WEO Methodology Errata & Corrigenda

CG-001 — WEO Methodology Rationale V1.5, §9.5.2 and §9.5.3

Applies to: WEO Methodology Rationale V1.5 (19 July 2026), DOI 10.5281/zenodo.21278281 · Applied to the self-hosted edition: 29 July 2026 · Status: Substantive correction to published calibration and validation figures. This corrigendum sits outside the Errata Register, which is reserved for corrections of expression. The methodology itself — the Criterion 1 mandatory gate, the mechanism specificity sub-test with its indicative table, tolerance clause and extension mechanism, and the Criterion 2 industry-tag binary test — is unchanged.

Sequence: June 2026 — corpus audit ranks 74 PPs, flags 10 LOW. · 19 July 2026 — Rationale V1.5 publishes the audit's retroactive impact as settled (10 removals). · 28 July 2026 — V1.5 provisions instantiated across the corpus; re-derivation against the published test removes 6 and retains 4. · 29 July 2026 — divergence recorded as CG-001; self-hosted edition corrected.

Background. §9.5.2 and §9.5.3 report the June 2026 Type 7 corpus audit (74 Parallel Policy connections ranked by mechanism specificity) and state its retroactive impact as settled. When the V1.5 Criterion 1 provisions were instantiated across the corpus on 28 July 2026, each of the ten LOW-ranked connections was re-derived directly against the published test and each event's verified source packages. The re-derivation established that the audit ranking had applied a stricter standard than the published sub-test: the ranking's LOW justifications weighed divergences of scale, vehicle structure and actor type that the sub-test's tolerance clause expressly tolerates, and recorded that no shared characteristic was identified during the exercise, where the published test asks whether one can be identified from verified sources. For four of the ten, one can — in two cases a characteristic listed in the Manual's §9.4.3 indicative table itself.

Corrections to §9.5.2.

• The statement "All 10 LOW-specificity PPs fail the sub-test" is corrected: six of the ten failed and were removed — five through the specificity sub-test and one through the mandatory gate. Four satisfied the sub-test on documented sub-category characteristics identified from verified sources (drawn from the indicative table or its extension mechanism) and were retained, with the shared characteristic recorded in each connection's specification_overlap field.
• "10 removals (13.5% of the 74-PP corpus)" is corrected to 6 removals (8.1%). "Zero MEDIUM or HIGH PPs are affected" stands.
• In the §9.5 sensitivity analysis, "2 cross-category CCs… both were removed" is corrected: of the two connections the audit identified as cross-category, one was confirmed cross-category and removed through the mandatory gate; the other was reassessed at instantiation as a same-category (Framework/Governance) match by parity with an existing corpus exemplar, satisfied the sub-test, and was retained. "8 of 50 Infrastructure-dominant PPs" describes the audit ranking; at instantiation the sub-test removed five of the fifty and preserved forty-five.
• The §9.5 Validation Commitment figure "74 PPs ranked, 10 removed, 2 confidence reclassifications" is corrected to "74 PPs ranked, 6 removed, 4 LOW-ranked PPs retained on documented sub-category characteristics, 2 confidence reclassifications". The confidence-reclassification figure is unaffected.
• The specificity distribution table stands as the record of the June ranking exercise. The correction severs the published equivalence between the ranking and the sub-test outcome.

Corrections to §9.5.3.

• "72 of 74 existing Type 7 PPs share at least one industry tag" is corrected to 73 of 74; the tag-overlap table is corrected to 63 same primary tag (85.1%), 10 bridged by secondary tag (13.5%), 1 zero overlap (1.4%); "97.3%" is corrected to 98.6%. These figures reflect the tag set current at instantiation, which governs under the Manual's §9.4.3 re-verification rule.
• The two-connection narrative is corrected: the single connection without a shared tag was already assessed as Criterion 2 NOT MET under V1.4 and was removed under the Criterion 1 corrections above. The connection previously described as dropping from 4/4 to 3/4 in fact shares a secondary industry tag under the current tag set, satisfies Criterion 2, and stands at 4/4. The closing finding — that the industry-tag formalisation produces zero additional removals beyond those attributable to the Criterion 1 changes — stands.

Corpus state. Following the 28 July 2026 instantiation, the Type 7 corpus comprised 70 connections, all conformant with the V1.5 schema and all at CC-A confidence. These corrections are applied to the self-hosted edition and will be folded into the next DOI-published version of the Rationale. The V1.5 deposit is the archival record; this corrigendum is the public record of the divergence.

Evidence annex. Removed (6): CC-185-188-7 (mandatory gate, cross-category), CC-126-31-7, CC-126-39-7, CC-128-126-7, CC-46-49-7, CC-126-52-7 (specificity sub-test). Retained (4): CC-139-46-7 ("State-backed dedicated AI investment fund"), CC-194-67-7 ("AI safety institute establishment and network participation"), CC-45-200-7 ("State-subsidised domestic sovereign AI compute capacity programme"), CC-175-177-7 ("Government-funded sovereign AI compute access for domestic researchers, startups and industry"). Zero-overlap connection: CC-46-49-7. Secondary-bridge connection previously reported as zero-overlap: CC-44-45-7. All ten connection records are verifiable against the live corpus via the REST API and MCP interface.