Korean applications get an evidence portfolio, not a longer resume

Context

The resume system had been built across roughly fifteen local commits — case films, resumeData.ts / resumeDataKo.ts, the ATS PDF generator, a Notion-style /ko/resume, and a monochrome /ko/resume/toss. None of it had ever been pushed. Production was serving only the old public/resume.pdf; curl -sL returned 404 for /portfolio, /ko/portfolio, /resume, /ko/resume and /ko/resume/toss.

The open question was what a Korean applicant actually submits. Korean hiring at service companies expects a compact document, while the strongest material — five animated case films of SK AX work — does not survive being flattened into a PDF.

Goals (ranked)

1. A submittable Korean document that reads in ten seconds and does not depend on assets that cannot be deployed.
2. Real evidence reachable from that document without inflating it.
3. One source of truth per language, so a number changed once does not disagree across surfaces.
4. No fabrication — status stated as-is, no invented metrics.

Options considered

A. One long Korean resume (Notion-style /ko/resume) as the submission. Pros: already built; toggles embed case-film GIFs inline; single URL to send. Cons: depends on public/films/*.gif, which is gitignored (.gitignore:45, added in 28f3b3a) and therefore 404s in production — five broken images on the primary submission surface. A scrolling web page is also not what a Korean submission portal accepts. Cost: zero. Reversibility: two-way. Evidence: git ls-files public/films → 0 files.

B. Reuse the English /portfolio for Korean applicants. Pros: no new route; the five case films already exist there. Cons: a Korean recruiter lands on English prose. The English page is a film archive, not an evidence page — it leads with mechanism rather than result. Cost: zero. Reversibility: two-way. Evidence: existing /portfolio copy is English-only.

C. Fixed 2-page PDF for submission + /ko/portfolio as the evidence surface. — chosen Pros: the PDF uses tracked JPGs only, so it never depends on ignored assets; the web page carries the animation the PDF cannot. Both render from lib/resumeDataKo.ts. Cons: two surfaces to keep in sync; a second portfolio route to maintain. Cost: one new route + one shared component. Reversibility: two-way. Evidence: output/pdf/유대선_이력서_서비스기업_2p.pdf renders; build prerenders ko/portfolio.html.

D. Do nothing — keep serving the old public/resume.pdf. Pros: no work. Cons: the live PDF is from June, contains a phone number, and links nowhere. Every new route stays 404, so no new submission can reference the site at all. Cost: zero. Reversibility: two-way.

Chosen + why

Option C. The deciding constraint was asset deployability: the one option that put the strongest material on the submission path (A) was also the only one that would ship broken. Splitting the surfaces lets the PDF stay dependency-free while the web page carries the films.

components/casefilm/portfolio-showcase.tsx is shared: English renders all five case films, Korean renders three (ERP↔MES, row-level lock, HTTP 413) plus links for the rest. One component, two locale-specific case lists.

Trade-off accepted

lib/resumeData.ts and lib/resumeDataKo.ts are independent objects, not a translation layer. Changing a figure requires editing both and cross-checking. This was accepted deliberately — the two languages emphasise different achievements, so a shared schema would have forced artificial symmetry.

Pre-mortem (six months out)

1. The two resume data files drift and a recruiter sees different numbers on the Korean page and the English PDF.
2. /ko/resume stays orphaned with five broken GIFs and someone finds it via search.
3. Neither portfolio route is in app/sitemap.ts, so the evidence page is never indexed and only ever reached by a link pasted into an application.

Second-order effects

Enables: submitting the new Korean PDF, which was explicitly blocked on the portfolio being publicly reachable. Forecloses: nothing — the old PDF stays served until a separate decision.

Decision criteria to flip

If a Korean submission portal rejects the 2-page PDF, or if GIF hosting is solved (Vercel Blob or committing the ~8.8MB of GIFs), option A becomes viable again.

Success measure

The three submission-path routes return 200 in production and the PDF's internal links resolve to real pages rather than 404s.

Verified by

• curl -sL -o /dev/null -w "%{http_code}" before the deploy: /portfolio 404, /ko/portfolio 404, /ko/resume/toss 404, / 200. The apex issues a 308 to www, so -L is required or every route reports 308 and hides its real status.
• Production build of the working tree, run in an isolated copy with symlinked node_modules to avoid touching a shared .next: exit 0. app-paths-manifest.json contains /(public)/ko/portfolio/page and /(public)/portfolio/page; both portfolio.html and ko/portfolio.html were prerendered.
• tsc --noEmit and next lint exit 0. No lib/source.ts or lib/storage.ts in any "use client" import graph.
• No credentials, tokens, phone numbers, or addresses in the pushed range. The phone number stays in the RESUME_PHONE env var and never enters the repo.

Known gaps at deploy time

• /portfolio and /ko/portfolio are absent from app/sitemap.ts.
• app/(public)/ko/resume/page.tsx:194 and lib/resumeDataKo.ts:65 still point Korean readers at the English /portfolio; the latter also hardcodes an absolute https://daeseon.ai origin, which contradicts 7be80a9's move to relative links.
• /ko/resume renders five /films/*.gif that are gitignored and will 404. The route is effectively orphaned — the header sends Korean readers to /ko/resume/toss.
• The old public/resume.pdf is still served and still contains a phone number.

Discussion artifacts

docs/HANDOFF-resume-portfolio.md — sections 2 (locked decisions) and 3.5 (deferred GIF delivery).

Commit

0d5427d — feat(resume): 한국어 증거 포트폴리오 /ko/portfolio + 본선 2쪽 이력서 — 첫 공개 배포