# Independent method review

Review date: 2026-09-23 (America/New_York). This is an independent computational check by a second agent, not human peer review.

## Frozen question and scope

The existing [Statecrafts protocol](https://statecrafts.com/research-updates/2026-09-19#ru26-icio-protocol) asks for originating value added embodied in a partner-sector gross-export bundle. The frozen extension is nine cells: JPN, KOR and DEU, each in C26, C27 and C28, in 2022. No cells may be selected or dropped because of their estimates. The full global network must be retained.

## Definition checks

- The [regular ICIO ReadMe](https://webfs-sti.oecd.org/files/STI-PIE/ICIO/2025/ReadMe_ICIO_small.xlsx) separates VA at basic prices (Structure!B19/D28), TLS on products (B17/D27), and output (B21/D29); units are current USD million (ReadMe!C11). Its six final-demand items include direct purchases abroad (Structure!D37). Use country/industry strings, not numeric row IDs: some numeric IDs in its RowItems sheet are duplicated.
- The [2025 TiVA guide](https://stats.oecd.org/wbos/fileview2.aspx?IDFile=2143f34e-6feb-41a9-abaf-cb52132608c4), Table 2.1 (p. 6) and section 4.2 (p. 12), defines its value-added decomposition using basic-price VA **plus** intermediate-product net taxes. Our primary VA-only numerator follows the original protocol; taxes therefore need a separate reconciliation. An explicitly labelled VA+TLS sensitivity is useful, not interchangeable with the primary estimate. Sections 4.4 (p. 16) and 7.1 (p. 38) support foreign intermediate plus final uses, including nonresident direct expenditure, and the single-export-cell origin decomposition. Its discussion also notes valuation, re-export and bilateral-balancing adjustments: these are not customs export totals.
- The [2025 annex](https://webfs-sti.oecd.org/files/STI-PIE/ICIO/2025/ICIO2025annex.pdf), Tables 1–2, confirms all country and sector codes. C26 includes computer, electronic and optical products; C27 electrical equipment; C28 machinery and equipment n.e.c. They are broad industry aggregates, not product-specific semiconductor/battery/machine-tool or military-capability measures.
- The [OECD release page](https://www.oecd.org/en/data/datasets/inter-country-input-output-tables.html) identifies regular versus extended tables and the January 2026 revision. Regular tables omit the China/Mexico firm-heterogeneity splits used for published OECD GVC indicators; agreement with a published TiVA cell is not an exact acceptance test without resolving that difference.

## Independent calculation plan

Build the matrix from CSV labels, use `A[i,j] = Z[i,j] / x[j]`, and separately sum the selected export rows over every foreign intermediate and final-demand column. Confirm those denominators against output minus domestic uses. Do not add the output total or separately add tax rows to the export denominator.

Instead of solving the analysis agent's nine forward production systems, solve **transposed** systems for each origin's VA and TLS coefficient masks. For origin `o`, solve `(I-A).T h_o = 1_o * (VA/x)`. Its share in exporter-industry `k` is `h_o[k]`; its monetary numerator is `h_o[k] * exports[k]`. Solve all 81 origin masks, plus separate TLS and input-balance-residual masks. This independently recovers every origin rather than checking just CHN and USA.

Required tests: source hashes and archive-member provenance; complete and unique labels; finite necessary data; zero-output treatment; row/column output identities; sign inventory without clipping; correct foreign-only denominator including DPABR; transposed solve residual; all-origin VA/TLS/rounding reconciliation; all nine CSV cells; all 729 origin rows; chart numeric labels and plotted components versus CSV. Tolerances must be explicit. Any unexpected missingness or material imbalance is a blocker, not an invitation to fabricate zeros or normalize shares.

## Interpretation boundaries

These accounting shares do not estimate the output lost under a disruption, physical chokepoints, ownership, alternate-supplier qualification, inventories, military usefulness, or wartime access. They do not measure 2026 conditions or a trend. A small value share may be essential; a large one may be replaceable. Gross export bundles can contain the same upstream value across multiple border crossings; summing the nine bundles would not identify unique coalition value added. No confidence interval is inferred from one deterministic table.

## Acquisition qualification

The current OECD ZIP is inaccessible through the direct download route during this run. The [World Bank reproducibility catalogue, record 511](https://reproducibility.worldbank.org/catalog/511/study-description), explicitly identifies its included `data_raw/2022_SML.csv` as the 2025 regular ICIO January 2026 update, accessed March 2026. If used, state this archival delivery route, hash the archive/member, and verify archive integrity. The catalogue's vintage claim is supporting provenance, **not** proof of byte identity with the presently hosted OECD ZIP. World Bank verification of its own paper does not validate this Statecrafts study.