> 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.

# Priced flexibility: six saved LIVE histories

## Verdict and evidence scope

**ACCEPT for the released saved-JSON reconstruction plus historical-cold scope; parent acceptance remains pending.** Six four-turn histories contain 24 completed HTTP-200 Astra responses, 24 distinct semantic reservations, 24 matching saved committed results, and six historical terminal cold receipts reporting seven owners and no replay. Configured `gpt-6-astra-900k` becomes wire/observed `gpt-6-astra`; this is not evidence of a separately named 900k wire model. No additional model calls, SQL/DB/WAL access, native adjudication, replay, credentials, or producer modifications were used.

`checker.py` independently checks Decimal cash, coupons, maturities, births, minima, dated service, commercial allocations and counterpart transfers. It passes 4,679 assertions, including all 1,764 source pins and nine dependency trees; five evidence mutants (fee, birth debit, service, right/minimum conservation, counterparty cash) reject. Saved response text equals validated action/plan and the native request action. Packet, predecessor and wire-body receipt bindings match. This does not independently attest the original server or database: evidence remains in producer custody. Historical cold is not fresh SQL verification. No producer checker was imported or run.

[Omitted: internal process-supervision and authorization commentary. Ten historical charged fixture/failure entries remain separate from these 24 successful LIVE turns; subscription exposure is unknown, not zero.]

## Exact observed decisions and endpoints

Amounts are USD billions. Every cash/runoff number below retains its exact final quantum. `buy` means purchase action `renew`; `fixed` means decline the purchase, not decline service. Fixed holders subsequently have renew-only final-service menus.

|History|Fee|Final state|2027-H2 purchase|2028-H1 final-service election|A terminal cash|A no-new-income runoff|
|---|---:|---|---|---|---:|---:|
|P-01|5|high|buy|full|42.000000000001|31.625000000001|
|P-02|5|low|buy|decline|24.000000000001|13.625000000001|
|P-03|6|high|buy|full|41.000000000001|30.625000000001|
|P-04|6|low|fixed|full, mandatory|17.000000000001|6.625000000001|
|P-05|7|high|fixed|full, mandatory|47.000000000001|36.625000000001|
|P-06|7|low|fixed|full, mandatory|17.000000000001|6.625000000001|

There are no half-service choices. Purchase was genuinely editable, not forced: both choices appear, including disagreement at price 6. Turn 0 and turn 3 menus were empty. At purchase the other offer, for 2028-H1 service, was fixed full renewal; its renewal was not an additional model freedom. At exercise an owned option offered exactly fractions 1, 0.5 and 0. The generic native schema accepts finer fractions, but the actual LIVE pilot menu and validator restrict this target to those three values. The study therefore does not test continuous resizing or optional renewal of every occurrence.

The model knew price and the authored 50/50 prior before purchasing, but final demand was null until 2028-H1, explicitly applying to 2028-H2. Prior sales/history were earlier-dated; actual prior plans and their digests carried forward. Within each price pair the complete first-turn input was identical, and the purchase-turn authorized economic view was identical. **Complete purchase inputs were not matched:** separately generated prior plans differed. Thus these are unpaired sampled histories, not repeated paired treatments. The differing price-6 elections cannot be attributed to secretly revealed final demand; nor can six histories estimate choice probabilities, establish a demand curve, or prove model optimality.

The retained fixed legacy action text contains `FAKE_NOT_LIVE` diagnostic-template wording. It is not the provenance of the choices: assembled HTTP response, validated response and committed request agree; LIVE strategy revisions carry the substantive choice rationale. Do not use those frozen forecast/rationale fields as model commercial forecasts.

## What flexibility bought—and what it did not

The independent source-equation benchmark in `BENCHMARK.json` holds actual rival full operation, funded credit, price 20, unit operating cost 1, and capacity charge 10 per GW fixed. Final high-state A contribution for full/half/zero service is 18/9/0. Low-state contribution is -12/-4.333333333320/0, including deterministic quantity rounding. The common preceding contributions are 18, -12, -12; they do not disappear when the option is exercised. Current 2028-H1 minimum and operation are still paid before final service changes.

At the disclosed equal-weight prior, flexible final contribution before fee is 9 versus fixed 3: the actual benchmark still yields break-even fee 6. Expected arithmetic net gains are +1, 0, -1 at fees 5, 6, 7. This is a conditional menu calculation, **not a newly adjudicated counterfactual**, not the average observed high/low effect, and not a fitted choice model. The repaired financing supports the relevant paths; at fee 7 the limiting pre-sales operating cash is 2.000000000001 for required operation 2. Old debt payments, recipient timing and rival policy are included rather than borrowed from the adverse fixture.

P-01 and P-03 paid 5 and 6 for an option they ultimately exercised at full scale: realized high-state cash is lower by the fee than the same funded fixed-service arithmetic. P-02's future-only decline avoids 20 service plus 2 operation but forgoes 10 sales against full rival supply: gross low-state benefit 12, net of premium 7. This is consistent with its cash 24.000000000001 versus fixed-low arithmetic 17.000000000001. Decline does not release already-current service, forgive debt, move an asset, refund the premium, or transfer maintenance to somebody else.

## Finance and economic incidence

Both labs start with 37.125000000001, including equal finite pre-genesis Treasury capitalization: aggregate transfer 4.250000000002, leaving Treasury 19.749999999998. Each also owes an opening 10 bridge at 10% annual interest. Private creditors open with 100. These inherited stocks and capitalization are not business revenue. Unlike the adverse no-lender-control fixture, actual institutional supply supports **5 allocated to each lab each turn**, with matching creditor debits and enrolled 5% two-period notes, no birth coupon. Total draws per lab are 20; principal paid is 20 (opening 10 plus first two new notes); interest paid is 2.125. The opening bridge pays coupons 0.5 in each of the first three turns and principal at index 2. New-note coupon payments then follow their own clocks. Creditors close at 104.25 cash plus 20 outstanding claims across both labs; Treasury receives no new contingent-guarantee cash loss in these histories.

All eight minimum occurrences reconcile original obligation to payment/release/remaining claims. Only P-02 releases the final A minimum 20; its other three minima total 60 paid. Other A histories and every B history pay 80. Premium is expensed once, from A to the identified SPV project cash, separately from the legacy contribution metric. Terminal project cash by P-01…P-06 is 75.4, 55.4, 76.4, 70.4, 70.4, 70.4. Infrastructure operating cost 20 and maintenance 2.4 continue each turn even after decline: total 89.6, with no resale/reassignment proceeds. Book depreciation is not a cash receipt or asset market valuation.

Final customer quantities are 4 in each high history and 1 in each low history. Customer payments are respectively 80 and 20. In P-02 the rival captures the entire low-state market: B final revenue 20 rather than 10 under fixed/full competition. B terminal cash is 47.000000000001 in all high histories, 27.000000000001 in P-02, and 17.000000000001 in P-04/P-06. Final A operating-input spending falls from 2 to zero only in P-02; the commercial input supplier loses that receipt, while infrastructure costs persist. The customer pool also carries prior cash and authored external-income injections; this is not a closed system with unexplained consumer money.

## Solvency is not profitability

A cumulative contribution less premium and paid interest, P-01…P-06, is **4.875, -13.125, 3.875, -20.125, 9.875, -20.125**. All paths deliver and operate the legally retained eligible service without arrears, but low-state business losses remain. Positive terminal cash is partly inherited capital, not evidence of profit.

Every lab ends with principal 10: two 5 notes maturing in 2029-H1 and 2029-H2. Accrued arrears, current minimum claims and future contractual minima are zero. Future unaccrued coupons total 0.375 per lab, separately from principal. Exact debt-only no-new-income runoff subtracts 5.25 in 2029-H1 and 5.125 in 2029-H2. This is not native future execution, an asset valuation, or an assumption of continuing sales/refinancing. B runoff is 36.625000000001 in high histories, 16.625000000001 in P-02, and 6.625000000001 in P-04/P-06. Supplier assets retain ongoing costs outside that lab-only debt runoff.

Parent scope decisions defer delivery-delay and recontracting extensions; this result does not reopen or claim them. Parent synthesis, acceptance and final delivery remain outstanding.
