LLM Accuracy Evals

How much does verified cricket evidence improve LLM answers?

We test each AI model twice on 250 sampled questions drawn from a 1,000-question cricket benchmark bank. Pass A: no data — the model answers from training knowledge alone. Pass B: CricketStudio's verified stats injected as context. The gap shows exactly how much structured cricket data improves AI accuracy.

1,000 questions in bank250 sampled per run4 models tested2 rounds — without data, then with CricketStudioClaude Haiku judgeWeekly · Mon 06:00 UTCBenchmark: cricket-qa-v2
Highest accuracy · 2026-07-15
Gemini 3.1 Flash Lite
18.8%
89.6%
+70.8 pts improvement

Highest accuracy across all models when CricketStudio data is injected as context.

Biggest lift · 2026-07-15
Gemini 3.1 Flash Lite
+70.8 pts
accuracy points gained with CricketStudio
18.8%89.6%

Started lowest, gained the most. CricketStudio data is the equaliser.

Rankingswithout data vs with CricketStudio context
Without dataWith CricketStudio OKF
1
G
Gemini 3.1 Flash Lite
Google
18.8%
89.6%
+71pp
2
O
GPT-4o
OpenAI
21.2%
88.8%
+68pp
3
P
Perplexity Sonar Pro
Perplexity
42.0%
88.8%
+47pp
4
A
Claude Sonnet 5
Anthropic
14.4%
83.2%
+69pp
Accuracy by question typeraw knowledge → with CricketStudio
T3
Single-entity facts
500 questions · One player, team, or venue in a specific context
GGemini
13.2%84.9%+72pp
OGPT-4o
14.5%86.2%+72pp
PPerplexity
38.2%86.2%+48pp
AClaude
11.8%77.6%+66pp
T4
Career arcs
250 questions · Multi-season evolution and peak/decline patterns
GGemini
23.4%100.0%+77pp
OGPT-4o
35.9%95.3%+59pp
PPerplexity
43.8%95.3%+52pp
AClaude
12.5%93.8%+81pp
T5
Compound conditions
150 questions · Two or more filters applied simultaneously
GGemini
28.6%92.9%+64pp
OGPT-4o
0.0%85.7%+86pp
PPerplexity
35.7%85.7%+50pp
AClaude
0.0%85.7%+86pp
T6
Causal & debate
100 questions · Why/who-is-better — rewards reasoning, not just recall
GGemini
40.0%90.0%+50pp
OGPT-4o
40.0%90.0%+50pp
PPerplexity
70.0%90.0%+20pp
AClaude
50.0%90.0%+40pp
Run history
2026-07-15Gemini 3.1 Flash Lite250 Q
2026-07-14Perplexity Sonar Pro250 Q
How it works
Pass A — Raw knowledge
  • Question sent with no context — model answers from training data alone
  • 250 sampled from a 1,000-question bank — cricket-qa-v2
  • Four types: single-entity facts, career arcs, compound conditions, causal debate
  • This score reflects what the model "knows" about cricket without any help
Pass B — With CricketStudio OKF
  • Same question, but CricketStudio's OKF knowledge bundle injected as context
  • Context source: okf.cricketstudio.ai/llms.txt
  • All OKF facts derived from ball-by-ball data with explicit provenance
  • The "Improvement" column shows how many accuracy points CricketStudio adds per model
What kinds of questions?

The bank holds 1,000 questions across four difficulty levels; each weekly run samples 250 of them. Even the “simple” ones stump most LLMs without CricketStudio data — they require exact ball-by-ball aggregation that no model carries in training.

T3Single-entity factsbank: 500

Stats about one player, team, or venue in a specific context. Sounds straightforward — but requires exact ball-by-ball aggregations that LLMs don't carry in training.

“What is Shubman Gill's head-to-head record vs Yuzvendra Chahal across their IPL career?”
“When captains win the toss at Dubai International Cricket Stadium, which choice wins more often?”
T4Career arcsbank: 250

How has a player evolved across multiple IPL seasons? Peak vs decline questions need season-by-season breakdowns — impossible without structured time-series data.

“How has Deepak Chahar's bowling economy changed across his IPL career? Which was his most economical season?”
“How has Trent Boult's wicket-taking evolved across 11 IPL seasons?”
T5Compound conditionsbank: 150

Two or more conditions stacked together. Requires filtering the dataset to only the matches where both X and Y are true simultaneously.

“What are the win odds when a team wins the toss AND elects to bowl at Ekana Cricket Stadium?”
“Has RCB's powerplay batting strike rate improved since the Impact Player rule was introduced in 2023?”
T6Causal & debatebank: 100

Who is better? Why does X outperform in Y conditions? Good answers define the metric, time window, and era before concluding — not just picking a name.

“Who is better in the IPL — Virat Kohli or Rohit Sharma?”
“Which season was Munaf Patel at his bowling peak in the IPL?”
Scoring
  • Judge: Claude Haiku — scores each response binary 0 or 1
  • Partial answers counted correct when the core fact is accurate
  • Models run in parallel batches, rate-limited to avoid throttling
  • Run weekly · Mon 06:00 UTC · results committed to this repo
  • Methodology published — see benchmark design

Benchmark: cricket-qa-v2 · Judge: claude-haiku-4-5-20251001 · Benchmark methodology· Citation policy