pub struct RunRecord {
pub model: String,
pub timestamp_display: String,
pub timestamp_file: String,
pub backend: String,
pub cases_dir: String,
pub system_prompt: String,
pub report: EvalReport,
}Expand description
One persisted eval run: an EvalReport plus the run metadata the HTML report needs to label,
sort, and group runs across models and over time. Serialized to
results/{model}_{timestamp_file}.json.
Fields§
§model: StringHuman-friendly model label for this run (an alias like qwen2.5-7b, the GGUF filename for a
local file, or the remote model id). Used as the comparison/grouping key.
timestamp_display: StringWall-clock start of the run for display, e.g. 2026-06-18 14:03:21 (local time).
timestamp_file: StringThe same instant in a filesystem-safe sortable form, e.g. 20260618-140321 — used in the
results filename and as a stable per-run sort key in the report.
backend: String"local" (a llama-server we spawned) or "remote" (the configured provider).
cases_dir: StringThe case directory this run scored against (so a report can flag runs over different sets).
system_prompt: StringThe system prompt this run used, copied from EvalReport::system_prompt so the persisted
record carries it at the top level too. #[serde(default)] keeps old result JSON loadable.
report: EvalReportThe full eval report (per-case outcomes + aggregate stats).