PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: Merge-safe at the current head: the guard sits at the exact projection seam the ticket root-caused, all five AC rows are delivered or honestly scoped (AC4's end-to-end receipt is explicitly deferred to the stacked #15657 with named ownership), CI is green at 98ae3a4b94, and every load-bearing evidence claim I could falsify independently held up. No return cycle needed.

Peer-Review Opening: Thanks for the clean leaf split out of #15657 — root-causing the starvation to projection-by-existence rather than patching widths is exactly the right instinct, and the AC table's two-cause golden accounting made this auditable claim by claim. Notes below; nothing blocking.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: Ticket #15837 (full body incl. live-DOM probe receipts); changed-file list; exact-head source of src/dashboard/DockLayoutAdapter.mjs (projectEdgeZoneNode :580, projectEdgeBand :667, projectTabsNode :797, call-site region :602-644) at 98ae3a4b94; apps/agentos/view/fleet/cockpitDockDocument.mjs (all five right-zone items verified autoHidden: true); resources/scss/src/dashboard/Container.scss :184-201 (the 17.5rem cross-extent); repo-wide grep for edge-band JS consumers (zero outside the adapter — SCSS hooks only); golden history on dev (last regen 02f2572e19 / #15653, strictly preceding #15680's merge 8357f436ff / #15726). Prior-art sweep: Memory Core query on the adapter/edge-band decision space returned a clear miss — no ADR or prior session governs this seam beyond the adapter's own documented sizing rationale.
• Expected Solution Shape: Guard at the projection seam keyed on liveness, not existence — skip the band only for a tabs projection with an empty live flow; split-node bands exempt (committed geometry); call sites null-guarded; witnesses red→green for the new class plus pins for both unchanged classes. Must NOT hardcode zone identity (no FM-cockpit-specific branching); the skip must be derivable from the projected config alone. Test isolation: pure model→config projection witnesses, no harness needed.
• Patch Verdict: Matches. The guard at projectEdgeBand :678 discriminates on config.dockNodeType === 'tabs' && !(config.items?.length > 0) — derived entirely from the projected config, zone-agnostic; the four call sites (:607/:615/:623/:639) null-guard their pushes; split exemption holds because split nodes never enter the guard. The three witnesses map one-to-one onto the three expected classes, and witness (a)'s red-proof is mechanism-sound (on the unfixed adapter the band was pushed unconditionally, so band would be defined).
• Premise Coherence: Coheres with verify-before-assert (the fix is built on live-DOM probe receipts, and the PR body's own evidence claims are independently falsifiable — I re-derived the +280px and the golden-drift provenance rather than accepting them) and friction→gold (dogfood friction from #15657 converted into a contained leaf fix with the inversion receipt deliberately staged in the stacked lane). No value conflicts.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #15837
• Related Graph Nodes: Blocks-toward #15657 (vessel-narrow layout, stacked next); lineage #15649 (measurements that recorded the symptom); golden co-cause #15680; ADR 0029 (reveal rides the absolute overlay, never push); apps/agentos/view/fleet/cockpitDockDocument.mjs (the all-railed document).

🔬 Depth Floor

Challenge (three, all non-blocking):

1. The cockpit-vessel-314 byte-stability mechanism is under-explained against on-viewport arithmetic. The ticket's own probe places the band at x51 w280 — i.e., most of it (x51→314) is on the 314 viewport, and pre-fix it was the dominant visible occupant while the split sat at w0. Naively, removing it changes what that viewport paints (empty band region → visible split content). The diff does confirm the golden file is untouched (only 3 PNGs changed), and your 5/5 green receipt implies the rendered scene matched within the harness threshold — so the outcome is consistent, but the stated mechanism ("the band sat off-viewport at that width") reads imprecise. Accepted as author-owned receipt because (a) the fix's correctness is carried by the unit witnesses, not this golden, and (b) #15657's first move inverts this receipt to a fit-assertion — if the golden was silently stale, the stacked lane surfaces it within days on your own PR. If the real mechanism is "empty band paints nothing visually distinct at that width" (transparent strip, zero-state content), one clause in the PR body would close the loop for the record.
2. JSDoc nit: projectEdgeBand's @returns {Object} (:663) now lags the contract — null is a first-class return the call sites depend on. {Object|null} when you next touch the file (or fold into #15657's branch); not worth a cycle here.
3. On your explicit review question — the split-band exemption is correct and this PR should NOT be widened. A fully-railed split band still represents committed geometry (child extents, splitter positions the owner committed through operations); collapsing it projection-side would destroy state the owner never asked to destroy, and unlike the tabs case there is no rail-only truthful projection of a split. No follow-up ticket needed from me: if a real document ever produces a user-visible dead split, the repair is an owner-facing collapse operation, not a projection guard — that would be product scope on its own ticket, motivated by a live document, not hypothetically filed now.

Rhetorical-Drift Audit (per guide §7.4):

• PR description: "+280px arithmetic receipt" independently re-derived (sips: 1254→1534, exactly the band's 17.5rem); "all five right-zone items auto-hidden" verified in cockpitDockDocument.mjs :29-39; two-cause golden accounting verified via merge order (golden last regen #15653 precedes #15680's merge #15726) and prose provenance ('static roster · offline' arrived in FleetGrid.mjs via #15638; spine-banner zero-state prose via #15290, both before #15680 changed the explicitly-wired zero state).
• Anchor & Echo summaries: the new adapter constraint comment (:670-677) uses precise mechanical terminology; no snapshot anchors.
• [RETROSPECTIVE] tag: none attached by author; my own below is scoped to what shipped.
• Linked anchors: #15649 (symptom measurements), #15680 (independent drift co-cause), ADR 0029 (overlay-never-push) all establish what the body claims of them.

Findings: Pass — with the single mechanism-imprecision noted in challenge 1 (outcome verified at file level; not drift).

🧠 Graph Ingestion Notes

• [KB_GAP]: None — the ticket and adapter comments already document the liveness-vs-existence distinction at the seam where a future reader needs it.
• [TOOLING_GAP]: None — visual goldens remain author-owned non-CI evidence; no new gap surfaced.
• [RETROSPECTIVE]: The sizing-rationale inversion is the reusable insight: projectEdgeBand's fixed cross-extent exists to protect the center from an unsized band — and an empty band inverts that purpose, eating the geometry itself. "Existence is not liveness" at projection seams is the generalizable lesson; the guard derives the skip from the projected config alone, zone-agnostic, which is why the FM cockpit needed no special-casing.

🧱 Conciseness Rule — Collapsed-N/A Audits

N/A Audits — 📑 📡 🔗

N/A across listed dimensions: internal @protected static projection seam (no public/consumed config, MCP, core-API, or CLI surface → no Contract Ledger required); no OpenAPI surfaces touched; no new conventions, skills, tools, or primitives introduced (bug fix within the existing projection contract).

🎯 Close-Target Audit

• Close-targets identified: Resolves #15837 (newline-isolated at body top); single commit subject carries the parenthetical (#15837) — standard format, no magic keywords in commit body.
• #15837 labels: bug, ai — confirmed not epic-labeled.

Findings: Pass

🪜 Evidence Audit

• PR body contains an Evidence: declaration line: "L2 achieved (unit red→green on the exact projection seam + full-suite blast radius + regenerated visual goldens on the live app) → L2 required" — accurate: #15837's ACs are projection-shape + golden criteria with no runtime-AC surface beyond CI's reach.
• Achieved evidence ≥ required; the one residual (AC4's end-to-end 314 fit receipt) is explicitly listed under Post-Merge Validation with the stacked #15657 named as its owner.
• Two-ceiling distinction held: the 314 byte-stability is framed as deliberate staging, not unprobed surface.
• No evidence-class collapse: the golden regen is claimed as author-owned non-CI receipt, not promoted to CI framing; I validated the receipt's arithmetic independently rather than reconstructing the run.
• Deployment causality: no external/runtime receipt is used as a merge gate.

Findings: Pass

🧪 Test-Evidence & Location Audit

• Execution evidence: exact-head required CI green at 98ae3a4b94 (unit, integration-unified, components, lint ×3, lint-pr-body, Analyze, CodeQL — all pass; mergeStateStatus CLEAN). Author non-CI receipts present and current-head-appropriate: 484/484 dashboard + 508/508 agentos blast radius, visual 5/5 on regenerated goldens, 1/35 red-proof on the unfixed adapter (mechanism-verified, not just believed).
• Reviewer falsifier: no test re-run — named concern (challenge 1) targets a receipt's stated mechanism, not a behavior CI or the unit suite fails to establish; the arbiter is #15657's stacked inversion, already scheduled.
• Test location: test/playwright/unit/dashboard/DockLayoutAdapter.spec.mjs — canonical unit placement for the adapter; the createTabsBandModel fixture mirrors the FM secondary-rail shape without importing app code.

Findings: Pass

📋 Required Actions

No required actions — eligible for human merge.

📊 Evaluation Metrics

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

• [ARCH_ALIGNMENT]: 95 — Guard at the exact projection seam, discriminating on node type with the split exemption; zone-agnostic derivation from the projected config; call-site null-guards; the sizing-rationale comment makes the invariant self-documenting. −5: the method's @returns contract annotation lags the new null return.
• [CONTENT_COMPLETENESS]: 95 — Model fat-ticket body: AC table with per-row disposition + evidence, explicit deltas-from-ticket, two-cause golden accounting, evidence declaration, scoped post-merge validation. −5: same JSDoc lag; ticket's witness spec (a)/(b)/(c) maps exactly onto what shipped.
• [EXECUTION_QUALITY]: 95 — CI green at exact head; three witnesses covering all three band classes with a mechanism-sound red-proof; blast radius 484+508; golden arithmetic independently re-derived by this reviewer (1254→1534). −5: the 314 byte-stability mechanism is under-explained against on-viewport arithmetic (challenge 1) — receipt accepted, arbiter named.
• [PRODUCTIVITY]: 100 — All five AC rows delivered or honestly scoped; AC4's residual has a named owner and a stacked vehicle; no scope creep into #15657's layout work.
• [IMPACT]: 70 — Deletes a real dead 280px gutter on the FM dogfood flagship and unblocks #15657's vessel-narrow lane; framework-adapter correctness with contained blast radius. Not organism-core (memory/neural-link), hence below the 80+ band.
• [COMPLEXITY]: 55 — One guarded seam + four call sites + three witnesses + golden regen with two-cause accounting; moderate reader load, high context demand (the dock projection model).
• [EFFORT_PROFILE]: Quick Win — High ROI (unblocks a stacked layout lane, removes dead geometry, deletes no committed state) at contained complexity.

Clean leaf split, honest evidence, and a guard that will outlive the cockpit document that motivated it. Merging this unblocks #15657 — over to @tobiu's human gate.