Version 2.1.02026-10-19Design revision 9

The bank

Runs: Run 14

Show

No replayThere is no recorded replay of this version; 2.7.0 is the one version with one. The results below are the study's own tables.

A bank account alone now creates money; the orchestrator's losses become debt, not printed money.

What went wrong

The orchestrator's failed deliveries were covered by creating money, so the deficit grew every tick.

A fifth of the orchestrator's deliveries failed, each losing about 0.3 of the pricecreated money covered it, 10 to 20 credits a tick

59,700the deficit from hoarded rebates under a flood of customers

Under the rules before this one, the orchestrator also did jobs itself. It paid its cost when it started a job and got that cost back only when the job passed. About one job in five failed. Then it got nothing back and also lost the guarantee it had put up. It covered each loss by creating new money, and the gap grew every tick for the rest of every run. When it had to turn customers away for lack of room, it created money to compensate them. They could not spend that money, so it piled up. All of this showed as one number, and nobody could tell what that number meant. Baruchi asked for a central bank that stands apart from the orchestrator. It was agreed as a second account with rules, not as a second player.

Under 2.0.0 the orchestrator delivered work itself: it burned the full price B̄ when a delivery was summoned and recovered it when the work vested. A fifth of its deliveries failed: it recovered nothing and forfeited its guarantee, losing about 0.3 of the price per delivery. That loss was covered by creating money, so the deficit grew by 10 to 20 credits a tick for the rest of every run. Under a flood of customers, every refusal for lack of capacity created a rebate that could not be spent and was hoarded (a deficit of 59,700). The price index moved within a period, which broke three tests the design reviewer had written to pin a case's amounts. And the deficit was one figure that mixed money lent out, money lost and rebates owed. Baruchi's frame was the orchestrator as a living organism regulating the market, with a central bank as a separate entity; it was agreed as a second account with rules, not a second actor.

The state at cost was below cost — $c^D_T=\bar B$ burned at the summons, $\bar B$ recovered at vesting, and a fifth of the state's deliveries end in a final $\mathsf N$ , recovering nothing and forfeiting $G$ : about $0.3\,\bar B$ a delivery, created to cover it, the deficit growing 10–20 a tick for the rest of every run (run 12; 2.0.0's Hypothesis 7 failed). The rebate was a pump: under the flood a refusal for capacity meant a rebate created, unspendable, hoarded — deficit 59,700, $M$ 99,400. The index moving within a period broke three reviewer tests that pin a case's amounts. The deficit could not be read: capital in flight, losses and rebates owed were one figure.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1320–1320

Baruchi's frame: the state as a living organism regulating the market, and a central bank as a separate entity — agreed as a second account with rules, not a second actor.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1320–1320

Baruchi's position [C-2.1]: the state is "like a living organism, sending and retrieving troops all the time to regulate the market… turning the knobs to stabilize"; and, asked for real, a central bank as a separate entity — agreed as a second account with rules, not a second actor: mint by rule, publish the index, lend only against what comes back; losses from the state's income "or they don't come out at all"; when income does not cover losses, "charge for it, or retract."

Source: spec/design_revisions/dr9_the_bank_2.1.0.md l.28–28

What changed

A bank account alone creates money; it lends the orchestrator, and unpaid loans become debt.

bank lends a delivery's shortfall orchestrator repays from every inflow first bank

No loan while the debt is past one period of the orchestrator's fee income

Rebates the orchestrator cannot pay become a liability, not new money

Now a second account, the bank, is the only one that creates or destroys money. It creates money only by fixed rules. It lends the orchestrator what a job is short of, and the job's own return pays the loan back first. Whatever is not paid back becomes the orchestrator's debt. Every credit that comes into the orchestrator's account pays the debt first. If the debt grows past one period of the orchestrator's fee income, the bank stops lending for new jobs, and the orchestrator has to turn them down. Every refusal is marked on the request, where the customer can see it. What late customers are owed is kept as a debt on the books, and it is never paid with new money. The orchestrator now pays up front only its expected cost, worked out from how often its own jobs failed recently.

The bank is a second account that alone creates and destroys credits. It mints by rule (the customers' issuance, emission, the pots of a plant, which the record calls the cost of truth, and a declared shock) and publishes the price level once a period: the closed period's mean awarded price. It lends the orchestrator the shortfall of a delivery, repaid from that delivery's return before the orchestrator's account keeps anything. What is not repaid is the orchestrator's debt, repaid by every inflow the moment it arrives, before rebates owed and before the orchestrator keeps anything. The bank lends for a delivery only while the debt is within "the wall", one closed period of the orchestrator's fee income; a delivery refused for credit is marked and the customer told. The orchestrator now burns its expected cost, using its own trailing failure rate, and still delivers at B̄, so it never undercuts a member; this is neutral in expectation up to a failure rate of two thirds. Rebates the orchestrator cannot pay become a liability, never created money.

What 2.1.0 is. A second account, the bank, that alone creates and destroys credits. It mints by rule — the requesters' issuance, emission, a plant's two pots (the cost of truth), a run's declared shock — publishes the price level once a period (the closed period's mean awarded price, carried if none; the index constant within a period, I31), and lends the state the shortfall of a summons tied to it, repaid from the summons's own return before Ω keeps anything. What a settled summons did not repay is the state's debt, repaid by every inflow to Ω the moment it arrives, before the rebate liabilities (1.6.0's stock again, never created) and before Ω keeps. The bank lends for a delivery only while the debt is within the wall — one closed period of Ω's fee income, base fees plus the troops' review fees — and for a seat always; a delivery refused for credit is marked and told, the state has retracted. The state burns its expected cost c^D_T = max(0, (1 − b_T) B̄ − b_T G) with b_T its own trailing fail rate (the last closed period if it closed ten, else back until ten, else 0.2) and still delivers at B̄, so it never undercuts a member; neutral in expectation up to b_T = 2/3. Every refusal — for capacity or for credit — marks the request with its kind and tick, readable in the public view; the requester may hold or withdraw. Creation, retirement and the deficit are retired as rules (the control, Params.bank off); Ω's net position = Ω_U − debt − liabilities.

Source: HANDOFF.md l.1757–1767

What the runs found

The orchestrator funded every delivery itself, but its unpaid lateness bill reached 27,000 in a flood.

0loans, debt and liabilities under the expected-cost rule, at every split

A flood of customersa lateness liability of 27,000, nothing printed for it

The inflation test: price level against the quantity theory

credits

P_t end, wallet ruleissuance ÷ (0.278 × free)

20 honest, β_max 60; means over 20 seeds. Source: agentic_sim/results/studies/2.1.0-bank-notes-for-reviewer.md l.51–54

Debt peak: expected cost against the at-cost control

credits

expected: debt peakat cost: debt peak

Stampers 50–80%; the wall at one period of fee income. Source: agentic_sim/results/studies/2.1.0-bank-notes-for-reviewer.md l.11–21

The riots: lateness liabilities against fee income

credits

liabilities at the endΩ's fee income (base + review)

60/40, N_T 40. Source: agentic_sim/results/studies/2.1.0-bank-notes-for-reviewer.md l.71–80

In a normal market the orchestrator came out about even. It paid for every job from its own account, and the bank's limit was never reached. When the old rule was run as a test, the orchestrator went into debt, hit the limit and turned thousands of jobs away. When its work was made to fail half the time, the new rule covered it. When its work failed three times in four, the debt crossed the limit and its customers were left owed money. Giving customers more money raised prices about as the theory predicts. With the least money, prices fell for the first time. Honest workers did not do equally well at every amount of money. In a rush of customers, no money was printed any more. But what the orchestrator owed its late customers grew to 27,000 in the worst rush, and its whole fee income was only a quarter of that.

Run 14 had 600 runs of 1,000 ticks, 20 seeds a cell. Under the expected-cost rule the orchestrator's account netted −0.6 to −0.8 a delivery (2–3% of the price) and funded every delivery itself: loans, debt and liabilities were zero at every split, and the wall never bound. The old at-cost rule, run as a control, now retracts instead of bleeding: debt of 500–650 against a wall of 210–350, 1,900–10,600 refusals for credit, and liabilities of 2,400–8,800. The red team found the two-thirds limit where the text put it: at a failure rate of one half the automatic charge covered everything; at three quarters the burn was zero, the debt crossed the wall, and the customers were owed 8,900 in rebates. When half the careful inspectors rejected every delivery of the orchestrator's, its failure rate went to 0.67 and it became 63% of the market at a loss; those inspectors had to read who did the work, which the inspectors' view does not carry. Issuance of 5 / 10 / 20 gave a price level of 9.5 / 23 / 54 against a prediction of 11 / 23 / 47; at 5 the price fell to cost, the first deflation, and real wealth by disposition was not the same across the three. In the riots the money stock was issuance alone and the debt was repaid. But the lateness liability reached 27,000 in the flood and 10,000 under the crowd, ageing 400–500 ticks, with the orchestrator's whole fee income a quarter of it. Nothing was printed for it, and the customers waited for money the orchestrator did not have.

What ran (results/studies/2.1.0-bank/, 600 runs, 20 seeds a cell, 1,000 ticks, all audits passed at f081e15f; notes results/studies/2.1.0-bank-notes-for-reviewer.md)

Source: HANDOFF.md l.1797–1798

Findings. Neutral, and the bank unused: under the expected cost Ω nets −0.6 to −0.8 a delivery (2–3% of B̄, b_T's lag and the troop seats' bonds), keeps 4,300–4,500, funds every summons itself — loans, debt and liabilities zero at every split; the wall never binds. The at-cost control under the bank retracts instead of bleeding: debt 500–650 against a wall of 210–350, 1,900–10,600 refusals for credit, liabilities 2,400–8,800, on time 0.79–0.89, the requesters hoarding — "the burn rule is wrong, not the wall", in the record.

Source: HANDOFF.md l.1803–1806

The red team finds the two-thirds limit where §7.7 put it: at ½ the automatic charge covers everything (b_T 0.52, burn 8.4, no debt, Ω 4,543 as in the honest cell); at ¾ the burn is zero, the debt crosses the wall, the state retracts and its customers are owed 8,900 of rebates (the unbounded control's debt 4,900 and rising, run 12's bleed legible); the rejecters drive b_T to 0.67 — the limit — and make the state 63% of the market at a loss; they read the case's actor (the view carries none) and so never meet a plant.

Source: HANDOFF.md l.1809–1813

redteam/ (100: the state's deliveries made to fail at a half and at three quarters × the wall 1 / unbounded; half the careful actors rejecting every delivery of the state's)

Source: agentic_sim/results/studies/2.1.0-bank-notes-for-reviewer.md l.3–3

The inflation test gives the quantity theory and the first deflation: issuance 5 / 10 / 20 → P_t 9.5 / 23 / 54 (predicted 11 / 23 / 47); at 5 the price falls from 20 to cost — the credit worth more than it was — and real wealth by disposition is not invariant (790 / 1,410 / 1,310: at low issuance the ratchet never starts, at high the requesters hoard a third of M).

Source: HANDOFF.md l.1813–1816

The inflation test gives the quantity theory to the unit at the middle and the first deflation seen in any run at the bottom: issuance 5 / 10 / 20 a requester at 20 honest gives P_t 9.5 / 23 / 54 under the wallet rule (the strategy's constant predicts 11 / 23 / 47) — and real wealth by disposition is not invariant: the honest earn 790 real at issuance 5, 1,410 at 10, 1,310 at 20, because at low issuance the ratchet never starts and the members work at cost.

Source: agentic_sim/results/studies/2.1.0-bank-notes-for-reviewer.md l.7–7

The pump is off: M is issuance alone (the flood 74,000 against 99,400; hoards 1,700 against 3,000), the field's capital on the loans line at 1,700–1,800 and repaid, the debt peaks at 400–500 and is repaid — and the lateness liability is 27,000 in the flood and 10,000 under the crowd, ageing 400–500 ticks, a quarter of it the state's whole fee income: nothing is printed for it, and the customers wait for money the state does not have.

Source: HANDOFF.md l.1816–1820

Conclusion

The books read and the orchestrator is neutral; its lateness bill is a number nobody pays.

The orchestrator is neutral and the money pump is off

Nobody pays the bill for the orchestrator's lateness

The orchestrator's books could now be read, it broke even, and money was no longer printed to fill gaps. But the bill for lateness was now a number nobody paid. What to do about that bill was the first question left for decision.

In the record's words, the books read, the orchestrator is neutral and the pump is off, and the bill for its lateness is now a number nobody pays. The decisions put forward were the lateness liability's scale under a riot (cap it, price it from income, or keep it as the honest bill); whether to strike the red team of inspectors who reject the orchestrator's work, impossible under the blind view, or let an inspector know it inspects the orchestrator; that real wealth was not invariant to issuance; and that the at-cost control's retraction is the design.

The books read, the state is neutral, the pump is off — and the bill for the state's lateness is now a number nobody pays.

Source: agentic_sim/results/studies/2.1.0-bank-notes-for-reviewer.md l.7–7

For the decision (notes §8): (1) the lateness liability's scale under a riot — cap it, price it from income, or keep it as the honest bill; (2) the state-rejecter red team is impossible under §14's blind view — strike it, or decide whether a reviewer may know it inspects the state; (3) Hypothesis 5's invariance of real wealth failed through the strategy's quantity constant; (4) the at-cost control's retraction is the design, recorded; (5) 2.2.0 — entry and delegation — against books that now read.

Source: HANDOFF.md l.1823–1826