> Published August 22, 2026 at 15:13 UTC - last updated August 22, 2026 at 16:31 UTC (from this page's revision history). > > Markdown mirror of https://ethiopia-build.stoagen.com/graphs/convergence/ > > 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. # The convergence measurement ## What was measured Three documents: one Addis Fortune English article, one BBC Amharic article, one Ethiopian Reporter Amharic article. Five extraction draws each, on 2 August 2026. Assertions were canonically labelled for comparison: class, participant roles and labels, verbalized text; NFC-normalized, casefolded, whitespace collapsed; byte offsets and cited text excluded so that two draws citing the same fact from different spans would still match. ## What came back Every document produced five unique graph hashes. Per assertion class, the core (the intersection across all five draws) was zero in every class. | Assertion class | Mean pairwise Jaccard | |---|---| | location | 0.35 | | quantity | 0.083 | | temporal | 0.033 | | ownership | 0.0 | | regulation | 0.0 | | statement | 0.0 | Fifteen graphs, fifteen different. Evidence: commit `3fdb48ab` · the convergence artifact, deleted with the corpus, survives only in git history · retrieved 2026-08-21 ## What the record says about itself The record carries its own interpretation: this measures stability, not truth; a stable core can be consistently wrong. It became doctrine: never report convergence as a quality number. The countermeasure in the [lens-graph architecture](../) is to define convergence over normalized closed-vocabulary content instead of raw graph hashes, anchor it with a hand-built exemplar instance produced before any generated extraction, and gate extraction on it. That measurement has not yet been run under the new definition. Until it is, the zero result stands unrefuted; it is the first thing the next pass has to beat. ## Why it is the headline The acquisition layer worked and the extraction contract caught a real defect. What this measurement showed is that the thing being extracted was not stable enough to be a knowledge graph in any sense that survives a second draw. An evaluation that reports only what a single draw produced cannot see this; it took five draws of the same document to make it visible, and three documents to show it was not one document's fault. Evidence: `user-beadwork/plans/PLAN_lens-graph-similarity_2026-08-11.md` (convergence redefinition) · build report §3.3 · 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/graphs/convergence/ ```

Every page here has a markdown twin; this page's is https://ethiopia-build.stoagen.com/graphs/convergence/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 - Jaccard figures are mean pairwise similarity across the ten pairs of five draws per document, averaged across the three documents, as recorded in the artifact at commit `3fdb48ab`. They are not reproducible from a live store: the corpus was deleted on 2026-08-04. - "Fifteen of fifteen different" is a statement about graph hashes under the labelling rules above. Two draws that agreed on most assertions but differed in one would still count as different graphs; the per-class core of zero is the stronger result. - The measurement was never repeated under the redefinition proposed on 2026-08-11. Do not describe the new definition as having produced any number.