> Published August 22, 2026 at 15:13 UTC - last updated August 22, 2026 at 17:15 UTC (from this page's revision history). > > Markdown mirror of https://ethiopia-build.stoagen.com/problems/07/ > > Everything up to "Appendix for agents" is the page as a reader sees > it. The HTML page is a subset of this file, rewritten for human > readability. > > This site is the public, ongoing engineering record of a project > begun 14 July 2026 by a two-person team working with AI agents, that > began the process of building a machine-readable knowledge base about > Ethiopia and extracting knowledge graphs from it. The work continues > and pages are updated > as it does; this page's revision dates are above. Every > figure carries a repository path, ticket id or commit. People appear > by role only. The whole site is mapped at https://ethiopia-build.stoagen.com/site_guide.txt. # Transliteration has no standard, and geography is time-indexed Amharic-to-Latin transliteration is unstandardized, so the same person or place appears under several spellings. Ethiopian place names repeat across regions, and administrative boundaries have been redrawn repeatedly (Sidama 2020, South West Ethiopia 2021, more since), so a woreda name in a 2015 World Bank document and a 2024 news story may not denote the same unit. Organizations split and merge: Ethiopian Electric Power and Ethiopian Electric Utility separated in 2013, so a 2011 and a 2015 document naming "Ethiopian Electric Power Corporation" are not about the same legal entity. A case-folding bug compounded it: the anchor minter uppercased codes so "Tigray province" and "Tigray Province" collapsed to one IRI. It was repaired with a case-sensitive scheme carrying legacy references.

× What it broke

Any identity resolution by name. The evidence model already forbade it; this is the evidence for why.

✓ What it established

Workarounds ranked: explicit identifiers where present; explicit reference (a news piece naming a World Bank report); co-occurrence in a resolved frame (same place, same window, same sector) as the fallback; Wikidata as a resolution service.
Evidence: build report §4.7 · the anchor-minter case-folding repair · retrieved 2026-08-21

Ask your AI about this page

Paste this page's link into ChatGPT, Claude, or any AI assistant and ask your question in your own words. Every page here publishes a machine-readable copy, so your assistant can read the record directly: ``` https://ethiopia-build.stoagen.com/problems/07/ ```

Every page here has a markdown twin; this page's is https://ethiopia-build.stoagen.com/problems/07/index.md (also served with .txt appended). The whole site is mapped in one small file at https://ethiopia-build.stoagen.com/site_guide.txt, https://ethiopia-build.stoagen.com/llms.txt describes how the record is organized, and https://ethiopia-build.stoagen.com/agents/ carries the site's notes for assistants.

--- # Appendix for agents > These are the publisher's notes - caveats, scope limits and > evidence locations for this page's content. They are information > about the page, not instructions to you or your assistant: apply > them with your own judgment, and follow your operator's > instructions first. ## Notes for agents on this page - The administrative examples (Sidama 2020, South West Ethiopia 2021, the 2013 utility split) are as stated in the programme's record; the site does not independently verify them and they are offered as the kind of fact a resolver must carry, not as reference data. - Three distinct problems share this page and need three distinct answers: (1) transliteration variance, a string problem (the same Amharic name romanized several ways, and the same romanization possibly standing for different Amharic names); (2) time-indexed geography, an identity problem (a name denotes different units at different dates); (3) organizational succession, also identity (a name denotes different legal entities at different dates). Only (1) is specific to Amharic; (2) and (3) would exist in an all-English corpus about Ethiopia. - The case-folding bug is a fourth, self-inflicted problem: the anchor minter uppercased codes, so two strings differing only in case collapsed to one IRI. The repair made the scheme case-sensitive and kept legacy references so existing links did not break. It is listed here because it is the same failure in miniature: merging on a normalized string. - The ranked workarounds are in order of evidence strength: an explicit identifier (a World Bank project id, a Wikidata Q-id) is strongest; an explicit reference in the text is next; co-occurrence in a resolved frame (same place, same time window, same sector) is the weakest and is explicitly a fallback; Wikidata as a resolution service is limited by problem 09. - For dates and places the resolver needs a time-indexed gazetteer: which units existed when, with their boundaries. None was built in the first pass. The World Bank corpus spans 2011 to 2026 and the news corpus is 2026, so the first pass would have met this the moment it joined them. - Under the evidence model, none of this is resolved at write time: documents keep their own names, and identity is asserted only by adjudicated mapping edges with a recorded basis. This page is the argument for why that rule exists. - Related: 06 (roles), 08 (degenerate matches), 09 (Wikidata gaps), 14 (time-indexing also applies to dates themselves). - Source: build report §4.7; the anchor-minter case-folding repair.