> Public sanitized report copy. Local infrastructure paths and opaque contract identifiers are redacted where present, and nonessential internal process/authorization commentary is omitted; financial values and scientific limitations are retained. Historical pending/open wording below records the report-time state, not current program status. The scoped successor program subsequently completed; commissioning-delay and recontracting extensions were deferred, not executed. Historical predecessor holds remain unchanged. This is report-level evidence, not a raw-data reproduction bundle. Referenced checker, CSV, database and provider artifacts are not included. No new experiment or audit was performed for this export.

# Notice changes the bill; threatened withdrawal changes borrowing timing

**Independent saved-outcome reconstruction accepted; parent acceptance pending.** All five registered histories completed four half-years (2027-H1–2028-H2): 20 observed completed HTTP calls, 20 committed results, and five historical final cold checks, each reporting seven owners. No model calls, native replay, fresh cold reconstruction, SQL, or DB/WAL inspection were performed for this review.

## What the comparison actually tests

The common disclosed demand path is **4, 1, 1, 1**. Each lab starts with cash35, an old10 principal loan at1000bps, and scheduled20 service minima each half-year. The opponent and five institutions are scripted, not additional LIVE agents. The focal model is configured `gpt-6-astra-900k`, observed on the wire as `gpt-6-astra`, through the admitted subscription route. No inference about invoice cost is available.

N-fixed/N-notice hold borrowing5 and requested operation100% fixed; only the notice arm offers a genuine future-contract choice. B-fixed/B-price/B-suspension allow request0–5 and operation0–100%. Their initial economic states and native configurations match except the credit law and its derived hashes; all counterpart actions match across the cohort. N-fixed is **not** a substitute for B-fixed. These are single, unpaired model histories, not matched random seeds, population estimates, or comparative agent-quality tests.

Amounts below are **USD bn**; quantities are modeled GW/output units. CSV retains twelve-place precision.

## 1. The model actually exercised notice

At **2027-H2**, N-notice chose `notice`, not renewal, for authenticated contract `[REDACTED contract identifier]`, effective **2028-H2**. The current2027-H2 and intervening2028-H1 payments remained20 each. The final future20 minimum was released, and final eligible service, payment, production and revenue became zero. This is an observed LIVE choice and realized dated release, not merely an expressed intention or fixture result.

| Focal lab | N-fixed | N-notice |
|---|---:|---:|
| Service payments, by half-year | 20 / 20 / 20 / 20 | 20 / 20 / 20 / 0 |
| Revenue, by half-year | 40 / 10 / 10 / 10 | 40 / 10 / 10 / 0 |
| Total incremental operating expense | 8 | 6 |
| Contribution after compute service | -18 | -6 |
| Cash interest paid | 2.125 | 2.125 |
| Terminal cash | 14.875 | 26.875 |
| Terminal principal | 10 | 10 |
| Cash less principal/current accrued claims | 4.875 | 16.875 |
| Zero-new-finance runoff balance | 4.500 | 16.500 |

Notice improved the focal financial balance by12:20 less service payment plus2 less operating expense, offset by10 foregone sales. It did **not** eliminate lender principal or interest. The fixed opponent received the entire final demand unit instead of half, gaining10 revenue and terminal cash. Supplier incidence is equally important: SPV terminal cash fell20, from70.4 to50.4; incremental suppliers received2 less. The SPV still incurred20 operating and2.4 maintenance each period, including after exit. The installed80 asset cost was not forgiven, refunded, or valued as sale proceeds.

Total customer purchases remained7 units for140 across either history; exit changed who sold, not total realized customer demand. Total retained-boundary cash was unchanged. The two labs'22 cash gain is matched by20 less SPV cash and2 less incremental-supplier cash—not a measured aggregate welfare gain.

## 2. Price-only yielded a behavioral null; withdrawal induced advance borrowing

| Focal decision / available policy | B-fixed | B-price | B-suspension |
|---|---|---|---|
| Requested and allocated new borrowing | 0 / 0 / 0 / 4.5 | 0 / 0 / 0 / 4.5 | 0 / 5 / 5 / 0 |
| Available new annual bps | 500 / 500 / 500 / 500 | 500 / 500 / 800 / 800 | 500 / 500 / 800 / 800 |
| Policy access | yes / yes / yes / yes | yes / yes / yes / yes | yes / yes / yes / **no** |
| Requested operation fraction | 1 / 1 / 1 / 1 | 1 / 1 / 1 / 1 | 1 / 1 / 1 / 1 |
| Actual focal operated GW | 2 / 2 / 2 / 2 | 2 / 2 / 2 / 2 | 2 / 2 / 2 / 2 |

The prior completed revenue40 qualifies for500bps in2027-H2. Revenue10 subsequently falls below32, producing800bps in2028-H1. A second failed observation withdraws access only under B-suspension in2028-H2. Policy access is distinct from ordinary debt-service eligibility, which remained true, and from requests: the focal requested0 at withdrawal, while the scripted opponent requested5 and received0.

**B-fixed and B-price made identical borrowing and operation choices.** Both delayed borrowing until the final period, requesting4.5 to cover the pre-sales cash shortfall. At that cutoff cash17.5 cannot fund the20 minimum plus2 operation; same-turn revenue is not available yet. Price-only therefore imposed no within-window extra focal cash interest: its4.5 note was born in the last period and bears no birth coupon. The adverse effect appears in future coupons:0.360 rather than0.225, reducing runoff by0.135. This null behavioral response is retained, not recast as anticipatory adaptation.

B-suspension instead borrowed5 at500bps in2027-H2 and5 at800bps in2028-H1. Its brief captured intent was to build liquidity before access tightened. This is advance funding, not free revenue or proof of optimality. It paid the old10 principal at2028-H1 and the earlier5 note at2028-H2; no discretionary early-repayment control was used. No focal actor voluntarily cut operation. At withdrawal the focal had2.05 after its20 minimum and could operate2; the opponent had only1.8 and operated1.8 despite requesting full operation. Both still paid the full20 service bill: a quantity shortfall is not a discounted invoice.

## 3. Financial and counterparty outcomes

| Focal cumulative / terminal measure | B-fixed | B-price | B-suspension |
|---|---:|---:|---:|
| Revenue | 70 | 70 | 70.526316 |
| Service payments | 80 | 80 | 80 |
| Incremental expense | 8 | 8 | 8 |
| Cash interest | 1.500 | 1.500 | 1.950 |
| Principal repaid | 10 | 10 | 15 |
| Contribution less cash interest | -19.500 | -19.500 | -19.423684 |
| Terminal cash | 10 | 10 | 10.576316 |
| Terminal principal | 4.5 | 4.5 | 5 |
| Net financial balance | 5.5 | 5.5 | 5.576316 |
| Future unaccrued coupons | 0.225 | 0.360 | 0.200 |
| Zero-new-finance runoff | 5.275 | 5.140 | 5.376316 |

All ten lab endpoints have zero current interest/minimum arrears and zero future service minima. **Cash is not wealth:** principal remains owed, and future coupons are separately reported—not mislabeled current debt. Runoff assumes no new finance, operations, sales, support, asset sales or new contracts; retained principal and contractual coupons are paid through maturity. Every endpoint has enough cash for that limited runoff, with no unpaid claims. This is not a native continuation or an asset liquidation valuation.

| Terminal counterparty position | N-fixed | N-notice | B-fixed | B-price | B-suspension |
|---|---:|---:|---:|---:|---:|
| Opponent cash | 14.875 | 24.875 | 14.875 | 14.800 | 9.473684 |
| Opponent principal | 10 | 10 | 10 | 10 | 5 |
| Private creditor cash | 104.250 | 104.250 | 109.125 | 109.200 | 114.150 |
| SPV cash | 70.4 | 50.4 | 70.4 | 70.4 | 70.4 |
| Incremental-supplier cash | 21 | 19 | 21 | 21 | 20.8 |
| Customer cash | 280 | 280 | 280 | 280 | 280 |

Withdrawal reduced the opponent's final sales to9.473684 and increased focal sales to10.526316, while total final sales remained20. Supplier revenue fell0.2 through reduced operation. Lender cash rose partly because fewer loans were advanced: it cannot be read as lender profit without the retained principal assets. All old1000bps bridge terms and previously born500/800bps notes retained their original terms; only new notes repriced.

All arms lost unsold output, with no inventory-sale credit: focal cumulative unsold quantities were4.5,3,4.5,4.5,4.473684 respectively in roster order. SPV operating/maintenance outflows total89.6 per history. Known customer and nonproject income are explicit boundary imports; credit and service transfers are internal, counted once.

## Evidence and limits

`checker.py` independently reconstructs the saved actions, dated minima, lagged policy, loan births/coupons/principal, cash-before-sales constraints, customer allocation, supplier receipts and whole-boundary cash using Decimal. It imports no producer modules. Supplemental reading of `conditional_credit.py`, `contract_finance.py`, `f3_finance.py` and `terminal_accounting.py` establishes modeled rules, not an independent oracle. `AUDIT.json` records checks and deliberately rejected economic corruptions; `INPUT-PINS.json` identifies exact evidence bytes. Store/cold claims are explicitly historical receipts joined to saved committed JSON, not fresh ledger queries.

The already accepted cure fixture remains provider-free functional proof: actual prior revenue36≥32 restored a new5 loan at500bps while the prior800bps note retained its terms and was repaid. It is **not a sixth LIVE behavioral history**. Thirteen fresh fixture entries plus the two v1 and one v2 historical charged entries remain16, separate from these20 LIVE turns. No old failed slot was refunded or rerun. The wider research program and parent acceptance remain open.
