coldstart

Calibrate AI-generated items before you have the pretest seats to do it the old way.

Automatic item generation and LLM-assisted writing produce items faster than pretesting can calibrate them. coldstart predicts each new item's difficulty from its features, uses the prediction as a robust prior, finds the families where prediction fails, and tells you how many responses each item needs.

library(coldstart)

sim  <- cs_simulate(seed = 1)                     # legacy + new items with features
it   <- sim$items; tr <- it$set == "train"

pr   <- cs_predictor(it$b_legacy[tr], sim$features[tr, ], it$family[tr])
pred <- predict(pr, sim$features[!tr, ], it$family[!tr])   # mean + honest SD

cs_plan(pred, target_sd = 0.3)                    # responses needed per item

resp <- cs_responses(sim, n_per_item = 25)        # or your pretest data
cal  <- cs_calibrate(resp, pred)                  # t-prior Bayes vs baseline
chk  <- cs_check(cal, setNames(it$family[!tr], it$item[!tr]))
cal  <- cs_calibrate(resp, cs_distrust(pred, chk))  # drop priors that failed

Installation

From CRAN:

install.packages("coldstart")

Development version from GitHub:

install.packages("pak")
# pak::pak("edidatasolutions/coldstart")
install.packages("coldstart", repos = c('https://edidatasolutions.r-universe.dev', 'https://cloud.r-project.org'))

Features can be anything numeric: embeddings from any text model, cognitive attribute codes, content metadata. The package does not call a model itself.

Design choices

• Honest predictive SD, two kinds. Out-of-fold RMSE for families seen in training, and leave-one-family-out RMSE for new templates.
• Robust prior. The default Student-t (df = 4) lets the data override a bad prediction instead of being dragged toward it.
• Family-level trust check. A chi-square test of prior-data conflict per family that allows for the prediction error all items of a family share (for a new template, its unknown family effect); cs_distrust() withdraws priors from failing families.

Validation (known truth, 100 replications)

Predictive SDs are honest: stated 0.54 vs actual RMSE 0.53 (seen families), 0.68 vs 0.64 (unseen family); 90% intervals cover 90.8% and 90.2%.

RMSE of difficulty, seen families:

responses per item	baseline	predicted prior
15	0.69	0.41
25	0.52	0.36
50	0.35	0.29
100	0.24	0.22
200	0.17	0.16

With the prior, 25 responses do about what 50 do without it.

Drifted ("rogue") template family (+1.2 logits vs its history): the prior alone hurts (0.63 vs 0.60 at n = 25). cs_check flags the family in 80% of replications at n = 15, 94% at n = 25 and 98–100% at n ≥ 50. After cs_distrust(), RMSE is back at or slightly below baseline (0.58 at n = 25).

False flags (nominal 1% per family): honest seen families 0.3–2.1% (1.3% at n = 200 pooled over 300 replications); the new, unseen family 0–1%. Since version 0.2.0, the check accounts for error shared within a family. Treating items as independent, as version 0.1.0 did, flagged the unseen family in 3–7% of replications.

Planner: a target posterior SD of 0.30 needs a median of 40 responses per item with the prior vs 57 without; achieved SD 0.303, RMSE 0.301.

Status

Done: cs_simulate, cs_responses, cs_predictor (+predict), cs_calibrate, cs_plan, cs_check, cs_distrust. Next: 2PL (discrimination priors), sequential updating as responses arrive, ability uncertainty for pretest examinees (currently treated as known from operational scoring), and non-linear predictors.

Getting help and contributing

Questions and bug reports: https://github.com/edidatasolutions/coldstart/issues. See CONTRIBUTING.md for how to report problems, get help, or contribute code.