Run and resume explicit-evidence linking sessions

Sessions retain common E1–E3 results under linking$estimator, including estimator ID/version, per-spoke continuation, diagnostics and exact provenance. Pair selection and stopping remain external. No estimator is selected by default. Adaptive Phase B D-optimal execution is unavailable for all estimators pending a separate selector validation study. Posterior covariance does not establish selector validity. Use these explicit-evidence sessions for E1–E3 comparisons; E3-MCMC remains an explicitly requested audit/reference engine.

Usage

start_link_session(input, status = "active")

resume_link_session(state, input, status = NULL)

Arguments

input

A prepare_link_input() object, or (for start_link_session) a list of such objects with distinct spokes and identical hub evidence/judge.

status

Explicit reporting status: active, probe, or frozen. This labels controller status only: all observations in input$cross remain active estimation evidence. Held-out probe outcomes must not enter that table.

state

A linking session from start_link_session() or load_link_session().

Value

A pairwiseLLM_link_session with exact data-only results and a stage log.

Details

Resume requires the same estimator, version, item order, Phase A, judge, numerical configuration and unchanged old cross-evidence prefix. New rows may only be appended. Passing the identical input is an exact no-op (including MCMC); extended evidence refits using the saved numerical mode, never the previous posterior as a new prior. A frozen spoke cannot append evidence until the caller explicitly sets status = "active".

With multiple spokes, each estimator fits its own hub/spoke pair. E2/E3 hub posteriors can differ across spokes. Item summaries retain link_spoke_id and do not average or silently select one hub posterior. Ranks are per fit.

See also

prepare_link_input(), fit_link(), start_link_session(), save_link_session()

Other linking: fit_link(), predict_link(), prepare_link_input(), save_link_session(), summary.pairwiseLLM_link_result()