I've included 24 carbon-tax vignettes: three frames crossed with two speaker types, with four versions per cell. They form a worked example with the simulated ratings in repllm_synthetic and repllm_human.
Format
A tibble with 24 rows and 5 columns:
- material_id
Identifier,
"m01"to"m24".- frame
Framing condition:
"economic","moral", or"scientific".- source
Speaker condition:
"expert"or"citizen".- version
Realisation number within the condition, 1 to 4.
- text
The generated vignette.
Details
Two texts contain their own condition labels, so
check_manipulation_leakage flags them. Length and readability
pass the default thresholds. These data let you inspect the checks without
making a model call.
Examples
data(repllm_materials)
# Tier 1: local checks, no API key required
validate_auto(repllm_materials, condition_col = "frame")
#>
#> ── Automatic validation ──
#>
#> ✔ Length balance: max deviation 4.8%
#> ✔ Readability: grade-level spread 0.4
#> ! Manipulation leakage: 2/24 materials name a condition term.
#> ✔ Distinctiveness: Jaccard overlap in [0.1, 0.11]
#> ! Needs attention: "leakage"
#> Automatic validation
#> Materials: 24 across 3 conditions
#> length PASS
#> readability PASS
#> leakage WARNING
#> distinctiveness PASS
#> Overall: needs attention
# The two materials that name their own condition
leak <- check_manipulation_leakage(repllm_materials$text,
repllm_materials$frame)
#> ! Manipulation leakage: 2/24 materials name a condition term.
leak$flagged[, c("condition", "term")]
#> # A tibble: 2 × 2
#> condition term
#> <chr> <chr>
#> 1 economic economic
#> 2 scientific scientific