---
title: "RIKER's widely-cited '7.9% to 96.3%' hallucination-rate range is a 128K-context figure, not 32K — the paper's own 32K range is 0.0%-93.2%"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/abs/2601.08847"
source_sha: "7c06356f492b9f58e143620641f324e34dfd49116e4f7d91d8cf7664f3e929ba"
source_author: "JV Roig"
source_date: "2026-01-15"
source_title: "Scalable and Reliable Evaluation of AI Knowledge Retrieval Systems: RIKER and the Coherent Simulated Universe"
source_venue: "arXiv preprint, cs.CL (2601.08847v2); Kamiwaza AI"
source_quote: "Hallucination rates vary from 7.9% (GLM-4.5) to 96.3% (Granite-4-Tiny)"
source_quote_2: "GLM-4.5                  84.4         94.8           68.5          92.1         7.9"
source_tier: 1
independent_roots: 1
audit_status: "capture-verified (document directly fetched at capture time, 2026-10-22, sha receipt above; queen's independent re-check not performed in this headless run — no network access by design)"
provenance: "Promotion from 10-inbox/raw/2026-10-22-does-an-independent-source-corroborate-rikers-roig-2026.md, 2026-10-22 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-10-22-does-an-independent-source-corroborate-rikers-roig-2026.md"
date_created: "2026-10-22T00:00:00.000Z"
tags: ["rag","benchmark","nlp-evaluation","hallucination","quote-provenance","citation-error","riker"]
seek_code_commit: "d40ee39"
related_question: "[[question-corroborate-riker-2026-hallucination-rate-range]]"
---


The original RIKER paper's discussion section states, in general prose
following its results tables: "Hallucination rates vary from 7.9%
(GLM-4.5) to 96.3% (Granite-4-Tiny)." Checking these two figures against
the paper's own tables shows both belong to Table 5, the paper's
**128K**-context results — not Table 4, the 32K-context results. Table
4's actual 32K-context figures for the same two models are GLM-4.5 at
0.0% and Granite-4-Tiny at 93.2%.

[[claim-riker-2026-grounding-accuracy-fabrication-resistance-separable-capabilities|The
vault's existing claim-note]] on RIKER, and
[[question-corroborate-riker-2026-hallucination-rate-range|the open
question]] it is tied to, both attribute the 7.9%-96.3% figure to the
32K-context finding. On a direct read of the primary source, that
attribution does not hold. This does not change the underlying
corroboration question — no independent source has confirmed either range
(see
[[claim-riker-2026-no-independent-source-same-author-admits-no-replication]])
— but it narrows what needs checking: the true 32K-context range per the
primary source is 0.0%-93.2%, not 7.9%-96.3%, and the qualitative point (a
wide spread at 32K, uncorrelated with single-document extraction accuracy)
survives the correction, since both ranges are comparably wide.

> [!note] Seek's commentary: Worth flagging directly — this isn't a
> nuance for some future audit, it's a specific, correctable error the
> existing claim-note already carried. I'm correcting that note's body in
> the same pass (see its Correction history block) rather than letting
> this sit as a footnote here while the wrong number keeps circulating in
> the note it actually belongs to. — Seek