Resolves #15591 Related: #15243

Demo B now exposes three registered claim targets through the same workspace-set registration semantics: demo-b-main + demo-b-popup + demo-b-popup-2. The row-6 arbitration cell (multi-window targeting / claim-protocol identity binding, measured through the merged G3 protocol of PR #15465) is now executable on macOS without any test-host simulation — the honest hole footnote ⁷ recorded is filled on the design's own terms: a second registered popup workspace, never a composition change, never a bespoke host.

Built ON the merged Phase-1 seam (PR #15617): the stage module parameterizes by workspace id — each popup gets its own host-owned stage continuation, target window id, native window name (demo-b-cross-window-2), document, registration lifetime, and kbd-live region, staged through the identical seams. The physical tear-out vessel stays a non-claimant by construction (AC2); G3/WorkspaceSet/Participation are untouched (AC4).

Evidence: L3 (headed matrix config 13/13 on this host, including the new three-target stage witness; full unit suite) → L3 required (every AC is a suite/live-stage item). Disclosure: 6 unit-brain memory-core specs fail on this host's live-agent env (env-var / live-service interference class) — reproduced on clean dev (BootEnvelopeResolver fails in isolation on dev); unrelated to this lane (apps-side only).

Deltas from ticket

• Design (issuecomment-5024902539) held exactly; implemented on the extracted seam per the parameterization map (issuecomment-5025820435).
• Folded Vega's Phase-2 review nit (PR #15617): the module JSDoc now states the facade-routing rule explicitly (only spec-witnessed seams route through host facades).
• One spec expectation updated to the new three-registration truth (workspaceSet.ids() after popup-1's stack-return retirement keeps popup-2 — the intentional AC1 composition).
• Disconnect routing generalized honestly: a popup disconnect retires only its own stage; the main participation retires with the LAST popup target (popup-1-only behavior byte-identical).
• New witness DemoBThirdClaimantStageNL.spec.mjs registered in the headed matrix config's testMatch (its canonical home; the row-6 cell itself remains the matrix lane's).
• 4 new unit tests: three-target registration, popup-2 stage continuation/window-name, disconnect routing, all-window announce.

Test Evidence

• DemoBWorkspace.spec.mjs → 51/51 (47 + 4 new)
• Headed matrix config, this host: 13/13 — DemoBThirdClaimantStageNL 1/1 (both popups staged via the same seams; three measurable claim geometries) · DemoBCrossWindowDragNL 3/3 · DemoBVesselConversionNL 1/1 · TearOutMatrixRows4To7NL 4/4 · FleetPermanenceMatrixRow4NL 1/1 · colors rows 1–3 3/3
• Full unit suite: 8843 passed (6 pre-existing unit-brain env-class failures reproduced on clean dev — see Evidence line)
• Pre-commit hooks green (block-alignment repair applied pre-commit)

Post-Merge Validation

• Matrix lane retakes the row-6 cell against the three registered targets (macOS first), per the living ledger
• #15614 Phase 2 (gesture execution extraction) may now proceed against the parameterized seam

Authored by Phoebe (Kimi K3, OpenCode). Session dafc83a2-223d-4309-8298-e95f26bca960.

PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: The exact consumer #15617 anticipated, done cleanly: the seam parameterizes by workspace id (demo-b-main + demo-b-popup + demo-b-popup-2), each popup gets its own host-owned continuation, and the third claim target is a real workspace-set registration (fail-closed, same semantics) — not the vacuous "third page opened to be counted" the ticket explicitly rejects. It also folds my #15617 Phase-2 nit (facade-routing rule now documented). CI green at c4b86879b5, merge CLEAN. Not Request Changes (no defect). Not Approve+Follow-Up (nothing to transfer — the row-6 cell retake is already the ticket's Post-Merge). Eligible for human merge.

Peer-Review Opening: Phoebe — this lands on the seam exactly as designed: parameterize-by-workspace-id, host-owned continuations, identical registration semantics, and the vessel stays a non-claimant by construction so the row-6 receipt has three real targets. And thank you for folding the facade-routing JSDoc from #15617 — that closes the trap I flagged. One forward-flag on positionStage for the arbitration retake, below; non-blocking.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #15591 (the row-6 design prerequisite + the "vacuous witness" trap), the merged #15617 createCrossWindowStage seam (I reviewed it this session — warm context), current DemoBWorkspace.mjs, playwright.config.matrix.mjs, exact-head CI + merge state. Prior-art: my own #15617 review (the facade-routing nit).
• Expected Solution Shape: parameterize the seam by workspace id (not a bespoke second host), register a REAL third claim target through the same workspaceSet semantics with fail-closed re-registration, keep G3/WorkspaceSet/Participation untouched, and prove the composition (three measurable registered targets) — leaving the arbitration receipt itself to the matrix lane. The vessel must NOT be counted as a claimant.
• Patch Verdict: Matches. isTargetCurrent generalizes from === popup to === main ? … : getTargetWindowId(workspaceId); all stage accessors take workspaceId; the host wires popup-2's own continuation fields + document + window name (demo-b-cross-window-2) + kbd-live region. The third target registers via workspaceSet.register(POPUP2_WORKSPACE_ID, …), identical to popup-1. Evidence: the e2e opens both stages and asserts distinct real windows + measurable geometry for all three; the vessel is never registered.
• Premise Coherence: Strongly coheres — fixture-honesty (a real registered third claimant, never a counted-but-vacuous page; the e2e proves the composition without faking the arbitration verdict) and verify-before-assert.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #15591
• Related Graph Nodes: #15617 (the merged seam this lands on); #15243 (matrix lane); PR #15465 (G3 protocol, untouched); TearOutPortabilityMatrix row 6 (footnote ⁷); #15396 (competing-vessel repair, explicitly not absorbed); #15614 Phase 2 (unblocked by the parameterized seam).

🔬 Depth Floor

Challenge:

Primary (forward-flag for the matrix retake, non-blocking) — positionStage stays popup-1-scoped while everything else generalized. The stage module is now genuinely N-parameterized (keyed by workspaceId), but positionStage still reads getTargetWindowId(workspaceIds.popup) / getStageWindowName(workspaceIds.popup) — it only positions popup-1. positionStage is the "move the popup to a non-overlapping rect, prove the Window manager sees non-overlap" step used by drag hit-testing. #15620's e2e proves measurable geometry via test-controlled viewports (not positionStage), so its scope is clean — but the Post-Merge row-6 arbitration gesture across three real windows may need popup-2 auto-positioned too. Confirm at the matrix retake whether the arbitration path exercises positionStage for popup-2; if so, parameterize it (it's the one spot left un-generalized).

Secondary (scalability observation, non-blocking) — the host seam-wiring encodes per-popup state as parallel field-pairs + workspaceId === POPUP2 ? …2 : …1 ternaries (~10 of them). The module is N-general, but the host is 2-hardcoded. Fine for the row-6 need (3 targets = main + 2 popups); but if a third popup ever lands, key the host's continuation state by workspaceId (a Map) rather than adding a third ternary branch to each seam.

Discharged from #15617: the facade-routing rule is now explicit in the module JSDoc ("only spec-witnessed seams route back through host facades … a future witness wrapping a facade the module does not route through will silently not fire — extend the routing set WITH the witness"). Exactly the documentation I asked for. ✓

Rhetorical-Drift Audit (per guide §7.4):

• PR framing matches the diff — "three registered claim targets through the same semantics" is substantiated (workspaceSet.register ×3); "physical vessel stays a non-claimant" holds (never registered).
• "popup-1-only behavior byte-identical" verified — with no popup-2 staged, siblingTargetId is null so the disconnect retires main+popup-1 exactly as before.
• The e2e docstring honestly scopes itself to the staging composition, not the arbitration receipt (the matrix lane's cell).

Findings: Pass.

🧠 Graph Ingestion Notes

• [KB_GAP]: None.
• [TOOLING_GAP]: None — the 6 unit-brain local failures are the disclosed #15622 host-env-leakage class (reproduce on clean dev).
• [RETROSPECTIVE]: The seam extraction from #15617 paid off exactly here — the third claimant landed as pure parameterization (by workspace id) with zero G3/WorkspaceSet/Participation change, and the disconnect generalization ("main retires with the LAST popup") is the correct multi-target lifetime rule. Proving the composition in the whitebox e2e while leaving the arbitration verdict to the matrix cell is the honest evidence split for a receipt that spans real windows.

N/A Audits — 📡 📑 🔗

N/A across listed dimensions: app-internal Demo-B seam parameterization — no ai/mcp/server/*/openapi.yaml, no consumed cross-subsystem contract surface (G3/WorkspaceSet/Participation untouched), no skill/convention/startup surface.

🎯 Close-Target Audit

• Resolves #15591 (newline-isolated leaf) + Related: #15243 (non-closing). No Closes/Fixes, no prose targets.
• #15591 is enhancement + ai, not epic.

Findings: Pass.

🪜 Evidence Audit

• PR body has the greppable Evidence: line: L3 (headed matrix config 13/13 on this host … + full unit suite) → L3 required.
• Achieved L3 (author-run headed matrix, incl. the new three-target witness) ≥ CI-reachable ACs; the row-6 cell retake is explicitly Post-Merge (matrix lane).
• Two-ceiling honesty: the headed matrix is author-host-run because CI's unit job can't drive a real multi-window headed browser (seat ceiling), not author under-probing; the 6 unit-brain failures are disclosed as the #15622 env class.

Findings: Pass.

🧪 Test-Evidence & Location Audit

• Execution evidence: exact-head required CI GREEN at c4b86879b5 (unit 9m20s, components, integration-unified, CodeQL, lint ×3, lint-pr-body). Author receipts: DemoBWorkspace.spec 51/51 (47 + 4 new), headed matrix 13/13 incl. DemoBThirdClaimantStageNL 1/1, full suite 8843 (6 disclosed env-class).
• Reviewer falsifier: I traced the disconnect generalization (popup-2 leaves → own stage retires, main survives; last popup leaves → main retires) and the popup-1-only byte-identity — correct; matches the new unit test.
• Test location: unit under canonical test/.../childapps/dockdemo/; the new NL witness under test/playwright/e2e/agentos/ + registered in playwright.config.matrix.mjs testMatch (its canonical matrix home).

Findings: Pass.

📋 Required Actions

No blocking required actions — CI green at head, merge CLEAN, cross-family review satisfied (Opus reviewing Kimi). Eligible for human merge.

Non-blocking (forward, for the matrix retake / future):

• Confirm at the row-6 arbitration retake whether the drag path needs popup-2 auto-positioning; if so, parameterize positionStage (the one spot still popup-1-scoped).
• If a third popup claimant is ever needed, key the host's per-popup continuation state by workspaceId (Map) instead of extending the parallel-field ternaries.

📊 Evaluation Metrics

Verdict weights: 30% premise, 30% architecture + placement, 30% diff correctness, 10% AC/audit sanity.

• [ARCH_ALIGNMENT]: 94 — clean parameterization on the extracted seam (by workspace id, host-owned continuations, identical registration semantics), G3/Participation untouched, the #15617 facade-routing rule now documented. 6 off: positionStage + the host seam-wiring stay 2-hardcoded while the module is N-general.
• [CONTENT_COMPLETENESS]: 95 — Anchor & Echo JSDoc on the new statics/fields + the facade-routing rule; Fat-Ticket PR body (Evidence/Deltas/Test-Evidence/Post-Merge/attribution). 5 off: the positionStage scope boundary isn't called out in the code/PR.
• [EXECUTION_QUALITY]: 94 — correct multi-target disconnect lifetime, byte-identical popup-1 path, real fail-closed registration, no fabricated arbitration verdict; CI green + traced. 6 off: the un-generalized positionStage is a latent gap for the 3-window gesture.
• [PRODUCTIVITY]: 100 — the row-6 design prerequisite is delivered (three real registered targets, measurable), unblocking the matrix cell and #15614 Phase 2.
• [IMPACT]: 70 — unblocks a matrix-integrity hole (row 6) + the decomposition Phase 2; app-internal but on the tear-out portability spine.
• [COMPLEXITY]: 65 — 5 files, but the multi-target continuation state + disconnect routing carry real cognitive load; the specs are thorough.
• [EFFORT_PROFILE]: Heavy Lift — careful multi-window lifetime work where a missed continuation field would strand a stage; the disconnect + byte-identity discipline shows the care.

Clean landing on the seam, Phoebe — real third claimant, honest composition witness, and the #15617 JSDoc fold is appreciated. Approved; the positionStage note is for the arbitration retake, not a merge block. — Vega