One field catalogue for every fund page

[taavipertman-glitch]

Summary

Replaces #69 and #86, and carries everything they had that master lacks, so both can be closed. Step 1 of TODO — Fund page document publishing.md (private tuleva repo, work/investeerimistegevus/docs/, raised as TulevaEE/tuleva#511).

All four fund pages now take their documents — and the CO2 intensity figure published beside them — from one catalogue in helpers/acf/fund-disclosures.php, with a scope table declaring per page template whether each is required, optional, or absent.

Field names are identical everywhere a field exists, so nothing downstream branches per fund: fund → page slug, document → field name. What differs per fund is whether the field exists at all.

Why a rebuild rather than a rebase of #69

#69 predates ~19 commits of master and moved one field. Master has since added upcoming prospectus, terms and NAV procedure rows to all four pages — every one of them as a hardcoded template URL, including on TKF100, the fund that is supposed to be ACF-driven (#88, #93, c9bcf9b). Wiring a new ACF field by hand was slower than editing the template, so the template kept winning. Generating the field set from a catalogue removes that incentive.

Absent is a real state

Two states are not enough. if ($url) cannot tell "no CO2 intensity is calculated for TKF100" apart from "TUK75's Põhiteave link silently vanished".

	wp-admin	REST	Page	Gap check
absent	no field	key absent	nothing	silent
optional	field	key present	renders when set	silent
required	field	key present	renders when set	reported by tuleva_fund_disclosures_missing()

TKF100 is a UCITS fund: it carries investor_rights_file and its own nav_procedure_file. The pension funds carry neither and share one NAV procedure document. The pension funds carry fund_co2_intensity; TKF100 does not. A field outside a fund's scope returns '' and never reaches the fallback — an empty pension field must not borrow TKF100's document. There are tests for exactly that.

Turning a field off for one fund is one line in the scope table. Each fund has its own page template and ACF binds groups by template, so the granularity is per fund, not per fund type. The savings template still carries the CO2 markup, so publishing a TKF100 figure one day is a scope-table edit and nothing else.

What came from #86

The CO2 intensity figure, as a catalogue entry rather than its own Sustainability field group. It is not a document, but it has the same problem — published on the fund page, updated on a cadence, not the same set for every fund — so the catalogue is renamed from documents to disclosures rather than growing a member its name excludes.

Text, not number: 133.80 must not render as 133.8. The three literals were inconsistent about this (83.68 and 83.73 bare floats, '133.80' quoted); all three are now strings through the same path.

group_fund_savings_details does not get show_in_rest as it did in #86 — with CO2 in the catalogue, nothing writes to that group over REST. Exposing the fee fields there is a separate decision, and FEE-UPDATES.md is a separate process.

What came from #69

The generate_report_link() rewrite and its 171 lines of tests. Output is identical for the three live report URLs — verified by running both versions — but the function no longer emits unescaped HTML, keeps an absolute URL's own host instead of discarding it and re-prefixing get_site_url(), reads the reporting period from the YYYY-MM in the filename rather than the upload folder, and no longer produces labels like 12.-1. from intval() on an unmatched preg_match. All four matter more now that the URL arrives from an ACF field: FundReportMapping.buildPdfFilename() always embeds the period, while the upload folder is just the month the file happened to be uploaded, and those diverge on a re-upload — see 2bf6f27, where exactly that broke a URL.

Bugs this fixes

• show_in_rest was missing on every field group, so ACF dropped the acf key from every REST write and returned 200 anyway. GET /wp-json/wp/v2/pages/35292 and /17533 both return "acf": [] on production right now — TKF100's automated path has never worked, and neither scripts/update_acf.py nor onboarding-service's WordPressMediaClient would have noticed.
• nav_procedure_upcoming_file was read but registered nowhere. fund-savings-details.php read it; no field group defined it. It always returned null, so that row could only ever come from the hardcoded URL.
• Two conventions for the same effective date — a hardcoded string on the pension pages, the hand-typed media library title on TKF100. Both now read …alates-DD.MM.YYYY from the filename, which the publishing script controls. TUK75-tingimused-kehtivad-alates-01.01.2027-1.pdf (a media re-upload suffix) is in the test cases.

No visible change

Rendered markup is identical on all four templates while the fields are empty, which they are in production. Verified by rendering each component with WordPress stubbed out, from master and from this branch, and comparing whitespace-normalised output — identical for all four, CO2 figures included. Rendering again with fields set shows the ACF value winning, the effective date following the new filename, and an out-of-scope field staying invisible.

Field keys TKF100's values are stored under are pinned in the catalogue, so moving those fields into the generated group does not orphan them. A test asserts each of the ten.

Notes for review

• tuleva_fund_disclosures_missing() has no caller yet — it is what makes required mean something, and the admin notice in step 5 of the plan is its consumer. It is tested.
• The three pension templates still render the four firm-wide documents (sustainability, adverse impacts, remuneration, pension NAV procedure) from helpers/extras.php constants. Those are one document for all four funds and move to the options page in step 3.
• Fallback URLs stay in the templates until the fields are populated (step 2) and verified. Once a page's field is set, editing the literal changes nothing visible — CLAUDE.md now says so in as many words.

🤖 Generated with Claude Code

[taavipertman-glitch]

Codex review found a live regression that my own verification was structurally unable to catch. Fixed in aade1fd.

The regression. These templates use a non-breaking space inside the gettext msgid (in Estonian). Retyping those rows replaced it with an ordinary space in five strings per pension template, so the lookup missed. In et.po that msgid translates to a single space — the suffix is meant to be invisible in Estonian — so the pages would have rendered a literal "(in Estonian)" and "(in Estonian, effective from 01.01.2027)" on all three pension fund pages.

Why the render comparison missed it. I compared master's and this branch's rendered markup with whitespace normalised via \s. Python's \s matches U+00A0, so the single character that changed was precisely the one being collapsed. The comparison now folds ASCII whitespace only and counts non-breaking spaces on both sides: 11 per pension template, matching master.

Three latent faults, also fixed — each a case where an empty value produced markup instead of nothing:

• paired prospectus/terms rows printed both anchors behind one || guard, so a missing half renders href="" once the fallbacks come out
• the report row and CO2 block rendered unconditionally, so an empty value would link to the site root and print a figure with no number
• the effective date was read from the filename with no validation. The pension funds' NAV procedure is published as ...kehtib-alates-02.03.3026.pdf — a typo for 2026, harmless only while nothing renders the date. Now range- and calendar-checked, falling back when implausible.

Verified by rendering a pension template with every field out of scope: no empty hrefs, no report row, no CO2 block, only the five firm-wide rows. 69 tests pass; all four pages still render identically to master.

Codex's remaining finding is the one already tracked as step 2 of the plan: no ACF assertion here has been observed against a real WordPress — pinned keys preserving values, location rules binding, show_in_rest accepting the write are all reasoned, not watched. Codex could not check it either; ACF Pro is not in the checkout.