Validation Readiness Gate — 2026-05-05

This gate records what is ready for new validation runs and what remains blocked. It exists to prevent aggregate-only, placeholder, queued-job, or fallback-count artefacts from being promoted as hardware validation.

Scope

The current promoted hardware evidence remains limited to:

• April 2026 Phase 1 DLA parity on ibm_kingston, backed by raw counts, job identifiers, integrity checks, and the committed reproduction harness.
• Legacy March 2026 ibm_fez rows only where a committed artefact is named; these are not broad-advantage or frontier validation.

No new IBM validation run may be promoted until a preregistered manifest exists before submission and the count-to-statistic reproduction path is committed.

Local readiness checks run on 2026-05-05

Phase 1 DLA parity raw-count reproducer

Command:

PYTHONDONTWRITEBYTECODE=1 /home/anulum/.local/bin/python scripts/run_dla_parity_suite.py --verify-integrity

Result:

Check	Value
Dataset circuits	342
Sub-phase runs	4
Backend	ibm_kingston
Fisher statistic	chi2=123.4001, df=16
Significant depths at 0.05	7 / 8
Peak asymmetry	+17.48% at depth 6
Mean asymmetry	+9.25%
Published claims checked	57, all within tolerance
Classical reference	qutip, max_abs_leakage=0.000e+00
Classical zero-leakage invariant	true within 1e-10

Decision: the Phase 1 DLA parity dataset is reproducible from committed raw counts and remains the promoted raw-count hardware dataset.

Broad-advantage guardrail smoke check

Command:

PYTHONDONTWRITEBYTECODE=1 /home/anulum/.local/bin/python scripts/bench_quantum_advantage_classical_matrix.py --sizes 4,6 --output /tmp/scpn_quantum_advantage_guardrail_smoke.json

Result:

n	dim	classical ODE ms	exact dense ms	exact sparse ms	Rust ODE ms	GPU
4	16	11.159962	0.533185	1.245445	0.008856	unavailable
6	64	41.192475	47.946798	5.254019	0.060972	unavailable

Decision: the matrix path executes on the current commit without overwriting committed artefacts. This is a smoke check only; the committed May 3 matrix remains the provenance artefact for the broader n=4,6,8,10,12 evidence gate.

Promotion rules for the next IBM validation run

A new IBM result is not promotable unless all items below are satisfied:

Gate	Requirement
Preregistered manifest	Observable, backend, circuit family, shots, depths, reps, mitigation state, abort criteria, and statistical tests are committed before submission.
Submission record	Every submitted circuit batch records backend, job identifier, transpilation settings, shots, and timestamp.
Retrieval record	Raw counts are retrieved from IBM and stored with a manifest that binds each count set to a job identifier and circuit identifier.
Reproducer	A committed script recomputes every promoted statistic from raw counts and exits non-zero on tolerance or invariant failure.
Ledger row	docs/hardware_status_ledger.md names the evidence class and artefact paths before any summary page cites the result.
No fallbacks	All-zero counts, queued placeholders, aggregate-only JSON, and submit-now-retrieve-later files are quarantined, not evidence.

Current readiness decision

The project is prepared for local validation and preregistration work. It is not prepared to submit or promote a new IBM hardware validation run until the Phase 2 manifest is committed and the IBM credit or promo window is live.

Phase 2 preregistration update

The Phase 2 preregistered manifest is now recorded in docs/campaigns/ibm_phase2_preregistered_manifest_2026-05-05.md, with the machine-readable inventory in results/ibm_phase2_preregistration_2026-05-05.json.

The first live command is narrowed to blocks A+G only so the run spends the minimum QPU time needed for high-statistics n=4 replication plus readout baseline. Blocks B-F are deferred until the primary raw counts are retrieved and reviewed.

This closes the preregistration document gate only. IBM submission remains blocked until the credit window is live and the exact reduced --confirm-promo-active --skip B C D E F command is approved immediately before execution.

2026-05-05 live-attempt update

The first reduced A+G live attempt was cancelled after the hardware transpilation path produced deeper circuits than the reduced dry-run manifest. IBM job ibm-run-ca8b9612732b84dc is recorded as CANCELLED, with IBM metadata reporting 0 quantum seconds and 0 usage seconds.

This cancelled job is quarantined execution evidence only. It supplies no raw counts and closes no validation claim. The follow-up script now performs a hardware-backend transpilation precheck before submission and requires the live max depth to remain within the manifest budget.

2026-05-05 completed run update

The reduced A+G Phase 2 run completed on ibm_kingston:

• Main job: ibm-run-7da8644af35021fb, 600 circuits, 4096 shots, IBM-reported quantum seconds 660.
• Readout job: ibm-run-6f9990bba1d90a12, 12 circuits, 8192 shots, IBM-reported quantum seconds 27.
• Raw-count dataset: data/phase2_dla_parity/phase2_reduced_ag_2026-05-05T121357Z.json.
• Reproducer: scripts/analyse_phase2_dla_parity.py --verify-integrity.

The verified reproducer reports Fisher chi2 140.671952, Fisher p 3.773718e-20, and 6/10 depths significant at Welch p < 0.05. This promotes only the reduced n=4 DLA parity replication plus readout baseline. It does not promote n=6-12 scaling, GUESS calibration, broad quantum advantage, or frontier claims.

2026-05-05 B-C scaling continuation update

The preregistered B-C continuation completed on ibm_kingston:

• Job: ibm-run-1f46ebd0da8912ff, 280 circuits, 4096 shots, IBM-reported quantum seconds 305.
• Raw-count dataset: data/phase2_scaling_bc/phase2_scaling_bc_2026-05-05T124722Z.json.
• Reproducer: scripts/analyse_phase2_scaling_bc.py.

The verified reproducer reports mixed scaling evidence:

• n=6: Fisher chi2 46.531552, p 1.883218e-07, 2/4 significant depths, with negative significant asymmetry at depths 8 and 20.
• n=8: Fisher chi2 29.420107, p 2.675193e-04, 3/4 significant depths, with positive middle-depth sign.

This promotes only mixed n=6,8 scaling evidence. It falsifies a simple monotone scaling story and does not promote D-E larger scaling, GUESS calibration, broad quantum advantage, frontier, multi-QPU, or live-loop claims.