Inspection is membership

Finding [S4]. Run 4 (October 17, 2026; 300 runs under 1.2.0, truthful appeals and no disputes throughout). Honesty held everywhere: false approvals four- to five-fold below 1.1.0 at every split from 50/50 to 80/20, and under a stamper share drifting to 90% they stopped growing once the dial saturated (83 against a control of 650). It was slow: a case waited for a reviewer 47 times longer than in the control, vesting ended 40% lower, and $\Omega$ could not fund a third of the plants the dial called for at $m_{\max}$. And suspension from duty was a reward: at 60/40 the eight careful actors were the only eligible reviewers, the dial sized the reserve to 7–8, they were on duty almost always — 500 to 800 reviews each at a fee of 2 — while the twelve suspended stampers, freed from duty, took the awards at 20 (delivery income 3,311 against 1,063 per actor; stampers richer in 19 of 20 seeds). Selection could not help, because it runs among free actors and the honest were not free. $\Omega$ banked the raised tax (+5,898 at 60/40) levied for the effort the honest were supplying; and a plant tied up 45 credits of which 30 was $\Omega$ paying itself.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1180

The third use of §19's path, written the same day from run 4's finding and Baruchi's answer to it: "if you get thrown out of inspection you are banned from receiving a new task; to get tasks back you have to prove you are doing good inspection for a while."

Source: HANDOFF.md l.759–761

Suspension now suspends market membership: no new award while suspended or on probation; the held commitment is finished by the ordinary rules. After T_susp the actor is a probationer — initial reviews only, audit odds pinned at 1, re-suspended by any plant failure, readmitted when its one trailing record passes θ_duty with n_min new plant outcomes judged (a full stamper: four rejected bad plants, about fifty honest reviews; its panel seats return only once its approvals roll out of the window). At most half the eligible members are on duty; probationers fill the initial seats first, members hold the audit and challenge seats. The dial's raised tax goes to a duty fund paid to the settling reviews by role weight as an uplift, instead of to Ω. Plant capital counts the pots only (15, not 45). Newcomers' odds are pinned at 1 until their bad-plant record is determined (the 1.2.0 taper was passed by a third of stampers at once). A duty levy — keep winning awards against a cash levy — was considered and rejected (Appendix B). The slowness run 4 measured is deliberately not treated here; Baruchi has a direction for it, reserved.

Source: HANDOFF.md l.761–771

Mechanism. The contest was no longer about detection but about who carries the reviewing load and what membership of the market requires. Suspension now suspends membership: no new award until the actor has inspected well, under full audit, for long enough to pass. Probation is the supervised way back; the members' duty cap keeps half the honest free to work while probationers take the initial seats; the dial's raised tax goes to the reviewers through the duty fund; plant capital counts only the pots.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1182

It changes the meaning of an existing state (suspension) and adds a cap and a fund; it alters no settlement, verdict or challenge rule.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1184

Smaller: probation pays a stamper a little (+1.2 a review) and the award is what deters;

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1188

Found at implementation — two readings, flagged for confirmation (§12.1, Appendix H). (1) The members' cap ⌈E_t/2⌉ is 3 at six actors — below the X₀ = 4 of Appendix A's initialization and §20's worked example, so the baseline's own worlds, and every reviewer test built on them, would violate the revision at tick 0 (the first suite run failed on "C is not on verification duty" across the board). The engine lets the cap never fall below the reserve's floor V_max + 1, as the panel floor yields to the cap; the difference from the literal text is confined to memberships of four to seven, where the literal cap would leave two or three members free. One line (World._member_cap). (2) An outgoing probationer whose turn ends with no free probationer stays, rather than hand its seat to a member whom the next tick's "probationers first" would displace back. One branch of the rotation. Both are the implementer's; the alternative to each is one edit. Confirmed as patch 1.3.1 (October 17, 2026; the sentences of design revision 4 rc.1's Appendix A), applied to spec/agentic_economics_loop_spec_v1_4_0.md; 1.3.0 to history.

Source: HANDOFF.md l.783–791

agentic_sim/results/studies/1.3.0-membership/ (and 1.3.0-membership.zip beside it): tipping/ (80 runs), adaptive/ (60), drift/ (80), redteam/ (140: camouflage 80, reformer 60) — 360 runs, 20 seeds per cell, 20 actors, every final audit passed, engine 91a49e5. Appeals truthful and no disputes throughout. The §6 reproduction is test_v13. 1.2.0 figures below are run 4's (same seeds, same driver, --rules 1.2.0).

Source: HANDOFF.md l.814–817

The tipping point (run 2 of the revision's §8) — the sign flips at every split.

stampers	1.3.0: stamper − careful (se), richer	1.2.0	1.3.0 false approvals (of them by probationers)	1.2.0 FA	vested 1.3.0 / 1.2.0 / 1.1.0	awards unfilled for want of members (ticks of 1,000)	fund paid to reviewers
50%	−2,132 (100), 0 of 20	−2 (51)	53 (7)	53	1,924 / 2,399 / —	816	4,099
60%	−2,397 (63), 0 of 20	+215 (34), 19 of 20	69 (15)	67	1,727 / 2,412 / 2,675	869	4,027
70%	−2,486 (37), 0 of 20	−10 (25)	90 (14)	89	1,439 / 2,396 / 2,739	878	3,040
80%	−2,428 (36), 0 of 20	−902 (25)	113 (16)	121	1,176 / 2,024 / 2,813	899	2,003

Source: HANDOFF.md l.819–826

Hypothesis 1 holds, decisively: a stamper earns award income only until its first handful of plants and then nothing, so the honest end richer by 2,100–2,500 per actor at every split, in 20 of 20 seeds. The channel is the award — stampers' delivery income falls from 3,311 (1.2.0, 60/40) to 1,221, their awards from 207 to 74 — and the fund adds 340–380 per careful actor. Honesty is unchanged (false approvals within a few of 1.2.0's); probationers account for 7–16 of them per run (Hypothesis 5: their approvals are all audited, and what reaches a customer is what an audit missed or no member was free to take). Plain stampers were never readmitted: they do not reform. The cost is the throughput Hypothesis 3 named and did not treat: vesting falls 20–42% below 1.2.0 and 36–58% below 1.1.0, because awards go unfilled in 800–900 of 1,000 ticks for want of free members — at 60/40 eight members with half of them on duty cannot deliver three tasks a tick (Q11).

Source: HANDOFF.md l.828–836

Adaptive dispositions (run 3): an honest equilibrium, faster. Actors start 50/50 and every 100 ticks take the richer disposition with probability 0.2; 3,000 ticks.

cell	stamper share at 500 / 1,000 / 2,000 / 3,000 ticks	reached ≤ 10% stampers	mean tick	vested	FA
1.3.0, fund on	0.21 / 0.15 / 0.04 / 0.01	20 of 20	865	7,148	91
1.3.0, fund off	0.20 / 0.12 / 0.01 / 0.00	20 of 20	770	7,216	90
1.2.0	0.36 / 0.29 / 0.15 / 0.07	19 of 20	1,468	6,796	92

Hypothesis 2 holds: an actor whose income is zero switches faster than one whose income is merely lower; the population reaches an honest state in about 800 ticks against about 1,500 under 1.2.0, and vesting recovers with it (7,100–7,200 over 3,000 ticks, above 1.2.0's 6,796). Hypothesis 4 does not: the fund does not speed the convergence (770 ticks without it, 865 with), because the gate decides the switch and the fund's uplift is small beside the award. The fund's effect is distributional — 8,048 paid to reviewers over 3,000 ticks instead of banked by Ω (Ω net 2,889 with the fund, 9,588 without).

Source: HANDOFF.md l.838–852

Drift (run 4): honest at both m_max, slower still. Stampers 50% → 90% over 1,500 ticks.

cell	FA (of them by probationers)	vested (1.2.0: 2,518; 1.1.0: 4,209)	stamper − careful	plants unfunded	fund paid
m_max 4, Ω₀ 300	97 (24)	1,506	−3,038 (120)	49	3,087
m_max 4, Ω₀ 100	91 (23)	1,529	−3,181 (156)	119	3,215
m_max 2, Ω₀ 300	158 (67)	1,820	−2,945 (203)	0	1,627
m_max 2, Ω₀ 100	152 (65)	1,856	−3,151 (213)	34	1,676

Both settings keep the loop honest (1.1.0 control: 650) and both end with two members on duty, two probationers, and member seats sized below one. m_max 2 costs about 60 more false approvals and gains about 300 vested cases; two thirds of its false approvals are probationers' (65 against 23 at m_max 4), because the pinned audits need a panel-eligible member free to take them and at m_max 2 fewer are seated — the probation safety of Hypothesis 5 binds on member availability before it binds on the cap.

Source: HANDOFF.md l.854–867

Red team (run 6). Camouflage (10% random rejections) under the gate: at 50/50 −1,997 (91) against −2,132, false approvals 37 against 53, false rejections 25 against 16 — an edge of about 130 per actor, no safety; at 90/10 −1,096 (173) against −1,321. The reformer — a stamper that reviews carefully on probation and stamps again once readmitted — is readmitted about 20 times a run (mean cycle 150–170 ticks), reaches one or two probationer false approvals per run, and ends 1,700–1,970 per actor poorer than the careful at 50/50, 60/40 and 90/10 (0 of 20 richer): the cycle's yield is the fee and uplift of about 40 probation reviews a run against the awards lost in between.

Source: HANDOFF.md l.871–877

What run 5 leaves. The gate does what it was for: the contest is decided at every split and the population converges to honesty in under a thousand ticks. Three things for the next design decision, in order: (1) the slowness — awards unfilled 80–90% of ticks at 60/40 and worse above; the shrinking of the market to its honest core is now the cost, and the one Baruchi has reserved a direction for; (2) probation safety binds on member availability — at m_max 2 two thirds of false approvals are probationers' whose pinned audits found no free member; member seats are sized on demand but the members are busy delivering; (3) the two implementer's readings above (the cap's floor; the outgoing probationer stays) need confirming or replacing.

Source: HANDOFF.md l.879–886