2026-08-08 — ระบบบ้าน polish: the student card, รุ่น, สาย-first อาจารย์, CSV

Archived out of STATE.md the day it shipped. Migrations 0123 and 0124, both applied and proved live. Commits: git log --oneline around c8949d1.

SHIPPED 2026-08-08 — ระบบบ้าน: the card, รุ่น, อาจารย์-per-สาย, CSV hardening

The import bug worth remembering: an upsert that sends every column wipes the ones the file did not have. A corrected name-list omitting sai would have cleared ~1,800 สายรหัส and every house placement, while the preview said "แก้ไข 1,800" — both halves computed from the same wrong column set, so they agreed. parseStudentsCsv now returns presentColumns and both the payload and the diff are scoped to it. Write-up + 3 tests in docs/mistakes/app-state.md.

Import and export now share ONE vocabulary — the table's column names, with the spreadsheet spellings (sai, nickname_th, ชื่อ, อีเมล) as aliases. The generated nickname left the export: the importer read it back as nickname_imported, so a round trip promoted a student's own ชื่อเล่น into the university's column. The handover spec (docs/house-data-spec-th.md) was rewritten around "3 things must be right, we normalise the rest": kkumail, split name columns, UTF-8. Leading-zero สาย is no longer fatal — Excel strips leading zeros only, so left-padding is lossless and the house (last digit) is invariant; it is a loud warning now, not a refusal. New fatals: non-UTF-8 (with an .xlsx-specific message), and a single combined "ชื่อ-สกุล" column.

Admin: click a สาย to manage its อาจารย์ — search by สาย or อาจารย์, filter to the ones with none. Same sai_advisors rows as the อาจารย์-first editor.

sai_locked explained in the UI — it bars ONE student from self-editing their สาย even while the global switch is open; they can still file a request.

Earlier the same day — the card, รุ่น instead of ชั้นปี (0123–0124)

• "ต้องกดหลายครั้งถึงจะขึ้น" is fixed. renderMyHouse() added a delegated listener to #homeMyHouse — a node that survives every re-paint — and the handler toggled d-none. Every auth event added one more listener, so the panel opened only on odd-numbered paints. Listeners now go on the nodes each paint creates, and panel state is ONE variable. Write-up + src/js/house/my-house.test.js pin both shapes.
• The card now reads like ตำแหน่งของฉันในทีม SAMO — same shell, same label→value list (ชื่อ-สกุล · ชื่อเล่น · รหัสนักศึกษา · รุ่น · สาขา · สายรหัส · บ้าน · KKU Mail). It REUSES my-seat.css on purpose; my-house.css holds only the crest, accent, advisors and roster. แจ้งข้อมูลไม่ถูกต้อง is an in-card form now, not two prompt()s (which Chrome can suppress silently).
• ชั้นปี is gone from ระบบบ้าน; รุ่น (MD50) replaces it. cohort − 2515, off the รหัสนักศึกษา. No clock, no year_override, no academic_year setting. student_year() dropped in SQL and its JS mirror, same commit.
• ยืนยันข้อมูล removed — a timestamp nothing branched on.
• เพื่อนร่วมบ้าน removed, อาจารย์ในบ้าน added (0124). Students do not need a list of classmates; they need to know who the อาจารย์ are. The card now shows their own สาย's อาจารย์ and then every other อาจารย์ in the house tagged with their สาย (from house_advisors, already in the payload). get_house_roster() is DROPPED, not just unlinked — a read path that publishes 1,800 students to any signed-in caller is removed at the database, not in the one button that called it. ระบบบ้าน names no student to any other student.
• 0123 APPLIED. Live proofs, both directions: payload keys carry cohort_year and no year/verified_at; a patch sending {"year_override":3,"verify":true,"nickname_self":"…"} wrote the nickname and ignored the other two (probe rolled back via raise).
• students.year_override, students.verified_at, students.is_listed, house_settings.academic_year, house_settings.roster_visible are vestigial, not dropped — nothing reads or writes them, and they carry comment on column saying so. Drop when the real data has landed if still empty.

SHIPPED 2026-08-08 — ระบบบ้าน polish (0123 + 0124)

Applied and proved live. Full detail + every proof: docs/state-archive/2026-08-08-house-polish.md. Headlines:

• บ้านของฉัน reads like ตำแหน่งของฉัน — one label→value list, same stylesheet. The "must click many times" bug was one delegated listener added to the surviving host per re-render, plus a toggle that read its own class.
• รุ่น (MD50 = cohort − 2515), never ชั้นปี. No clock, no override; SQL's student_year() and its JS mirror both deleted. No ยืนยันข้อมูล.
• ระบบบ้าน publishes อาจารย์, never students. get_house_roster() DROPPED; the card lists the อาจารย์ of every สาย in the house instead.
• An import now writes only the columns its file carried. Sending every column would have cleared ~1,800 สายรหัส from a file that merely omitted them.
• One CSV vocabulary (the table's), spreadsheet spellings as aliases. Leading-zero สาย is recoverable, so it warns instead of refusing; non-UTF-8 and a combined "ชื่อ-สกุล" column are the new fatals.
• Admin: click a สาย to manage its อาจารย์.
• 0125 — a student owns their identity, and not their สาย. They self-edit ชื่อ · นามสกุล · ชื่อเล่น · รหัสนักศึกษา · สาขา (a chooser over team_majors, refused server-side if off-list). สายรหัส is NOT self-editable at any level — it decides the house, so the route is a request an admin approves. Four of those are import-owned columns, so students.self_edited text[] + a BEFORE UPDATE trigger preserve them on any write stamping a new last_import_batch: admin > student > import, enforced on the TABLE. Proved live both ways (self-edit sticks through a simulated import; cohort_year still updates; an off-list สาขา raises). sai_self_edit_open / sai_locked / sai_self_edits are now vestigial — ระบบบ้าน has no admin settings left.
• 0126 — a student row can arrive with NO NAME. first_name_th is nullable and full_name is NULL (not '') when empty, so the ask to Data Analytics is now 4 columns: kkumail, student_id, sai, major — 1,800 names never leave their department. The student types their own (0125). A COMBINED "ชื่อ-สกุล" column is still refused: no name column names nobody, one combined column renames everybody whose surname has a space. Also fixed: a self-edited duplicate รหัสนักศึกษา hit students_sid_uniq and surfaced a raw 23505 — now a Thai sentence, pre-checked AND caught in an exception handler for the race.
• KKU SSO probed live (UAT) — it CAN supply the identity. One real student login: user.profile returns studentId + studentCode (653070317-0, already our canonical form), type: STUDENT, Thai + English names and facultyName — none of studentId/studentCode is in the vendor manual, which also documents an immutableId the API does not return and calls mailemail. It cannot supply สาขา (levelName is the degree level) or ชื่อเล่น. UAT is backed by the REAL directory, so this transfers to prod. Blocking: our registration is UAT-only (prod login answers "Cannot find the CREDENTIAL"), so a production app must be requested; and citizenId comes back on both calls and must never be stored. Probe: node tools/sso-probe.mjs.
• KKU SSO assessed, not built — docs/KKU-SSO.md. It is a login improvement, NOT a data source: no roster endpoint, no สายรหัส, no สาขา, and it returns citizenId we must never store. The CSV is still required. Credentials in .env.local (KKU_SSO_*); manual docs/KKU-SSO-MANUAL.md.

The live proofs, both directions

• get_my_student_record() payload keys: carries cohort_year, and no year / year_override / verified_at / roster_visible.
• A patch sending {"year_override":3,"verify":true,"nickname_self":"…"} wrote the nickname and ignored the other two — probe rolled back by ending the DO block in a raise, so production was never changed.
• get_house_roster is gone from pg_proc while get_house_summary is still there (the positive control that proves the sweep can find things).
• house_advisors returns one row per (อาจารย์, สาย) for the caller's house.

0125–0126 — the self-edit boundary, and the nameless row

• 0125: a student owns ชื่อ · นามสกุล · ชื่อเล่น · รหัสนักศึกษา · สาขา, and their สายรหัส not at all. Four of those are import-owned columns, so students.self_edited text[] + a BEFORE UPDATE trigger preserve them on any write stamping a new last_import_batch (admin > student > import). สาขา is validated against team_majors, whose write gate widened to house. Consequence: sai_self_edit_open / sai_locked / sai_self_edits are vestigial and ระบบบ้าน has no admin settings at all any more.
• 0126: first_name_th nullable, full_name NULL rather than '' when empty — the import file may deliberately carry no names. Recommended ask is now kkumail, student_id, sai, major. Also: a self-edited duplicate รหัสนักศึกษา used to surface a raw 23505 from students_sid_uniq; now a Thai message, pre-checked and caught in an exception handler for the race.

Live proofs (all rolled back via a terminal raise): a self-edit survives a simulated import while cohort_year still updates; an off-list สาขา raises; a patch carrying sai_code changes nothing; a nameless row inserts and reads back full_name = NULL; a duplicate รหัส and a blanked name both raise in Thai.

Vestigial columns left in place

students.year_override, students.verified_at, students.is_listed, house_settings.academic_year, house_settings.roster_visible. Nothing reads or writes any of them; each carries a comment on column saying so. Dropping a column is the one step re-running a migration cannot undo, so it waits until the real data has landed and they are confirmed empty.