Version 2.2.02026-10-19Design revision 10

The rebate from income

Runs: Run 15

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.

Lateness claims are paid from each period's fee income; what it cannot cover is forgone, not owed.

What went wrong

Rebates owed for lateness piled up far beyond the orchestrator's whole fee income.

26,839owed after the flood of twelve customers, against fee income of 7,852

Under the last rules, a customer whose job was late was owed a small payment for every tick of waiting. In the worst rush of customers, these debts grew to almost 27,000. That was far more than all the fees the orchestrator took in during the whole run. The debts kept growing and kept getting older. The record put the cause in how the payment was sized, not in the bank: it grew with the number of customers waiting times how long they waited. The customers' real protection was the notice that told them their request was refused. The payment's job was to punish the orchestrator in money.

Version 2.1.0 had expected a stock of rebates owed that would peak at a few thousand and age a period or two. Run 14's flood of twelve customers left 26,839 owed, and the crowd of thirty 9,968, ageing 420–520 ticks and still growing, against a whole run's fee income of 7,852 and 5,832. The record names the cause as the rebate's sizing, not the bank: a claim of so much a tick on every late request scales with the crowd's size times its wait. The customers' real protection is the mark: every refusal is told, and a customer who withdraws gains nothing by it. The rebate's job is the orchestrator's own punishment in money.

2.1.0 §6 expected a liability stock that peaks at a few thousand and ages a period or two; run 14's flood of twelve left 26,839 (accrued 32,503), the crowd of thirty 9,968, ageing 420–520 ticks and growing, against a whole run's fee income of 7,852 and 5,832. The cause is the rebate's sizing, not the bank: a claim of $r_s$ a tick on every late request scales with the crowd's size times its wait; the state's means scale with $N_T$ . The customers' real protection is the mark — every refusal told, the withdrawing requester gaining nothing by withdrawing; the rebate's job is the state's own punishment in money, as 1.6.0 said ("punish the network somehow — earn less or something").

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1344–1344

What changed

Each period's fee income pays that period's claims pro rata; the rest is forgone.

orchestrator fee income, after the debt the period's pool pro rata, at the close requester

What the pool cannot pay is not owed, not carried and not created

Lateness is measured the same way as before. But now the orchestrator's fee income, after it pays its debt, goes into a pot for each period. At the end of the period, the pot pays the late customers' claims, shared out in proportion. Whatever the pot cannot cover is dropped. It is not owed, not carried over, and no money is created for it. Each request shows what it was paid and what was dropped. So there is no pile of unpaid debt any more.

The claim stays as the measure of lateness. The orchestrator's fee income, after the debt, accrues to the period's pool, a lock on the orchestrator's account that holds fee income only. At the period's close the claims stopped in that period are paid pro rata, and the rest is forgone: not owed, not carried, not created. Each request and each period shows the claims, what was paid, what was forgone, the forgone share and what the orchestrator retained. There is no liability stock. Two rulings were carried: the inspectors' view stays blind to who did the work, and 2.1.0's fifth hypothesis was restated to say that printing money is neutral in work and not in distribution.

What changed. The claim is unchanged as the measure; $\Omega$ 's fee income after the debt accrues to the period's pool, a lock on $\Omega$ , fee income only; at the period's close the claims stopped in the period are paid pro rata and the rest is forgone — not owed, not carried, not created — and read on each request and in the period (claims, paid, forgone, the forgone share, $\Omega$ retained); no liability stock (§2.4, §3.1, §10.4, §12.3, §18, §20). Two rulings carried: the reviewer's view stays blind to the actor (§14); Hypothesis 5 of 2.1.0 restated — printing is neutral in work and not in distribution, issuance still not a dial.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1346–1346

What the runs found

No liability existed at any close; the flood forwent 28,000 of a 34,200 bill.

0.87of every flooded period's claims, forgone

The floodthe orchestrator ended 1,500 up instead of 27,000 down

The riots: the clock's bill, paid and forgone

credits

claims (the clock's bill)paidforgone

60/40, N_T 40; means over seeds. Source: agentic_sim/results/studies/2.2.0-rebate-notes-for-reviewer.md l.15–18

Forgone share by period in the riots

share of claims

the crowd of thirtythe flood of twelve

Periods in the order the table lists them. Source: agentic_sim/results/studies/2.2.0-rebate-notes-for-reviewer.md l.15–19

The order: forgone share by period with the state in debt

share of claims

at costfailing at ¾

60/40, the wall at one period. Periods in the order the table lists them. Source: agentic_sim/results/studies/2.2.0-rebate-notes-for-reviewer.md l.44–47

In 260 runs, no debt to late customers was left at the end of any period. In the worst rush, the customers' claims came to 34,200, and 28,000 of that was dropped. The orchestrator ended that rush 1,500 up instead of 27,000 down. When a crowd of customers left, the next period paid in full. In a rush, the orchestrator hardly needed the bank's loans any more. In a quiet market every claim was paid in full, and nothing else changed beyond one period's fees. When the old cost rule was run as a test, its debt took every fee first, and by the eighth period late customers got nothing. Customers who tried to game the pot, by posting many small requests or by withdrawing and posting again, got the same as everyone else.

Run 15 had 260 runs on run 14's seeds. No liability existed at any close. The flood forwent 28,000 of a 34,200 bill, 0.87 of every flooded period's claims; the orchestrator's account retained nothing in those periods and ended 1,500 up instead of 27,000 down. The crowd's three periods forwent 0.84, 0.94 and 0.93, and the period after paid in full. The bank's loans in the riots fell to a peak of 62 against 1,700–1,800, because the returns of capital stay unlocked and fund the next deliveries. In the quiet market every claim was paid in full; the orchestrator retained 186–200 less, with loans of 78–103 a run in the first period only, and price, share, honesty and wealth were the same within noise. Under the at-cost control the forgone share climbed from zero to 1.0 by the eighth period as the debt took every fee first. Nobody gamed the pool: two customers posting many small requests collected 9 each, like everyone, and the withdrawing customer's rebates were the same as the holder's.

What ran (results/studies/2.2.0-rebate/, 260 runs on run 14's seeds, all audits passed at 148069fe; notes results/studies/2.2.0-rebate-notes-for-reviewer.md)

Source: HANDOFF.md l.1866–1867

Findings. The stock is gone and the bill is read: no liability at any close in 260 runs; the flood forgoes 28,000 of a 34,200 bill — 0.87 of every flooded period's claims (§6 said 0.84) — Ω retaining nothing in those periods and ending 1,500 up instead of 27,000 down; the crowd's three periods forgo 0.84 / 0.94 / 0.93 and the period after pays in full, Ω retaining again at once. The loans line empties in the riots (peak 62 against 1,700–1,800): the returns of capital stay unlocked and fund the summonses. M, hoards, on time and lateness within variance of run 14 (Hypothesis 1).

Source: HANDOFF.md l.1871–1875

The quiet market does not notice beyond one period's fees: every claim paid in full; Ω retained lower by 186–200, loans of 78–103 a run in the first period only; price, share, honesty and wealth identical within noise (Hypothesis 2 as restated). The order shows: the at-cost control's forgone share climbs from zero to 1.0 by the eighth period as the debt takes every fee first; the three-quarters failure's to 0.80; the wall's retraction unchanged (Hypothesis 3). Nobody games the pool: two small-request requesters collect 9 each like everyone (a side effect: posting at exactly B̄ hands the request to the state, share 52%, P_t 28.9); the withdrawer's rebates are the holder's (Hypothesis 4).

Source: HANDOFF.md l.1875–1880

Conclusion

Nothing in the rules changes; accepted with one ruling on advancing inspectors' reimbursement.

An inspector's advanced reimbursement may draw on the pool during the period

Nothing in the rules had to change after these runs. The share of the bill that was dropped turned out to be a reading that works. One ruling was added: when an inspector has to be paid back in advance, that money may come out of the pot during the period. Next came the supply side: letting new workers join, and letting workers hand work on.

The notes say there is nothing in the rules to change: the forgone share is a reading that works, the quiet market does not notice, and the order does what it says. The run was accepted with one ruling: when an inspector's reimbursement has to be advanced, the advance may draw on the pool during the period, and the pool at the close is net of it, so that verification is paid before the orchestrator's blame. It was to be implemented with 2.3.0, the supply side: entry and delegation.

Nothing in the rules. The forgone share is a reading that works; the quiet market does not notice; the order does what it says.

Source: agentic_sim/results/studies/2.2.0-rebate-notes-for-reviewer.md l.69–69

Accepted as recorded (October 19). One ruling for Appendix O: a §4.3 reviewer advance may draw on the pool during the period, the pool at the close net of it — verification before the state's blame; to be implemented with 2.3.0, when test_review_06 and test_r4_08 return from the expected failures. Until then the engine runs 2.2.0 as accepted (the advances wait for the unlocked balance). Next: 2.3.0 — the supply side, entry and delegation; the draft follows.

Source: HANDOFF.md l.1883–1886

A §4.3 reviewer advance — Ω advancing a reviewer's reimbursement when the allowance is exhausted — may draw on the period's pool during the period, the pool at the close being net of it: verification before the state's blame.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1352–1352