Building the Ethiopia knowledge base

Chapter 03 · the rules underneath

The evidence model

Five layers that may never collapse into one note, and the ruling of 25 July 2026 that nothing receives an absolute truth state. Identity is never resolved at write time. These are the rules every generation of graph was built under.

Five layers

Five layers that may never collapse into one note:

  1. The publication observation. A surface exposed an item at a retrieval time.
  2. Source-supplied metadata and content. What the publisher said and served.
  3. Attributed claims. Hard-pinned to the single status attributed-unverified.
  4. Program-produced derivatives. OCR, translation, transcription, summaries, each carrying source hashes, method, version, producing task, timestamp and review state.
  5. Assessment. Corroboration, contradiction, bias, each requiring its own evidence.

Story clusters across publishers carry the constant cluster_is_not_truth_determination: true.

ethiopia-program/docs/NEWS_CORPUS_CONTRACT_V1.md · ethiopia-program/docs/schemas/ · retrieved 2026-08-21

The evidence-relative ruling

On 25 July 2026 the operator issued the ruling that reshaped intake: stop evaluating supplied material with a binary true/untrue or verified/unverified gate. Nothing receives an absolute truth state. Preserve claims, sources, provenance and context; record support, tension, conflict, uncertainty and interpretation at the claim level; keep conflicting material in the corpus; reserve "verification" for source identity, capture integrity and provenance. Rights, access, privacy, security and publication controls are separate, non-epistemic gates.

The first source record accepted under it, the same day, was a user-supplied Civilization VI wiki clipping, taken in with its provenance gaps listed rather than filled.

2026-07-25 ruling · the SB Ingestor seat's intake contract, encoded into four surfaces · retrieved 2026-08-21

Identity is never resolved at write time

Cross-document identity comes only through adjudicated mapping edges with a recorded basis, never through name or label match, because "Productive Safety Net Project" and "Commercial Bank of Ethiopia" in two documents may denote different referents, and because write-time cosine merging is order-sensitive, destructive, and erases corroboration counts.

This rule is why the entity matching finding and the transliteration problem are findings rather than bugs: the design refused to merge on names, and the record shows what happens when a corpus is left unmerged.

build report §3.4 · oo-ld-corpus/AGENTS.md · 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/evidence-model/

Every page here has a markdown twin; this page's is https://ethiopia-build.stoagen.com/graphs/evidence-model/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.

Published . Times come from this page's revision history and can be checked against it.