Long-term sustainability

This document tests whether the economy in 14 and 33 holds up over five years, not just the first. The numbers come from modelos/sustentabilidade.py (output in modelos/sustentabilidade-saida.md). That model reuses economia.py and tokenomics.py without changing any of their assumptions. Nothing here is measured data: the result shows what breaks first, not how much the game will earn.

Update (11/03/2026): the proposals in this document were adopted in the v2 design (38), with adjustments marked in each one. The findings still stand as a record of what led to v2; 14 says what applies today.

There are three questions, each with a short answer:

1. Gold over 60 months

Finding 1: the old controller only reacts to growth. With 15% or more bots, the stock per player rises until it sits above the band (4.8 to 20.7 weeks of income). After that it stops growing. Since growth stays below 2% per month, the controller does nothing and the game is stuck at a high inflation level. With no bots the opposite happens in the 20 thousand and 200 thousand scenarios: gold stays below the band (0.6 and 0.8 weeks), and the controller does not act either.

Finding 2: chasing the band solves most of it. The controller now raises the sinks when the stock is above 3 weeks and lowers them when it is below 1 week, on top of the 2% per month rule. With the lever cap going from x1.6 to x2.0, the scenarios inside the band at month 60 go from 4 to 8 of 12. The x2.0 lever only shows up in scenarios with 15% or more bots.

Finding 3: with 30% bots, nothing in the controller fixes it (8 to 14 weeks even with x2.0). At that level, raising sinks only punishes humans. The defense is anti-bot, not price.

Proposal:

1. The controller in 33 §10 now chases the band, with a lever between x0.8 and x2.0 and a step of 5% per month. Adopted (38 §I): BAND_CTRL is on by default in economia.py, and 33 §10 was redone.
2. The bot target in 33 §10 becomes an operational trigger (adopted, 38 §I, 14 §11):
   • above 5% estimated bots in active accounts, bans get priority;
   • above 10%, no new price sink goes in before a ban wave, so humans are not punished for what bots do.
3. With 0% bots, the controller that chases the band swings now and then at the lower edge (up to 4% growth in a single month). This is an effect of the 5% step, not inflation, and it is fixed with a smaller step in alpha.

Limits: the scenarios hold the population change for all 60 months. The 1 million scenario at −6% per month ends with ~24 thousand active players, which is a stress test and not a forecast. The sinks and tiers from Chapter 2 onward are not in the model (33 §11). Neither is the return of dormant players with saved gold.

2. Token: money in against money out

A token with no sale and no issuance to the fund only holds up if the money going in covers the money going out:

• In: players buying token to pay for cosmetics and land tax, and the buyback.
• Out: players selling what they receive from the season, the treasury selling what it receives, and team, partners and testers selling what they unlock.

With the assumptions in 14, the balance before the fixed schedule is positive in all three scenarios: US$ 3.7k, 37k and 186k per month. The calculation includes the conservative assumption that every player sells 100% of what they receive from the season.

The problem is in the team unlock phase. With half of the fixed schedule sold in the month it unlocks, the price the flow supports in the peak month is:

The reference price in 14 (US$ 0.02) does not hold with 200 thousand players while the team is unlocking. The biggest outflow is the treasury selling the 70% of token spent on cosmetics and its share of the market fee: that is US$ 68.5k of the US$ 123.3k that leaves each month (200 thousand active players).

Proposal:

1. The treasury does not sell token on the market to pay bills. Studio bills are paid with card and PIX revenue (§3). This nearly triples the price the flow supports. Adopted (38 §E), with one change: there is no longer a buyback fund; treasury token only goes to liquidity, to cover a creator shortfall, and to burn, and the season comes from the reserve, up to 25% of the burn.
2. Stretch the unlock: team with a 12-month cliff and 72 more months of unlock; partners over 60 months. The monthly peak drops from 5.9M to 4.8M. Adopted (38 §F); with partnerships at 7% (they were 10%), the v2 peak is 4.3M.
3. Take the "reference price" out of all communication. In the model it only converts dollars to token. The question that matters for sustainability is whether the balance is positive, and it is. Adopted (38 §5).
4. After the fixed schedule ends, the treasury and the reserve only release up to the month's burn. Net supply is zero or negative, and the positive balance supports the price on its own. So the token is only at risk while the fixed schedule runs (5 years in v1, 7 with the longer unlock), and items 1 and 2 target that window. Replaced in v2: releasing "up to the burn" undid the burn (38 §1); now reserve and treasury release at most 50% of the previous quarter's burn, and net supply falls whenever there is activity.

Items 1 and 2 went into 14 with v2 (11/03/2026).

3. Studio with the token worth zero

The hardest test is to assume the token is worth zero. Since every cosmetic can also be bought with card and PIX (14 §3), studio revenue does not depend on the token. It depends on conversion:

The assumption in 14 (5% paying US$ 15 per month) is optimistic for a cosmetics-only game. With 2% paying US$ 8, each active player costs more in server than they pay, and growing only makes the math worse.

Proposal:

1. Survival metric: gross cosmetics revenue per active player. It needs to pass ~US$ 0.20 per month to cover the server and ~US$ 0.40 to pay 5 people at 200 thousand active players. It is measured in beta, and only then do we decide whether the token comes in (14 §12). Adopted as the token launch gate (38 §I). v2 finding: with the token live, what the player pays in token is burned and does not become cash, so the studio catalog is card and PIX only (decided on 11/03/2026) and the floor of US$ 0.20 per active player in card and PIX becomes an alarm (14 §9).
2. The 20% buyback comes out of net revenue above server cost, not gross revenue. If the month does not cover the server, there is no buyback, and the season fund shrinks along with the studio instead of sinking it. Replaced in v2: there is no buyback at all tied to revenue (a form of profit sharing, 38 §1); buying comes from creator payments (38 §C).
3. If the metric in item 1 is not reached in beta, the game goes on without a token, with cosmetics on card and PIX. Everything that is gameplay (gold, instanced land, professions) works the same, because none of it depends on the token (14 §1). Adopted (38 §I).

4. Outside the models

Points that decide the long life of an MMO and that none of the three models measure:

• Content: a chapter every 6 to 9 months (25). Delaying a chapter hurts the population more than any parameter here.
• Retention: the 12% monthly churn rate in economia.py is an assumption.
• Land: the Harberger tax (15) has no final percentage yet (0.5% to 1% per month). The tax revenue in tokenomics.py (US$ 0.30 per active player, reused in tokenomics_v2.py) is a guess.
• Legal: any real token depends on the legal opinion (16, 34 §10).

5. Decision order

1. Turn on the controller that chases the band (§1). Done (11/03/2026, 38 §I).
2. Decide items 1 and 2 of §2 (treasury does not sell, longer unlock). Done in v2 (14, 38 §E and §F).
3. Adopt the buyback on margin (§3, item 2) and the survival metric as the gate for the token. Gate adopted; buyback removed in v2 (38 §C and §I).
4. Measure in alpha and beta: bots, conversion, spend per payer, share paid in token, gold per hour. After that, redo the models with real data (38 §7).