COR-2026-001
Fixed
Own articles still cited data vendors on our excluded-source list
What we claimed / the state of affairs
We apply a hard rule to sources: secondary commercial data vendors may not be used as a data source (internally graded R4). Nevertheless, among our 66 in-house industry articles some still named or cited such vendors (Longzhong, Sci99, Mordor, MarketWatch, GII Research and others). This contradicted our own rule and had not been disclosed.
Basis for overturning it
Every one of the 66 in-house articles was scanned to locate the named vendors and classify the context as either “cited its data” or “mere mention”. Result: 13 articles matched, all of them ‘cited its data’; zero were mere mentions.
When
Detected and remediated: 11–12 October 2026
What was done
Following our existing practice for excluded sources, the 13 articles were marked as withdrawn but not deleted: their identifiers were written to the article library's exclusion list (with reason, evidence and date), which the observations index build consumes before regenerating the list page.
Current status (including what is still open)
Fixed · measured: https://trustbaselab.com/insights/ returns HTTP 200 and states 53 articles (66 − 13 = 53); each of the 13 excluded vendors returns zero hits.
Related record IDs
rcb-art-026、rcb-art-028、rcb-art-050、rcb-art-051
COR-2026-002
Open
Under a bilingual claim, the observations index exists at one Chinese URL only
What we claimed / the state of affairs
We claim bilingual delivery and provide a language toggle. However, the observations index exists only at /insights/, whose lang is zh-CN and whose canonical points to itself — there is no equivalent English list page, so English readers landing there from an English entry point see Chinese.
Basis for overturning it
Measured HTTP status: /insights/ → 200; /zh/insights/ → 404; /insights → 301. There is no /zh/-prefixed equivalent and no English counterpart.
When
First recorded: 12 October 2026 (re-confirmed the same day)
What was done
Not yet fixed. Fixing it requires a coordinated change to URL structure and i18n routing — a separate publishing-layer task. This round records it rather than shipping a half-change.
Current status (including what is still open)
Open, recorded as-is: the page works; the defect is the mismatch between the bilingual claim and the URL structure. Because self-consistency is part of our stated method, this entry is published before it is fixed — an archive records, it does not only collect closed items.
COR-2026-003
Partially converged
Three different public record totals: 2,347 / 2,348 / 2,415
What we claimed / the state of affairs
Several public surfaces each report a “total record count” and the numbers disagreed. Measured on 12 October 2026: the MCP data service returned total_records = 2,347; the homepage status bar showed 2,348; the authoritative dataset data.json reported meta.total_records = 2,415. For a data infrastructure selling verifiability, the total is the single easiest number for an outsider to spot-check — three coexisting values are themselves an untrustworthy signal.
Basis for overturning it
Recomputed table by table (definitions and commands are in corrections.json):
(1) the live MCP snapshot (data_version 0.12, exported_at 2026-10-11T18:44:15): 12 tables summing to 2,347, with domestic = 37;
(2) the homepage snapshot: the same 12 tables but with domestic = 38, summing to 2,348;
(3) the authoritative local dataset: 13 tables (including the fitness domain, 67 records), summing to 2,415.
When
First recorded 12 October 2026; the definition below was fixed the same day
What was done
One definition, fixed: the public total equals meta.total_records in the authoritative dataset data.json, and must equal the sum of all keys in meta.counts; records flagged excluded=true are not counted. Every public surface must be driven by that field instead of computing its own.
Current status (including what is still open)
Partially converged, recorded as-is: the definition is fixed, but two discrepancies remain and require a data release — (1) the fitness domain (67 records) is not yet published on our public surfaces, so the authoritative total (2,415) still exceeds the public total (2,348) by 67; (2) the live MCP snapshot is one domestic record behind the homepage (37 vs 38). Until that release lands, the public figure remains 2,348 and this archive does not claim it has converged.
Related record IDs
rcb-cn-001 (示例)、cb-trade-*、tbm-*