docs(hba): close out §2026-07-04 A/B/C — Construction window + Person forward link + Leave-as-Resource

Design pass (AskUserQuestion) + build happened in bim-ootb this session (PR #645, awaiting merge).
Updates RESUME_HR_BIM_ASSET.md with the full closeout, PROGRESS.md's one-liner, and
HRBIMAssetGuide.md's "Jump straight to the ERP record" table + the three affected sections.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: red1oon

PROGRESS.md

@@ -23,8 +23,13 @@
 - §PCLOSE-RACE (T7-RACE's TOCTOU shape found again in `erp_period_close.js`, fixed) — PR #640 MERGED.
 - W-HBA-CLICKTHRU: first REAL click-through witness for HBA's 6 deep-links (found 3/6 dead — PR #641 MERGED);
   fixed 2 of them (missing AD_Window seed rows, PR #643 open); Tenancy/316 left to the Construction-window spec.
-- Spec'd 3 HBA next-steps, NOT started: Construction AD_Window, Person bidirectional link, Leave-as-Resource.
-  Full spec `prompts/RESUME_HR_BIM_ASSET.md` §2026-07-04; memory `project_hba_construction_window_person_linkthrough.md`.
+- HBA §2026-07-04 A/B/C ✅ DONE (design pass + build, same session): "Construction" AD_Window minted over
+  M_Warehouse (proto-cloned from window 139) + Find-panel Room-lens "iDempiere ↗" link · Presence forward
+  link into AD_User(108) · Leave TAKE now also compiles a real `S_ResourceUnAvailable` row ("Resource ↗" into
+  window 236). Witnessed (10/10 new + 4 extended, full 41-file suite zero regression) + 2 live Playwright
+  smokes (8/8 real HHS building, 9/9 fm_panel.html), bim-ootb PR #645 (`lane/hba-construction-clickthru`,
+  awaiting merge). Doc updated: `docs/HRBIMAssetGuide.md`. Full detail: `prompts/RESUME_HR_BIM_ASSET.md`
+  §2026-07-04 (closeout appended).
 
 ## Kernel op-log timebomb lane — T1/T2/T3/T6/T7 ✅ SHIPPED; T4+T5 remain
 - Findings: `prompts/KERNEL_TIMEBOMB_AUDIT_2026-07-03.md`. Shipped: T3+T6 (#623 v10) · T2+T1 roster (#630 v11)
@@ -49,12 +54,21 @@
   `mo_ifc_<name>` key fork before `_openBuffer`; diag-proven before/after, no regression on the two pre-existing
   witnesses (W-E2E-WALK 8/8, W-E2E-WALK-ALL 10/10). Pushed `lane/modeller-ifc-open` (`e6acb56`, synced with
   origin/main), user approved → **PR #642 opened 2026-07-04** (bim-ootb, awaiting CI/merge).
-- NEXT (once #642 merges): continue `prompts/ARC_GEO_FETCH_SPEC.md §NEXT` item 2 — onboard
-  Hospital/Clinic/LTU/HHS_Office as Modeller residents + migrate SH/DX/SC into the canonical `IFC/` folder.
-  3DGrid (Move Grid / stretch-recompose) has its own coverage in `witness_e2e_move.js` (per
-  `feedback_test_real_user_path_not_seams.md`'s templates) — NOT re-verified this session (out of scope: this
-  session's walking tests targeted STR/MEP disc-walk only, not grid-move). If grid-move needs a fresh
-  cross-resident pass too, that's a separate follow-up, not covered by W-E2E-WALK-IFCOPEN.
+- 3D-Grid cross-resident pass DONE (2026-07-04, user follow-up question): `Bonsai.grid` is a DECOUPLED
+  authoring grid — no production code wires an opened building's own STR-walked grid into it (grepped clean),
+  so "test grid-stretch on building X's real structural grid" isn't meaningful; what generalizes IS the
+  author-a-grid→drag→recompose mechanism after a real (not just Duplex) open. New
+  `modeller/tests/witness_e2e_gridstretch_multi.js` — 20/21, **1 known RED**: on SampleCastle (`.db`-open),
+  `#b-clear` resets the scene+op-log but not `str_walker_outliner.js`'s own STR-walker module state, so a
+  later grid-drag on an unrelated synthetic wall still fires SampleCastle's real STR re-walk, whose ops
+  collide on `kernel_ops.id` and roll back (SAFE — all-or-none, no corruption, just a loud rejected toast +
+  console errors). Root cause traced (step-tagged diagnostic), confirmed reproducible 2×, NOT an async-timing
+  artifact. **NOT fixed** — the fix is a `#b-clear` state-reset-contract design call (reset every walker
+  module, not just THREE+oplog?), left open, pushed as-is (`834fbc4`) so the PR carries an honest red witness
+  rather than a silently-passing green one.
+- NEXT (once #642 merges): (1) decide + fix the `#b-clear`/STR-walker state-leak above; (2) continue
+  `prompts/ARC_GEO_FETCH_SPEC.md §NEXT` item 2 — onboard Hospital/Clinic/LTU/HHS_Office as Modeller residents
+  + migrate SH/DX/SC into the canonical `IFC/` folder.
 
 ## Codebase quality audit (2026-07-02) — TRIAGED 2026-07-03; §2/§5 DONE (bc #20, bim-ootb #618) → archived
 - ⛔ BLOCKED (user call): are `migration/DV_*_rules.sql` mined-rule files EXEMPT from append-only, or enforce?

docs/HRBIMAssetGuide.md

@@ -90,6 +90,10 @@ is variations on it.
    automatic level-of-detail ladder (dot → mini → full) so a full floor never turns to soup. The screenshot above
    catches the ladder mid-transition: a full-figure cluster close to camera, mini-dome clusters further back.
 4. Headcount comes from **signed check-ins** — a room with no check-in has no avatar (never a faked person).
+5. The **Presence roster** (the drawer's own list of every check-in for the period) carries an **open ↗** link
+   per person straight to their `AD_User` record — the first bidirectional click-through for a Person in this
+   module (every other pane already had one). No reverse direction (an ERP record flying you back into the 3D
+   model) exists yet — that's a genuinely new capability, tracked separately, not built here.
 
 ### Spot equipment that needs service
 1. Open the drawer → tap **Assets / IoT**.
@@ -139,8 +143,10 @@ from the real AD dictionary, never guessed:
 | Dashboard → Resources | Resource | `S_Resource` |
 | Payslip → concept line | Payroll Movement | `HR_Movement` |
 | Leave → unpaid entry | Payroll Concept Catalog | `HR_Concept` (the "Leave without pay" concept it feeds — Leave has no native table of its own) |
+| Leave → "Resource ↗" | Resource | `S_Resource` (a leave TAKE also compiles a real `S_ResourceUnAvailable` blackout on the employee's own resource — see [Leave — balance & statement](#leave--balance--statement)) |
 | Tenancy → subscription | Subscription | `C_Subscription` |
 | IoT → billing line | Sales Order | `C_Order` |
+| Presence → roster row | User | `AD_User` (the employee's own identity record — see [See who is physically here right now](#see-who-is-physically-here-right-now--with-people-you-can-walk-up-to)) |
 
 No link appears on a row that doesn't (yet) resolve to a real record — the same non-invent discipline as
 everywhere else in this module: an absent link is honest, never a dead one.
@@ -191,6 +197,11 @@ everywhere else in this module: an absent link is honest, never a dead one.
    genuine addition, not a reinvention of something that already existed. So an unpaid row's **open ↗** doesn't
    point at a fabricated "leave record" window; it points at the real **"Leave without pay" payroll concept**
    the unpaid days feed into once payroll runs — the honest, real thing an unpaid entry actually compiles onto.
+3. Leave is *also* a resource-availability fact, not just a payroll deduction — a taken leave marks the employee
+   unavailable the same way a room's maintenance blackout does, via the real native `S_ResourceUnAvailable` table
+   (the child "Unavailability" tab of the Resource window). A **Resource ↗** link at the top of the pane opens
+   the employee's own Resource record, where that blackout shows up in the standard iDempiere Resource Schedule
+   — no separate leave-record UI to maintain, the same window Dashboard's Resources already deep-link into.
 
 ### Tenancy — AD compile
 1. Open the drawer → tap **Tenancy / AD**. Every lease and strata charge in the building compiled onto real
@@ -202,6 +213,14 @@ everywhere else in this module: an absent link is honest, never a dead one.
 2. Row click flies the camera to that unit (the same shared fly-to used by the Presence roster above); the
    **open ↗** link is separate and deep-links straight to the `C_Subscription` record. A record whose unit
    doesn't resolve to a real room in this building is **skipped**, never fabricated — the footer says so.
+3. The building itself compiles onto `M_Warehouse` — but the only *stock* iDempiere window over that table is
+   139 "Warehouse and Locators", warehouse-industry language for what's conceptually a building. A second,
+   purpose-named window — **"Construction"** — now sits alongside it over the *identical* `M_Warehouse` row
+   (the same convention iDempiere itself already uses for `C_BPartner`, which has 10 distinct windows over one
+   table). Window 139 is untouched; "Construction" is just a role-appropriate second door onto the same record.
+   In the Viewer, tapping a room or storey in the **Find** panel's Room lens surfaces an **iDempiere ↗** link to
+   this building's Construction record (building-grain — every room in one building shares the same warehouse
+   record, there's no per-room window).
 
 ### BIM BOM — assembly & recipe
 1. Open the drawer → tap **BOM**. Every room in the building is an **assembly**: the room itself plus

prompts/RESUME_HR_BIM_ASSET.md

@@ -1693,3 +1693,68 @@ resource-availability record, the same way a room's blackout does.
 
 Neither A, B, nor C is started. Log here, not in `MEMORY.md`'s per-session churn, so the next session that
 touches HBA finds this without re-deriving it.
+
+## ▶ 2026-07-04b — A/B/C ✅ ALL DONE (the fresh Sonnet-with-user design pass + build this section asked for)
+
+The design pass happened first (AskUserQuestion, 4 questions covering the real forks left open above), THEN
+implementation, in `/tmp/wt-hba-a2c` off fresh `origin/main`, branch `lane/hba-construction-clickthru`, bim-ootb
+PR **#645** (pushed, not yet merged — this repo's normal PR flow, `main` has required-status-checks).
+
+**Design answers that shaped the build (supersede the open questions above):**
+- A coexists with the floating panes (doesn't replace them). "Further BIM model tabs" was a misreading —
+  Construction is purely an **iDempiere-side** AD_Window; the Viewer-side click-through is the actual ask,
+  answered as "in Find, at a storey/room, zoom to iDempiere."
+- C: don't invent a new Viewer panel — the blackout must show up in the **standard iDempiere Resource
+  Schedule** (native tab 416 under window 236). This session's Leave-side scope = click-through only.
+- B: forward link only (Presence → AD_User). Reverse (ERP record flying back into the 3D model) explicitly
+  deferred — genuinely new mechanism, no precedent anywhere in the codebase, needs its own spec.
+
+**A — "Construction" AD_Window.** `scripts/seed_hba_construction.js` (NEW): proto-clones window 139's
+AD_Window + all 5 AD_Tab rows + all 60 AD_Field rows under a new id band (`CONSTR_BASE=7,800,000`), Name→
+'Construction', window 139 byte-identical/untouched. Idempotent (2nd run = 0 additions), self-witnessed (10
+checks: tab/field/column parity, `ad_parser.js getWindow()`'s own query resolves it — no manifest edit needed
+since it's a brand-new id). `hba_lens.js AD_WINDOWS.CONSTRUCTION=7800000`. `navigate_find.js` gained a second
+shared link element (`#find-construction-open`, mirrors the existing `#find-erp-open` Project-link pattern)
++ `_surfaceConstructionLink(guid,label)`, wired into both `_roomSelect`/`_roomGroupSelect` — building-grain
+honestly (every room/storey in one building shares the same warehouse record; HBA absent → honest no-op,
+never fabricates a warehouse id). Live smoke `viewer/tests/poc_construction_link_live.js` (NEW, Playwright,
+real HHS_Office_Federated + Find's Room-lens axis toggle + a real tree-row click): 8/8, confirms
+`window=7800000&record=990000` (the real seeded warehouse id), 0 pageerrors.
+
+**B — Presence forward link.** `viewer/hba_lens.js openPresenceDrawer`: each roster row (was a `<button>`,
+now a `<div>` so a nested `<a>` is valid) gains an "open ↗" to `AD_WINDOWS.USER=108` (AD_User), record=
+`official.ad_user_id` (always populated on every real `MODELS.Official` row, unlike the nullable
+`c_bpartner_id`) — `stopPropagation()` so the link doesn't also trigger the row's `flyToZone`. Witnessed:
+`witness_erp_govern_wire.js` WIRE6 (+1, 9/9 total). Live smoke (below, shared file with C): 9/9.
+
+**C — Leave-as-Resource.** NEW `hr_bim_asset/ad_leave.js`: `toUnavailableRow(takeOp, s_resource_id, seedId)`
+projects a signed `TAKE` op onto the real `S_ResourceUnAvailable` columns (datefrom=the TAKE's own date,
+dateto=datefrom+(days-1), pure arithmetic on the supplied ts — never `Date.now()`); `compileLeaveUnavailability`
++ `readUnavailability` (lens read, mirrors `ad_attendance.readPresence`'s JOIN shape). `scripts/seed_hba_erp.js`
+extended (§8, before the self-witness block): seeds 6 real rows (EMP001/EMP002 × 3 demoLog TAKE ops each),
+proto-cloned from the stock row (id=100 "Training class") for exact column parity; self-witness confirms the
+JOIN through Resource↔AD_User is lossless (6/6). `viewer/hba_leave.js renderStatement` gained a "Resource ↗"
+link to `AD_WINDOWS.RESOURCE=236`, record=`A._hbaEmpResourceMap[empId]` — a NEW standalone export off
+`hba_lens.js _regovern` (the same real-row-sourced map `_hbaAttendanceSpec` already builds internally, now
+also exposed for reuse); honest no-link when ungoverned (no ERP db loaded → no map → no fabricated id).
+Witnessed: `witness_ad_leave.js` (NEW, 10/10 — non-invent gate, date-range math, honest-skip, determinism) +
+`witness_leave_pane.js` LP5b-d (+3, 13/13 total — absent-when-ungoverned, present-with-correct-href-when-
+governed). Live smoke `viewer/tests/poc_presence_leave_link_live.js` (NEW, Playwright, `fm_panel.html`): 9/9 —
+Presence roster's 3 rows all link to `window=108`, Leave pane's "Resource ↗" links to `window=236&record=
+990109` (EMP001), 0 pageerrors.
+
+**Regression:** full 41-file HBA node witness suite (was 40 baseline + `witness_ad_leave.js`) — zero failures
+throughout the build, checked after every file edit.
+
+**Docs:** `docs/HRBIMAssetGuide.md` updated — the "Jump straight to the ERP record" table gained Leave→Resource
+and Presence→AD_User rows; new paragraphs in the Presence, Leave, and Tenancy sections explaining the
+Construction window / Resource-Unavailable / forward-link additions. **Not done:** no new screenshots this
+session (the 3 additions are click-through mechanics, not new visual panes — flagged rather than silently
+skipped, matching this doc's own established convention for partial doc passes).
+
+**Genuinely still open (not started, not this session's scope):**
+1. **B's reverse direction** (ERP record → fly back into the 3D model) — zero precedent anywhere in the
+   codebase; needs its own spec (what triggers it, cross-tab/window messaging or same-tab-only) before any code.
+2. Recapture `hba_fm_drawer.png`-style screenshots for the 3 new affordances (Construction link, Presence
+   open↗, Leave Resource↗) if this guide gets a fresh screenshot pass.
+3. bim-ootb PR #645 needs a human merge decision (branch protection + required status checks on `main`).