# Ey atom-gap cavity · final independent validation :::{admonition} No passing final artifact yet :class: warning The previous 500 nm Ey campaign is frozen at a gray beta-4 checkpoint and is not a final design. Its best conservative ringdown estimate was Q=130.91 with Purcell=3.039, while 92.6% of designable pixels remained gray. The optimizer is stopped. This page will be replaced automatically only after a literal binary result passes every independent validation gate below. ::: ## What the final page will require The final validator refuses a checkpoint unless it is marked complete, uses the Ey octant symmetry `(1, -1, 1)`, has exhausted the exact-binary projection schedule, and contains only literal air/dielectric endpoints. The entire hard 1 µm gap and all fixed-air cells must be exactly zero. Fresh validation then supplies: - a normalized-LDOS spectrum and fitted pole; - a 4.5 ps atom-Ey ringdown; - an independent 4.5 ps total-energy ringdown; - a late-time volumetric electric phasor and oriented mode volume; - radiation loss separated into x, y, and z boundary channels. The reported Q is the minimum of the three lifetime estimates. The fits must agree to within 20% and pass residual, time-window consistency, resonance alignment, and atomic-detuning gates. Purcell is computed from that conservative Q and the atom-oriented mode volume. A result is public only if Q≥1,000 and Purcell≥5. ## Fresh campaign launch boundary The future `multistart_v5` namespace has an immutable, CPU-only preflight manifest but contains no optimizer, screen, spectrum, or shifted-Q artifact. The preflight proves the 500 nm single extrusion, `(1, -1, 1)` Ey octant, 140×100 two-dimensional mask, exact 1 µm gap, six equal-budget cold starts, finite eight-hour/300-update/400-attempt resonance acquisition, a bounded beta schedule ending in a literal binary stage, source hash, protected frozen artifacts, disk capacity, and at least 10,000 MiB of free GPU memory. Its immutable contract fingerprint is `09d05fa3a651158d3bfde03152183665145c9a10aadf12e17eaf13ff397e83f6`. Earlier v1--v4 manifests are preserved as stale evidence and cannot launch the current source tree. Both the launcher and production driver require an explicit authorization variable, a nonempty campaign ID, and a single-owner process lock. When a beta transition requests a same-mask spectrum, the driver performs that audit and continues; it does not mistake `needs_audit` for success or failure. Only a genuinely complete checkpoint triggers fresh final validation, rendering, and publication. A pole found by the beta=2 cold-start screen is discovery evidence only. The shifted optimizer begins at beta=4, so handoff requires another independent spectrum of the actual beta>=4 mask. This prevents a material projection from changing between the audit and shifted-Q optimization. At exact completion, the same-stage spectrum and literal trust mask must share one SHA-256 fingerprint. The checkpoint is atomically canonicalized to that mask instead of its pending Adam proposal, and the final validator independently checks the fingerprint, Q/Purcell evidence, campaign, slab, symmetry, and exact geometry before allocating the GPU. The [live campaign dashboard](scratch_atom_ldos_dashboard.md) already follows this v5 namespace. Until the first FDTD evaluation exists, it shows only the immutable physical contract and an explicit pending state—never the stopped legacy mask or a made-up schematic. As the equal-budget screen runs, it fills with the actual beta=2 seed geometries and measured LDOS after each atomic, resumable seed boundary. The optimizer remains paused. Its eventual explicit launch command is: ```bash FDTDX_AUTHORIZE_SCRATCH_GPU=1 \ ./scripts/start_scratch_atom_campaign.sh multistart_v5 500nm ``` ## Reproducible final command ```bash FDTDX_SCRATCH_VARIANT=500nm \ FDTDX_SCRATCH_POLARIZATION=ey \ FDTDX_SCRATCH_CAMPAIGN_ID= \ FDTDX_SCRATCH_VALIDATE_FINAL=1 \ uv run python -m benchmarks.cases.invdes_scratch_atom_ldos.final_validation uv run python scripts/render_scratch_atom_final_validation.py \ --artifact benchmarks/artifacts/invdes_scratch_atom_extruded_500nm_ey__final_validation.npz ``` The renderer fails closed unless every persisted gate passes. A successful page will contain the exact fabrication mask and cross-section, full field slices, LDOS spectrum, directional loss, both ringdowns, metrics, named gates, and the command above.