Pastes are kept in this browser only.
The demand model turns a price into units, and back: picking the price in a market is the same as picking how many units to sell there. So the planner gets each market's revenue as a curve instead of a fixed price, and one solve picks the units in every market, the production on each shift, hires, machines and distribution together. Each market's price is then the one at which the model's demand is the units sold.
revenue = (backlog × the lower of its old price and the new one + new demand × price) × share kept after returns
The line is the demand each price needs to earn as much this quarter as the grid does now; above it earns more. Production is re-balanced for each case. Click the chart to test a point.
For the product tested, the planner is re-run shipping 50%, 75%, 125% and 150% of what the grid's plan ships this quarter. Each run is proven best and re-balances everything else: overtime, shift II, hires, layoffs, machines, raw material, the other products and later quarters. Taking out what the product's units earn leaves what the rest of the plan costs at that many units:
cost(n) = plan cost + value of skipped units − the product's skipped units this quarter × earns a unit
Between two runs the cost is a straight line; outside them the nearest line is carried on, and past a run where the plant was full no more units can be made. Then, for any price and demand:
earnings = sales after returns − shipping − cost(units shipped)
earns a unit = price × (q − r·(1 − q)) − variable shipping to that market
A unit costs the same to make whatever market it goes to, so markets differ only in what they earn. Extra units do not all cost the same, though: some fit in hours already paid for, some need overtime, some a new hire or a machine.
The cost ladder comes from re-planning with fewer of the product's units allowed to be skipped this quarter, in four steps down to none. Between two steps, the change in the plan's cost, less what the extra units earn, is what they cost to make, including what it does to other products and later quarters (a hire made now is still there next quarter). Each rung is one step: its units, what each costs, what the plant needs for them, and which markets they go to.
The planner gives the cheaper rungs to the markets that earn more. A skipped market's extra units are the ones on the rungs that go to it:
loss a unit = its rung cost − what it earns break-even price = (its rung cost + variable shipping) ÷ (q − r·(1 − q))
Raising one market's price moves it up the queue: it takes cheaper rungs, and the market that now earns least is skipped instead. To make more units in total, the market last in the queue has to earn more than the next rung costs. Units still skipped with none allowed don't fit at any price.
The planner treats the target units as fixed, so demand at a new price is your call: check it against last quarter's demand and backlog, which were seen at last quarter's price.
| New hire(s) | Retire / layoff | New machine(s) | Raw materials | Bank loan |
|---|
| Line | Plan | Pasted form |
|---|
| Plan | Pasted form |
|---|
| Segment | Machine h | of | Labour h | of |
|---|
| Production | Submitted | Cheapest |
|---|
| Statement line | Model | Actual | Check |
|---|
Each product and market is a newsvendor problem, repeated every quarter: how many units to have in the market before the quarter's demand D is known.
Choose Q* so that P(D ≤ Q*) = Cu / (Cu + Co)
This minimises Co·E[(Q − D)+] + Cu·E[(D − Q)+]. If D is normal with mean μ and standard deviation σ:
Q* = μ + z·σ, z = NORMINV(Cu / (Cu + Co))
margin = price × (q − r·(1 − q)) − (raw material + labour + variable shipping)
Cu = (1 − c) × margin
Co = storage per unit per quarter (last quarter of the game: unit cost + half a quarter's storage)
The game is a P-system with a review period of one quarter. The order-up-to level in a market is T = μ + z·σ + backlog, and the shipment is T minus the stock already there. Production must be shipment ÷ shippable share, less factory stock.
μ = last quarter's demand, σ = √( mean of (Dt − Dt−1)² )
Backlog conversion: the share of unmet demand that comes back next quarter.
| Product | Raw material | Labour | Variable shipping |
|---|
Change in demand per unit of each lever, with one standard error. Runs live in
data/runs/; each lesson restarts from the same quarter, so each is one experiment.
Per market; it reaches every product sold there.
Per product; it reaches every market.
| Run | Saved | Periods | What was set differently | Pages | Profit | Score |
|---|
To add a run, see data/runs/README.md.
| Test | Lever | Question | What to set |
|---|
Every product and market of every run is one observation:
y = ln(demand ÷ demand in the same quarter a year earlier) x = ln(your price ÷ reference price)
The reference price is the average of all firms' prices in that market if the run saved the Prices by Firms page; otherwise your own price in the quarter the restart begins from.
Price and promotion: y = levelrun, product + bproduct·x + g·promotion/10,000
Advertising and level: y = levelrun + levelproduct, market + bproduct·x + g·promotion/10,000 + k·advertising/10,000 + l·product level
Firm 5, P9–P16. From data/FINAL_PLAN.md and the workstream reports in
data/final/. Every forecast is the planner's demand model and unconfirmed unless marked. Scores
are the final score after P16 from the 8-quarter simulator, the other firms repeating MOCK unless named; two
paths within ±0.05 of each other are a tie.
Decided 2026-10-22 from P12's pages (data/runs/FINAL/) with
CAP=0,0,0 WAGE=105 QUALITY=0.980 DIV=500000 SHARE=124,377,189,277 node analysis/final_next.js:
Best prices and Plan ahead 3, no new machines in any quarter, wage 105%. Demand is the refit model with two
corrections: each product's average miss on P10–P12 (PA +0.09, PB +0.08, PC +0.12, PD −0.07), and last year's
P8→P9 season taken against the market (our P9 repositioning had been read as season: PA M1 −0.76 → −0.13).
Each price is then set on the spread of outcomes (±0.20): a unit short is a sale lost, a unit left over is
sold in P14 at its marginal revenue less storage. That puts the odds of selling out at about 10% in every cell.
| Pricing | M1 | M2 | M3 | M4 | Advertise |
|---|---|---|---|---|---|
| PA | 6,190 | 6,210 | 6,330 | 6,440 | 30,000 |
| PB | 8,640 | 8,780 | 8,920 | 9,010 | 30,000 |
| PC | 12,420 | 12,470 | 12,710 | 12,920 | 120,000 |
| PD | 15,980 | 16,180 | 16,630 | 16,680 | 30,000 |
| Promote | 80,000 | 80,000 | 80,000 | 80,000 |
| Distribute | M1 | M2 | M3 | M4 |
|---|---|---|---|---|
| PA | 68 | 59 | 170 | 85 |
| PB | 46 | 44 | 118 | 79 |
| PC | 42 | 43 | 119 | 129 |
| PD | 38 | 36 | 32 | 42 |
| Production | Shift I regular | Shift I overtime | Shift II regular | Shift II overtime | R&D |
|---|---|---|---|---|---|
| PA | 336 | 0 | 0 | 0 | 0 |
| PB | 315 | 0 | 0 | 0 | 0 |
| PC | 0 | 0 | 342 | 0 | 0 |
| PD | 0 | 0 | 23 | 122 | 0 |
| New hires | Retire/layoff | New machines | Raw materials | Bank loan | Bond | Treasury | Dividend | Wage % |
|---|---|---|---|---|---|---|---|---|
| 30 | 12 | 0 | 2,000,000 | 2,145,229 | 0 | 0 | 500,000 | 105 |
| P13 result | Expected (model) | Page |
|---|---|---|
| PA / PB / PC / PD sold (mean) | 291 / 222 / 255 / 114 | Product Status |
| Left in the markets (mean), sold in P14 | 85 / 60 / 73 / 32 | Product Status |
| Costs before revenue | 6,817,998 | Cash Flow |
| Cash before revenue | 8,366 (exact if typed exactly; P12 came in at the accounts' figure less $1) | Cash Flow |
| Sales revenue (+ 160,000 raw-material discount) | 8.87M | Cash Flow |
| Cost / profit before tax / tax | 6.91M / 2.12M / 0.53M | Firm Status |
| Bank loan, repaid at the quarter's end (interest 48k) | 2,145,229 | Cash Flow |
| Credit left after it | 1,498,021 of 3,643,250 | Firm Status |
| Ending cash, after the 500,000 dividend | 5.52M | Cash Flow |
| Net assets / bonds outstanding | 22.85M / 5,325,000 | Major Indices |
| Capital profit ratio / profit per capita | 0.075 / ~5,000 | Firm Status |
From a score subagent's report (2026-10-22, rebuilt from Major Indices and Market Shares P10–P12 with
analysis/score_model.js), with the dividend cut to what the credit allows. Score = 0.4 × this quarter's +
0.6 × last overall, so P16's final = 0.13 S12 + 0.09 C13 + 0.14 C14 + 0.24 C15 + 0.40 C16.
| Index (weight) | Our z | F7 z | Gap × weight | Lever |
|---|---|---|---|---|
| Accumulated dividends (12%) | 0.11 | 1.66 | −0.186 | Pay every quarter, more from P14 on |
| Profit per capita + capital profit ratio (25%) | 0.087 against 0.167 | −0.080 | Profit; no new bonds; idle workers off | |
| Accumulated tax (12%) | 1.21 | 1.32 | −0.013 | Follows profit |
| Profit (20%) | 0.67 | 0.71 | −0.008 | The main lever: +100k a quarter ≈ +0.011 to +0.016 final |
| Net assets (16%) | 1.30 | 0.96 | +0.054 | Falls with dividends |
| Market share (15%) | 1.06 | 0.44 | +0.093 | PA cells near the cap; a PB unit is worth ~3× a PA unit |
confirmed current scores ours 0.747, F7 0.886 (from the published P11/P12 scores); the first five rows rebuilt from the pages. The two ratio indices' split is model.
| P13 | P14 | P15 | P16 | |
|---|---|---|---|---|
| Dividend | 500,000 (the most the credit allows: P14 borrows ~1.31M of the 1.50M left) | 70% of after-tax | 40% of after-tax | 60% of after-tax |
| Profit needed | 2.11M (this form) | ≥ 2.0M a quarter to finish 1st if the others keep their trend | ||
| Workers | 30 hires, 12 retire, for the 305 machines | replace retirements; lay off any worker idle for the rest of the game | no hires | |
| Other | no treasury, no bonds, no machines | no treasury | hold cash for P16 | one raw-material batch of ~3.5M units at the 92% tier (+~140k profit); keep cash + credit ≥ ~8.7M |
| Share values in the planner | PA ~124, PB ~377, PC ~189, PD ~277 $ a unit in P13 (×3 by P16), not the plan's PA 300 / PB 220 | |||
| Others | Our profit | P13 | P14 | P15 | P16 |
|---|---|---|---|---|---|
| Repeat P12 | 2.1M a quarter | 0.794 #2 (F7 0.882) | 0.930 #1 | 1.022 #1 | 1.118 #1 (F7 0.866) |
| Their trend | 2.1M a quarter | 0.759 #2 (F7 0.849) | 0.821 #1 | 0.812 #1 | 0.788 #1 (F2 0.748) |
| Payout race | 2.1M a quarter | 0.756 #2 | 0.816 #1 | 0.807 #1 | 0.778 #1 (F2 0.700) |
| Their trend | 1.8M from P14 | 0.759 #2 | 0.752 #2 | 0.702 #3 | 0.657 #2 (F2 0.772) |
Decided 2026-10-20 from P10's pages (data/runs/FINAL/): the refit demand model, each product's
prices raised until demand + backlog fits what can ship, accounts by analysis/accounts.js (exact to the dollar
on P10's costs before revenue). Machines and hires are sized from demand, then cut to what the cash and credit carry.
| Pricing | M1 | M2 | M3 | M4 | Advertise |
|---|---|---|---|---|---|
| PA | 6,070 | 6,110 | 6,190 | 6,230 | 30,000 |
| PB | 8,590 | 8,800 | 8,900 | 8,950 | 120,000 |
| PC | 11,470 | 11,540 | 11,850 | 11,950 | 120,000 |
| PD | 12,200 | 12,200 | 12,500 | 12,500 | 0 |
| Promote | 80,000 | 80,000 | 80,000 | 80,000 |
| Distribute | M1 | M2 | M3 | M4 |
|---|---|---|---|---|
| PA | 55 | 52 | 75 | 99 |
| PB | 45 | 56 | 109 | 77 |
| PC | 54 | 56 | 113 | 102 |
| PD | 0 | 0 | 0 | 0 |
| Production | Shift I regular | Shift I overtime | Shift II regular | Shift II overtime | R&D |
|---|---|---|---|---|---|
| PA | 298 | 0 | 0 | 0 | 0 |
| PB | 264 | 0 | 0 | 0 | 0 |
| PC | 0 | 0 | 324 | 162 | 0 |
| PD | 0 | 0 | 0 | 0 | 0 |
| New hires | Retire/layoff | New machines | Raw materials | Bank loan | Bond | Treasury | Dividend | Wage % |
|---|---|---|---|---|---|---|---|---|
| 118 | 10 | 43 | 2,028,002 | 0 | 0 | 0 | 0 | 110 |
| P11 result | Expected (model) | Page |
|---|---|---|
| PA / PB / PC sold | 287 / 281 / 317 | Product Status |
| Cash before revenue | 193,342 (exact if typed exactly) | Cash Flow |
| Revenue | 7.97M | Cash Flow |
| Profit (before tax) / tax | 1.79M / 0.45M | Firm Status |
| Ending cash, after the 43 machines | 3.34M | Cash Flow |
| Net assets | 21.34M | Major Indices |
| Capital profit ratio | 0.065 (others ~0.027) | Major Indices |
| Units | P11–P12 (262 machines) | P13–P16 (305), split like demand | All demand at the market average |
|---|---|---|---|
| PA | 298 | ~302 | 315 |
| PB | 264 | ~355 | 370 |
| PC | 486 | ~553 | 579 |
| PD | 0 | 0 | 328 |
| Machine hours | 340,280 of 340,600 | 395,610 of 396,500 | 626,917 (483 machines) |
| Alternative | New machines | What it takes | Why not |
|---|---|---|---|
| Cover all demand at the market average, PA–PD | 221 (483 working) | 22.1M | Over three times what P11 can raise |
| The most P11's cash can pay | 51 | Ends P11 at 2.5M | ~1.4M credit left after P13: too thin for its cash-shortage check |
| Bond for the whole room (3.62M) | 74 | 2.72M still owed at P16; capital profit ratio −0.04 to −0.09 a quarter | The 31 extra machines earn about what the ratio loses, with more risk |
| PD launched in P11 | 43 | 500k R&D, ~0.6M loan; P11 profit −0.2M | Only 193k left before revenue. Moved to P12 |
| A safety bank loan | 43 | 2.25% interest, uses credit | Costs before revenue are exact; type the cells carefully instead |
| Chosen: cash only | 43 | Ends P11 at 3.34M | Keeps ~3M of credit after P13 |
Projection from P11's forecast, give or take ~1M a quarter. Resize every quarter from its own pages.
| Quarter | Opens with | Bank loan | Ends with, before machines | Credit left |
|---|---|---|---|---|
| P11 | 6.80M | 0 | 7.64M (3.34M after the 43) | 8.51M |
| P12 | 3.34M | ~3.6M (~4.2M with PD) | ~5.6M | ~4.9M |
| P13 | ~5.6M | ~1.8M | ~8.0M | ~3.1M |
| P14 | ~8.0M less P13's machines | 0 to ~2.5M | ~10.5M less P13's machines |
| PD share, every market | Market share index | Score a quarter |
|---|---|---|
| 10% | +0.19 | +0.029 |
| 20% | +0.42 | +0.063 |
| 25% | +0.53 | +0.080 |
| PD price | PD units | PC left | PC loses | PD adds | Net a quarter |
|---|---|---|---|---|---|
| 17,500 | 78 | 365 | 376k | 798k | +423k |
| 16,500 | 118 | 303 | 598k | 1,090k | +492k |
| 15,500 | 178 | 211 | 977k | 1,466k | +489k |
Decided 2026-10-19: PC first, PD from P11. The whole P9 form below is typed by hand: don't press Best prices in P9 (the planner can't launch a product). Amber cells differ from what the planner would give.
| Cell | Type | Why |
|---|---|---|
| Whole form | the P9 form below | The planner gives PC no demand curve before its R&D is paid |
| PC price | 12,000 in every market | Between the model's price (11,160–11,580) and what MOCK P10/P11's flatter PC demand asks (12,220–12,860). Being wrong either way then costs ~25–30k instead of 100–120k |
| PD | not launched (R&D 0) | P9 is short of workers: one new product gets them all (PC 144–188 sold instead of 108). PD starts in P11, when the P9 machines work. The model calls the start quarter a tie |
| New machines | 60 | The expansion path: 60 in P9, 30 in P11, 30 in P13 (Machines). The solver would say 0 |
| Wage rate | 110% | A reading: P9's Product Status shows whether quality moves |
| Bank loan | the solver's figure (expect 0) | Cash before revenue stays about 3.8M |
| Pricing | M1 | M2 | M3 | M4 | Advertise |
|---|---|---|---|---|---|
| PA | 6,610 | 6,650 | 6,850 | 6,910 | 0 |
| PB | 8,480 | 8,530 | 8,680 | 8,760 | 30,000 |
| PC | 12,000 | 12,000 | 12,000 | 12,000 | 30,000 |
| PD | 12,200 | 12,200 | 12,500 | 12,500 | 0 |
| Promote | 40,000 | 40,000 | 80,000 | 80,000 |
| Distribute | M1 | M2 | M3 | M4 |
|---|---|---|---|---|
| PA | 6 | 9 | 24 | 33 |
| PB | 48 | 50 | 84 | 98 |
| PC | 40 | 40 | 48 | 74 |
| PD | 0 | 0 | 0 | 0 |
| Production | Shift I regular | Shift I overtime | Shift II regular | Shift II overtime | R&D |
|---|---|---|---|---|---|
| PA | 8 | 26 | 0 | 0 | 0 |
| PB | 0 | 0 | 244 | 62 | 0 |
| PC | 193 | 0 | 4 | 56 | 300,000 |
| PD | 0 | 0 | 0 | 0 | 0 |
| New hires | Retire/layoff | New machines | Raw materials | Bank loan | Bond | Treasury | Dividend | Wage % |
|---|---|---|---|---|---|---|---|---|
| 75 | 5 | 60 | 2,000,000 | solver's (0) | 0 | 0 | 0 | 110 |
The planner's demand model with the held-out P9 correction, sales by the game's rules, accounts by
analysis/accounts.js (wage 110%, 60 machines). Computed 2026-10-19 before P9 was played. The model's
error after the correction was about ±18% a cell when held out.
| P9 result | Expected (model) | If PC is less price-sensitive (MOCK P10/P11) | Page |
|---|---|---|---|
| PA sold | 109 (17/18/42/32) | 109 | Product Status |
| PB sold (incl. P8 backlog) | 273 (47/49/82/95) | 273 | Product Status |
| PC sold | 144 (25/27/38/54) | 188 (34/38/47/69) | Product Status |
| PC left in the markets | 54 | 9 | Product Status |
| New backlog | none | none | Product Status |
| Revenue | 4.79M | 5.32M | Cash Flow |
| Profit (before tax) | 0.91M | 1.03M | Firm Status |
| Tax | 0.23M | 0.26M | Cash Flow |
| Ending cash, after the 60 machines | 2.33M | 2.82M | Cash Flow |
| Net assets | 19.50M | 19.59M | Major Indices |
| Machines bought | 60, none cut | 60 | Firm Status |
| Quality | ≥ 0.981 if 110% works | Product Status |
| Pricing | M1 | M2 | M3 | M4 | Advertise |
|---|---|---|---|---|---|
| PA | 6,610 | 6,650 | 6,850 | 6,910 | 0 |
| PB | 8,620 | 8,660 | 8,970 | 9,100 | 30,000 |
| PC | 12,270 | 12,500 | 12,580 | 12,610 | 30,000 |
| PD | 15,820 | 16,290 | 16,180 | 16,320 | 30,000 |
| Promote | 40,000 | 40,000 | 80,000 | 80,000 |
| Distribute | M1 | M2 | M3 | M4 |
|---|---|---|---|---|
| PA | 6 | 9 | 24 | 33 |
| PB | 44 | 43 | 69 | 81 |
| PC | 22 | 21 | 29 | 40 |
| PD | 20 | 14 | 22 | 34 |
| Production | Shift I regular | Shift I overtime | Shift II regular | Shift II overtime | R&D |
|---|---|---|---|---|---|
| PA | 0 | 0 | 0 | 0 | 0 |
| PB | 255 | 0 | 0 | 0 | 0 |
| PC | 22 | 0 | 88 | 34 | 300,000 |
| PD | 0 | 0 | 71 | 42 | 500,000 |
Same last row as above. Expected (model): sold PA 109, PB 231, PC 108, PD 87; profit 0.90M; ending cash 2.64M; net assets 19.49M.
Run Best prices, then Plan ahead 3, on P9's pages for prices, spend, production and distribution. Then type the amber cells over it.
| Cell | P10 |
|---|---|
| New hires | 40 (keep the solver's figure if it is higher) |
| New machines | 0 (next: 30 in P11, 30 in P13) |
| Wage rate | 110% if P9's quality ≥ 0.981 (115% if ≥ 0.991), else 105% |
| R&D | 0 (PD launches in P11) |
| Raw materials | the plan's (P9's 2M batch covers most of P10) |
| Bank loan | the solver's figure (P10 opens near 2.3M; the bank tops it up) |
| Bond | 0 |
| Dividend | 0 (cash goes to P11's machines and material) |
| Treasury | 0 |
| Planner input before Best prices | P10 |
|---|---|
| Quarters left | 7 |
| Share value PA / PB / PC | 70 / 75 / 100 |
| Others' expected change | 0 (unless expanding, below) |
| Demand growth | 0 |
| PC price effect | 5.5, or 4.5 if PC sold > ~165 in P9 |
| If P9 shows | Then in P10 |
|---|---|
| Quality < 0.981 | Wage back to 105% |
| PC sold > ~165 (expected 144) | PC price effect 4.5 in the planner: PC is less price-sensitive |
| ≥ 4 others with Ending Cash < 5.0M, or ≥ 6 at Workers rank 1 | The others are expanding: their expected change −2% (P10), −4% (P11); follow about half |
| ≥ 6 other PC sellers | Earlier PC entry than MOCK: the planner reads it from Prices by Firms |
| Who sells PD, and at what price (Prices by Firms) | PD's P11 price: 20–26% over the others' PD average |
Overall = 0.4 × this quarter's score + 0.6 × the last overall, from 0 at P8. Each quarter's own score then weighs in the final:
| Quarter | P9 | P10 | P11 | P12 | P13 | P14 | P15 | P16 |
|---|---|---|---|---|---|---|---|---|
| Weight | .011 | .019 | .031 | .052 | .086 | .144 | .240 | .400 |
confirmed Formula rebuilt on all 12 firms for MOCK P9–P11 (workstreams A, B, E).
"Held out" means the test used data the method was not fitted on.
| Test | What it checked | Result | Status |
|---|---|---|---|
| Score formula rebuild (A, B, E) | Every MOCK firm's published score, P9–P11, from Rules.md | 0.4/0.6 timing, z capped at ±3, share = sum over 16 cells ÷ 16. Error 0.002–0.006 rms, all from guessed headcounts; other timings fail by thousands on F10 | confirmed |
| MOCK replay chain (A) | P9–P11 rebuilt from the forms and the rules' accounts | Matches every P9 Cash Flow line; P10 off by one PA return (a page quirk) | confirmed |
| MOCK counterfactuals (A) | One decision changed at a time, carried P9 → P11 | Pricing to clear in P9 +0.110; P11 dividend 500k +0.021; the full dividend −0.053 | model dividend rows confirmed |
| P9 demand correction (F) | Each restart's P9 corrected by the other five runs | Error 0.241 → 0.178 rms; better in 6 of 6 runs (MOCK 0.153 → 0.041); holds with every demand effect moved | held out P9 only, same news |
| P9 form robustness (F) | The PC-only form typed in other worlds | Others ±2 points ≤ 26k; demand ±10% ≤ 63k; PA 10 / PC 4 slopes 118k (reverse 100k) | model PD form not run there |
| Other firms' scenarios (B) | MOCK per-firm units, workers, cash, dividends → three futures | F12's units rebuild its revenue to the dollar; teams repeated P9 within ~1% in 4 runs | scenario |
| 8-quarter simulator (C) | Machines × hires × PD start × finance, P9–P16 | See Machines, PD, Dividends. Stresses: demand ×0.8/×0.9/×1.1, PD ×0.6/×0.3, PC/PD slopes 4, others paying early or expanding | model noise ±0.05 |
| Product LP (D) | One steady quarter: PC/PD value, machine and worker values, level and wage break-evens | PC made and sold the quarter its R&D is paid. PD curve at MOCK P11 sellers' prices: 443 units against their ~483 | confirmed PC timing · one check PD |
| Lever values (E) | What profit, dividends, treasury, bonds, workers and share are worth each quarter | The dividend method on P10's pages alone picked 514k for P11; the hindsight best was ~500k | model one out-of-sample decision |
| Planner version check (F) | Old and new core.js on the same inputs | Same cells; Plan ahead 3 gives an identical objective | confirmed |
| Tax after a loss | Rules.md against accounts.js on MOCK F8 | The credit comes off the tax, not the profit; accounts.js fixed | confirmed |
Plan: R&D 300,000 in P9, PC 12,000 in every market, 30k advertising from launch. P9's workers go to PC alone (PD from P11).
| Avenue | Result | Verdict |
|---|---|---|
| Launch in P9 | +0.23–0.30M of P9 profit over launching in P10; the 300k is the same either way. Paid back during P10 | plan |
| Launch in P10 | Loses that P9 profit | no |
| Price at or under the market (MOCK P9: 9,524–10,000) | Sold out, with backlog that sold in P10 at P9's price. The same 53 units at their clearing price: +0.093..+0.154 | no |
| The PC-only P9 form (PC 11,160–11,580) | 0.011–0.048 below the PD form; costs 118k if PC's slope is −4 | fallback |
Tests: D's LP, A's counterfactual, F's robustness. confirmed launch timing · model value.
Plan (2026-10-19): R&D 500,000 in P11, when the P9 machines start working, priced 20–26% over the others' PD average (MOCK's sellers were at 104–134% by P11). P9's workers go to PC. The table below is from the earlier runs: PD in P9 or P11 is a tie.
| Avenue | Score | Verdict |
|---|---|---|
| PD from P9 + 60 machines | 1.708 | plan |
| PD from P11 + 60 machines | 1.697 | tie |
| PD from P9, no machines | 1.487 | no |
| No PD, no machines | 1.310 | no |
| No PD, 60 machines | 1.299 | no |
| PD sells, of its shown demand | Scores | Then |
|---|---|---|
| About as forecast | 1.708 | Keep the plan |
| 60% | 1.51 with machines · 1.37 PD alone · 1.31 neither | Keep PD, buy no more machines |
| 30% | Machines tie (1.278 each); PD alone −0.03 against no PD | Re-plan PD on its real P9 curve, buy nothing more |
Tests: C's simulator, D's LP and stress. weak PD has never been sold; its curve is our shown demand, checked once (443 against ~483 units).
Plan: sell all four every quarter. When machines bind, cut PD, then PC, PB, PA last. When labour binds, cut PA first.
| At MOCK P11's market price | PA | PB | PC | PD |
|---|---|---|---|---|
| Machine hours a unit | 200 | 290 | 420 | 650 |
| Margin per machine hour | 25.2 | 20.8 | 16.7 | 14.5 |
| Margin per labour hour | 22.9 | 33.6 | 35.0 | 39.3 |
Rules arithmetic (confirmed) at a model price.
Plan: 60 machines in P9, 30 in P11, 30 in P13: 280 working from P15, then P15–P16 harvest with no more machines. Hires the quarter before machines work: 40 in P10, 43 in P12, 43 in P14, else replace the 3% retiring, none in P16. Machines are paid from the quarter's own cash; each next quarter covers its costs before revenue with a bank loan. No bonds. Keep at least ~3M of the 11M credit unused. PD starts in P11 (the expansion runs launched it in P9; the start quarter was a tie in every earlier pair).
| Quarter | P9 | P10 | P11 | P12 | P13 | P14 | P15 | P16 |
|---|---|---|---|---|---|---|---|---|
| Machines bought | 60 | 0 | 30 | 0 | 30 | 0 | 0 | 0 |
| Machines working | 160 | 160 | 220 | 220 | 250 | 250 | 280 | 280 |
| Hires | 75 | 40 | ≈ 8 | 43 | ≈ 8 | 43 | ≈ 8 | 0 |
| Units sold (model, flat demand) | 537 | 597 | 683 | 664 | 734 | 896 | 872 | 953 |
| Profit (model), M | 0.9 | 0.9 | 2.0 | 1.8 | 2.0 | 2.1 | 2.4 | 2.5 |
Every path launches PC and PD in P9; hires keep about 1.15 workers per working machine. "Loans" = later quarters borrow their costs before revenue so cash goes to machines, keeping ~3M of credit; a bond only for what that can't cover. Same solver settings for all.
| Machines bought | Total | Flat demand | +2% a quarter | Debt | Verdict |
|---|---|---|---|---|---|
| 60 P9, 30 P11, 30 P13 | 280 | 1.791 | 1.877 | loans 7.7M, 3.3M credit left; bond 0.6M | plan |
| 85 P9, 60 P11 | 305 | 1.750 | 1.875 | bond 6.2–6.4M + loans | tie with growth |
| 120 P9 | 280 | 1.755 | 1.871 | bond 6.0M + loans | no |
| 85 P9, 40 P11, 40 P12 | 325 | 1.751 | 1.864 | bond 6.8M + loans | no |
| 60 P9, 30 P12 | 250 | 1.774 | 1.859 | none / loans | no |
| 60 P9 only | 220 | 1.727 | 1.812 | none | no |
| 100 P9, 50 P11, 50 P12 | 360 | 1.620 | 1.851 | bond 10.8M + loans | no |
| 85 P9, 60 P10, 60 P11 | 365 | out of credit | 1.086 | 20M of loans needed | can't fund |
| 120 P9, 60 P10, 40 P11 | 380 | out of credit | out of credit | 40M+ of loans needed | can't fund |
60 in P9 first on every path. Hires 43 (for 30) or 77 (for 60) the quarter before the new machines work. All 14 runs with the same solver settings and the plan's finance rule, so they compare like for like.
| After the 60 in P9 | Demand flat | Demand +2% a quarter | Bonds needed | Verdict |
|---|---|---|---|---|
| Nothing more (buy once) | 1.727 | 1.812 | 0 | no |
| +30 in P12 | 1.774 | 1.848 | 0 (0.8M with growth) | plan |
| +30 in P11, +30 in P13 | 1.759 | 1.857 | 1.9M in P11 | if demand grows |
| +60 in P12 | 1.759 | 1.833 | 2.4–3.9M in P12 | tie |
| +30 in P13 | 1.755 | 1.842 | 0 | tie |
| +30 in P11 | 1.752 | 1.827 | 1.9M in P11 | tie |
| +60 in P11 | 1.732 | 1.844 | 4.3M in P11 | no |
| 90 in P9 instead of 60, nothing later | 1.739 | 1.837 | 3.4M in P9 | no |
| 120 in P9 instead of 60, nothing later | 1.738 | 1.851 | 6.5M in P9 | no |
| Machines bought | PD from P9 | PD from P11 | Bond needed | Verdict |
|---|---|---|---|---|
| None | 1.487 | 1.439 | 0 | no |
| 60 in P9 | 1.708 | 1.697 | ≈ 0.2M | plan |
| 90 in P9 | 1.739 | 1.706 | 3.3M | tie |
| 120 in P9 | not run | 1.703 | 6.4M | tie |
| 60 in P9 + 60 in P10 | not run | 1.696 | 3.3M in P10 | no |
| 90 in P10 instead | 1.719 | 1.671 | no | |
| 60 in P10 instead | not run | 1.657 | no |
| World (PD from P11) | 60 | 90 | 120 | None |
|---|---|---|---|---|
| Base | 1.697 | 1.706 | 1.703 | 1.439 |
| Our demand ×0.8 | 1.317 | 1.300 | 1.235 | 0.931 (no PD) |
| Our demand ×0.9 | 1.523 | — | 1.505 | — |
| Our demand ×1.1 | 1.790 | 1.818 | 1.816 | — |
| PD demand ×0.6 | 1.514 | 1.503 | — | 1.365 (PD) |
| Others pay early | 1.670 | 1.648 | 1.640 | — |
| Others expand early | 0.923 | 0.887 | 0.916 | 0.562 (no PD) |
Tests: C's simulator (analysis/final_C_run.js, second purchase re-run 2026-10-19;
growth is the scenario option growth). model Can't see demand at 1.5–2× our
volume or the others' reaction to us. The buy-once path reads 1.727 here against 1.708 in the earlier table:
solver tolerance, which is why only same-run comparisons count.
Plan: 75 in P9 (the cap), 40 in P10, then replace the 3% retiring (≈ 8 a quarter) and hire the quarter before new machines work. No hires in P16.
| Avenue | Result | Verdict |
|---|---|---|
| 75 hires in P9 | +0.26 against 30 hires. Labour binds at P8: 95–100% of hours used, machines 74–87% | plan |
| Hires to ≈ 1.15 × working machines | 220 machines from P11 → ~253 workers → 40 hires in P10 | plan |
| Workers ≈ 1.0 × machines | The early machine paths were labour-short and scored lower | no |
| Layoffs to raise profit per capita | The dismissed still count in the denominator (Rules.md) | no |
| Lay off workers P16 leaves idle | Saves ~3,460 each, within the 10% limit including retirements | P16 only |
Tests: C's capacity grid, E's levers. confirmed labour binds at P8 · model value.
Plan: no bonds. Bank loans are cheaper (9% a year, one quarter at a time) and stop when you stop drawing. If a machine bill is more than the cash, buy fewer that quarter and the rest the next.
| Avenue | Result | Verdict |
|---|---|---|
| No bond, 60 machines from cash | P9 ends at about the 2.5M floor; the bank tops up P10 for a quarter if needed | plan |
| Bonds to fund 90 or 120 machines in P9 | 3.3M / 6.4M of bonds for a score that ties with 60 | no |
| Bond for machines bought later | Only the part the spare cash can't pay | if needed |
| Bond to hold a cash reserve | An early reserve rule issued 13% bonds for nothing; fixed | no |
| Bond to pay dividends | Costs more than the dividend earns | no |
A bank loan can't pay for a machine. Rules.md: the loan arrives at the start of the quarter and is repaid with interest at the end of the same quarter, and the cash order repays it before tax, machines, dividends and treasury. A bond's money stays (5% of it repaid a quarter). A loan can pay for machines only the long way round: our own cash buys the machines, and the next quarters borrow their costs before revenue. That is cheaper than a bond (9% a year, one quarter at a time), but every draw uses up the 11M credit for the rest of the game.
| Same path, financed with | Bonds (plan) | Loans, 3–4M credit left | Loans, credit used up |
|---|---|---|---|
| 60 in P9, +30 in P12 (needs no bond) | 1.774 | 1.766 | 1.762 |
| 60 in P9, +60 in P11 (4.3M of bonds) | 1.732 | 1.761 | 1.784 (0.2M left) |
| 120 in P9 (6.5M of bonds) | 1.738 | 1.755 (1.5M left) | over the credit |
| +1M bond held to the end, issued in | P9 | P11 | P13 | P15 | P16 |
|---|---|---|---|---|---|
| Final score | −.030/−.034 | −.027/−.031 | −.023/−.026 | −.014/−.015 | −.004/−.005 |
Tests: E's levers, C's finance replay. model
Plan: loans fund the expansion: each quarter's cash buys machines, and the next quarter borrows its costs before revenue (type the solver's bank loan). About 8M over the game; keep at least ~3M of the 11M credit unused.
Plan: 0.75 × the planner's profit forecast × 40%; 50% while our Accum. Dividends are under 80% of the top other firm's, 30% while over 120%. Only from cash beyond max(2.5M, next quarter's costs before revenue − the loan we accept). Starts P10–P11; P16 at 50–60%.
| Avenue | Against the rule | Verdict |
|---|---|---|
| The 40% ± 10% rule | Within 0.013 of the best per-quarter shares in all five scenarios | plan |
| Pay nothing | −0.10 to −0.22 | no |
| Pay the full limit every quarter | −0.17 to −0.24 (MOCK P10 + P11: −0.053) | no |
| About 3M late, P14–P16 | −0.03 to −0.08 | no |
| Start in P9 instead of P10 | ≤ 0.003 either way; P9's cash sits at the floor after the machines | tie |
| Start in P12 | −0.02 to −0.05 | no |
| Best schedule found (up to 80–100% in P10, 50–60% in P16) | +0.02–0.04 over the rule on C's paths | hindsight |
Tests: E's rules over five scenarios, A's MOCK counterfactuals, C's finance replay. model
Plan: what is left after dividends and machines, above next quarter's needs before revenue. Never in P16.
confirmed interest is income (P8 Cash Flow) · model value.
Plan: 110% in P9 as a reading. Keep it if P9's quality reads ≥ 0.981, else back to 105%.
| Wage | Cost or effect (P9 plant) | Pays if | Verdict |
|---|---|---|---|
| 100% | Saves 77k but quality falls to ~0.950: −35k the first quarter, −81k after | never | no |
| 105% | Quality 0.974 (MOCK). Every MOCK firm stayed here | — | fallback |
| 110% | +77k a quarter (97k at 217 machines) | quality reaches 0.984–0.986 | plan |
| 115% | +5 points more | quality 0.993–0.998 | if ≥ 0.991 |
Tests: D's LP. unmeasured wage → quality above 105%.
Plan: no steps. If a quarter's plan leaves machines and workers idle, the bets worth taking: PC to level 2 (from P10) or PB to level 2. Never PA.
| Break-even demand gain, step in | P9 | P10 | P11 | P12 | P13 | P14 | P15/16 |
|---|---|---|---|---|---|---|---|
| PA (120k) | 15% | 18% | 21% | 25% | 31% | 41% | 62% |
| PB (150k) | 4.6% | 5.3% | 6.1% | 7.4% | 9.2% | 12% | 18% |
| PC (150k) | — | 2.8% | 3.3% | 3.9% | 4.9% | 6.6% | 9.9% |
Tests: D's LP. confirmed costs · unmeasured effect.
Plan: Best prices, then Plan ahead 3, every quarter from P10 on that quarter's pastes. Quarters left = 16 − the last played quarter.
| Avenue | Result | Verdict |
|---|---|---|
| Price to the units we can ship when sold out | MOCK P9 left +0.09 to +0.11 by pricing 3–10% under the market while selling out | plan |
| PA well above the market | +25% over the others in MOCK P10 sold 7–8 units in M1/M2 | no |
| Promotion ≤ 80k a market | Above 80k is unmeasured and fell short the one time tried (−0.018..+0.023) | plan |
| The others cut: follow about half | Following costs ~160k a quarter; holding prices loses 520–660k | if they cut |
| The others cut: hold our prices | 0.161 against 0.604 following (yardstick path, others expanding) | no |
| Share value per unit, $ | P10 | P11 | P12 | P13 | P14 | P15 | P16 |
|---|---|---|---|---|---|---|---|
| PA | 70 | 130 | 200 | 300 | 400 | 490 | 560 |
| PB | 75 | 100 | 150 | 220 | 300 | 370 | 430 |
| PC, PD | 100 | 105 | 170 | 250 | 340 | 420 | 490 |
Advertising per product: 0, 30k or 120k (the measured steps); PC and PD at 30k from launch. Tests: A's counterfactuals, B's scenarios, E's share values. model
Plan: 2,000,000 in P9 (covers P9–P10), one batch ≥ 2,000,000 in P11, and one large batch in P16 sized so cash + credit left ≥ 1.1 × P16's costs.
| Extra raw material in P16 | 0 | 2M | 4.5M | 8M | 10M | 12M | 14M |
|---|---|---|---|---|---|---|---|
| Score | 1.660 | 1.679 | 1.695 | 1.712 | 1.722 | 1.718 | 1.623 |
| P16 bank loan | 0 | 0 | 1.8M | 5.0M | 6.9M | 8.7M | over credit |
| Scenario | What the others do | Our plan scores | We see it |
|---|---|---|---|
| repeat_mock (base) | MOCK P9–P11, then their P1–P8 price trend | 1.697 | Others' P9 prices within ±1% of MOCK P9 |
| payout_early | Every active firm pays 60% from P9, 90% in P16 | 1.670 | ≥ 5 payers or ≥ 500k paid in P9 |
| expand_early | 7 firms hire 75/60/40 and buy ~55 machines in P9; their prices fall 2–6% from P10 | 0.923 | ≥ 4 with Ending Cash < 5.0M, ≥ 6 at Workers rank 1; P10: ≥ 5 above the default Machines rank |
| Firm | Style |
|---|---|
| F1 | Buys ~60 machines in P9 every run, barely uses them; losses P9–P10 |
| F2 | Cuts PB in P10 and leads PB volume; machines in P10; dividends from P10 |
| F3 | PB price leader (7,150–8,300); hoards cash, no dividends |
| F4 | No hires in P9; PC from P10, PD from P11; dividends 20k/100k/500k |
| F6 | Biggest hirer in P10, no machines, no PC/PD; profit sliding |
| F7 | MOCK's winner. PC leader, 75 hires in P9, dividends every quarter (700k in P11) |
| F8 | PD from P9; lost 1.65M in P10, back to 1.38M in P11 |
| F9 | PC and PD from P9, both dear; no dividends |
| F10 | PD leader (~12,300 in P9, then 15,000–20,000); 150 workers all game |
| F11 | ~50 machines in P9; PC from P10; dividends 50k/450k |
| F12 | Repeats the P8 form every quarter |
Tests: B's per-firm rebuild and scenarios. scenario Can't see the others' backlog, PD totals beyond ±30%, or machines bought in P11.
MOCK: 0.791, 2nd of 12 (F7 1.013). P9 alone decided it: F7 − F5 = +0.298 (P9) − 0.069 (P10) − 0.006 (P11).
| Change to what we played | Score after P11 |
|---|---|
| P9 PA/PB priced to clear (+2% to +13%), not 3–10% under the market | +0.110 |
| P9 PC: the same 53 units at their clearing price | +0.093..+0.154 |
| P9 PC volume 150–185 by making no PA | +0.045..+0.084 |
| P10 PA raise not made | +0.009..+0.010 |
| P11 promotion capped at 80k | −0.018..+0.023 |
| P11 dividend 500–600k instead of 0 | +0.021 |
| The full allowed dividend in P10 and P11 | −0.053 |
| Clearing prices + P11 dividend + PC price, together | 1st (1.018..1.058) |
Tests: A's chain and counterfactuals. model except the dividend rows (confirmed).
Found while planning; patches are in data/final/patches/, not applied.
| Gap | Effect | Workaround |
|---|---|---|
| Best prices can't launch a product | No curve for PC or PD before their level-1 R&D | Type the P9 form by hand |
| A machine's cash after revenue isn't modelled | Its 100k at the quarter's end, the tax and next quarter's 2.5M minimum are missing | The finance rules on this tab; check Feasible Decisions |
| Later quarters planned on this quarter's workforce | Undervalues machines: buys 0 in P9 | Take machines from this tab |
| "Costs this quarter" includes later storage | About 139k too high in P9 | Display only |
| Unknown | The plan assumes | If wrong | First seen |
|---|---|---|---|
| PD's real demand | Our shown PD demand, slope 5.75 | At 60%: still best. At 30%: machines tie, PD −0.03 | P11 PD sales |
| The final's P9 news | Same as MOCK (M3 recovering) | M3 over-supplied in P9; the stock sells in P10 | P9 news, before submitting |
| PC's price slope | 5.5 | At 4: type 4.5 from P10 | P9 PC residual |
| Demand growth | None in the base; +2% a quarter tested | With growth, buy earlier: 30 in P11 and 30 in P13 | P10–P11 demand at the same price against the market |
| Demand at 1.5–2× our volume | The same model | ×0.8 / ×1.1: same ranking | P11 on |
| Wage → quality above 105% | Unknown; break-even 0.984–0.986 | ≤ 77k | P9 Product Status |
| R&D level effect | None | Not used | — |
| Market-level effect | −3.4 | Sizes how much expansion hurts | — |
| Leftover raw material at $1 | One observation (MOCK P10) | P16 batch worth 0, not +0.06 | P16 |
| The others' reaction to us | Fixed scenarios | — | Each quarter's Prices by Firms |