Unverified50% confidenceEventExact time
After whitespace normalization of the extracted text, the value score dropped from 24/24 to 6/24, while the citations remained 4/4 fully correct, and all values were placed into the wrong fields.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGemma 4 26B terminated with a runtime error before producing a score, because the scoring code assumed at least one sentence would be returned in the correct structured format.68% similarUnverifiedAfter adjusting the prompt, all 24 values extracted by the model (6 fields × 4 rows) were correct, and all 4 citations were also correct.64% similarUnverifiedIn a held-out evaluation on 32 briefs, the new rules read 22 of the explicit budgets (25) correctly, versus 11 for the old regex; and 25 of the explicit deadlines (27) correctly, versus 8 for the old regex.64% similarUnverifiedThe executor, independently of the 60-character prompt, reached 88.9% (32/36) field accuracy simply by reading the full original report; the reviewer added 1 further correct field on top of that, bringing the pipeline as a whole to 33/36.62% similarUnverifiedThe format instruction on the last line of the prompt is the key variable determining the output quality of small models: on math tasks, after unifying the last-line format, differences in question text style caused at most a 5-point fluctuation, while changing the format line could produce a 45-point gap.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1071932API
curl https://kongchang.com/api/v1/knowledge/claims/1071932MCP
get_claim(id=1071932)