Skip to content

Longitudinal Retention and Transfer Probe

Long-horizon monitor — instantiates Critical-Window Intervention Timing

Tests, well after the window has closed, whether what was acquired actually persisted and transferred to real-world use — the long-horizon check that separates durable uptake from a gain that faded.

Looking good at closure is not the same as having stuck. The Longitudinal Retention and Transfer Probe re-tests the acquired configuration long after the window has shut, asking the two questions an at-closure check cannot: did the gain persist once support was withdrawn, and does it transfer beyond the context it was trained in? Its defining move is the long horizon paired with an explicit transfer test — it deliberately measures untrained, real-world use months or years later, because a configuration that survives in the clinic but never reaches daily life has not really been acquired. It is the mechanism that catches the window investment that looked like a success the day it ended and quietly unravelled afterward.

Example

A rehabilitation unit delivers intensive upper-limb therapy in the early weeks after a stroke — the window when the brain's spontaneous recovery makes therapy most potent. At discharge the patient scores well on a clinic reaching task; uptake looks good. But the question that matters is whether that gain still helps them button a shirt at home a year on. The Longitudinal Retention and Transfer Probe schedules follow-ups at roughly 3, 6, and 12 months, and — critically — it tests transfer, not just retention: not the trained clinic reach, but untrained everyday tasks.

Its follow-up signal can surface the uncomfortable patterns an exit assessment hides: clinic gains that regress once therapy stops, or gains that hold on the trained task yet never show up in real life. Read against the drift a no-therapy course would have shown, that signal is what tells the program whether the window's investment actually stuck — and feeds back into whether the protocol needs a dedicated stabilization phase.

How it works

Its distinguishing feature is measuring late and testing transfer, not merely retention:

  • Schedule follow-ups past closure. Set measurement points well after the window shuts, spaced to catch slow decay.
  • Test retention and transfer separately. Re-test the acquired configuration on both the trained task (retention) and untrained, real-world contexts (transfer) — the harder, more honest bar.
  • Emit a longitudinal signal. Track the trajectory over time, so decay or failure-to-transfer shows up as a trend, not a single reading.
  • Guard the sample. Chase non-returners, because the people who drop out are rarely the ones doing well.

Tuning parameters

  • Horizon and spacing — how long after closure and how often to re-test. Longer catches slow decay but loses participants.
  • Retention vs. transfer weighting — how much to test the trained task versus untrained generalization. Transfer is the harder and more truthful measure.
  • Attrition handling — how aggressively to pursue non-returners. Lax follow-up lets survivors flatter the result.
  • Signal threshold — how much decay or failure-to-transfer counts as "didn't stick" and triggers a protocol change.
  • Ecological fidelity — a clinic proxy versus real-world measurement; the latter is truer but costlier.

When it helps, and when it misleads

Its strength is that it is the only check that catches a window investment which looked successful at closure but did not last or did not matter in the wild — separating durable acquisition from a transient bump.

Its failure modes are the classic traps of follow-up measurement. Differential attrition makes the survivors look better than the cohort ever did;[1] regression to the mean can masquerade as durable gain when an unusually low starting point simply drifted back up. The classic misuse is measuring only easy retention of the trained task and declaring transfer without ever testing it. The discipline is to test untrained transfer explicitly, chase the non-returners, and read every gain against the drift an untreated course would have shown.

How it implements the components

The probe fills the long-horizon verification components — persistence and reach, not delivery:

  • durability_and_transfer_check — the delayed test of whether the acquired configuration both persisted past closure and generalized to untrained, real-world contexts.
  • longitudinal_follow_up_signal — the repeated long-horizon readout that reveals decay or failure-to-transfer as a trend over time.

It verifies long-term outcomes but does not deliver the immediate stabilization support or apply the at-closure uptake criterion (stabilization_support, uptake_and_stabilization_criterion — Stabilization and Consolidation Schedule); it checks whether that stabilization ultimately held.

References

[1] Survivorship (attrition) bias — when the individuals who remain in a follow-up are systematically different from those who drop out, the measured outcome reflects the survivors rather than the original group, typically overstating success. It is the reason a durability claim is only as good as the fraction of the cohort still being measured. ↩