# Capital Loop research integration — October 24, 2026

## Outcome

The reviewed evidence ledger contains 2,400 uniquely identified evidence records: all 1,737 frozen/live baseline records and all 663 supplied research records. All 1,737 original stable relationship IDs were recovered from the ten deployed JSON packages. Three descriptions differed slightly from their sole matching package record; their date, parties and amount matched uniquely.

Default public records: **1,978** (1,529 baseline and 449 new research). Other records remain accessible separately: **123 update facts, 224 contextual/corrective records, 75 watchlist records**. The counts describe records, not unique transactions, verified sources, or completed payments.

## Evidence preserved

The live baseline CSV has SHA-256 `c100bfa4ab812b19fc95d387c27074a7a786e9711a52fa10b31e864c949ba8ca`, exactly the frozen-baseline hash cited in the research sidecars. Raw desktop files and live snapshots remain unchanged. Input hashes, complete research dispositions, baseline ID mapping and correction ledger are retained in the output and supporting files.

All four supplied research CSVs have unique research IDs and matching reconciliation entries. Their counts are W1 127, W3 193, W4 181 and W5 162. Source verification metadata comes from the supplied research. This integration does not independently certify every cited article or every prior claim.

## Material repairs

1. Oracle's $638B company-wide RPO is now Oracle → aggregate customers/RPO context, with the unsupported OpenAI-specific attribution removed.
2. The $850B data-center lease aggregate no longer appears as Meta → Microsoft; the $830B sector capex estimate no longer appears as Microsoft → Amazon. Both are contextual sector records.
3. The SpaceX/xAI $1.25T merger pair (baseline records 912/1595) is one default-public record with both sources retained. The duplicate remains context with `duplicate_of`.
4. The overlapping Reuters $700B sector capex pair (378/487) is one contextual estimate; the duplicate remains linked context.
5. The $1T analyst valuation expectation was removed from the confirmed IPO-filing monetary field (1547). Filing evidence and valuation speculation are explicitly separated.
6. Anthropic/Nscale Monarch W3-0120 and W4-0115 are one economic contract with dated supporting terms. The new-event record uses the prospectus maximum of approximately $44.6B with delivery/availability conditions and uncommitted required financing. The update carries no second monetary amount. No artificial $400M reduction is inferred from rounded earlier $45B reporting.
7. Bird term-loan/revolver and SoftBank USD/EUR note components now have distinct descriptions, preserving shared financing parents without repeating identical summary prose.
8. Baseline amount classification now prioritizes the relationship being measured over accompanying context. A $122B fundraise mentioning $852B post-money is classified as funding; Microsoft's $135B stake value is classified as paper equity value. Infrastructure spend, debt/equity mixtures and ambiguous financial amounts have distinct labels.
9. Focused review of high-value same-party/same-amount candidates resolved 18 additional repeated-event records beyond the original two. Their citations and alternate date/description histories are attached to retained records; originals remain contextual with `duplicate_of`. There are now 20 linked duplicate records and 25 correction-ledger entries. This review is not exhaustive event-level deduplication. Same dollar amounts with distinct transaction structures remain separate, including xAI equity versus GPU-SPV financing and NVIDIA's SB Energy prepaid forward versus additional conditional IPO equity.

## Coverage limits

- **W2 is absent:** July 20–August 11 has handoff leads but no complete supplied ledger. It must not be described as comprehensively covered.
- W5 capture ended October 20 at 20:07:53 America/Chicago. Late October 20 and October 21–24 are not established by this package.
- Corporate confirmation verifies the announcement and stated stage; it does not establish payment completion, contract enforceability, or future delivery.
- Source-sharing is not a duplicate rule. One filing can describe several distinct investments, contracts, facilities and obligations. Only confirmed same-event examples and exact record duplicates were suppressed.
- 123 update facts remain separately labeled. Baseline antecedent matches are candidates, not proof of exact economic overlap. They must not be appended as fresh independent cash totals.
- Baseline amount categories are classifications from original text, explicitly marked as such. Baseline transaction-stage metadata was absent; the output says so rather than inventing it.

## Checks completed

The unified file parses; all 2,400 IDs are unique; every required UI field is present; source URL arrays contain valid HTTP(S) values; no default-public record retains a duplicate marker. The output deliberately provides no aggregate monetary total.


## Complete dashboard integration

The original Capital Loop dashboard is retained: animated six-stage loop, named relationships, entity network, capital lane map, selected-view breakdowns, evidence ladder, table, filters, shared-view links and source inspector. The October native graph file feeds every original renderer. New controls filter by research window and distinguish new events, linked updates, context and unresolved leads.

Network weights and default lanes use record counts. Optional largest-record mode shows a maximum individual USD figure within the selected amount basis, not a sum. No mixed-dollar grand total is presented. This prevents incompatible valuations, funding, guarantees and commitments from being combined while preserving the visual dashboard.