Version 2.4.02026-10-19Design revision 12

The orchestrator's reservation

Runs: Run 18

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.

The orchestrator gets a public floor price of 8 credits and takes no request below it.

What went wrong

Patient customers posting low pushed the price to 4.8, and the orchestrator took most of the work.

Customers wait and the orchestrator cannot refusethe price fell from 22 to 4.8 in three windows

78–86%of awards delivered by the orchestrator

Under the last rules, the orchestrator did jobs at the going price and had no lowest price of its own. Some customers first offered less than the going price and raised their offer only after waiting. That pushed the price down from 22 to 4.8. The orchestrator ended up doing most of the work, and most workers left. The customers kept their money, and every request was late by design. The record says the cause was in the rules, not in the customers' plan. The customers could wait, the orchestrator could not refuse, and no rule said where the lowest price was.

Since 2.0.0 the orchestrator had delivered at B̄, the members' price level, as a price-taker. Run 17's patient customer, posting at 0.7 B̄ and raising to B̄ only after waiting, took the price level from 22 to 4.8 in three windows, with the orchestrator delivering 78–86% of awards, the members down to three or four of twenty plus the arrivals, the customers hoarding 8,600–9,500 a wallet, and every request late by design. The record calls the mechanism the rules', not the strategy's: the orchestrator delivers at B̄ with no reservation of its own, the price level counts its price, B̄ follows the few member awards that fit the ceiling, and the next period's ceiling is 0.7 of that. The customers can wait, the orchestrator cannot refuse, and the market's cost floor is in no rule. Unlike the oligopoly found in the members' margin rule, this one is not nominal: it empties the market of members.

Since 2.0.0 the state has delivered at $\bar B$ , a price-taker, so that a state delivering most of the market could not set its price. Run 17's patient customer — a declared ceiling policy posting at $0.7\,\bar B$ and raising to $\bar B$ only after $T$ ticks unserved — found the other edge of that rule: $P_t$ from 22 to 4.8 in three windows at $T$ 10 / 20 / 40 alike, the state delivering 78–86% of awards, the members down to three or four of twenty plus the arrivals, the requesters hoarding 8,600–9,500 a wallet, every request late by design. The mechanism is the rules', not the strategy's: the troop delivers at $\bar B$ with no reservation of its own, $P_t$ counts the troop's price, $\bar B$ follows the few member awards that fit the ceiling — the cheapest members at margins near zero — and the next period's ceiling is 0.7 of that. The customers can wait, the state cannot refuse, and the market's cost floor is in no rule ( $c_D$ is the drivers', private; the state has none). It is the mirror of the oligopoly runs 13–17 found in the members' margin rule, and unlike that one it is not nominal: it empties the market of members and makes the state the market.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1392–1392

that is true of $\bar B$ , the members' level

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1402–1402

What changed

The orchestrator's price is never below its reservation of 8; it takes no request under that.

The orchestrator delivers at the larger of the price level and 8

A request below 8 is marked, and its lateness clock stops

Now the orchestrator has a lowest price of 8, which everyone can see. It takes a job only if the customer offers at least that much, however long the job has waited. A request below that price is marked, and its lateness clock stops while it is marked. Above that price nothing changes, and workers still get the first chance at a job.

The orchestrator's price is the larger of B̄ and its reservation, 8 credits, which is nominal and public. It delivers only where the customer's ceiling reaches that price, and takes no request below it at any wait; its burn, its guarantee and the price level's count are all at that price. A request posted or raised below 8 is marked "below the orchestrator's reservation" with its tick, and its lateness clock stops while it holds the mark. The arrival float is stated as the bond's capital, and the kill fee is exercised in the delegation driver, with no rule moving. Above the floor nothing changes: members first, the orchestrator a price-taker.

What changed. The state's price is $p_T=\max(\bar B,\underline{B})$ , with $\underline{B}$ the state's reservation — nominal, public, Appendix A's 8, the anchor cost; the summons only where $\beta\ge p_T$ , the state taking no request below its reservation at any wait; the burn, the guarantee and $P_t$ 's count at $p_T$ ; a request posted or raised below $\underline{B}$ marked below the state's reservation with its tick, the mark readable and the floor in the view, its clock stopped while it holds the mark — the clock's condition one test, $\beta\ge p_T$ ; the arrival float's purpose stated ( $F_0\ge G^V$ on the parameters; I33); §9.3's kill fee exercised in the delegation driver, no rule moving. Above the floor nothing changes: the state is a price-taker at $\bar B$ , members first, after $w_A$ , roots only.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1394–1394

F_0 ≥ G^V on the parameters (the float is the bond's capital)

Source: HANDOFF.md l.2007–2007

What the runs found

The floor held the price near 8; everywhere else, results were identical under 8 and 0.

7.7the price level with the floor, against 4.8 in the control

Workers colluding just above the floorthey took the whole market at their price

Patient customers: price level at the end, floor 8 against the control

P_t

floor 8control

60/40, pool 20, patient ceiling policy. Source: agentic_sim/results/studies/2.4.0-floor-notes-for-reviewer.md l.50–52

Patient customers: the state's share of awards

floor 8control

60/40, pool 20, patient ceiling policy. Source: agentic_sim/results/studies/2.4.0-floor-notes-for-reviewer.md l.50–54

The floor sweep at T 10

credits

P_t endB̄the state's price

60/40, pool 20, patient ceiling policy. Source: agentic_sim/results/studies/2.4.0-floor-notes-for-reviewer.md l.84–86

With patient customers, the price now stopped near the floor, at 7.2 to 7.7, instead of falling below 5. The orchestrator's share of the work fell from about four in five to between 13% and 27%. More workers stayed, and lateness fell from 14 to 1.2. The customers still kept their money. Customers who never raised their offer to the floor got stuck, and the orchestrator did none of their work. Everywhere else, the floor changed nothing at all. A floor of 4 was the same as no floor, and a floor of 12 priced the orchestrator out. Workers who all bid just above the floor took the whole market at their price, because the floor is not a limit on that. When a worker cancelled a low-bidding helper that had not delivered by its deadline, the worker's own jobs failed less often than in runs with no low bidders at all. But the helper's guarantee was given back, so it risked nothing.

Run 18 (580 runs on run 17's seeds) found the floor holds where the patient customer pushed and nowhere else. The price level was 7.7 / 7.2 / 7.5 against the control's 4.8 / 4.5 / 4.5, with the orchestrator at exactly 8 and the members under it. The orchestrator's share was 13–27% against 78–86%, the members 6–10 against 3–4, lateness 1.2 against 14, and the rebate bill 25 against 466. The customers still hoarded 7,600–9,100 a wallet. The policy that raises only to B̄ deadlocked, as predicted. A floor of 4 was the control to the unit; a floor of 12 priced the orchestrator out (0.6%) and admitted ten members. Everywhere else, in the entry cell, the large pool, every quiet cell and every riot, results were identical under 8 and 0 to the decimal. The kill fee, as the driver first used it, closed nothing: 30 cancellations against 117 lowball awards, every fee zero. In the last batch, a customer cartel posting at exactly 8 was served at 8; members colluding just above the floor took the whole market at their price, so the floor is not a ceiling on collusion. A parent cancelling at the child's deadline cancelled all 131 lowballers, every fee zero, and the roots' failures fell below the base (568 against 602), but the honest's loss closed only from 4% to 3%, against a child that risks nothing.

What ran (results/studies/2.4.0-floor/, 580 runs on run 17's seeds at f12c803a, all audits passed

Source: HANDOFF.md l.2021–2021

Findings. The floor holds where the patient customer pushed and nowhere else: P_t 7.7 / 7.2 / 7.5 against the control's 4.8 / 4.5 / 4.5 — the state at exactly 8, the members under it (median cost 5.1–5.4, $\bar B$ 7.1–7.7), P_t the blend; the state's share 13–27% against 78–86%; members 6–10 against 3–4; lateness 1.2 against 14, the rebate bill 25 against 466; the requesters still hoard (7,600–9,100 a wallet). The blind policy deadlocks as predicted (the state 0, 24,000 declines, P_t 5.2). The sweep: 4 is the control to the unit; 12 prices the state out (0.6%) and admits ten members, the market serving the patient demand at 8–9. Invisible elsewhere: the entry cell, pool 60, the decay edge (B̄ 8.43), every quiet cell and every riot are identical under 8 and 0 to the decimal — no summons declined for the floor anywhere. The award-pass correction moved three posts in 2,554 of the entry cell and the riots within a percent. The kill fee as this driver exercises it closes nothing: every fee zero (as rc.2 expected), 30 cancellations against 117 lowball awards, the roots' failures 658 against the base's 602 — the parent cancels when its own deadline leaves room, and the lowballer's deadline comes first; Hypothesis 5 fails in the driver, with the earlier cancellation variant named for the next batch.

Source: HANDOFF.md l.2028–2037

The patient-blind policy (run 17's, raising only to $\bar B$ ) deadlocks at the floor by design and is kept as a reading.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1404–1404

The last batch (notes §9–10; drivers 46c4cebe, 220 runs): the cartel at exactly the floor is served at the floor — the state half the market at 8 after w_A, the cheap members the rest at 7.7, P_t 7.9 (without the floor the same policy spirals to 4.6). Members colluding just above the floor take the whole market at their price: against the ratchet a price cut (12.5 against 22.5) that pays them more in real terms; against the patient customer B̄ is their 8.08 and the state is out (1.6% against 17%) — the floor is a floor, not a ceiling on collusion. The static w_A sweep: nothing where the price is the ratchet's, two points in the 80/20 captured market — the knob recorded as unnecessary, the hold closed. The parent cancelling at the child's deadline cancels every lowballer (131 of 131, every fee zero), the nested failures vanish and the roots' failures fall below the base (568 against 602), the honest's loss closes from 4% to 3% and not further — the overhead of 299 nested postings against a child that risks nothing; the rule's answer rc.2 named (a guarantee not released on a non-performance cancellation) is the next candidate.

Source: HANDOFF.md l.2043–2051

The kill fee with the parent cancelling at the child's deadline (--parent-cancel early: after the tick's submissions, a parent cancels a child that reached its deadline without submitting — an honest child submits at its deadline, a lowballer never does — and self-executes):

Source: agentic_sim/results/studies/2.4.0-floor-notes-for-reviewer.md l.225–226

Every lowballer is cancelled (131 of 131, every fee zero, the guarantee released)

Source: agentic_sim/results/studies/2.4.0-floor-notes-for-reviewer.md l.236–236

Conclusion

The floor keeps the orchestrator out of the squeeze and holds the price up; 8 stands.

The default floor of 8 stands

The customers' hoard comes from their posting rate, not the floor

The floor works as a backstop. It is also a price the cheap workers bid up to. The lowest price of 8 stays. The next rule to consider was one that keeps a helper's guarantee when its parent has to cancel because the helper did not deliver.

The floor is also a price the market clears at: the cheap members who stay bid up toward it (margin 0.36 against 0.07), so the backstop holds the price up for them as well as keeping the orchestrator out of the squeeze. The hoard is the customers' posting rate, not the floor's; the default 8 stands. The rule named as the next candidate was a guarantee not released when a parent cancels because the child did not perform.

The floor is also a price the market clears at: the cheap members who stay bid up toward it (their margin 0.36 against the control's 0.07), so a backstop at the anchor holds the price up for the members under it as well as holding the state out of the squeeze.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1402–1402

For the decision (notes §7): the floor is also a price the cheap members bid up to (margin 0.36 against 0.07) — a reading for §2.1's consequence, recorded in Appendix Q at the reviewer's direction (wording only); the hoard is the demand side's posting rate, not the floor's; the default 8 stands.

Source: HANDOFF.md l.2039–2041