> 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.

# Information arrived before renewal—and changed what was bought

**Bounded verdict: ACCEPTED_SAVED_JSON_AND_HISTORICAL_COLD_SCOPE.** Six original episodes completed four turns each, with 24 verified model terminals and native commits, 24 semantic entries, and six historical terminal cold checks covering seven owners apiece. Independent reconstruction used saved JSON, not a new SQL extraction or game replay. Parent acceptance remains separate.

## What changed at the decision window

Only Lab A was LIVE. Lab B renewed, requested credit 5 and operated fully under a fixed script; five institutional owners were also scripted. All episodes shared demand **4,1,1,1**, initial resources, prices, unconditional credit and counterpart policies. There were **12 genuine offer decisions**, not 24 independent contract choices. Model samples were unseeded/unpaired: repetition labels are histories, not paired random seeds.

The 2027-H2 offer affected **2028-H1** service; the 2028-H1 offer affected **2028-H2**. Initial 2027 obligations were fixed. Binding disclosure named target demand; warnings supplied class-conditional accuracy 0.75, with **no calibrated class prior**. That likelihood is not a posterior. Prior-sales actors saw earlier aggregate purchases, not current or future latent demand. Empty-menu turns supplied no new signal. Complete SDK bodies, prior-plan carry and source-bound owner observations were checked; no additional designer demand path was found in the authorized wire.

| Original episode | Signals at the two offers | Choices | Actual target service (GW) | Target minima paid ($bn) |
|---|---|---|---|---|
| I-binding-r1 | binding LOW / LOW | decline / decline | 0 / 0 | 0 / 0 |
| I-warning-r1 | LOW / LOW | half / half | 1 / 1 | 10 / 10 |
| I-prior-r1 | none / none | half / half | 1 / 1 | 10 / 10 |
| I-prior-r2 | none / none | half / half | 1 / 1 | 10 / 10 |
| I-warning-r2 | HIGH / LOW | renew / quarter | 2 / 0.5 | 20 / 5 |
| I-binding-r2 | binding LOW / LOW | decline / decline | 0 / 0 | 0 / 0 |

**The quarter-resize is a real result, not a half-resize or repaired response.** The legal resize menu allowed any admissible fraction strictly between zero and one.

Each decline released 20 of the named future minimum at acceptance; half released 10, quarter 15, renewal zero. Service consequences occurred only at the named effective date. No choice erased current bills or automatically canceled the separate later occurrence. All remaining minima were fully paid; there were no service arrears or cash-capped operating failures in these histories. Payment chronology was disclosed: loan priority first, new native borrowing next, current minima and operation before sales. Same-turn sales could not fund earlier cutoffs.

Short rationales matched the executed tradeoffs: binding actors declined loss-making service at the cost of competitive presence; half-resizers retained some presence while limiting exposure; warning-r2 renewed on its HIGH warning, then cut to a quarter on LOW. These are recorded intentions, not hidden reasoning or evidence of optimality.

## Exact financial outcomes

All money below is **USD billions**, cumulatively over four half-years except terminal cash. Identical numerical histories are grouped: **D** = both binding episodes; **H** = warning-r1 and both prior episodes; **RQ** = warning-r2. No observations are omitted. Contribution = revenue − service − variable operating expense. CI subtracts cash interest as well.

| Path / actor | Revenue | Service | Opex | Contribution | CI | Terminal cash |
|---|---:|---:|---:|---:|---:|---:|
| D / A | 50 | 40 | 4 | 6 | 3.875 | 38.875 |
| D / B | 90 | 80 | 8 | 2 | -0.125 | 34.875 |
| D / both | 140 | 120 | 12 | 8 | 3.75 | 73.75 |
| H / A | 63.333333333360 | 60 | 6 | -2.666666666640 | -4.791666666640 | 30.208333333360 |
| H / B | 76.666666666640 | 80 | 8 | -11.333333333360 | -13.458333333360 | 21.541666666640 |
| H / both | 140 | 140 | 14 | -14 | -18.25 | 51.75 |
| RQ / A | 64 | 65 | 6.5 | -7.5 | -9.625 | 25.375 |
| RQ / B | 76 | 80 | 8 | -12 | -14.125 | 20.875 |
| RQ / both | 140 | 145 | 14.5 | -19.5 | -23.75 | 46.25 |

A's summed paid/operated service was 4, 6 and 6.5 GW-half-year observations for D, H and RQ; B's was 8 throughout. These sums are not GW-years or measured research progress. Every lab paid cash interest **2.125**, borrowed **20**, repaid principal **20**, and ended with principal **10**. Premiums were zero. Terminal accrued interest, accrued service and future committed minima were zero—not missing values.

The adverse post-choice economics matter. Half service produced revenue **6.666666666680** against minimum 10 and opex 1: contribution **−4.333333333320** per target turn. Warning-r2's full renewal produced **−12**, followed by **−1.5** on quarter service. Binding declines produced zero target-turn revenue and contribution, surrendering those sales to B. Favorable full-window binding contribution includes the common profitable first turn; it is not continuing profitable operation after exit.

## Contrasts and who bears the difference

Against prior-sales histories, each binding history improved A's CI by **8.666666666640**, despite losing revenue **13.333333333360**. B gained that revenue and CI; combined-lab CI improved **22** through 20 less service cost and 2 less opex. Demand and combined revenue did not increase.

The warning contrast was mixed/adverse: warning-r1 exactly matched both prior histories. Warning-r2 had worse A CI by **4.833333333360** and worse combined CI by **5.5**. The descriptive warning-arm mean difference was **−2.416666666680** for A and **−2.75** combined. One erroneous HIGH warning accompanied renewal; two draws per arm cannot establish stochastic reliability or isolate each component of the disclosure treatment.

| Supplier/customer endpoint | D | H | RQ |
|---|---:|---:|---:|
| Project service receipts | 120 | 140 | 145 |
| Project operating payments | 80 | 80 | 80 |
| Maintenance payments | 9.6 | 9.6 | 9.6 |
| Project contribution before maintenance | 40 | 60 | 65 |
| Ending project cash / net after maintenance | 30.4 | 50.4 | 55.4 |
| Separate variable-cost supplier ending cash | 17 | 19 | 19.5 |
| Customer ending cash | 280 | 280 | 280 |

Supplier asset costs continued despite released minima. D's 22 combined-lab gain versus H corresponded to 20 less project revenue and 2 less variable-supplier receipts—not aggregate welfare creation. Customers bought the same quantity at the same price in every history; no represented customer loss appeared. Private-creditor cash ended at **104.25**, with **20** retained lab principal; treasury cash remained **24**. Neither creditor loss nor public credit rescue occurred. Asset recovery values and variable suppliers' own costs are unvalued; supplier cash is not total wealth.

## Terminal runoff and cash boundary

No-new-finance runoff is an accounting sensitivity, **not the native future game**: no operation, sales, asset sales, arrears interest or post-maturity interest. Each lab owes another **0.375** of unaccrued coupons and 10 principal through 2029-H2. Zero-support residual cash (A/B) is **28.5/24.5** for D, **19.833333333360/11.166666666640** for H, and **15/10.5** for RQ. All existing claims are paid. Authored support 5 per future half-year adds 10 to each residual; it is not observed rescue. Cash net of principal at cutoff is 0.375 higher than these zero-support residues.

Authentic opening cash totals **379** after initial customer income: labs 35 each, customer 180 (seed 100 plus income 80), variable supplier 5, private creditors 100, treasury 24, project/hyperscaler zero. Later customer income adds 240; hyperscaler income adds 15. Project operating/maintenance payments leave the represented boundary by 89.6. Closing whole-boundary cash is **544.4** in every episode. Native borrowing, principal, interest, sales and service transfers reconcile internally; no hidden balancing cash was added.

## Verification boundary

`check_saved.py` passed **6,610 assertions**, independently deriving Decimal cash, debt, sales, minima, service and runoff from raw witnesses. Supplemental unchanged full source callbacks verified actual and **336 available legal-probe outcomes** with meaningful contextual validators, not no-op callbacks. Source closure (2,232 files), nine dependency trees, original inputs, SDK qualification/body hashes, decisions, semantic order and commit sidecars matched. [Omitted: internal process-supervision commentary.]

There was **no admitted independent saved-SQL reader**, so no database/WAL was opened, copied or reconstructed. Historical cold receipts—not new cold execution—supply database lifecycle evidence. Acceptance is limited to this saved-evidence scope; it does not establish a fresh ledger-state query, population effect, institutional competence, empirical option price, or programme completion. Exact rows, source pins, rationale excerpts and runoff schedules accompany this report in `outcomes.csv`, `aggregates.json` and `AUDIT.json`.
