Version 2.6.12026-10-21Erratum 2.6.1 (Appendix E, E5)

The erratum and the one re-read

Runs: The 2.6.1 re-read

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.

Planted cases now draw their price ceilings from real requests and raise them as real requesters do.

What went wrong

A plant's requester never raised its price ceiling, while 35% of real requests were raised.

35%of real requests raised their ceiling while waiting; planted cases, none

Plants never raised their ceilinghow long they waited still set them apart

Planted cases whose answer the orchestrator already knows are meant to wait in the market as real work waits. A check still found a small give-away.

First, planted cases were posted a tick late, so they waited longer than real work. That was fixed. Then a smaller difference was left. About a third of real buyers raised their price limit while waiting. The planted cases never raised theirs.

At run 24, the probe that tells planted cases (cases whose answer the orchestrator already knows) from real ones found the separation in a timing detail: the orchestrator posted plants after the tick's award pass, so a plant waited a tick longer than every real request. That was corrected in the driver.

On the corrected construction, what a case's own fields added beyond the public context fell to about a third (+0.016 to +0.019, from +0.052), still not within the criterion. The largest remaining piece was the time a request had been seen available: the plant's requester never raised its ceiling, while 35% of real requests were raised.

The window closed nothing: the case's separation sat in the seating gap (rv_gap the one field adding beyond the context) because the orchestrator's instrument posted the plants after the tick's award pass, so a plant waited a tick more than every real root (1.13 against 0.41) and was seated first as the oldest of the tick's submissions (4.4 % seated later against 16.2 %). Corrected in the driver: the plants due are posted in the requesters' step at a drawn position among the tick's postings

Source: HANDOFF.md l.2259–2262

On the corrected construction the criterion is not met at a third of the distance: the case's fields add +0.016 ± 0.004 (appeals off) and +0.019 ± 0.005 (on) at twenty test seeds, from +0.052; no field above +0.007 alone; the largest piece is the time a root was seen available — the plant requester never raises its ceiling while 35 % of real roots are raised.

Source: HANDOFF.md l.2263–2265

The test the reviewer set for the construction is whether the case's own fields add separation beyond the context, which is public by rule (the plant rate's rhythm)

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1448–1448

What changed

Plants draw ceilings from real requests and raise them as real requesters do; the difference is minted.

real requests' ceilings drawn planted case raised as requesters raise waits as a real request waits

bank mints the raise planted case

Now a planted case starts from a price limit copied from what real buyers first posted. Then it raises its limit the way real buyers do, at the same moment in the same step.

The record also wrote down a rule for every step the orchestrator takes for a planted case: whatever a real case does, a planted case does in the same step, at a random place among the real cases doing it, with its numbers copied from recent real cases.

Erratum 2.6.1 changes how a plant's price ceiling is drawn and how it waits. The ceiling is drawn from real requests' ceilings as posted, and the plant's requester raises the ceiling as the population's requesters raise theirs, so the plant waits as a real request waits. The difference is minted.

The posting correction was recorded as the third conformance correction of the orchestrator's instrument, with a principle: at every step of a request's life the instrument takes for a plant (posting, raising, submission), the plant takes the step in the same pass a real request would, at a drawn position among the tick's real requests taking it, with its parameters drawn from the trailing real requests'.

The posting correction recorded as the third conformance correction of the instrument, with the principle: at every step of a root's life the instrument takes for a plant — posting, raising, submission — the plant takes the step in the same pass a real root would, at a drawn position among the tick's real roots taking it, its parameters drawn from the trailing real roots' at that step; the award is the market's already. Erratum 2.6.1 (E5): the plant's ceiling drawn from the real roots' ceilings as posted (the window filled at post_root), and the plant requester raising it as the population's requesters raise theirs (raise_ceiling admits a market plant, the difference minted; the driver's raise step covers the plants; plants_raised), so the plant waits as a real root waits; I39 unchanged in substance.

Source: HANDOFF.md l.2275–2281

The driver's raise_below_market — the population's requesters' step, in the posting pass before the awards — covers the plants in the working set with the same rule and timing (--ceiling-policy, --patient-ticks), so the plant takes the step in the same pass, at the same moment, as the real roots taking it. One asymmetry stays and is declared: a requester's raise is refused when its wallet cannot lock the difference; the plant's is minted and never refused.

Source: agentic_sim/results/studies/2.6.1-reread-notes-for-reviewer.md l.17–20

What the runs found

Plants raised as often as real requests; what still set them apart stayed at or below 0.02.

36%of plants raised, against 34% of real requests

0.011added by a case's own fields with appeals on; 0.020 with appeals off

How much a case's own fields tell plant from real, beyond the public context

AUC gap (raw − context)

appeals offappeals on

Initial reviews; train seeds 1–20, test 21–40. The ruled floor was 0.02. Appeals on is the population from 2.6.1 on. ± is the standard error over the seeds. Source: agentic_sim/results/studies/2.6.1-reread-notes-for-reviewer.md l.51–56

Raised before the award: plants against real requests

% of roots awarded

plantsreal roots

Five careful runs in run 16's world, appeals on. Source: agentic_sim/results/studies/2.6.1-reread-notes-for-reviewer.md l.30–35

One check was run, as the reviewer ruled. Planted cases now raised their price about as often as real ones.

One difference stayed. A real buyer cannot raise its price if its wallet is too short, so some real cases wait a long time. The orchestrator's money for a planted case is created when needed, so a planted case is never refused. Because of this, a real case waited three ticks or more in 12 awards in 100, and a planted case in 1.

A case's own details still gave away a little more than what is public anyway. The amount was not zero, but it was under the limit the reviewer had set in advance.

The ruling allowed one re-read, with the criterion fixed in advance: within two seed-level errors of zero, closed; a residual at or below 0.02, accepted as the construction's floor and closed with its number; no third pass.

Under the erratum the plant's requester raised as often as the population's (36% against 34%). The one asymmetry left is the wallet: a real requester's raise is refused when it cannot lock the difference, while the plant's raise is minted. So a real request waited three ticks or more in 12% of awards, and a plant in 1%.

On forty seeds, the case's own fields added +0.011 ± 0.002 beyond the context in the population going forward (appeals on) and +0.020 ± 0.003 with appeals off. Neither is within two errors of zero; both are at or below 0.02.

(5) Then one re-read on the same recipe (train 1–20, test 21–40, both populations): within two seed-level errors of zero → closed; a residual ≤ 0.02 → accepted as the construction's floor, closed, recorded with its number; no third pass.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1452–1452

The re-read (one, as ruled: train 1–20, test 21–40, both populations). Under the erratum the plant requester raises as often as the population's (36 % against 34 %) and the plant's tail to w_A is gone; the one asymmetry left is the wallet — a requester's raise is refused when it cannot lock the difference, the plant's is minted — so a real root waits three ticks or more in 12 % of awards and a plant in 1 %. The probe on forty seeds: the case's fields add +0.011 ± 0.002 beyond the context in the population from here (appeals on) and +0.020 ± 0.003 with appeals off — not within two errors of zero, at or below 0.02: closed at the construction's floor, recorded (from 0.052 at run 22, 0.016–0.019 at run 24); the largest piece the time a root was seen available, the wallet asymmetry.

Source: HANDOFF.md l.2284–2289

reading (roots awarded, five runs)	plants, run 24's correction	real roots, run 24's correction	plants, 2.6.1	real roots, 2.6.1
raised before the award	0 %	35 %	36 %	34 %
ticks in the market before the award, mean	1.40	0.63	0.33	0.79
awarded the tick posted / after 1 / after 2 / after 3 or more	86 / 7 / 2 / 5 %	77 / 9 / 6 / 9 %	76 / 19 / 5 / 1 %	71 / 10 / 7 / 12 %
seated later than the tick it submitted	18.4 %	14.5 %	16.5 %	13.1 %

Source: agentic_sim/results/studies/2.6.1-reread-notes-for-reviewer.md l.30–35

Conclusion

Closed at the construction's floor, 0.011 and 0.020, with the wallet asymmetry named.

A real requester's raise is refused when it cannot pay; the plant's raise is minted

The question was closed. The small leftover give-away was written down with its size and its name, the wallet difference. The reviewer said a fix for it would come only if a learner ever made money from it. Later, after run 35, the reviewer ruled that a planted case's raise is refused as often as real buyers' raises are, and granted after the same delay.

Accepted on October 21: closed at the construction's floor, 0.011 and 0.020, named the wallet asymmetry. A mimic of a requester's wallet for the plant was noted as a later design item, to be taken up only if a learner monetises the floor. The order stood: 2.7.0 next.

Later, after run 35, the reviewer adopted a further correction: the plant requester's raise is refused at the trailing real requesters' rate and granted at their delay.

Accepted (October 21, 2026): closed at the construction's floor, 0.011 / 0.020, named the wallet asymmetry; a drawn-wallet mimic a later design item only if a learner monetises the floor; the withdrawal covered in advance by the Appendix S principle. Next: 2.7.0 (the arrival loan) on its acceptance; Q1's learned line once under it with appeals and re-arrival on; Q3 calibrated at ×4.

Source: HANDOFF.md l.2291–2293

After run 35 (October 26, 2026) — the raise refused.

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1468–1468

The reviewer's ruling: "The raise conformance item adopted: the plant requester's raise refused at the trailing real requesters' rate and granted at their delay; then one probe read on the learned population's collection for the floor under stress."

Source: spec/agentic_economics_loop_spec_v2_9_0.md l.1468–1468