Status: bounded public-source pass completed and combined September 27, 2026. Two Sol evidence lanes, fresh independent Astra review and coordinator integration produced the internal investigation. Historical cash reconciles; exact private recourse, present permit/guarantee outcome and a dated future funding gap remain unknown. No substantiated fraud finding or trading instruction. Full review and return conditions are linked in the output; the broader AI-host funding commission remains open.

What was noticed?

Jan requested an Oracle deep dive on September 27, describing a possible subprime/Enron analogy. September 24–25 reporting about a Project Jupiter force-majeure notice and data-center financing makes the new question concrete: whose payment obligations survive a power or completion delay? The analogy separates into credit/duration mismatch, accounting judgments, and substantiated misconduct; none establishes the others.

Why might it matter?

The answer can change our assessment of Oracle equity risk, project lenders, sponsors and suppliers, and distinguish an avoid/watch decision from an actionable bearish thesis. It may also identify who actually collects cash from the AI buildout. No security is selected for a trade here.

What evidence is real, and what is still only a pointer?

The coordinator and scoping analyst inspected the August 31, 2026 10-Q and September 10 issuer release. They support a financing/commitment investigation and a countercase involving customer funding. September 25 reporting supplies the Jupiter lead. The original commercial lease, notice, lender covenants and current guarantees are not yet established. A county tax/incentive lease is not automatically the commercial lease.

Full source inventory, exact filing footholds, failed retrievals and six detailed work packages: scoping record.

Which existing research does it update or duplicate?

Extend the September 23 Oracle profile, which incorporates the archived September 11 financing work; do not redo a general company profile. Cross-link the unresolved AI-host funding commission without closing it from one company case. The separate AI-issuance/rates task retains its macro scope. This commission owns the Oracle/Jupiter increment.

Bounded work and acceptance

One analyst owns the investigation; a fresh reviewer checks the conclusion and accounting. A first pass is timeboxed to one working day, with source gaps reported instead of open-ended access hunting. Claim and disposition through the existing commission-status command.

1. ORCL-J1 — contracts and recourse. Identify borrower, sponsor, lessor, tenant, guarantors, lenders and notice terms. Table each payment trigger, due date, relief provision and recourse with an exact source clause or unavailable. Distinguish a notice from accepted relief, deferral from forgiveness, and late capacity from default. Stop with conditional outcomes when private clauses are unavailable.
2. ORCL-J2 — funding bridge. Reconcile latest filings/issuances: cash, debt maturities, commenced and uncommenced leases, guarantees, capex commitments and customer prepayments. Keep debt principal, lease present value and undiscounted commitments separate. Code-compute base, construction-delay and customer-payment-delay cases; avoid double-counting prepayments in operating cash and adjusted capex. Report earliest funding need and actual bridge, or bounds.
3. ORCL-J3 — customer/backlog quality. Separate RPO, AI/cloud subsets, earned revenue and received cash. Trace only documented customer financing/payment/guarantee links; identify cancellation/refund/collectibility and owned-versus-customer-supplied equipment. Announced funding is not closed funding. Missing concentration disclosure is unknown, not a made-up OpenAI share.
4. ORCL-J4 — accounting tests. Inspect relevant revenue, financing component, capitalization, useful-life, impairment, lease, consolidation/VIE, related-party and guarantee notes. Separate economic risk, lawful estimates requiring sensitivity, and evidence of inconsistency/misstatement. Do not call disclosed uncommenced leases hidden debt or infer fraud from a funding gap. Shorter GPU-life sensitivity requires documented ownership and accounting basis.
5. ORCL-J5 — first-loss map. Draw actual money and recourse edges; test power delay, customer default, refinancing stress and collateral-value decline. Distinguish a listed asset manager from its managed funds, lender losses from interest-rate marks, and delayed supplier orders from cancellations. No unsupported contagion map.
6. ORCL-J6 — bull rebuttal and decision. Test legacy cash generation, prepayments/customer GPUs, funded completion and enforceable demand as counterarguments. Classify supported credit risk, equity risk, accounting concern or indeterminate; attach decisive missing evidence and dated catalysts. A bearish trade additionally needs timing, priced expectations, premium/borrow/carry and executable liquidity. Without that, stop at research/avoid/watch, not a ticket.

What should be written if the evidence clears?

One internal investigation, a reproducible cash-bridge worksheet/script, an evidence/claim table and loss-allocation diagram; no general enterprise valuation or market-wide securitization essay. Use actual production date for the output filename. Every material finding has source date, measurement basis and confidence/unknown status. Include a section that could falsify Jan's concern. Link the produced event with --commissioned-by so this assignment enters review; independent review precedes any publication or strategy implication.

Stop boundaries: no paid/private lender access or new subscription, broker action, account read, backtest, short/option position or commit. A missing key contract is an explicit limitation. Expand to a systemic-risk study only if sourced common creditors/collateral/covenants show the need.