Repository navigation
Phase 3: CORRELATED-NOISE-01 Stage 1 hardware results and analysis amendment A-3 - #24
Merged
Merged
Conversation
…, committed after it) Amendment A-3 to CORRELATED-NOISE-01 Stage 1. It adds local 1/f and local-OU null models, a seeded random train/held-out split, the DD pulse timing the circuits use, echo-only held-out calibration and a stagger-change held-out score. It is validated on synthetic counts (correlated_noise_01_A3_validation.json). These files are committed exactly as they were when A-3 was recorded, at 2026-11-16T23:02:15Z (the amendments text at 23:02:56Z). The CN01 job was created later, at 23:34:53Z. The files were not committed or pushed before the job: it was submitted from another checkout, whose runner does not check for A-3. The record pins the lock, the QPY and the analysis code by SHA-256, and all of them still match. The only change made after the job is a CI import guard in tests/test_phase3_cn01_a3.py, because CI does not install qiskit-ibm-runtime. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rakesh Job dbtp67dav06s73cfbsm0: 244 PUBs x 2,048 shots, 139 s of QPU time charged (149.3 s estimated). The owner submitted it from another checkout of main, using the locked QPY and lock. A read-only re-query matched the saved counts to IBM's stored result for all 244 PUBs. Both analyses are reported, and they agree: no non-local noise mechanism is supported, and the Stage-2 gate is closed. - As locked: H3 is selected in-sample (dBIC 10,263) but fails the held-out test (p = 0.82), so the preregistered kill criterion falsifies the transferable H3 claim. A-3's validation shows that this analysis selects H3 on purely local noise. - A-3: H0 is selected, so the outcome is local noise only. - C_DFS = +0.43 pp, 95% CI [-0.13, +1.00] pp. No contrast is significant. A-3 was hashed before the job but published after it, so the owner decides which analysis is primary. See CORRELATED_NOISE_01_RESULTS.md and the note added to PHASE3_PREREG_AMENDMENTS.md. The raw counts have the IBM Cloud instance and plan IDs removed (listed in redacted_fields). phase3/cn01_a3_bound_diagnostic.py is diagnostic only and not preregistered: it shows that A-3's 40 kHz white-floor cap on three models changes none of its decisions. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
CORRELATED-NOISE-01 Stage 1 has run on
ibm_marrakesh:dbtp67dav06s73cfbsm0, 244 PUBs × 2,048 shots, 139 s of QPU time charged against a 149.3 s estimate.main, using the locked QPY and lock. Nothing was resubmitted: IBM lists only this job and TESSERACT-MAP-02's since 21:00Z.Verdict: both analyses agree. No non-local noise mechanism is supported (correlated dephasing, correlated memory, spectator ZZ or pulse collision), and the Stage-2 gate is closed.
The locked analysis's H3 selection carries no information about correlation: A-3's synthetic validation shows it selects H3 on purely local 1/f or OU noise in 100% of replicates. C_DFS is +0.43 pp, with a 95% CI of [−0.13, +1.00] pp, and no contrast is significant.
Full write-up:
docs/phase3/CORRELATED_NOISE_01_RESULTS.md.Needs the owner's decision: which analysis is primary
A-3 was hashed at 23:02:15Z, before the job was created at 23:34:53Z. But it was not committed or pushed before the job, because the job came from a checkout whose runner does not check for A-3. That means A-2's own condition ("not submitted until A-3 is committed and pushed") was not met, and the evidence that A-3 predates the data is local only. The CN01 lock and QPY were public from 22:21Z, when #15 opened. That PR merged at 23:02:29Z, without A-3.
Both analyses are reported and they give the same verdict, so the choice changes the framing, not the conclusion. Neither is labelled primary here. A dated note in
PHASE3_PREREG_AMENDMENTS.mdrecords the three A-2/A-3 statements that did not hold for the run.Commits
a7674832, amendment A-3. These are the files as they were when A-3 was recorded: the analysis, the validation, the runner, the A-3 tests and the amendments text. The record pins the lock, the QPY and the code by SHA-256, and all of them still match. The only change made after the job is a CI import guard in the A-3 test.8e020758, CN01 results. This adds:Caveats
instance_idandplan_id. They are removed from the committed raw counts and listed inredacted_fields. The owner keeps the unredacted original locally.phase3/cn01_a3_bound_diagnostic.py; not preregistered, written after the data): A-3 caps the white dephasing floor at 40 kHz for H0_1f, H3_local and H3. The v-qubit floor fitted here is 57 kHz, so those three models fit badly. With the cap raised to 60 or 200 kHz, they collapse onto H0 and A-3's selection is unchanged. So the cap changes no decision, but those three BICs reflect the cap, not the data.Verification
pytest tests/ phase2/on this branch: 2,197 passed, 26 skipped, 0 failed.tests/test_phase3_cn01_hardware_artifact.pychecks that the artifacts agree with each other and with the lock, that the raw counts carry no account identifiers, and that both analyses reproduce from the archived counts.qiskit-ibm-runtime, so the two new CN01 test modules skip there. They need a local run.Not changed
🤖 Generated with Claude Code