**CARTE bug** (user-reported on iPhone): a CARTE gamer who contributed their deck to multiple gate slots could fill ≥1 role for ≥1 seat, navigate away (BYE → dashboard, CONT GAME → return, etc.), come back to the room — and the JS guard on .card-stack would wrongly fire "Equip card deck before Role select" + block further role picks, even though the deck was demonstrably in play on existing seats. Symmetric for the "stay in room during Role Select" variant the user thought we'd squashed before (the prior fix was 759ce8d for the multi-slot SELECT path, but the room VIEW context never got the same treatment) ; **root cause**: `select_role()` at epic/views.py:619-621 clears `user.equipped_deck` after the first role pick ("deck committed to room"). The room view's role-select context at epic/views.py:286 then passes `equipped_deck_id = user.equipped_deck_id` to the template — which is now None — and the template renders `data-equipped-deck=""` → JS guard at role-select.js:165 sees the empty string and fires the "no deck" warning. The deck IS in play; the context just isn't recognizing seat-level deck assignment as a deck source ; **fix** (epic/views.py:286ish): when `user.equipped_deck_id` is None, fall back to the deck_variant of any of the user's seats in this room (order_by slot_number for determinism). The guard now sees a non-empty id and the fan opens. Storage-side unchanged — seat.deck_variant remains the canonical "this deck is in play on this seat" signal, and the user's deck-third contribution per role (PC=levity brands+crowns / NC=levity trumps / SC=levity grails+blades / AC=gravity grails+blades / EC=gravity trumps / BC=gravity brands+crowns) flows from existing `select_role` logic that inherits deck_variant from the first seat ; **TDD trail** — 2 new ITs in `SelectRoleMultiSeatTest` (apps.epic.tests.integrated.test_views): T1 pins the context (`response.context["equipped_deck_id"]` equals the existing seat's deck_variant_id after `user.equipped_deck` clears); T2 pins the template (rendered `data-equipped-deck="<id>"` not `""`). Initial reds — `None != 2` + `data-equipped-deck=""` substring assertion. Fix lands both green ; **bundled: My Sign applet rename** — user clarified naming convention 2026-05-18: **applets** use the "My X" prefix (My Sign, My Sea, My Posts), **standalone pages** use the "Game/Dash/Bill X" prefix (Game Sign page, Game Sea page, Game Kit page). Sprint 4a's initial migration set the applet name to "Game Sign" — corrected after the user saw the gear-menu toggle list reading the wrong word. Applet template header link "Game Sign" → "My Sign" (user-edited); migration 0010 added to update the Applet row's `name` in already-migrated DBs (dev + staging); applets/0009 frontmatter + defaults updated to "My Sign" in case of a fresh migrate-from-zero; test seed helpers in billboard test_views.py + functional_tests/test_bill_my_sign.py updated to "My Sign". Slug stays `my-sign` (URL + selectors stable) ; **bundled: rootvars.scss** — user-modified mid-session (pre-staged) ; 1022 IT/UT green in 46s — no regressions; 4 ITs in SelectRoleMultiSeatTest green (2 pre-existing CARTE multi-seat ITs + 2 new return-trip context ITs)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User-reported gap on /billboard/my-sign/ — admin user's only deck was in-use as `TableSeat.deck_variant` in another room (Wonderbeard) → `equipped_deck` cleared → previous my-sign template showed "Equip a card deck first…" w. no actionable next step. User scoped fix: don't force equip, just nudge via a Brief banner "Look!—no deck is equipped. Navigate to the Game Kit to equip one (FYI) or (NVM) proceed with the Earthman [Shabby Cardstock] deck.", title "Default deck warning". NVM dismisses + picker proceeds against an Earthman card pile labeled in-copy as the temporary backup; FYI links to /gameboard/ (Game Kit equip). User-modified line text in template: "Paperboard" → "Cardstock" mid-session ; **helper fallback** (epic/models.py): `personal_sig_cards(user)` now falls back to `DeckVariant.objects.filter(slug='earthman').first()` when `user.equipped_deck` is None — same 16-or-18 card pile, just sourced from the canonical Earthman deck rather than the empty FK. No new DeckVariant row needed; "Shabby Cardstock" is purely UX framing (cards are the same TarotCard records the room sig-select uses). Preserves the existing helper signature so no callers had to change ; **view + template** (billboard/views.py + my_sign.html): view passes `no_equipped_deck` + `show_backup_intro_banner` flags. Template removes the old `{% if not equipped_deck %}` forced-equip branch — picker now renders unconditionally w. cards from the backup helper when no deck is equipped. Brief banner fires via `Brief.showBanner({...})` on DOMContentLoaded when `show_backup_intro_banner` is true — gets h2-overlay positioning + NVM behavior + portal styling for free (per [[sprint-baltimorean-note-unlock-may18]] portrait h2 measurement in note.js's `_alignToH2`). Added `<script src="note.js">` to my_sign.html since the page didn't load it before. Post-render JS tags the Brief w. a `.my-sign-intro-banner` class so FTs (and any future my-sign-specific styling) can distinguish this nudge from other Briefs on the page ; **TDD trail** — 4 new FTs in `MySignBackupDeckTest` (test_bill_my_sign.py): T1 banner renders w. "Default deck warning" title + "no deck is equipped" + "Shabby Cardstock" copy + both action btns visible; T2 picker still populates 16 cards from backup; T3 NVM click removes the banner from the DOM; T4 FYI href ends w. /gameboard/. Initial reds (`NoSuchElementException` on all 4) confirmed before implementation. Plus 1 new IT in `PersonalSigCardsTest` pinning the helper fallback (16 cards w. all `c.deck_variant.slug == "earthman"`) ; pre-existing change picked up: `static_src/scss/rootvars.scss` (user-modified mid-session) ; 1020 IT/UT green; 7 FTs green (3 picker happy-path + 4 backup deck) in 56s. Sprint 4a-follow complete — primary deferral from Sprint 4a (deck-source fallback UX) now landed. Unblocks Sprint 4b (My Sea gating w. --terUser link to /billboard/my-sign/ when no sig set)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User scope (per design conv this session): split the room's sig-select responsibility off into a standalone billboard-context "My Significator" applet — branded "Game Sign" on the surface. Same 18-card pile as room sig-select (16 middle arcana + Major 0 & 1 filtered by Note unlocks); polarity collapses to a single FLIP choice (the FLIP btn in the picker carousel toggles User.significator_reversed). Selection persists globally on the User model + propagates to the billboard's Game Sign applet ; **naming convention locked**: "significator" stays at storage (User.significator FK + User.significator_reversed) + room sig-select context (DRY w. existing template/JS); "Sign" / "Game Sign" is the billboard-surface branding (file my_sign.html, URL /billboard/my-sign/, URL names my_sign + save_sign, applet name "Game Sign", page wordmark "Game Sign", btn label SAVE SIGN). Action URLs don't carry a trailing slash per project convention (/billboard/my-sign/save vs the page's /billboard/my-sign/) ; **schema**: User gains 2 fields — `significator: FK → epic.TarotCard (nullable, on_delete=SET_NULL)` + `significator_reversed: BooleanField(default=False)`. Migration lyric/0006_user_significator_user_significator_reversed.py auto-generated; reversible. Applet seed in applets/0009_seed_my_sig_applet.py adds the row (slug='my-sign', name='Game Sign', context='billboard', default_visible=True, grid_cols=4, grid_rows=6), idempotent update_or_create, reversible unseed() ; **picker page** (my_sign.html): solo lift of `_sig_select_overlay.html` — sig-stage-card scaffold + sig-stat-block + 18-card grid + SAVE SIGN form. Stripped: countdown / WebSocket / polarity / multi-user / reservations. Empty-state branch covers no-equipped-deck (link back to Game Kit; full Brief-redirect + Earthman-Backup fallback deferred to a follow-up sub-sprint). Minimal inline JS: click .sig-card → mark .sig-focused + set hidden card_id + enable SAVE SIGN; FLIP btn toggles .is-reversed + the hidden reversed input. Stage-card preview (name/qualifier population + keyword swap on FLIP) deferred — Sprint 4a follow-up will lift stage-card.js's populator into a non-room context ; **applet partial** (_applet-my-sign.html): renders user.significator's corner-rank + suit-icon + name_title if set; `.my-sign-applet-empty` "No sign chosen yet." otherwise. Header `<h2><a href="{% url 'billboard:my_sign' %}">Game Sign</a></h2>` links to the picker ; **helper refactor** (epic/models.py): extracted `_sig_unique_cards_for_deck(deck_variant)` from `_sig_unique_cards(room)`. New public `personal_sig_cards(user)` parallels `levity_sig_cards / gravity_sig_cards` but pulls from `user.equipped_deck` instead of `room.deck_variant`. Same Note-unlock filtering. No behavior change to existing room callers (3-line wrapper preserves the room signature) ; **TDD trail** — user called out mid-sprint that I'd skipped FTs; pivoted to FT-first. test_bill_my_sign.py (new, 3 FTs): T1 picker renders w. wordmark + target card present in grid; T2 click card → SAVE SIGN enables → POST persists → applet shows the card; T3 fresh user → applet renders empty-state. Initial reds — (a) setUp's `personal_sig_cards(user)` returned [] because StaticLiveServerTestCase → TransactionTestCase flushes migration-seeded DeckVariant + TarotCard between tests; fixed w. `serialized_rollback = True` on the test class (per [[feedback_transactiontestcase_flush]]); (b) h2 wordmark assertion against `MYSIGNIFICATOR` failed against the renamed "Game Sign" + the letter-splitter spreading chars across <span> children — switched to whitespace-stripped substring check `GAMESIGN`; (c) `.fan-corner-rank` text is CSS-hidden so Selenium returns "" — replaced corner-rank assertions w. data-card-id selectors (already-proven reliable from the parent .sig-card lookup) ; ITs (+12, in apps.billboard.tests.integrated.test_views): MySignViewTest (6 — login redirect, 200 + template, 16-card pile, save persists, invalid card_id → 403, GET save redirects); BillboardAppletMySignTest (3 — applet rendered, empty-state w/o sig, card+reversed class w. sig). PersonalSigCardsTest in apps.epic.tests.integrated.test_models (3 — happy path 16 cards, no-equipped-deck → [], schizo Note unlocks Major 1) ; pre-existing change picked up by the commit: my_sea.html branding "Game Sea" (user-modified mid-session; was "My Sea" in Sprint 3 — divergence captured in MEMORY.md follow-up) ; 1020 IT/UT green (+12) in 46s; 3 FTs green in 24s. Sprint 4a unblocks Sprint 4b (My Sea gating w. --terUser link to /billboard/my-sign/) + Sprint 4c (FT helper for mocking the sig choice across other FTs)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User caught a missed surface on iPhone after the May-18b rename pass (1ccb045): the post.html `.post-attribution` spans still rendered "@disco the Ard!" instead of "@disco the Baltimorean" — six callsites across the post header (author / invitee / shared-self / created-by) plus `_my_buds_applet_item.html`'s bud row body. Same shape on display: navbar DON greeting is the only surface that should keep "Ayo, Ard!", per the May-18b architectural decision ; root cause: `User.active_title_display` at lyric/models.py:152 returned `self.active_title.display_title` ("Ard!" for Baltimorean) instead of `self.active_title.display_name` ("Baltimorean"). The Sprint-18b rename pass swapped the inline `attr_combo` + Brief.title to use `display_name`, but missed this property which is the indirection layer for `.post-attribution` callsites. Navbar uses `{{ user.active_title.display_title }}` directly (no helper-property indirection) so it stays at "Ard!" — that's the intended single Ard! surface ; fix: one-line swap in `active_title_display` from `display_title` to `display_name`. For stargazer / schizo / nomad these two are equal (the Note model's `display_name` property at drama/models.py:262 falls through to `display_title` unless the slug has an override in `_NOTE_DISPLAY[slug]["display_name"]`) — Baltimorean is the only current override w. `{"display_name": "Baltimorean"}`. So this is no-op for every non-Baltimorean Note ; TDD trail: +3 UTs in apps.lyric.tests.integrated.test_models.UserModelTest: `test_active_title_display_returns_earthman_when_no_note_donned` (smoke), `test_active_title_display_uses_display_name_not_display_title` (pins the Baltimorean override path — went red 'Ard! != Baltimorean' before the fix), `test_active_title_display_falls_through_to_display_title_for_non_overridden_slugs` (pins the no-op path for stargazer). Red → green confirmed. Surfaces auto-affected: post.html post-attribution × 5 callsites + `_my_buds_applet_item.html` bud row body (all use `{{ user.active_title_display }}`) ; 1008 IT/UT green in 46s (+3 from 1005)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User roadmap step (Sprint 3 of cluster): scaffold the My Sea applet on the gameboard + the standalone /gameboard/my-sea/ page where later sprints will host the gatekeeper / sig-select / sea-select reskin for solo-user draws. Shell-only — no draw flow yet; latest-draw rendering, mid-progress save, daily quota land in Sprints 4-9 ; **migration**: `applets/migrations/0008_seed_my_sea_applet.py` — RunPython that `update_or_create`s Applet(`slug='my-sea'`, name='My Sea', context='gameboard', default_visible=True, grid_cols=12, grid_rows=4). 12×4 wide horizontal banner so the Celtic Cross spread's 10 cards can render left-to-right in the applet aperture, scrollable like My Palette (per user spec). Reverse migration (`unseed`) deletes the row so the migration is reversible for staging rollbacks ; **applet partial**: `templates/apps/gameboard/_partials/_applet-my-sea.html` — same `{% applet_context %}` auto-discovery shape every other applet uses (`<section id="id_applet_my_sea" style="--applet-cols: {{ entry.applet.grid_cols }}; --applet-rows: {{ entry.applet.grid_rows }};">`). Header is a `<h2><a href="{% url 'my_sea' %}">My Sea</a></h2>` link (gold via global `body a` rule); body is a `.my-sea-scroll` container that either renders `.my-sea-card` cells from a `latest_draw_cards` context (TBD in Sprint 4-7) or a `.my-sea-empty` placeholder line "No draws yet." for fresh users ; **standalone page**: new `gameboard/views.py:my_sea` view + url at `/gameboard/my-sea/` (URL name `my_sea`) rendering `apps/gameboard/my_sea.html` — `{% extends "core/base.html" %}` shell w. letter-spread `<span>My</span><span>Sea</span>` h2 wordmark + `.my-sea-page__empty` placeholder paragraph "Your sea is calm. Draws will appear here." `page_class` doubled to `page-gameboard page-my-sea` so the body inherits the gameboard's landscape aperture treatment AND any future my-sea-specific styles can target a single class. Login-required like the rest of gameboard ; **tests (+6 ITs)**: GameboardViewTest gains 3 — `test_gameboard_shows_my_sea_applet` (cssselect pins #id_applet_my_sea), `test_my_sea_applet_renders_empty_state_for_new_user` (asserts ".my-sea-empty" text + no ".my-sea-card" rows), `test_my_sea_applet_header_links_to_my_sea_page` (h2 a href == reverse('my_sea')); new MySeaViewTest class — `test_my_sea_requires_login` (redirect to /?next=...), `test_my_sea_renders_200`, `test_my_sea_uses_gameboard_page_class` (page-gameboard + page-my-sea both in body class). Existing GameboardViewTest setUp already does `get_or_create` per-applet so no fixture change needed for the migration-driven my-sea row ; 1005 IT/UT green (+6 from 999) in 45s; visual verified in Claudezilla at iPhone-14 portrait — applet renders w. rotated "MY SEA" vertical label + "No draws yet." body; /gameboard/my-sea/ standalone page renders w. letter-spread wordmark + placeholder ; **next**: Sprint 4 — My Sea sig-select phase (single-significator pick for solo user, w. the parameterized hex CSS from Sprint 1 hosting the chair-less or single-chair variant)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User ask: ensure trinket FTs (test_trinket_carte_blanche.py, test_trinket_coin_on_a_string.py, test_trinket_backstage_pass.py) run in the room-stage CI bucket so they exercise the same room-template surface as test_game_room_* — relevant when a sprint touches the table hex SCSS / chair geometry / gatekeeper flow, since trinket FTs create rooms + walk thru the gate. Previously they fell into test-FTs-non-room by elimination (the non-room glob was `grep -v 'test_game_room_'`) ; main.yaml updates: non-room step's grep is now `-vE 'test_(game_room|trinket)_'` to exclude both clusters; room step's ls now globs BOTH `test_game_room_*.py` AND `test_trinket_*.py` ; companion memory tweak in feedback_ft_naming_prefix.md documents the new convention so future trinket FTs end up in the right bucket without re-asking ; pipeline behavior unchanged otherwise — both steps still depend_on test-two-browser-FTs only, parallel re-enable still blocked by the shared-sqlite issue from 2026-05-12 (cf. memo note)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User reported on iPhone: after depositing a COIN at a game's gatekeeper, the Kit Bag's Trinket slot still shows the COIN — even though the tooltip correctly carries the room attribution ("Ready 2026-05-25 / Billingsworth"). Expected behavior matches CARTE: the deposited token disappears from the Kit Bag Trinket slot because it's committed elsewhere & can't be re-used as the active trinket until released. PASS preserved — auto-admits w.o ever going thru the deposit path so it stays equipped ; **the real bug**: `debit_token` in epic/models.py's COIN branch set `current_room` + `next_ready_at` but never cleared `user.equipped_trinket`. CARTE's `drop_token` view (epic/views.py:440-442) explicitly unequips at deposit time via `user.equipped_trinket = None; user.save(update_fields=["equipped_trinket"])`; COIN had no parity. Fix: same 4-line unequip stanza now lives inside the COIN branch of `debit_token`, guarded by `if user.equipped_trinket_id == token.pk` so a fresh-purchased COIN deposit (not the equipped one) doesn't accidentally clear another trinket. PASS untouched — falls thru `debit_token` w.o entering any branch & never reaches this path; CARTE untouched too (its branch is `pass`, unequip happens at `drop_token` time before debit_token is even called) ; **the FT false-positive**: yesterday's Sprint 2 commit (d2491c5) shipped `test_coin_deposit_unequips_from_kit_bag_and_fills_one_slot` w. selector `#id_kit_bag_dialog .kit-bag-placeholder`. That selector was matching the **Dice** section's placeholder (Dice feature isn't built — `_kit_bag_panel.html` L23-29 renders `.kit-bag-placeholder` unconditionally), masking the bug whether or not the Trinket section was empty. Tightened to `.kit-bag-section--trinket .kit-bag-placeholder` w. comment explaining why a bare selector is unsafe ; template change in `_kit_bag_panel.html` L31: Trinket section gains a `kit-bag-section--trinket` modifier class so the FT (and any future selector that needs to single out the trinket section vs the deck/dice/tokens siblings) has an anchor. Mirrors the existing `kit-bag-section--tokens` class at L70 ; TDD trail: (1) tightened selector + reran → red on `NoSuchElement` (no `.kit-bag-section--trinket .kit-bag-placeholder` because COIN still equipped post-deposit, so trinket section renders the token card not the placeholder); (2) added unequip stanza to debit_token; (3) reran → green. 10 trinket FTs in 99s; 999 IT/UT in 46s — no regressions ; **generalizable trap**: when an FT waits for an element via a CSS selector, scope the selector to the section/container that uniquely identifies the assertion target — a class like `.kit-bag-placeholder` that's reused across multiple sections will silently pass even when the section you care about is in the wrong state. This is the second false-positive trap in two days (cf. d2491c5's wrong-selector trap where `.token-slot.claimed` was Carte-specific); pattern's worth noting
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User-driven roadmap step (Sprint 2 of My Sea cluster): give the COIN trinket parity w. CARTE's deposit-aware Game Kit surface — when COIN is deposited in a room (Token.current_room FK set via debit_token in epic/models.py:135), its Game Kit token should expose the room name on `data-current-room-name` so the mini-portal can render "In-Use: <room name>" on hover, and both DON / DOFF btns should go btn-disabled (token is committed elsewhere, neither equip nor un-equip is valid). Mirrors the 3-state branch CARTE has in the same template (deposited / equipped / unequipped); PASS not in scope — auto-admits w.o deposit ; template change in _applet-game-kit.html — line 40 (COIN's div) gains `data-current-room-name="{{ coin.current_room.name|default:'' }}"` + an extra `{% if coin.current_room %}…{% elif coin.pk == equipped_trinket_id %}…{% else %}…{% endif %}` branch that fronts the existing 2-way w. a deposited-state arm (both btns "×" btn-disabled). View-side wiring already in place — coin context var is the user's COIN token incl. its `current_room` FK; no Python change needed ; TDD trail — test_trinket_coin_on_a_string.py (new, 3 FTs): T1 hover equipped COIN → mini portal "Equipped" + main portal tooltip prose (Coin-on-a-String / Admit 1 Entry / "…and another after that…" / no expiry); T2 deposit flow — rails-click → slot 1 RESERVED → click `.btn-confirm` inside the reserved gate-slot (NOT `.drop-token-btn` which is Carte's carte_active path) → slot fills + COIN admits only 1 entry (slot 2 has no follow-up btn cf. Carte's 6) + kit-bag Trinkets section empty (COIN unequipped on deposit); T3 navigate back to /gameboard/ → COIN's `#id_kit_coin_on_a_string` has `data-current-room-name="Commitment Room"` + both DON & DOFF btns inside `.tt` are btn-disabled ; initial red run hit a Carte-specific selector trap — `.token-slot.claimed` (the Carte machine UI from `user_filled_slot or carte_active` branch in _gatekeeper.html L23) doesn't fire for COIN, which lands on `.token-slot.pending` (user_reserved_slot branch); diagnosed via screendump grep — slot 1 carried class "gate-slot reserved" + token-slot was "pending"; FT rewritten to wait for `.gate-slot[data-slot='1'].reserved` → click `.btn-confirm` (the OK btn rendered for the reserving user in _table_positions.html L7-15) → wait for `.filled`. T1+T2 then green; T3 stayed red on `data-current-room-name` AttributeError (None != "Commitment Room") which is the actual bug the template fix addresses ; test_trinket_backstage_pass.py (new, 4 skeleton FTs): T1 staff-user signal contract — `gamer.equipped_trinket_id == pass_token.pk` post-signal; T2 tooltip renders title/description/shoptalk/expiry (Backstage Pass / Admit All Entry / "'Entry fee'? …" / no expiry); T3 equipped PASS mini portal says "Equipped"; T4 PASS btn apparatus — DON × btn-disabled, DOFF active w. label "DOFF" (symmetric to COIN's equipped state cf. test_gameboard.py:207-220). DEPOSIT FLOW DEFERRED to future sprint w. TODO comment block — PASS magically auto-admits any gate w.o going through the `.token-rails` deposit path that CARTE / COIN share, so no `data-current-room-name` parity work applies; user explicitly chose "Auto-admits, never deposited — keep current behavior" for this sprint ; 10 trinket FTs green in 93s (carte 4 + coin 3 + pass 3 — wait, pass has 4: 4); full IT/UT 999 green in 46s — no regressions; coin context already passing `coin.current_room` correctly thru _game_kit_context (no Python change). Sprint 2 of [[project_my_sea_applet]] cluster — next: My Sea applet shell (Sprint 3+)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
Belt-and-suspenders for the iOS Safari auto-zoom-on-input quirk: Mobile Safari zooms the viewport when an `<input>`/`<textarea>`/`<select>` is focused & its computed font-size < 16px, and never zooms back out on blur. Two layers ; PRIMARY — SCSS prevention: new `input, textarea, select, [contenteditable] { font-size: unquote("max(16px, 1em)") }` in core.scss (Sass can't reconcile px/em units in compile-time max() so unquote() passes the CSS max() through verbatim — modern browsers handle natively). 1em inherits parent, max() floors at 16. ALSO floored `.form-control-lg` in _base.scss — was `font-size: 1.125rem`, which at rem=14 (small portrait, clamp(14px, 2.4vmin, 22px) hits its floor) computes to 15.75px → **0.25px** under iOS's 16px threshold → the "ever so slightly" zoom on New Game + New Post applets the user reported (both use `.form-control.form-control-lg`, specificity 0,2,0 beats my element-level 0,0,1 rule). Floor: `unquote("max(16px, 1.125rem)")` ; SECONDARY — JS fallback in base.html: rewritten from `setAttribute('content', ...)` toggle to full meta-element remove+re-add, which modern iOS handles more reliably than attribute mutations on the existing meta. Triggers on document-level `focusout` (bubbles natively, no capture-phase needed) for `input/textarea/select`; injects fresh viewport meta w. `maximum-scale=1.0, user-scalable=no` for 300ms (iOS reads as zoom violation → snaps to 1:1), then swaps back to the cached base content so pinch-zoom remains available elsewhere ; user observed horizontal scrollbar appearing when the page zoomed — that's the symptom the user actually cared about (broken layout, not aesthetic zoom). w. SCSS floor in place the zoom shouldn't trigger to begin with; the JS is purely for inputs that slip through (future custom controls, shadow DOM, etc.) ; iOS-specific behavior — Selenium+Firefox doesn't replicate the auto-zoom so no FT layer added. Verified by user manual iPhone test (post-fix retest pending after force-refresh)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
iOS Safari auto-zooms when a user taps an `<input>`/`<textarea>`/`<select>` whose font-size is < 16px, and does NOT auto-zoom back out on blur — the page stays zoomed even after the field loses focus. New ~10-line IIFE at base.html slots next to the existing h2-letter-splitter at the bottom of <body>: caches the page's `<meta name="viewport">` content, listens (document-level, bubbling `focusout`) for inputs leaving focus, then briefly appends `, maximum-scale=1.0` before reverting 100ms later — iOS reads the tightened constraint as a "zoom violation" and snaps the viewport back to 1:1, after which the revert frees the user to pinch-zoom manually anywhere else on the page ; chose `focusout` over `blur`+capture-phase since focusout bubbles natively (cleaner); skips if `.matches` isn't available (defensive for older browsers); skips silently if no viewport meta is present (defensive) ; no test layer — iOS-specific behavior that's awkward to FT (would need a real iOS Safari runner; Selenium+Firefox doesn't replicate the auto-zoom). Verified no conflict w. other focusout listeners (grep: only vendor JS — d3 / htmx / jquery / select2 — none of which listen at document scope on inputs/textareas/selects). Side-track addition between Sprint 1 (table hex layout fbe6c12... well, 7165974) and Sprint 2 (My Sea applet kickoff) per user ask
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
User report: hex felt smaller than the aperture even at portrait (off-centered, room to spare on top + bottom), and chair labels overlapped the hex edges at landscape — progressively worse as the hex grew at larger viewports. Three contributors stacked: (1) `.room-shell { max-height: 80vh }` capped the shell at 80% of viewport height even when the .room-page aperture had more room — at 1789×1031 this donated 228px (1053→825) of aperture height to dead margin; (2) scene design 360×300 was wider than tall (1.2 aspect) but landscape aperture is narrower than tall (~1.4), so the height cap bottlenecked scene-scale at min(aperture_w/360, aperture_h/300) instead of letting the hex grow; (3) chair font-size scales w. rem (clamp(14,2.4vmin,22)) but chair position scales w. --table-scale — at large viewports rem maxes at 22 so labels widen and push chair icons further from box-center toward the hex (visual "creep") ; fix: remove the 80vh cap (`max-height: 80vh` → `height: 100%` on .room-shell L340) so the shell stretches to fill the .room-page aperture; bump hex from 160×231 to 200×231 (regular pointy-top w. width = height × √3/2 = 200 * 1.1547 — comment in _room.scss updated); apothem of 200-wide pointy-top regular hex is 100px exact (200/√3 × √3/2), so `$pos-d` 110px → 140px gives 40px design-units of radial chair clearance (was 30); derived `$pos-d-x: round(140*0.5) = 70`, `$pos-d-y: round(140*0.866) = 121` for slot 2/3/5/6 diagonal anchors at 60° from horizontal (matches existing geometry approach); scene design height 300 → 320 to leave enough vertical headroom at large landscape that the rem-driven (font-size 1.6rem × scale) chair icons + labels don't clip the aperture top/bottom edges — at 1789×1111 w. scene_H=300 the AC/BC label tops sat AT aperture top (y=-21 vs aperture y=-22), bumping to 320 drops scale from 4.05 → 3.54 and leaves 76px of headroom; SCENE_H in room.js bumped to 320 to match (Math.min(w/SCENE_W, h/SCENE_H) sets --table-scale CSS var via transform: scale on .room-table-scene) ; visual verification via Claudezilla across three viewports (no test layer per user preference — layout regression coverage via spot-check on next room render) — iPhone-14 portrait 566×875: hex 243×281 → 314×363 (+29% wider, fills 55% of aperture width vs 44% before); mid landscape 1149×781: hex 333×385 → 493×569 (+48% wider, 56% vs 38% before); large landscape 1789×1111: hex 440×509 → 708×818 (+61% wider, 48% vs 30% before — the most dramatic improvement, matching user's "progressively worse the larger the hex grows" observation). Chair clearance now uniform 40 design-units radially across all scales; AC/BC labels stay 76px inside aperture top at the largest viewport ; dead `$seat-r`/`$seat-r-x`/`$seat-r-y` consts at L357-359 left in place (unused elsewhere in codebase but out of scope for this layout fix) ; full IT/UT 999 green in 46s — no regressions; .table-hex / .table-hex-border / .room-table-scene / .table-seat positioning consts are the only refs to these dimensions across SCSS & JS so no cascade beyond room layout. Unblocks Sprint 2+ (My Sea applet will share the same hex CSS, parameterized, per user's intent for future friend-invite up-to-6-person rooms)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
CAST SKY btn click handler in sig-select.js init() was bound to `Tray.open()` — wrong on two counts: (1) the tray was already played during the `polarity_room_done` → `Tray.placeSig` sequence (sig stage card slides into the tray cell before the overlay dismisses), so re-opening it on CAST SKY click pops the tray a second time; (2) Sky Select never opens — `_sky_overlay.html` is only `{% include %}`d server-side when `room.table_status == "SKY_SELECT"`, so during SIG_SELECT the partial + its `openSky` handler aren't in the DOM and `Tray.open()` is the only thing the click does. Bug surfaced symmetrically in both polarity rooms regardless of which finished first ; fix: replace `Tray.open()` w. `window.location.reload()` so the server re-renders the room w. table_status=SKY_SELECT — which surfaces the sky overlay partial + the `openSky` handler bound at _sky_overlay.html:192-193. Same pattern as `_onSkyConfirmed` in the sky partial (location.reload after sky save) ; testability hook mirrors `RoleSelect.setReload` (role-select.js:236): `var _reload = function () { window.location.reload(); };` at module scope, listener calls `_reload()` (closure looks up the var at click time so reassignment works), `setReload(fn)` exposed on the module's test API. SigSelectSpec.js adds `describe("CAST SKY click (post pick_sky_available)")` w. 2 specs — reload spy hit on click + `Tray.open` spy NOT hit on click; the negative assertion catches the original bug, the positive verifies the fix's intent. Existing 363 specs untouched ; Jasmine FT green in 8.6s; full IT/UT 999 green in 44s ; collectstatic mirror at src/static/apps/epic/sig-select.js refreshed in same commit so the served JS carries the fix (Django serves from STATIC_ROOT, not from app static dirs, in StaticLiveServerTestCase)
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Sonnet 4.6 <noreply@anthropic.com>
pipeline fix — `_post_gear.html` (commit 6a7464e) gated the NVM target on `{% url 'billboard:my_posts' user_id=request.user.id %}`, which exploded w. NoReverseMatch when an anonymous user (Percival ch.18 anonymous-post lab — ownerless `Post.objects.create()`) hit view_post (which has no @login_required); whole gear-include now wrapped in `{% if request.user.is_authenticated %}` since anonymous viewers can't DEL/BYE/back-to-my-posts anyway; AnonymousPostViewerTest pins the 200-render + gear-absence contract so future ownerless-post regressions surface in ITs (pipeline run #298 fixed) — TDD
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Opus 4.7 <noreply@anthropic.com>
.woodpecker/main.yaml — CI test-FTs step splits into parallel siblings test-FTs-non-room (22 files via `ls functional_tests/test_*.py | grep -v 'test_game_room_'`) + test-FTs-room (9 files via `ls functional_tests/test_game_room_*.py`); room cluster is the heaviest (~70% of the pre-split ~40-min wall-clock) and now runs concurrently w. the rest instead of in series; DAG explicit via depends_on on every step (Woodpecker mixes default-sequential w. depends_on awkwardly, so each step pins its prerequisite); collectstatic stays in test-two-browser-FTs only — the shared workspace propagates assets to both parallel FT steps, no race + no duplication; screendumps + build-and-push fan back in (depends_on both parallel steps); deploy-staging + deploy-prod depend on build-and-push
smoke-import: 31/31 FT modules green after the rename pass
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- billboard.views.view_post adds viewer_is_owner + other_recipients context vars. is_real_invitee = (auth AND post has owner AND viewer != owner). Anon viewers + ownerless-post legacy path fall through to owner-style rendering (which renders empty gracefully via the at_handle / display_name AnonymousUser guards).
- other_recipients = post.shared_with.exclude(viewer) when invitee; .all() otherwise.
- post.html .post-header branches:
• viewer_is_owner: existing prose ("just me, @owner …" / "shared between {recipients} & me, @owner …").
• sole invitee: "shared with me, @viewer the {viewer.title}" + "created by @owner the {owner.title}".
• multi invitee: "shared with {other_recipients}" + "& me, @viewer the {viewer.title}" + "created by @owner the {owner.title}".
- lyric_extras at_handle + display_name: guard against AnonymousUser (no .email attribute) — return "" rather than crash. Preserves the Percival ch. 18 anon-views-ownerless-post path.
- 12 new ITs in test_post_invitee_view (context vars: viewer_is_owner, other_recipients exclude/include; template prose: sole + multi invitee phrasing, owner unchanged).
- 878 IT regression + 8 post-html FT regression green (1 Marionette flake on multi-run that passes in isolation).
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Brief schema (billboard/0007): post FK becomes nullable + new room FK to epic.Room + KIND_GAME_INVITE enum value. to_banner_dict resolves post_url to reverse('epic:gatekeeper', room.id) when post is null and room is set.
- epic.invite_gamer view refactor:
• Accepts `recipient` (matches bud-panel field; legacy `invitee_email` still works for full backwards compat).
• Resolves via apps.billboard.views._resolve_recipient (email if "@" present, else username).
• RoomInvite stores the resolved User's email (or raw input if unregistered).
• Auto-adds inviter ↔ recipient to each others' buds (symmetric per Phase 2 spec) when recipient is a registered User.
• Spawns a Brief w. owner=request.user, kind=GAME_INVITE, room=room, post=null.
• Accept: application/json → {brief, recipient_display}; otherwise redirects to gatekeeper as before.
• Self-invite + blank recipient: 200 w. brief=null, no RoomInvite, no buds touch.
- _gatekeeper.html: gate-invite-panel block (lines 62-71) removed.
- new templates/apps/billboard/_partials/_bud_invite_panel.html: clone of _bud_panel.html w. data-invite-url + autocomplete from request.user.buds. JS posts to invite_gamer + Brief.showBanner. room.html includes it owner-only when not table_status and gate_status != RENEWAL_DUE.
- room.html scripts block now loads apps/dashboard/note.js so window.Brief is defined for the slide-down banner.
- Tests: new test_invite_gamer.py (14 ITs) covering ajax + legacy form-submit paths, recipient resolution, RoomInvite creation, Brief w. room FK + GAME_INVITE kind, symmetric buds auto-add, unregistered/self/blank silent no-op cases. New test_gatekeeper_bud_btn.py FT (9 tests) covers presence (owner-only), absence of legacy #id_invite_email, async invite flow end-to-end (RoomInvite, Brief, banner, panel close, username resolve, buds auto-add).
- test_brief.test_brief_owner_post_required relaxed to test_brief_owner_required (post is now nullable).
- test_room_gatekeeper.test_second_gamer_drops_token_into_open_slot updated to drive the bud-btn flow (drops the #id_invite_email/#id_invite_btn references).
- 866 ITs (+14) + 9 gatekeeper FTs + 28 existing room-gatekeeper FTs green.
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- _game-kit.scss #id_kit_btn landscape rule now uses `right: calc((var(--sidebar-w) - 3rem) / 2)` — same formula as gear-btn (_applets.scss) and bud-btn (_bud.scss). All three 3rem-wide circular btns now share the same horizontal-centre math against the fluid sidebar.
- Drops `@media (orientation: landscape) and (min-width: 1800px) { right: 2.5rem }` which was a leftover from the old doubled-8rem-sidebar regime; the rem clamp ceiling now caps the sidebar without per-breakpoint overrides.
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- portrait h2 spans get padding-inline-end / padding-inline-start (0.4em each) at the 45/55 boundary; box-sizing: border-box keeps the flex basis honest. Solves the "B I L L B O A R D" run-together where the L of BILL touches the B of BOARD.
- landscape h2 drops the `display: block` override + the > span resets that nuked text-align: justify and flex. Now inherits the portrait flex 45/55 + per-span justify + padding-inline. With writing-mode: vertical-rl, the flex axis runs vertically (45% bottom for BILL post-rotate, 55% top for POST/BOARD/etc.); padding-inline-end resolves to the bottom edge of the first span = natural gap between the two rotated words.
- Explicit h2 height: 80vh in landscape so the flex 45/55 percentages have a defined basis to resolve against (block height isn't auto-derived in writing-mode: vertical-rl).
- 8 layout/navbar FTs still green; assertions are categorical (position: fixed, etc.) not exact-px.
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- html { font-size: clamp(14px, 2.4vmin, 22px) } — single sliding scale; everything in rem (sidebar widths, h2 font-size, paddings) scales together. Phone rotation swaps width/height but vmin stays the same → 1rem stays the same → navbar/footer/h2 hold their size between portrait + landscape.
- :root --sidebar-w: 5rem (replaces the locally-scoped $sidebar-w SCSS var that lived inside @media blocks); --h2-col-w: 3rem for the rotated wordmark column in landscape. var(--sidebar-w) + var(--h2-col-w) are the only knobs that move the layout.
- Landscape container: margin-left = calc(var(--sidebar-w) + var(--h2-col-w)); margin-right = var(--sidebar-w). Applets are now clipped INSIDE the h2 column, so the rotated "BILLPOST" / "DASHBOARD" wordmark never has content bleeding behind it (the original complaint).
- h2 markup refactor across 13 templates: <span>BILL</span><span>POST</span> instead of <span>BILL</span>POST. Portrait styling: display: flex; first span flex 0 0 45% + --quaUser colour; second span flex 0 0 55% + --secUser inherited. Per-span text-align: justify + text-justify: inter-character keeps the inter-letter spacing within each span. Landscape resets the flex (single rotated wordmark, not split).
- Drop the four h2 font-size jumps (min-height: 400/500/800px) — single font-size: 3rem now scales fluidly via root rem. Drop the @media (orientation: landscape) and (max-width: 1100px) h1 override (rem-fluid handles cramped widths). Drop the entire @media (orientation: landscape) and (min-width: 1800px) sidebar-doubling block in _base.scss / _applets.scss / _bud.scss — the rem clamp ceiling already caps the size.
- _bud.scss + _applets.scss: bud-btn / bud-panel / bud-suggestions / gear-btn / applet menus all switch to var(--sidebar-w)-based positioning; landscape rules are single (no per-breakpoint duplication).
- Per-spec tradeoff: non-.btn-primary buttons (BYE / NVM / OK / kit-btn / etc.) inherit rem-fluid like everything else and will scale slightly w. viewport. User explicitly OK'd this — they don't need to stay px-fixed.
- 852 ITs + 24 layout/navbar/bud FTs green; existing geometry assertions are relative or categorical (not exact-px) so the rem clamp doesn't surface failures at the 800x1200 FT viewport.
Code architected by Disco DeDisco <discodedisco@outlook.com>
Git commit message Co-Authored-By:
Claude Opus 4.7 (1M context) <noreply@anthropic.com>
