Render the offline report in Spanish behind the portfolio's EN/ES catalog gates

[ChelseaKR]

What

exitdrill report receipt.json --lang es --out informe.html renders the offline HTML report in Spanish from a reviewed gettext catalog, with English as the default and the fallback; the result-state names, dimension questions, and every fixed limitation sentence are message IDs whose Spanish text is human-reviewed and pinned byte-for-byte, and a report with any untranslated cover string renders the English string and marks the page "partially translated" rather than dropping the sentence. explain (proposed separately) would share the catalog.

Why it matters

docs/I18N.md declares N/A on the grounds that the report is operator evidence, not a public interface — but the report is precisely the artifact an operator hands to a director, a board, or a funder, and civic organisations in this portfolio's audience include Spanish-speaking ones. This proposal deliberately reopens that declaration, as the declaration itself says it should when a delivery commitment to a bilingual audience appears. The constraint is the project's usual one: the limitation texts are load-bearing claims, so they are never machine-translated silently, and the report-content tests must assert the claims in whichever language rendered. Inside the freeze: strings only, no new evidence or contract.

Scope

• Wrap the report's reader-facing strings in a lookup; extraction template committed; --lang flag; catalog under src/exitdrill/locales/.
• The disclosure and offline/script-free gates run in both languages; a key-parity gate fails on a missing cover key.
• docs/I18N.md updated to "Applies: report surface only", with reviewer and date.

Out of scope

• CLI output, JSON documents, docs, or the canary READMEs; locales beyond es.

Done when

• --lang es on the demo receipt yields a page whose every limitation sentence matches the reviewed catalog exactly, and the English page is byte-identical to today's.
• A catalog missing one non-cover key renders the English fallback plus the marker; a missing cover key fails the build.
• The Spanish page passes the same no-script, no-subresource, and no-record-value assertions.
• --lang xx is a usage error listing the available locales.

Pointers

• src/exitdrill/report.py, tests/test_report.py, tests/test_disclosure.py, docs/I18N.md
• Portfolio i18n direction (EN/ES) as adopted in sibling repos such as family-greenhouse

Proposed with AI assistance.

[ChelseaKR]

Triaged 2026-09-06 against origin/main at 857a411. Not built, and deliberately so: two of its four acceptance criteria cannot be met by anyone who is not a Spanish reviewer, and the issue's own framing asks for a declaration change that is the maintainer's to make.

Nothing exists yet. src/exitdrill/locales/ is absent, report.py has no lookup or --lang, and git grep -n 'gettext\|--lang\|locales' over src/ returns nothing. No open pull request, no branch, no git log --all --grep '#135' hit. So the five-check test comes back clean: this is genuinely unbuilt, not already-shipped.

Two blockers, and they are different in kind.

1. A policy decision the issue names outright. docs/I18N.md declares i18n status: N/A and states the exact condition for flipping: "a public-facing surface, a hosted service, or a delivery commitment to a bilingual audience". The issue argues the third has arrived because the report is what an operator hands to a director or a funder. That argument is reasonable and it is not mine to accept — it changes what the project declares about itself, and the Scope bullet asks to rewrite that declaration to "Applies: report surface only, with reviewer and date". A reviewer and a date are two facts I would have to invent.

2. A named human reviewer. The issue is explicit that the limitation sentences are load-bearing claims and "are never machine-translated silently", and the first done-when is that every limitation sentence "matches the reviewed catalog exactly". An agent-produced es.po would satisfy the mechanical parity gate and fail the only thing the criterion is actually asking about. Shipping one would put unreviewed Spanish behind a green check on a page an operator hands to a funder — the precise inversion of what this repository's gates are for.

What is separable, if the declaration is flipped. Roughly two thirds of the issue is mechanical and reviewer-independent, and could land ahead of any Spanish text:

• wrapping the report's reader-facing strings in a lookup and committing the extraction template (the message IDs, not the translations);
• --lang with --lang xx as a usage error listing available locales;
• the key-parity gate, the cover-key/non-cover-key split, and the "partially translated" marker with English fallback;
• running the existing offline, no-script, no-subresource and no-record-value assertions over whatever page --lang renders, so the disclosure guarantees are not language-specific.

With es present but empty, all four of those are testable today: every string renders English, the marker appears, and a deliberately-missing cover key fails the build. That is a real negative control and it needs no translator. I have not opened it, because building the apparatus commits the declaration change in practice while leaving the record saying N/A, and because a catalog scaffold sitting in the tree is an invitation for someone to fill it in unreviewed. Say the word and it is a contained piece of work.

Prepared with AI assistance; reviewed before submission.