Resume Prompt
Open
docs/plan/launch-plan-stage-1.htmlin browser to verify split rendered correctly (hero / topnav / sidebar Stage 1 phases local + Stage 2/3/4 cross-file links / Foundational §1-§7 / Phase 1.1-1.11 / Open Questions). Then sync 3 files to~/code/offplan-online/preview/plan/:launch-plan-stage-1.html(NEW),launch-plan-v3-full.html(heavily modified, 3576→1268 lines),launch-plan-changelog.html(v4.4 entry from CONV-17/18, untouched today). Commit + push preview repo with message "plan(v4.4): split + cumulative — stage-1 extracted + Reservations removed + Closed pool + clarity". After push: start phase rework around Atelier prototype (/Users/diver/Downloads/admin-quick-build-atelier-standalone.html) — Phase 1.2 simplifies (no demo, Atelier IS entry), Phase 1.5 priority decision (AI floorplate tagging may need bump from 🟡 High → 🔴 Critical because Atelier copy promises it). Don't send to Roman until phase rework reflects Atelier direction.
Summary
Continued Sergey's read-through of v4.4 review document and applied multiple clarity polishes to Foundational sections (Partner type wording multi-select fix; tier shape qualitative scaffolding with naming-namespace flag; §4.3 Scoping rewrite; §5 Visibility lead + Custom preset surfaced + viewer-state row 4 fix; Foundational counter badges 8→7). Mid-session pivoted to a major refactor: extracted Stage 1 from launch-plan-v3-full.html into a new launch-plan-stage-1.html (2870 lines) using sed-based extraction with cross-link substitution; main file shrunk from 3576→1268 lines with stub callout. Priority order callout in stage-1 fixed (was missing Phase 1.3 ⭐ and 1.11 ⭐, plus stale 1.2 description). Sergey shared Atelier React prototype as the new direction for Phase 1.2 onboarding (replaces demo-mode); design opinion captured but plan not modified yet — phase rework is the next big item. Memory updated: divergence callout on plan_permission_tenancy_state.md (Block C1/C4 stale vs current plan), new feedback_stage1_only_focus.md, new project_conv19_resume_point.md. Split + clarity work is NOT YET synced to preview repo or pushed to git.
Changes
docs/plan/launch-plan-v3-full.html— Stage 1 extracted (3576→1268 lines); §1 Partner type «один → один или несколько»; §3 tier table 4 rows + namespace caveat; Open Questions Pricing tier reminder updated; §4.3 Scoping bullets rewritten; §5 Visibility lead + Custom preset surfaced + viewer-state row 4 corrected; Foundational counter badges /8 → /7 (×6); Stage 1 stub callout block inserted; sidebar/topnav Stage 1 refs → cross-filedocs/plan/launch-plan-stage-1.html(NEW) — full Stage 1 content extracted from main; cross-file refs to Stage 2/3/4 (launch-plan-v3-full.html#...); priority table fixed (added Phase 1.3 ⭐, 1.11 ⭐, cleaned 1.2 row); footer with back-nav links~/.claude/.../memory/plan_permission_tenancy_state.md— added⚠️ DIVERGENCE FROM PLANcallout flagging Block C1 (Closed pool replaces Shared default) and Block C4 (Reservations removed) as stale vs current plan~/.claude/.../memory/feedback_stage1_only_focus.md(NEW) — feedback memory: only Stage 1 in scope; restrict greps/reads to Foundational + Phase 1.x~/.claude/.../memory/project_conv19_resume_point.md(NEW) — resume checkpoint with split status, pending sync, Atelier phase-rework direction~/.claude/.../memory/MEMORY.md— index updated with two new memory pointersworkstreams/stage1-roman-integration.md— What's Next updated for CONV-19 state
Decisions
Decision 1: Split Stage 1 into separate file BEFORE pushing to Roman. When Sergey raised context-window concerns mid-session («не грузить твоё контекстное окно сильно»), recommended split-then-push instead of push-then-split. Rationale: pushing the unsplit file would give Roman a URL that breaks 30 min later when split happens; better to give him one stable artifact. Rejected: push-and-refactor-after (Roman's review would land on broken-link version). Rejected: full 3-file split (stage-1 + stage-2 + future) — over-scoped; minimal effective split is just Stage 1 since that's what's in active work. Stage 2 stays inline in main file until Stage 2 work begins.
Decision 2: Keep «type» term for Partner Studio/Agency/Developer instead of renaming to «роли». Sergey queried the «Имеет один type (multi-select)» contradiction in §1 glossary. Considered changing «type» → «роли» but rejected because «5 ролей» (Owner/Admin/Manager/Editor/Agent) is already heavily-loaded — overloading the term creates worse confusion across §1 / §4.2 / Phase 1.3.X callouts. Final wording: «Имеет один или несколько types (multi-select): Studio / Agency / Developer — можно совмещать (например, VV = Studio + Agency)». Embedded VV example inline so reader doesn't need to scroll for clarification.
Decision 3: Add tier shape (Tier 1/Tier 2/Custom) qualitative scaffolding NOW vs deferring fully to ADR 0008. Sergey noticed glossary references «Tier 1+ ($)» without any tier definition. Considered: (a) add cross-link only (lighter); (b) qualitative descriptors with no pricing (committal). Picked (b) because cross-link just defers the same question one click; reader still hits «what's a tier?» when they get to §3. Qualitative descriptors don't lock pricing strategy — ADR 0008 can rename Tier 1/2/Custom → Starter/Studio/Agency/Enterprise (existing namespace from 1.7.8) freely. Added explicit «namespace TBD» flag in Open Questions surfacing this naming conflict for Roman to decide.
Decision 4: §4.3 Scoping — keep 4-bullet structure, polish wording (don't restructure into table). Sergey said «просто перепиши чуть понятнее» — that's polish, not restructure. Earlier draft proposed table+privileged-override-note layout but rejected per user preference. Final form: 4 bullets in same shape but with explicit axis lead-in («в каких проектах membership»), bullet 2 renamed to clarify it's a sub-case of bullet 1 (not separate category), bullet 4 «privileged» → «bypass scope» plus contrast (who DOES NOT bypass).
Decision 5: Atelier prototype is direction signal, not yet plan-locked. Sergey shared admin-quick-build-atelier-standalone.html with intent that Phase 1.2 onboarding will be reworked around this builder pattern (replaces demo-mode). Captured design opinion + plan implications (Phase 1.2 simplifies, Phase 1.5 MCP/AI tagging becomes load-bearing for launch) but did NOT modify plan content yet. Phase rework is queued as separate task post-split-sync. Atelier itself has 5 design issues to resolve before locking into plan (hero placeholder vs uploaded inconsistency; «AI tags» copy is technology promise dependent on Phase 1.5; «Object Builder» wording; SKIP semantics; «3 uploads» as full trial vs first wizard step).
Next Steps
- Visual sanity check
docs/plan/launch-plan-stage-1.htmlin browser — verify hero / topnav / sidebar / Foundational §1-§7 / Phase 1.x / Open Questions all render; confirm sidebar Stage 2/3/4 links cross-navigate tolaunch-plan-v3-full.htmlandlaunch-plan-future.html - Visual sanity check
docs/plan/launch-plan-v3-full.html(modified main) — Stage 1 stub callout renders, sidebar Stage 1 links go to stage-1 file - Sync 3 files to
~/code/offplan-online/preview/plan/: copylaunch-plan-stage-1.html(NEW),launch-plan-v3-full.html(modified),launch-plan-changelog.html - Commit + push preview repo:
plan(v4.4): split + cumulative — stage-1 extracted + Reservations removed + Closed pool + clarity - Phase rework — update Phase 1.2 (Self-Serve Onboarding + Quick Build) to align with Atelier direction (real-time subdomain validation, live-preview iframe, multi-device switcher, trial countdown, lazy conversion); decide Phase 1.5 priority (AI floorplate tagging may need 🟡 High → 🔴 Critical bump)
- Decide on Atelier flagged issues (hero consistency, AI tags copy, Object Builder wording, SKIP semantics, 3 uploads scope) — apply to mockup or note in plan
- Send to Roman with explicit pointers: §3 tier shape (his pricing input needed); §4.4 Closed pool divergence from his earlier Shared default approval; Open Questions section
- After Roman approves → ratification: fill
plans/permission-and-tenancy-model.md, ADR 0009 (tenancy/permission), ADR 0010 (stock allocation Closed/Open dichotomy), update ADR 0005 (Google + email/password) - Send Ilya tech-estimate spec with 7 unknowns (wildcard SSL, multi-tenant DB, token storage, email provider, Super Admin Panel reuse, real-time sync, VV API audit)
- Carryover from CONV-17: rotate Perplexity API key
Open Questions
- Will Roman accept Closed pool default? (he previously approved Shared default — surface the change explicitly when sending review URL)
- Tier naming namespace (Tier 1/2/Custom vs Starter/Studio/Agency/Enterprise) — ADR 0008 must pick one
- Atelier «AI tags pages» promise — ship-blocking (requires Phase 1.5) or copy-toned-down? Decide before Phase rework writes it into plan
- Does «3 uploads» represent full trial scope or just first wizard step? Affects Phase 1.2 spec and continuation flow design
- Hero rendering in Atelier — placeholder always, OR replaced by uploaded user image? UX consistency check
- Cyprus vs Abu Dhabi — memory says Cyprus decided, plan Open Questions still parks it; reconcile before ratification
Context for next session
- Priority of work order: Visual check → preview sync + push → phase rework → Roman send. Don't skip visual check before push — split was sed-based, no manual review yet beyond grep sanity.
- Sergey's batch-mode preference confirmed and active: read → accumulate → «применяй» / «да» / «меняй» → I apply. Don't apply edits real-time mid-read.
- Sergey said «не перебирай Stage 2/3/4 каждый раз» — saved as feedback memory. When grepping/reading plan content, restrict to Stage 1 sections (now in stage-1.html file); avoid pulling Stage 2+ context unless explicitly asked.
- Sergey said «просто перепиши чуть понятнее» = polish existing structure, not restructure into tables/new layouts. Validated pattern.
- Atelier file is at
/Users/diver/Downloads/admin-quick-build-atelier-standalone.html(1.6MB React-bundled, opens in browser, has trial/paid mode toggle in TweaksPanel). Standalone — can be sent to Roman/Ilya as single file. - Memory
plan_permission_tenancy_state.mdhas DIVERGENCE callout at top — Block C1 (Stock allocation) and C4 (Reservations) interview decisions are STALE vs current plan. Use plan as ground truth, not those blocks. Memory's other blocks (A, B, C2, C3, D, E, F, G) are still valid. - Preview repo is at
~/code/offplan-online/preview— separate git repo, public. CONV-15 v4.3 currently published; v4.4 + split not yet synced. - Notion MCP query is broken (
invalid_request_urlon data_source query) — Workstream relation skipped this session. Sessions row create works fine. - Author property couldn't be set — Notion rejects bot user as Author («Cannot mention bots»). Property left empty in CONV-19 row.