ATLAS — IBM PHOENIX / SP 800-90B / LGI / DTC — EVIDENCE PACKAGE
================================================================
Date: 2026-09-09

1. SP 800-90B RESTART CAPTURE
Job: dag8u5vi3e6s738m31q0
Backend: ibm_phoenix (120 programmable qubits)
Matrix: 1000 restarts x 1000 eight-bit samples = 1,000,000 bytes
init_qubits: true; default rep_delay: 1 us
Matrix SHA-256: deb23d8fa9fcc18838f418ccf53fe60a6688329c197b61d4cd844beee6edeec4
ea_restart non-IID with H_I=7.287800:
  Sanity Check: PASS
  Validation Test: PASS
  H_r=6.638383; H_c=7.876484
  final min(H_r,H_c,H_I)=6.638383 bits/input byte

2. CONDITIONED 8-BIT OUTPUT MARKING
At a 2:1 input/output ratio, the conservative restart result supplies
2 x 6.638383 = 13.276766 estimated entropy bits per 8-bit output byte.
Therefore a vetted conditioning component may claim at most the output-width
ceiling of 8.000000 bits/output byte. This does NOT mean raw bytes have 8 bits.
It is an entropy-budget statement, not a NIST certificate or product approval.

3. SAME-CALIBRATION COMBINED PHOENIX JOB
Job: daga50hhvn6c73cqamdg
152 PUBs; 296,000 executions; measured quota consumption: 9 seconds.
Blocks: LGI/NSIT pilot, Floquet DTC pilot at 4Q and 12Q, 8Q QRNG replica.
QRNG replica SHA-256: 832e2ac7b7277225b2738b7a06903b5914a78e00729d7b1571abaec8dc6dd3a1
Replica non-IID (100,000 samples; below the recommended 1,000,000 warning):
  H_original=7.621810
  min(H_original,8*H_bitstring)=6.713106 bits/input byte

4. LEGGETT-GARG / NSIT PILOT
theta=pi/3: K3=1.3935, 16.30 sigma above 1; captured NSIT contrast fails at 4.30 sigma.
theta=pi/4: K3=1.2980, 12.62 sigma above 1; captured NSIT contrast is 1.66 sigma.
theta=pi/2 negative control: K3=0.8960; no violation; NSIT contrast 1.25 sigma.
Interpretation: strong, falsifiable LGI evidence and a clean candidate at pi/4.
Not yet a complete semi-device-independent certification: the pilot captured
only one NSIT contrast per theta rather than all three required equalities.

5. FLOQUET / DISCRETE TIME-CRYSTAL PILOT
All 4Q and 12Q curves peak at FFT bin 8/17=0.470588, the closest available
bin to 1/2. Alternating order parameters:
  4Q: epsilon .02 -> .8647; epsilon .17 -> .3857
 12Q: epsilon .02 -> .7532; epsilon .17 -> .3031
Interpretation: subharmonic response compatible with frequency locking against
the specified trivial-precession comparison. It is not, by itself, proof of a
thermodynamic DTC phase; interaction-off, disorder-off, denser epsilon, longer
time and finite-size controls remain necessary.

DIRECTORY MAP
01_SP80090B_RESTART     Raw matrix, metadata, official restart transcript,
                        and ten-block stability analysis.
02_PHOENIX_COMBINED_JOB Full IBM counts, QRNG binary, plan, metadata, LGI/DTC
                        analysis and ea_non_iid transcript.
03_SCRIPTS              Capture and analysis programs.
04_NIST_TOOLS           Local ea_restart and ea_non_iid executables used.

No statement in this package represents certification or endorsement by NIST,
IBM, Quantinuum, Google, or any government body.