Project student loan cohorts and model date-aware write-offs #2675

Problem

PR #2650 preserves each existing holder’s recorded student loan plan and the repayment amount passed to the projection helper. That fixes unsupported first-year relabelling and generic cancellation, but projected datasets keep ages fixed and no longer move existing holders across later cohort boundaries. The remaining transitions are Plan 1 to Plan 2 for England and Wales at the 2012 intake and Plan 2 to Plan 5 for England at the 2023 intake. This issue is the single successor for both remaining tasks from #2636: future cohort transitions and exact date-aware, plan-specific write-offs. PR #2650 has merged; it does not implement either remaining task.

Vahid Ahmadi’s 25 October 2026 audit used enhanced_frs_2024_25 at revision 2526edbb and the intake-at-18 proxy. It reported 0.53m, 1.57m and 2.99m England Plan 2 holders reaching the 2023 intake in 2026, 2028 and 2031 respectively, out of 7.14m recorded Plan 2 holders. For England and Wales Plan 1 holders reaching the 2012 intake, it reported none in 2025, 0.30m in 2028 and 0.68m in 2031, out of 1.52m recorded Plan 1 holders. These approximate counts describe that reviewer’s dataset audit; this issue does not provide an independent population or revenue estimate. The reviewer accepted documenting this gap and opening a follow-up.

Scope and source requirements

Treat forward cohort projection and legal cancellation as separate workstreams within this issue. A cohort replacement in a fixed-age projected population does not establish that an individual borrower’s loan has reached its legal write-off date.

Define a forward cohort projection that distinguishes actual borrower facts from modelling proxies. The current inputs lack funding agency, course start, first loan payment and first April due to repay. An intake-at-18 assumption misses mature starters, and present household region misses cross-border funding. Decide which verified producer fields or explicit, documented fallback assumptions the projection requires before changing assignments.

Use the official repayment-plan rules and cancellation rules to establish funding nation, relevant dates and any cancellation deadlines. Treat a cohort transition and a legal write-off as separate operations. A generic age or course-start-plus-term rule cannot establish an existing holder’s legal cancellation date. Exact date-aware write-off is tracked here alongside cohort projection. Establish its own sourced borrower facts and validation before implementation; do not use the cohort proxy to infer a legal cancellation date.

Acceptance criteria

• Preserve observed base-year plans and avoid retrospective or backwards relabelling from missing facts. Define how contradictory recorded plans and proxy dates behave.
• Implement and test the selected forward-boundary approach for the 2012 England/Wales and 2023 England transitions, including a projection spanning both boundaries.
• Retain Plan 4, Postgraduate and existing Plan 5 categories where the relevant funding/date facts do not justify a change; cover Northern Ireland, Scotland, unknown region, mature starters and cross-border movers explicitly.
• Preserve repayment values passed to uprate_student_loan_plans, separate average-earnings uprating and the existing reported-versus-modelled repayment precedence tracked in EFRS student_loan_repayments input overrides modelled repayment formula #1764. Any repayment-precedence change needs separate scope and evidence.
• Write meaningful failing synthetic regressions before the implementation, with independently sourced or explicitly assumed expectations. Cover before/at/after each boundary, base and past years, protected categories, older holders without proven cancellation facts, unequal stored repayments, repeated projections and row-order invariance.
• Establish and implement plan-specific cancellation using the funding, payment and repayment-due dates required by the official rules. Test before, at and after each applicable deadline. Specify and test the fallback when required facts are missing, including older holders whose cancellation date cannot be established. Do not reintroduce the removed generic write-off.
• Document producer fields, missing-data behaviour, fixed-age/intake assumptions and remaining legal limitations separately for cohort projection and cancellation.
• Coordinate plan-mix remeasurement with the existing owners of Microcosm #1095 and the known-defect/SLC dependency in Microcosm #1383. This issue does not transfer their dataset, calibration or producer work. Keep exact dataset revision, model head, weights and coverage labels on any future population evidence.

A PR implementing only one workstream must identify the remaining work and keep this issue open unless that work is explicitly transferred to a linked successor.

This consolidation updates issue tracking only. No implementation, producer migration, population rerun or encoder run has been performed for this follow-up.