <= currentRing). If the path to the next capital ran through a leftover, the game was permanently stuck. Nobody found it by playing. A 300-seed sim found it: 32 seeds locked at ring 3, 50 at ring 4, 60 at ring 5.

How to look for this class of bug:

1. Generate the world for many seeds (hundreds, they're cheap).
2. Play it with the real "can I attack / is it visible / neighbours" functions, using a bot that always takes the cheapest legal move.
3. Assert reachability at every step: the next required target must be reachable through legal moves.
4. Report the seeds and the step where it got stuck.

A template is in snippets/softlock_sweep.py. Fix leftovers by keeping them as easy "mop-up" targets, and add a "clear all" button for leftover ruins.