UK wave 1: SLC student loan forecasts, student support and repayments (#132)

[juaristi22]

The wave-1 remainder, unblocked by #136's academic_year merge — with #137 (region×age population; ISC landed via #141 batch 4), this completes #132's package inventory.

Closes #132.

Lane: ledger-source-ingestor — judges ledger-source-fidelity + ledger-boundary.

What

Six packages, 199 facts, new slc source directory — and the first real consumers of two fresh mechanisms:

package	facts	mechanics
slc-student-loan-borrower-forecasts-england-2025	116	EES permalink Table 6a on the ees_permalink_table_html lane: borrowers liable to repay / earning above threshold, by product group × plan × status, AY2024/25–2029/30. First academic_year consumer: fact periods store the opening year while the EES value columns are END-year named (value_2025 = AY 2024/25) — the exact leak the ALLOWED_PERIOD_TYPES caution documents. All values are forecasts → model_output + assertion: source_projection, so profiles opt in via assertion_policy (#154). Null publisher cells skipped per year, not zero-filled.
slc-student-support-england-2025	72	slcsp052025 Tables 3A (maintenance-loan Grand-total recipients ×1e3 and amounts ×£1m) + 4C(i) (ADG + PLA recipients and amounts), AY2013/14–2024/25. 'Grand total ' carries a trailing space in the workbook — guards pin the exact text. Cohort rows and the other 4C products are named backfill exclusions.
slc-student-loan-repayments-england-2025	8	The corrected slcsp012025 workbook uk-data cites, Table 1A "Net repayments posted during the financial year", FY2024-25 block: all seven plan columns + Total HE at publisher £m precision, fiscal_year 2024. Earlier year blocks → backfill lane.
slc-student-loan-repayments-{scotland,wales,northern-ireland}-2025	1+1+1	The country publications are HTML-only; archived pages parsed on the document-numbers lane (£203.3m / £229.1m / £181.7m, FY2024-25), sentence + element-index pinned.

Fidelity vs policyengine-uk-data (every delta enumerated)

• Exact matches: all EES Higher-education-total snapshot values (e.g. plan_2 above-threshold AY2024/25 = 3,985,000); Table 3A/4C counts; England repayments total 5,018,231,834.95 and plan 1; all three country figures.
• uk-data derivations kept populace-side: its plan_2 (2,778,253,361.64) and postgraduate (346,409,713.95) England targets are FT+PT and Masters+Doctoral sums of the publisher's columns — the columns land here, the sums reconcile to the penny; its END-year calendar keying of academic years (its 2025 = AY 2024/25) is a timing declaration over these facts; its whole-pound roundings of published £m values (e.g. ADG 2013/14 £33,050,090.68 → 33,050,091) are transcription artifacts.
• uk-data's EES scrape is runtime-only with hardcoded TESTING=1 snapshots; the archived permalink page is now the pinned artifact.

Gates

• validate-package + build-suite + agent_acceptance green for all six; spot checks byte-exact against artifact cells and the uk-data snapshot table.
• Bundle expectations at post-publication truth (39,418 facts / 66 packages = main + this branch), now proven against live R2 storage: the merged-bundle consumer-contract test passes with every fact carrying a real raw_r2_uri.
• Draft because the consumer export awaits publish-raw Published: credentialed chronicle publish-raw ran on this branch (post-Pin Cloudflare account for R2 publishing and fix publish-raw failures #148/Fill missing source_id/package_id so publish-raw covers every manifest #149) — all six SLC artifacts uploaded to ledger-raw under content-addressed keys and the manifests commit their real storage: blocks. No designed-red remainder. Known conflict-by-design with UK wave 1: ONS population source packages — mid-year estimates 2023+2024 and 2024-based projections (#132) #137/UK wave 2: national fiscal/benefit source packages (#133) #139/UK wave 2 continuation: DWP, VOA, scotgov, DfT and ONS national packages (#133) #141 on SOURCE_PACKAGE_ALIASES + the bundle constants stands; whichever merges second re-measures.

🤖 Generated with Claude Code

[vahid-ahmadi]

Six SLC packages covering all four nations with a registry-only code change — clean. Splitting England (which has the full forecast + support tables) from the three smaller repayment packages matches how SLC actually publishes.

1. academic_year looks like a new period type — is the vocabulary declared anywhere?

This PR introduces academic_year:2013 through academic_year:2029, which I don't see elsewhere in the existing by_period counts. If period types are a closed vocabulary validated at load, that needs a registration; if they're open, then the consumer side needs to know how academic_year:2024 aligns to a tax_year or calendar_year base period, or these facts will be unresolvable by any profile that asks for a normal year.

Related: academic_year:2025 through 2029 are borrower forecasts. #154 (open) adds the assertion: source_projection axis precisely for this, with observed_only as the default resolution policy. Without that field these forecast years look identical to the observed 2013-2024 history, and the first consumer to ask for "latest" quietly gets a projection. Worth sequencing against #154 — the SLC forecast table is a textbook case for it.

2. The golden-count test will collide with the other three open UK waves.

#137, #139, #141 and #155 all branch from the same base and rewrite the absolute totals in tests/test_chronicle_bundle.py to different values:

PR	fact_count	source_package_count	period_count
#137	42895	64	19
#139	40131	69	61
#155	39418	66	33

Each is correct in isolation and stale the moment a sibling lands. That's four rebases with hand-recomputed constants, and resolving the conflict by picking a side gives you a green test that has stopped checking anything. Worth filing as its own issue — per-package delta assertions would compose where one whole-bundle snapshot can't.

3. Minor

• The diff includes a couple of unrelated re-orderings in the by_source dict (hmrc/ici swapped, onem_rva_unemployment/ons swapped) that aren't caused by this PR's data. Harmless in a dict literal, but they add conflict surface against the three sibling branches that touch the same lines. Reverting them would make the rebase cheaper.
• db/data/slc/student_loan_borrower_forecasts_england_2025/ees-table-6a-permalink.html is committed as a single line (+1). If that's a minified permalink response it's fine — just checking it isn't a truncated download.
• Three of the four nations land as .html scrapes while England is .xlsx; if SLC publishes equivalent workbooks for Scotland/Wales/NI, those would be more stable inputs than parsed HTML.