Skip to content

S1 Feedback IBM Paper Plan

Date: 2026-05-20

Working Title

Monitored feedback versus open-loop control for Kuramoto-XY synchronisation on IBM dynamic circuits

Alternative:

A preregistered dynamic-circuit feedback test for small Kuramoto-XY synchronisation

Thesis

The next IBM paper should test a control hypothesis rather than another static characterisation hypothesis. The S1 campaign asks whether a monitored cross-shot feedback policy moves the live Kuramoto-XY order-parameter estimate toward a preregistered target better than a matched open-loop control under the same backend, layout, circuit family, shots, and repetitions.

This is not a quantum-advantage paper. It is a bounded hardware-control paper: dynamic-circuit feedback is useful only if the measured feedback arm improves the target-error metric relative to the matched open-loop arm.

Why This Is The Next Candidate

  • It is already preregistered and has a no-QPU readiness bundle.
  • It uses a small four-qubit dynamic-circuit payload.
  • It has a matched open-loop control arm, so a negative result remains interpretable.
  • It moves the programme from static leakage/tomography observations toward actual control.
  • The estimated execution budget in the preregistration package is 24.0 seconds before queue and calibration uncertainty.

Current Evidence Status

Completed:

  • S1 no-QPU readiness bundle.
  • S1 preregistration JSON and Markdown.
  • S1 latency benchmark artefacts.
  • S1 synthetic raw-count analysis rehearsal.
  • Provider dry-run payloads with submission_enabled=false.
  • Hardware-job dossier embedded in the preregistration package.

Fresh no-QPU refresh on 2026-05-20:

PYTHONDONTWRITEBYTECODE=1 python scripts/reproduce_s1_feedback_readiness.py

The refresh regenerated:

  • data/s1_feedback_loop/s1_feedback_loop_latency_summary_2026-05-06.json
  • data/s1_feedback_loop/s1_feedback_loop_latency_summary_2026-05-06.csv
  • data/s1_feedback_loop/s1_feedback_preregistration_2026-05-06.json
  • data/s1_feedback_loop/s1_feedback_preregistration_2026-05-06.md
  • data/s1_feedback_loop/s1_feedback_analysis_summary_2026-05-06.json

IBM submission status:

  • Paired feedback/open-loop IBM submitter is implemented behind the existing approval-gated scheduler.
  • Live IBM sampler results are converted into the preregistered raw-count package with r_live records.
  • Corrected dynamic-circuit payload uses unconditional monitor resets and conditional corrections only; IBM rejected reset operations inside conditional blocks during the first attempted feedback-arm execution.
  • Corrected paired run completed on ibm_kingston.

Fresh live no-submit IBM readiness on 2026-05-20:

PYTHONDONTWRITEBYTECODE=1 python scripts/probe_s1_ibm_metadata.py --backend ibm_kingston
PYTHONDONTWRITEBYTECODE=1 python scripts/prepare_s1_ibm_live_readiness.py --backend ibm_kingston

The live probe and transpilation artefacts are:

  • data/s1_feedback_loop/s1_ibm_metadata_probe_ibm_kingston_2026-05-06.json
  • data/s1_feedback_loop/s1_ibm_live_readiness_ibm_kingston_2026-05-20.json
  • docs/campaigns/s1_ibm_live_readiness_ibm_kingston_2026-05-20.md
  • data/s1_feedback_loop/s1_ibm_feedback_pair_readiness_ibm_kingston_20260520T123941Z.json
  • data/s1_feedback_loop/s1_feedback_raw_counts_ibm_kingston_20260520T123941Z.json
  • data/s1_feedback_loop/s1_feedback_analysis_summary_ibm_kingston_20260520T123941Z.json

Observed readiness:

  • backend capability status: ready;
  • backend availability: active with zero pending jobs at capture time;
  • corrected transpiled monitored payload depth: 717;
  • corrected transpiled operation counts: cz=183, if_else=3, measure=6, reset=3, rz=380, sx=363, x=2;
  • hardware submission: false;
  • corrected paired-run status: completed;
  • corrected paired-run jobs: d86qn3lg7okc73elg2eg, d86qn65g7okc73elg2hg;
  • preregistered analysis decision: null_or_negative;
  • feedback mean target error: 0.2148958333;
  • matched open-loop mean target error: 0.2116406250;
  • target-error improvement: -0.0032552083.

S1b direct-XY observable extension:

  • status: completed on ibm_kingston;
  • purpose: keep this in the same paper and test direct XY-sector correlators after the binary synchrony proxy saturated;
  • observables: XXI, YYI, IXX, IYY;
  • jobs: d86r1rqs46sc73f7c2g0, d86r1udg7okc73elggi0, d86r201789is739022q0, d86r21is46sc73f7c2o0, d86r252s46sc73f7c2tg, d86r26gp0eas73dlbkkg, d86r288p0eas73dlbkmg, d86r29p789is7390238g;
  • mean absolute feedback-control separation: 0.0205078125;
  • signed feedback minus control by sorted observable row: 0.0247395833, 0.0358072917, -0.0175781250, 0.0039062500;
  • interpretation: not a robust positive-control result, but channel-structured evidence that the binary proxy hid small direct-XY feedback/control differences.

Preregistered Job Shape

Field Value
System qubits 3
Monitor qubits 1
Total qubits 4
Classical bits 6
Dynamic rounds 3
Arms feedback and matched open-loop control
Circuits 2
Shots per circuit 1024
Repetitions 12
Estimated execution seconds 24.0
Hardware submission state completed on ibm_kingston

Primary Observable

The primary observable is target-order-parameter error:

abs(R_live - R_target)

The feedback arm is promoted only if it improves this metric relative to the matched open-loop control under the same hardware window.

Result Decision Tree

Raw-count outcome Paper framing
Feedback improves target error after the preregistered analysis Bounded evidence that monitored dynamic-circuit feedback can steer the tested small-system synchronisation observable on the selected backend/layout.
Feedback and open-loop are statistically indistinguishable Hardware-control null result; dynamic-circuit overhead/noise dominates at this scale.
Feedback worsens target error Observed. Negative control result; the feedback action is not robust under this backend/calibration window.
Capability or transpilation gate fails Provider-readiness paper note only; no hardware-control claim.

S1b does not replace the preregistered binary-proxy decision tree. It adds a same-paper diagnostic extension showing whether the null/negative result is uniform across direct XY-sector observables. The observed answer is no: feedback/control differences are small but not uniform across channels.

S1c shallow/gain-tuned extension:

  • status: completed on ibm_kingston;
  • purpose: same-paper shallow/gain-tuned continuation to test whether S1/S1b weakness is dominated by dynamic depth and feedback strength;
  • configuration: n_rounds=1, correction angle 0.06, base gain 0.4;
  • observables: XXI, YYI, IXX, IYY;
  • readiness artefact: data/s1_feedback_loop/s1c_xy_observable_readiness_ibm_kingston_20260520T131646Z.json;
  • execution readiness artefact: data/s1_feedback_loop/s1c_xy_observable_readiness_ibm_kingston_20260520T132455Z.json;
  • raw-count artefact: data/s1_feedback_loop/s1c_xy_observable_raw_counts_ibm_kingston_20260520T132455Z.json;
  • analysis artefact: data/s1_feedback_loop/s1c_xy_observable_analysis_ibm_kingston_20260520T132455Z.json;
  • maximum feedback transpiled depth: 237, compared with S1b feedback depth about 720;
  • estimated QPU seconds: 24.0.
  • jobs: d86rca1789is73902fc0, d86rcc2s46sc73f7cf50, d86rcdgp0eas73dlc2j0, d86rcf0p0eas73dlc2lg, d86rcgh789is73902fm0, d86rcias46sc73f7cfh0, d86rcjp789is73902fsg, d86rclgp0eas73dlc2sg;
  • mean absolute feedback-control separation: 0.0299479167;
  • signed feedback minus control by sorted observable row: -0.0364583333, -0.0319010417, -0.0221354167, -0.0292968750;
  • interpretation: the shallow/lower-gain policy does not rescue the feedback arm; all four direct XY channels move negative relative to matched open-loop control.

S1d policy-direction sweep:

  • status: completed on ibm_kingston;
  • purpose: same-paper discriminator for correction polarity, gain strength, and calibration-window sensitivity after S1c;
  • variants: current_shallow_positive (n_rounds=1, correction angle 0.06, base gain 0.4), polarity_flipped (n_rounds=1, correction angle -0.06, base gain 0.4), and weak_positive (n_rounds=1, correction angle 0.03, base gain 0.2);
  • readiness artefact: data/s1_feedback_loop/s1d_xy_observable_readiness_ibm_kingston_20260520T134614Z.json;
  • raw-count artefact: data/s1_feedback_loop/s1d_xy_observable_raw_counts_ibm_kingston_20260520T134614Z.json;
  • analysis artefact: data/s1_feedback_loop/s1d_xy_observable_analysis_ibm_kingston_20260520T134614Z.json;
  • maximum transpiled depth: 237;
  • estimated QPU seconds: 72.0;
  • jobs: 24 total, with 14 recovered from the first approved submission and 10 submitted in a resume pass without duplicating completed arms;
  • result: current_shallow_positive has mean signed feedback-control delta 0.0257161458 and mean absolute separation 0.0667317708, driven by a large YYI shift; polarity_flipped and weak_positive remain close to zero with mean signed deltas -0.0040690104 and -0.0050455729;
  • interpretation: S1d is diagnostic rather than a promoted controller result. It shows policy- and calibration-window sensitivity in the direct-XY sector; stronger feedback claims require a predeclared repeat or redesigned policy.

S1e confirmatory policy-sweep repeat:

  • status: completed on ibm_kingston;
  • purpose: same-paper statistical repeat of S1d to test whether the large YYI response in current_shallow_positive reproduces at higher repetitions;
  • variants: same as S1d, with repetitions increased from 3 to 5;
  • readiness artefact: data/s1_feedback_loop/s1e_xy_observable_readiness_ibm_kingston_20260520T140919Z.json;
  • raw-count artefact: data/s1_feedback_loop/s1e_xy_observable_raw_counts_ibm_kingston_20260520T140919Z.json;
  • analysis artefact: data/s1_feedback_loop/s1e_xy_observable_analysis_ibm_kingston_20260520T140919Z.json;
  • maximum transpiled depth: 237;
  • estimated QPU seconds: 120.0;
  • jobs: 24 completed IBM jobs;
  • result: current_shallow_positive has mean signed feedback-control delta 0.0345703125 and mean absolute separation 0.0607421875; its YYI response repeats at 0.1824218750 after S1d measured 0.1848958333;
  • interpretation: S1e upgrades the S1d YYI channel from a one-run curiosity to a reproducible channel-specific policy sensitivity. It still does not promote backend-general feedback control because the favourable response is not uniform across direct-XY channels or policy variants.

S1f quadrature mechanism check:

  • status: completed on ibm_kingston;
  • purpose: final same-paper mechanism discriminator for whether the S1d/S1e YYI response is stable and quadrature-localised;
  • configuration: current_shallow_positive only, n_rounds=1, correction angle 0.06, base gain 0.4, repetitions 5;
  • observables: YYI, XXI, XYI, YXI, ZZI;
  • readiness artefact: data/s1_feedback_loop/s1f_xy_observable_readiness_ibm_kingston_20260520T142213Z.json;
  • raw-count artefact: data/s1_feedback_loop/s1f_xy_observable_raw_counts_ibm_kingston_20260520T142213Z.json;
  • analysis artefact: data/s1_feedback_loop/s1f_xy_observable_analysis_ibm_kingston_20260520T142213Z.json;
  • maximum transpiled depth: 237;
  • estimated QPU seconds: 50.0;
  • jobs: 10 completed IBM jobs;
  • result: mean absolute feedback-control separation 0.0082812500; YYI moved negative (-0.0203125000) while XXI, XYI, YXI, and ZZI stayed small;
  • interpretation: S1f blocks a stable quadrature-localised positive-mechanism claim. The S1d/S1e YYI effect is reproducible within a calibration window but not stable enough across the later mechanism-check window to promote the tested feedback law as a robust controller.

Claim Boundary

Safe after successful raw-count analysis:

  • the preregistered S1 dynamic-circuit feedback payload was executed on the selected IBM backend;
  • the measured feedback arm did or did not improve target-error relative to a matched open-loop arm;
  • the result is bounded to the selected backend, layout, circuit family, shots, repetitions, and calibration window.

Blocked even after a positive result:

  • quantum advantage;
  • backend-general feedback control;
  • analogue-native feedback suitability;
  • sub-microsecond real-time control unless provider-side timing evidence is captured and analysed;
  • claims beyond the tested small-system payload.

Current Manuscript Source

Draft scaffold:

  • paper/submissions/submission_006_s1_feedback_control/s1_feedback_control_short_paper.md

The draft must remain pending until live backend metadata, live transpilation, explicit approval, raw-count artefacts, and preregistered analysis outputs exist.