Space Tycoon — Economic Balance Report, 2026 Q3

Published 2026-09-02. The first quarterly balance report under the "balance review cadence" commitment in the public policy (docs/POLICY.md): median corporate net worth, inequality (Gini), price stability of core commodities, faction balance, new-player retention and P&L distribution, published in the open. Every figure below comes from one of three sources and says which: the 50-year balance simulation (scripts/sim-50yr.ts, run 2026-09-02 on the unified clock), the live world's public APIs (/api/space-tycoon/leaderboard, /market, /market/share, read 2026-09-02), or the shipped constants in docs/BALANCE.md. Nothing is estimated. Where a number cannot be measured yet, the report says so and states how it will be measured next quarter.

1. Summary

• The clock defect is fixed and the balance tables are now the rate players experience. Until 2026-09-02 the engine credited one game-month of income every 60 real seconds while the calendar advanced one game-month every 6 real hours (360x). Every balance playtest had been run at the 6-hour rate, so the design tables were right and the live game was wrong. Post-mortem: dev log, 2026-09-02; engineering detail: docs/BALANCE.md "Clock unification".
• The economy is on the Pass-6 curve where the curve was not deliberately moved. The year-10 and year-30 late joiners land within rounding of the Pass-6 tables (joiner-y10 $18.74B at year 50 vs $18.70B recorded; joiner-y30 $80.9M vs $80.9M). The founding integrator is off the Pass-5 table by −28% on income and −51% on net worth at year 50; the identified contributor is D5 making the T5 research tree affordable (it now spends $230B on research it previously could not buy, and holds a $20B flagship with an $80M/month floor). The exact split from other August changes cannot be settled from the recorded tables and needs a counterfactual run next quarter. See §4.
• D5 flagship economics work in practice for the one archetype that reaches a flagship. The integrator's first Jupiter datacenter returned its $20B capex in 76 game-months on its own P&L line (D5's design band was 120–240 months at a 2.07x revenue stack; the sim runs a 3.0x stack). Seven of eight archetypes never buy a flagship in 50 years in either run — the deep-tier horizon is still canon (Pass 7), but the "money-losing asset" finding is closed.
• Mark-II refits partly fill the H3 void. With refit-aware play, decision cadence in years 10–30 rises from 1/1 to 9/11 months per decade for the mono-expander, 8/13 to 13/27 for the integrator, and 7/8 to 35/29 for the year-10 joiner; the year-30 joiner goes from 5/0 to 25/27. But the refit content itself is consumed in the first decade a portfolio matures (mono-expander: 22 refit-months in years 0–10, then 3, 2, 1, 2). The cadence gain in years 10–30 comes mostly from the income uplift un-stalling research, not from ongoing refit decisions. Mark III arrives only for the integrator, in years 40–50.
• Inequality falls with refits. Year-50 Gini across the eight archetypes: 0.730 without refits, 0.548 with; top-1 share of wealth 67% → 39% (Pass 5: 0.82 / 89%; Pass 6: 0.787 / 80%). Refits are cheap relative to the catalog's next rung, so the archetypes that were stuck (mono-expander, both joiners) gain far more than the one that was not.
• No runaway, no new stall. Money-supply sink coverage stays 95–104% per decade in both runs (cumulative net minted −$2.0B standard, +$28.3B refit-aware, over 50 years and ~$1.3T of flow). The industrialist's 40-year stall on deep_drilling ($8B) and the hoarder's flat line are unchanged from Pass 5 — known, not new.
• The live world has two corporations. Median net worth $236.9M, Gini 0.289, no player order-book fills in the trailing 30 days (NPC share of traded value: 100%). Retention and faction balance are not statistically meaningful at n=2 and are marked measured next quarter.

2. Methodology

Simulation. npx tsx scripts/sim-50yr.ts (standard) and npx tsx scripts/sim-50yr.ts --refit (refit-aware). 600 game-months (50 game-years = 150 real days at the unified 6 h/month clock), eight scripted archetypes in ONE shared world with every realism switch on: contended NPC absorption caps (order-book FIFO), the shared labor market at the D6 supply base, dynamic spot prices from combined flows with the real mean-reversion math, construction material settlement, the delivery-contract outlet (5/day), D5 flagship upkeep floors, D4 Mark refit multipliers, and the Pass-6 graduation glide for the late joiners. Deterministic — no wall clock, no random numbers; two runs diff clean.

"Refit-aware" means every archetype takes a Mark II (and, once its category's T3 gate tech is researched, Mark III) refit on any building whose preview payback — the same arithmetic the in-game refit card shows — is under 60 game-months, cheapest first, at most its construction-slot count per month, never spending more than half its cash on one refit.

What the simulation models with the real engine modules: the service revenue stack (saturation × shared demand pools × power × supply efficiency), price-linked mining with shared extraction pressure, consumption and storage integrity, NPC absorption caps, the contract outlet, crafting, corporate overhead, executive compensation on book net worth, labor-market payroll, dynamic spot, price campaigns, decommission recovery, serial money-gated research on the real tree, D4 refits, D5 floors.

Approximations, stated plainly: private revenue multipliers are a fixed 3.0x stack (research cap 2.0 × workforce service bonus cap 1.5) once the scripted research completes — the live engine's stack differs in composition (research bonus cap +50%, tier +20%, station +15%, commanders, legacy) and can be higher or lower; corporate tier is inferred from lifetime earnings only (T6/T7 legacy gates not modeled, reported tier caps at 5); buildings and refits complete instantly (their wall-clock timers are minutes to an hour against a 6-hour month); headcounts follow a rational-cap formula. Not modeled at all: megastructures, interstellar expeditions, story chapters, senate and factions, ships and lanes, hazards and insurance, espionage, takeovers, seasonal events, mentorship, player-to-player order-book trades, D6 takeover and slot-auction gates. The simulation says nothing about those systems.

Clock note. scripts/sim-50yr.ts and its harness have always stepped one game-month at a time (research advances 21,600 real seconds per step), so the derived 10,800 ticks-per-month changed nothing in these tables; the whole 600-month world runs in about three seconds. The tick-stepping runner that did need conversion, scripts/balance-archetypes.ts, now steps a 30-tick month through the engine's monthFraction override and reproduces its pre-unification numbers exactly (regression: src/lib/game/__tests__/sim-month-grid.test.ts).

Live world. Public leaderboard and market endpoints, read 2026-09-02 (UTC). No private data; the two corporations named below are the public leaderboard's rows.

3. The 50-year decade grid

Book net worth (cash + depreciated buildings and refits + inventory at base prices) and trailing-12-month net income at each decade end. Decision months = months with a build, a research completion, a refit, a decommission or a campaign declaration (b/r/f = build / research / refit months). Sink coverage = that archetype's money destroyed ÷ money created in the decade.

3a. Standard run (no refits)

Decision months per decade (b/r/f):

Sink coverage per archetype per decade: mono-expander 131/101/100/99/100%; integrator 95/90/107/98/103%; industrialist 73/123/72/115/115%; aggressor 118/99/100/95/83%; turtle 116/94/100/105/99%; hoarder 186/93/104/94/71%; joiner-y10 —/108/97/91/103%; joiner-y30 —/—/—/108/107%.

3b. Refit-aware run

Decision months per decade (b/r/f):

Mark levels held at year 50: mono-expander 32 buildings, all Mark II; integrator 34 (19 Mark I, 6 Mark II, 9 Mark III); industrialist 11 (4/7/0); aggressor 7 (1/6/0); turtle 8 (0/8/0); hoarder 6 (all Mark I); joiner-y10 38 (19/19/0); joiner-y30 22 (7/15/0).

Sink coverage per archetype per decade: mono-expander 119/97/98/98/102%; integrator 98/99/103/98/100%; industrialist 93/91/111/65/95%; aggressor 111/103/92/95/88%; turtle 111/99/92/36/44%; hoarder 188/95/119/65/101%; joiner-y10 —/97/96/101/97%; joiner-y30 —/—/—/98/91%. The turtle's 36–44% in its last two decades is a corporation that has finished its scripted research and does not reinvest: it banks ~$50–65M/month against an executive-compensation wealth tax that does not stop the pile growing. That is the passive archetype by construction, and a WATCH item (§9).

3c. Money supply (world totals)

Pass 5 recorded +$15.0B cumulative net minted over 50 years and $237B of research destroyed. Research destruction is up, not down, after the D5 ÷10 reprice — the integrator (and, refit-aware, the year-10 joiner) now buys the T5 tree it could never afford before. Campaign fees burned: $250M in each of decades 2–5 (standard), $250M–$500M (refit-aware).

4. Against the Pass 5 / Pass 6 tables

Pass 5 (docs/BALANCE.md, 2026-08) and Pass 6 were run at the same 6 h/month rate this run uses, so they should be close except where a shipped change deliberately moved them.

Verdict on D1 (clock). The economy is on-curve wherever the curve was not intentionally moved: every joiner figure Pass 6 recorded reproduces to the rounding digit, solvency and concentration are at or better than recorded, prices and the labor index are where the tables put them. Two founders are off the Pass-5 table by more than 25%: the integrator (−28% income, −51% net worth) and the turtle (−79% net worth). What can be identified for the integrator: D5 divided the 35 most expensive research nodes by ten, and the integrator now spends $229.75B on 162 techs and reaches T5 — Pass 5's integrator could not touch T5 at all (finding C2) — and it buys a $20B flagship at month 407 that pays an $80M/month upkeep floor. The standard run ends with 34 buildings; in the refit-aware run the integrator recovers most of the income ($830M/month, −16% vs Pass 5; NW $71.2B) while spending $326.9B on 200 techs. What cannot be identified: Pass 5 did not record the integrator's own research spend or building count, and the turtle (no mining, no flagship, T1+T2 techs whose prices D5 did not touch) moved too, so the split between "cash diverted into the tree" and the other changes shipped since August (Pass-6 duty-cycle opex, Pass-9 fee indexing, D6 labor supply, the manufacturing and early-fab waves, and the other archetypes' changed footprints in the shared pools) cannot be settled from the recorded tables. The constant that most directly moves the integrator figure is T5_RESEARCH_REPRICE_DIVISOR (10 → 5 would halve the cash the tree absorbs). We recommend NOT changing it this quarter: D5's research-destruction arithmetic assumed exactly this spend, the sink is now exercised instead of theoretical, and the joiners and every refit-aware archetype are on or above curve. Follow-up (engineering, not balance): a --pre-d5 counterfactual switch in the harness so next quarter's report can attribute the founders' gap by cause instead of by inference.

5. D5 — did flagship payback get fixed in practice?

"Own line" = that building's service revenue at the archetype's multiplier stack minus its operating cost and its D5-floored maintenance, accumulated from the month of purchase. First-copy self-payback at neutral multipliers (the harness's marginalCurve, same as D5's table) is unchanged at 1,683 months for datacenter_jupiter, 1,450 for mining_europa, 1,153 for mining_titan, 1,764 for fabrication_titan, 1,114 for deep_space_relay and 669 for mining_kuiper — the neutral figure is a thin generational asset by design, and the realised figure at a mature stack is inside (below) the 120–240 band. Verdict: D5 landed — the one corporation that reaches a flagship makes money on it. The other five income flagships and the three infrastructure flagships were built by nobody in 50 years in either run; the D5 WATCH ("does anyone build them once T5 is reachable") stays open, to be read from live telemetry rather than the sim.

6. D4 — does Mark II fill the H3 void?

Pass 5's H3 finding: decision cadence collapses to 0–3 months per decade by year 30 for every archetype but the integrator, because copy N+1 lands in a 0.35-floored pool and the catalog jumps from ~$2B to $8–80B.

Reading the b/r/f breakdown honestly: in years 10–30 the refit months themselves are few (mono-expander 3 and 2; integrator 0 and 0; joiner-y10 10 and 9). Most of the cadence gain is research completions that the higher refit-boosted income now affords, plus the joiners' faster ladder climb. The refit content is front-loaded: a portfolio takes its Mark IIs in the decade it matures and is then done until a Mark III gate opens (only the integrator reaches one, in years 40–50, 9 buildings). Verdict: Mark II fills the void for the archetypes that were starved (mono-expander, both joiners) and materially raises founder income (mono +$17M/month at y50, integrator +$117M/month), but it is a one-decade burst, not a standing rung. The industrialist and hoarder — mining-heavy portfolios whose lines at the 0.4 extraction floor do not clear a 60-month refit payback — are untouched (industrialist: 7 refits, all in years 0–10; hoarder: 0), so H3's mining-side residual and H4 remain: the industrialist still stalls on deep_drilling ($8B) for 40 years in both runs. No constant changed here; the next lever for those two is content (a mid-band mining rung or deposit), as Pass 5 said.

7. Inequality, concentration and tiers

Gini over book net worth across present archetypes (negatives clamped to zero), and the top corporation's share of positive net worth:

Tier concentration at year 50 (lifetime-earnings thresholds; T6/T7 not modeled): standard — T5 ×1 (integrator), T4 ×2 (turtle, joiner-y10), T3 ×4, T2 ×1 (joiner-y30); refit-aware — T5 ×1, T4 ×4 (mono-expander, industrialist, turtle, joiner-y10), T3 ×3.

P&L distribution at year 50 (trailing-12-month net, standard run): −$718K, $988K, $26.4M, $30.8M, $38.5M, $43.7M, $318.1M, $713.2M per month — a median of $34.7M against a top of $713.2M. Refit-aware: $18.4M, $23.9M, $38.8M, $52.1M, $94.4M, $329.1M, $481.0M, $829.8M — median $73.3M, every archetype profitable.

8. The live world (2026-09-02)

Read from the public leaderboard and market endpoints after the 2026-09-02 ÷360 rescale (see the dev-log post-mortem). Epoch 2 opened 2026-08-24.

Price stability of core commodities. Live: the raw-material floor (iron, aluminum, titanium) sits 4–5% below base after nine days of NPC production; one component (rare_earth) is 25% above base. Sim: no organic excursion beyond −10% in 50 years (iron 90% at y50, lunar_water 92%, aluminum 94%); the only prices that move meaningfully are those under a declared price campaign, which pin at the 30% band floor for the campaign window and recover after it (refit-aware run: lunar_water at 30% at month 119 and 36% at month 359, back above 90% at every other snapshot).

NPC market share. The measured figure above is the public free tier of market-share.ts (top participants by fill value, trailing 30 days). A resource-by-resource share and the 90-day quarterly window need a database read (computeServerTradeSummary, getResourceShare) and will be published from the live database next quarter. Two things were fixed on 2026-09-03 (both were open when this report published): the share endpoint used to mark the five __NPC_CORP_* industrial corporations isNpc: false (only the market maker __NPC_MARKET_MAKER__ was recognized) — it now uses one canonical predicate (npc-identity.ts's isNpcProfileId) shared by market-share.ts, market-orderbook.ts and flow-map.ts, so every UI reading the share endpoint labels NPC industrial volume as NPC, not rival-player volume. And per-side shares used to sum to 200% of value (buyer and seller each counted the full fill value against the single-counted market total); sharePct is now computed against the doubled trade-side total, so it sums to ~100% across participants. The pre-fix reading survives as a separate field, sideValuePct — needed by at least one real consumer (corp-pacts-server.ts's non-aggression-pact clause, calibrated to "holds 40% of traded value," not "holds 40% of trade-side credit"), which was switched to it so the fix does not silently change that pact's enforcement bar. Does the measured NPC share above change? No — it is unchanged at 100%. The fix only changes how value is apportioned among participants; this window had zero player fills (all five participants were NPC industrial corporations trading with each other), so the player/NPC split — 0% player, 100% NPC — was never affected by the per-participant double-counting bug.

Faction balance — measured next quarter. No public endpoint exposes faction standings, and the sim does not model factions. Method: aggregate per-faction reputation and standing counts across active profiles from the profile store; publish the six-faction distribution and the share of contracts by faction.

New-player retention — measured next quarter. With two corporations the figure is not meaningful. Method: cohort of profiles created in the quarter vs. profiles that synced in the following 7 and 30 days, from the profile timestamps; published as a percentage with the cohort size.

Starting-archetype fairness (idle coast). scripts/balance-archetypes.ts, 100 games × 3 starting archetypes × 24 idle months on the month grid: median net worth at month 24 — Cape Heritage $315.4M, Meridian Signals $288.0M (−8.7%), Tracking Consortium $265.2M (−15.9%). Spread 15.9%, the runner's "moderate" band; unchanged from its pre-unification reading by construction (same 30-tick month).

9. What changed this quarter, and what we watch

Shipped (all 2026-09-02, founder-approved; detail in docs/BALANCE.md):

• D1 clock unification — one game clock; ticks-per-month derived from the 6 h calendar; offline operations on the same clock with overhead and executive compensation charged; a state-derived money plausibility ceiling with a $500K/s backstop. D2 migration: every balance divided by 360, no compensation (everyone scaled equally). D3: subscriber perks that touched starting money, build/research speed and offline hours removed.
• D4 Mark-II/III refits — in-place upgrades at 1.5x / 2.5x baseCost, 1.6x / 2.4x own-line revenue, 2.2x / 3.6x maintenance; saturation still counts one unit; nine previously inert T3 techs gate Mark III.
• D5 flagship economics — upkeep floor 0.4%/month of baseCost above $20B, the 35 most expensive research nodes ÷10, eight flagship services retuned into a 120–240-month payback band at a 2.07x stack.
• D6 population gates — takeovers open at 10 active corporations (was 25); slot auctions on relative occupancy; labor supply base ÷5.

Watch list for Q4 (live telemetry, not sim):

1. First flagship built in the live world and its realised payback (D5 WATCH); whether anyone builds the three infrastructure flagships.
2. Refit uptake: share of eligible buildings at Mark II within 30 days of eligibility, and whether Mark III gates are researched.
3. Money supply: cumulative net minted per game-month across all profiles (the sim's refit-aware world mints +$28B over 50 years against ~$1.3T of flow — bounded, but positive in the last two decades where the standard world was negative).
4. Passive cash piles: corporations with no spend for 12+ game-months and growing net worth (the turtle's 36–44% sink coverage).
5. The labor index leaving the 0.80 floor at ~12 active corporations (D6), the first tender at 10, the first relative-occupancy slot auction.
6. Industrialist / mining-specialist stall (deep_drilling at $8B): the next content rung, not a repricing.
7. The "my income is frozen" risk: run-rate widgets now move every 6 hours.

Balance-adjustment notice. No balance constant was changed for this report. Per policy, minor adjustments carry 7-day notice and major ones 30-day notice with community comment; none is pending.

10. Reproducing this report

npx tsx scripts/sim-50yr.ts            # §3a, §4, §5, §7 (standard)
npx tsx scripts/sim-50yr.ts --refit    # §3b, §6 (refit-aware)
npx tsx scripts/balance-archetypes.ts  # §8 starting-archetype spread
curl https://spacenexus.us/api/space-tycoon/leaderboard
curl https://spacenexus.us/api/space-tycoon/market
curl https://spacenexus.us/api/space-tycoon/market/share

Section 11 of each sim run prints the decade grid, the flagship table, the concentration table and the Mark-level table exactly as used above.