Version 2.3.02026-10-19Design revision 11

The supply side

Runs: Runs 16 and 17

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.

New workers can now arrive and depart, and a worker can hand part of a job to another.

What went wrong

Prices followed how many free workers there were, and no rule let a new worker join.

Twenty workers and no rule to admit a twenty-firstfree workers stayed scarce and margins ratcheted prices up

22.5the price with 20 honest workers, against 10.9 with 40

Before this version, the list of workers was fixed at twenty. Nothing in the rules let a new worker join. Where free workers were scarce, their prices kept climbing. Adding workers brought the price down to cost, but there was no way to add them. Handing part of a job to another worker had never been used, because every worker had the same cost. Baruchi agreed that workers should join when they can earn, that newcomers should start on trial, and that workers should differ in their cost, which only each worker knows.

The record's finding: the price is set by the quantity, and the quantity by the members' strategy until members can enter. Where free members are scarce the margin rule ratchets prices up until a free member wins one award per 3.6 ticks; adding workers takes the price to cost (22.5 / 11.6 / 10.9 at 20 / 30 / 40 honest), but the twenty actors were given and nothing in the rules admitted a twenty-first. Honest earnings fall with headcount, so entry on earnings limits itself. Delegation had never run: with one cost a child bids no lower than its parent would. Baruchi's positions were entry on earnings, arrivals as probationers, and a wide competence spread, in cost only and private.

The price is the quantity's, and the quantity is the strategy's until members can enter: where free members are scarce the margin rule ratchets until a free member wins one award per 3.6 ticks (run 13); adding workers takes it to cost — 22.5 / 11.6 / 10.9 at 20 / 30 / 40 honest — but the twenty actors were given, and nothing in the rules admitted a twenty-first. Honest earnings fall with headcount (1,446 → 830 → 589 real), so entry on earnings is self-limiting. Delegation had never run: with one cost a child bids no lower than its parent would. A customer posting at the market price got the state (run 13, 15). The state's books were clean enough to read supply against (runs 14–15). Baruchi's positions: entry on earnings — agreed; arrivals as probationers — agreed; the competence spread — wide and swept, cost only, private (the design reviewer's answer).

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1368–1368

What changed

Arrivals get a float from the bank and start on probation; workers with no obligations may leave.

bank float newcomer passing record on planted cases worker

A worker with no live obligation may depart, taking its wallet

parent nested request and its pots child

Now workers can arrive and leave. A newcomer gets a starting sum from the bank and begins on trial. It is admitted once it has been tested on enough planted bad cases and its record passes. A worker may leave only when it owes nothing and nothing is owed to it, and it takes its money with it. Each worker has its own cost of work, which nobody else can see. A worker whose own cost is high can hand a job to another worker, and then it pays for that job's inspection itself. The orchestrator never does a handed-on job itself.

Two public operations, arrive and depart. An arrival gets a float minted by the bank and starts as a probationer; it is admitted on a set number of bad plants with a passing record. A member departs only with no live obligation, taking its wallet out. Competence is a per-actor cost, private, with nothing per actor in the view. A parent delegates when its cost exceeds the expected child's bid plus the overhead; the parent posts the nested request's pots, so a nested request is a full loop in money. The orchestrator never delivers a nested request.

2.3.0-rc.1 opened the door both ways: entry as a probationer with a bank-minted float, exit with the wallet out, competence as a private per-actor cost, delegation with the troops never nested, the outsider's view, run 15's advance ruling.

Source: HANDOFF.md l.1890–1891

What changed. Two public operations, arrive and depart; one new mint, the arrival float; admission on $n_{\min}$ bad plants with a passing record, counted apart from readmission; a departure with no live obligation, off duty as it leaves, the wallet out as departure, the identity archived; competence as the drivers' per-actor cost, private, the view exposing nothing per actor; the parent posts the nested pots (Q3 answered) and a nested request is a full loop in money; the troops never deliver a nested request, seats as on roots; no nested clock; the reviewer advance from the rebate pool (run 15's ruling).

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1370–1370

The drivers' delegation rule is declared: a parent delegates when its cost exceeds the expected child's bid plus the overhead (the nested fee, the pots' expected net cost, its care of 2), at that saving as the ceiling, carrying a longer τ; children bid as on roots; the parent judges by its disposition; a child rejected appeals truthfully; a child unawarded for $T_{\text{wait}}$ is withdrawn and the parent self-executes.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1378–1378

depart now refuses while an obligation to the actor is unpaid — a case of its own not yet final (the price at vesting, a court fee won), a nested case it posted whose pots have not refunded — "commitment" read to the money's end.

Source: HANDOFF.md l.1931–1933

What the runs found

Entry cut the price from 32.9 to 22.9; delegation ran only where costs varied widely.

22.9the price at the end with entry, against 32.9 in run 15

Patient customers posting lowthe price fell to 4.8, the orchestrator took 86% of the work

Price level at the end, with entry against run 15

P_t

2.3.0, the pool of 20run 15

Wallet rule, K_open 5, σ 0.43. Source: agentic_sim/results/studies/2.3.0-supply-notes-for-reviewer.md l.15–20

Reviews unassignable by window, with entry

case-ticks

60/4080/20

Windows in the order the table lists them. Source: agentic_sim/results/studies/2.3.0-supply-notes-for-reviewer.md l.15–26

Patient customers: price level by window, T 10

P_t

Run 17, batch C. Windows in the order the table lists them. Source: agentic_sim/results/studies/2.3.0-run17-notes-for-reviewer.md l.192–194

All twenty newcomers arrived early, and fifteen were admitted. Every stamper left, having earned nothing, and so did the costliest careful workers. The market ended with 19 members instead of 8. The orchestrator's share of the work fell from 39% to under 1%. Honest workers earned less, 1,237 against 1,928. The price fell, but only to about 23, not to cost: it settled where the workers' own way of raising their prices puts it. Newcomers helped with inspection only while they were on trial. After they were admitted, more inspections waited again. Handing work on happened only where costs differed a lot. Even then, three in four handed-on jobs found no taker, because the helpers asked more than the worker would save. Honesty did not change.

A second set of runs asked why inspections waited early on. The waiting seats were mostly audits and appeals. Almost no one present was allowed to sit on those panels yet, so the orchestrator's own inspectors took most of them. With that rule switched off, the workers filled the seats, but wrong approvals rose from 13 to 35. A pool of sixty newcomers brought the price down to 14.5. With no starting sum, nobody could be admitted. One result stood out. Customers who first offered a low price, and raised it only after waiting, pushed the price down to 4.8. The orchestrator then did 86% of the work, at a price below what almost every worker's work cost.

Run 16 (240 runs): at 60/40 all twenty candidates arrived in the first two windows and fifteen were admitted at a mean of 310 ticks. Every stamper departed on zero earnings and the dearest careful followed. The market ended with 19 members where run 15 had 8, the orchestrator's share at 0.7% against 39%, the honest earning 1,237 real against 1,928, and the price at 22.9 against 32.9. It dipped to 18 as the cohort was admitted, then climbed back to 22.5 as margins ratcheted to the constant run 13 found. Entry relieved inspection only until admission. Delegation ran only where the spread was wide: nothing at the narrow spread, 5% of awards at the widest (a fifth was predicted), with three quarters of nested requests dying because the child's bid exceeded the parent's saving; the gain from trade was 3–4 where a child was found. When the root customer paid for the child's inspection instead, parents delegated 2.4 times as much. With entry and a wide spread, the members' median cost fell from 8 to 6.4, yet the price rose with the spread, because it follows the free members' count, not their cost. Honesty was unmoved.

Run 17 (700 runs) looked at the inspection queue: before admission nineteen of twenty waiting seats were audits and appeals, almost no one present could sit on those panels, and the orchestrator's own inspectors took 605 of 710 waiting audits. With the determined-record gate off, their seats fell from 688 to 8 and false approvals rose from 13 to 35. A pool of sixty took the price to 14.5 and the honest to 810; a float of zero admitted nobody. Patient customers, posting at 0.7 B̄ and raising to B̄ only after a wait, collapsed the price to 4.8 and handed 86% of the work to the orchestrator below every member's cost but the cheapest, while the customers hoarded 9,000 a wallet. Faster margin decay took the price to 8.3 with the honest's real wealth unchanged at 1,256.

What ran (results/studies/2.3.0-supply/, 240 runs on run 15's seeds, all audits passed

Source: HANDOFF.md l.1926–1926

Findings. Supply finds its level, and the level is the ratchet's: all twenty candidates arrive in the first two windows, fifteen are admitted at a mean of 310 ticks, every stamper departs on zero earnings and the dearest careful follow (departed median cost 8.4 against the arrivals' 7.9); the market ends with 19 members where run 15 had 8, the state's share at 0.7% against 39%, the honest earning 1,237 real against 1,928 — and the price at 22.9 against 32.9: it dips to 18 as the cohort is admitted, then climbs back to 22.5 as the free members settle at nine and the margins ratchet to run 13's constant, the twenty-honest cell to the unit (Hypothesis 1 in direction, not in level). Entry relieves inspection only until admission: reviews unassignable are flat while the probationers carry the initial seats and rise once they are members who deliver and the stampers who sat on duty have left (Hypothesis 2 fails).

Source: HANDOFF.md l.1936–1942

Delegation runs where the spread is wide and the children refuse the price: nothing at σ 0.2, 5% of awards at 0.8 (predicted a fifth) — three quarters of nested requests die at T_wait because the child's ratchet bid exceeds the parent's saving; the gain from trade is 3–4 where a child is found; brokers are the dearest two or three members with 5% of the market; a careful parent's wrongful rejection is overturned on the child's appeal, a stamping parent approves bad child work (Hypotheses 4 and 5). The root control answers §8.3: the parents delegate 2.4 times as much (294 against 122) when the root pays for the child's inspection. Selection at the wide spread with entry: the members' median cost falls from 8 to 6.4 as the dear depart (9.3) and the cheap arrive (7.2) — and the price rises with the spread (21.8 / 22.5 / 23.5) because it follows the free members' count, not their cost (Hypothesis 3: half). Hoards over M 0.84 with delegation and without (Hypothesis 8: the circuit does not close at this scale). Honesty unmoved (Hypothesis 7).

Source: HANDOFF.md l.1942–1950

700 runs on run 16's seeds, all audits passed

Source: HANDOFF.md l.1960–1960

Findings. A: the queue before admission is panel eligibility, not the cap or the window. Nineteen of twenty waiting seats in the first three hundred ticks are audits and appeals; while they wait there are five members on duty idle but not panel-eligible, eight more off duty, free, funded and duty-eligible but not panel-eligible, ten probationers barred from panels — and 0.1–0.6 members in the decisive state (free, fit, under the cap, unseated), none held out by the cap (slack eight seats). The troops take 605 of 710 waiting audits at exactly w_R; after admission the state's inspection share falls from 46% to 0.1% and the queue is initial reviews filled a step later by the rotation's lag.

Source: HANDOFF.md l.1969–1973

The control (--panel-determined off) drops the troops' seats from 688 to 8 and raises false approvals from 13 to 35: the gate is what the troops bridge. The §12.1 question is answered no. B: pool 60 takes the price to 14.5 and the honest to 810 (the ratchet broken by turnover); F_0 = 0 admits nobody (no bond, no seat, no record); r and the arrival rate move nothing.

Source: HANDOFF.md l.1975–1977

the control with the determined-record gate off.

Source: HANDOFF.md l.1963–1964

Patient customers (--ceiling-policy patient: a request posts at 0.7 B̄ and is raised to B̄ only after T ticks unserved; T 10 / 20 / 40):

Source: agentic_sim/results/studies/2.3.0-run17-notes-for-reviewer.md l.190–190

C: patient customers collapse the price to 4.8 and hand 86% of the work to the state below every member's cost but the cheapest — the state delivers at B̄, the ceiling is 0.7 B̄, P_t counts the troops, the members depart, the requesters hoard 9,000 a wallet (the one result that may warrant a §19 decision: a reservation price for the state's delivery, or the clock reading a raise as a posting); the faster decay takes the price to 8.3 with the honest's real wealth unchanged at 1,256 (the ratchet's constant is nominal)

Source: HANDOFF.md l.1982–1986

Conclusion

Nothing in the rules changes; the patient customers' result was left to the reviewer.

The price supply finds is the workers' strategy's constant

Entry's relief for inspection lasts only until admission

Supply found a level, but it was the level set by how the workers raise their prices, not their cost. Nothing in the rules changed after these runs. The patient customers were the one result that might call for a rule, such as a lowest price for the orchestrator's own work. That decision was left to the reviewer.

The level supply finds is the strategy's constant; entry's inspection relief ends with admission; the pool's size is the lever's length; delegation's scale is the children's bids. Nothing in the rules. Run 17 answered the duty-cap question no. The patient customers were the one result that might warrant a decision, a reservation price for the orchestrator's delivery or the clock reading a raise as a posting, and it was left to the reviewer.

For the decision (notes §5): the level supply finds is the strategy's constant (a reading on the margin rule for 2.4.0's §8); entry's inspection relief ends with admission (a §12.1 question — the duty cap on an all-honest list); the pool's size is the lever's length (§8.5's knobs); delegation's scale is the children's bids. Nothing in the rules.

Source: HANDOFF.md l.1952–1954

For the decision (notes §5): nothing for §12.1; the float as a gate (§2.1); the cycler's and swarm's bill to the loop (§8.7's text); the kill fee unexercised against the below-cost child (a driver gap); patient customers against the state's delivery at B̄ (the reviewer's §19 call); the price level as the members' nominal choice.

Source: HANDOFF.md l.1989–1991