phanerozoic/qiskit-calibration-drift

Qiskit Calibration Drift

Calibration parameters from IBM Quantum Heron processors joined to ambient and space-weather conditions at the time of each measurement. Designed for time-series forecasting of qubit drift and for studying environmental coupling to superconducting calibrations.

A GitHub Action (source) polls backend.properties() on every available Heron device every 30 minutes, reads IBM's historical properties for every half hour since a device's previous poll, and appends new calibration events keyed on (backend, property, qubit_a, qubit_b, calibrated_time). From 2026-01-31 on, the dataset holds every calibration event visible in IBM's historical properties at each half hour and at each poll time.

Backends

Backend	Processor	Qubits	Active
ibm_fez	Heron r2	156	yes
ibm_kingston	Heron r2	156	yes
ibm_marrakesh	Heron r2	156	yes
ibm_torino	Heron r1	133	retired 2026-04-27

All backends are at IBM's Yorktown Heights, NY facility (lat=41.27, lon=-73.78).

Schema

Column	Type	Description
backend	str	backend name
property_family	str	normalized property group (e.g. sx_error, cz_error, T1, readout_error)
property	str	IBM API canonical name (e.g. sx_gate_error, cz_gate_length)
qubit_a	int64	qubit index, or first qubit of a coupled pair
qubit_b	float64	second qubit of a coupled pair (integer-valued); NA for single-qubit and qubit-level properties
value	float64	the calibration value
unit	str	unit string from the IBM API (us, ns, empty for dimensionless)
scope	str	one of qubit, gate1q, gate2q
is_failure_ceiling	bool	true when value ≥ 0.999 for an error or probability property — the calibration saturated at the ceiling and the underlying measurement is unreliable
is_new_measurement	bool	true for the first row of a (backend, property, qubit_a, qubit_b) and for each row whose value differs from that of the same unit's row with the next earlier calibrated_time; filter on this to drop heartbeat replicas
observed_time	timestamp UTC	time of the snapshot the row was read from: the live pull, or the half-hour mark of a historical snapshot
calibrated_time	timestamp UTC	IBM's published freshness timestamp for this property
snapshot_update_time	timestamp UTC	the last_update_date of the BackendProperties snapshot the row was read from; NA for ibm_torino rows observed before 2026-05-08
calibration_age_seconds	float64	observed_time − calibrated_time
chipwide_recal_event_id	str	identifier shared across rows that participate in a chip-wide recalibration event (≥20% of qubits updated simultaneously); NA otherwise
latitude, longitude	float64	IBM facility coordinate
solar_zenith_deg	float64	solar zenith angle at the site (>90 means night)
temperature_c, pressure_hpa, humidity_pct	float64	ambient weather
bz_gsm_nt	float64	interplanetary magnetic field Bz in GSM coordinates, nT
neutron_flux	float64	cosmic-ray neutron monitor count rate
kp_index	float64	planetary K index (3-hour), 0..9 in 1/3 steps
ap_index	float64	planetary equivalent-amplitude (3-hour)
Ap_daily	float64	daily mean ap
SN	float64	International Sunspot Number v2.0 (daily)
f107_observed_sfu	float64	observed F10.7 cm solar radio flux, s.f.u.
f107_adjusted_sfu	float64	distance-adjusted F10.7
solar_flux_sfu	float64	NOAA SWPC 10cm flux summary
dst_nt	float64	hourly Dst index, nT

Property families

Coherence. T1, T2 per qubit, in microseconds.

Single-qubit gate errors. sx_error, x_error, id_error, rx_error, rz_error. Reported separately by the IBM API; on a given measurement the four basis gates of the same qubit are typically bit-identical because IBM derives them from one calibration.

Slow single-qubit gate. xslow_error and its _length, on ibm_fez and ibm_kingston.

Two-qubit gates. cz_error (controlled-Z) and rzz_error (parametric ZZ rotation), stored once per ordered directional pair (qubit_a, qubit_b). For an undirected pair {a, b} both (a, b) and (b, a) rows appear and carry different values.

Readout. readout_error, prob_meas0_prep1, prob_meas1_prep0, readout_length, measure_gate_error, measure_gate_length. Readout-error values are quantized to multiples of 1/8192. measure_2_error is an alternative readout scheme present on ibm_kingston only.

Durations. sx_length, x_length, id_length, rx_length, rz_length, cz_length, rzz_length, readout_length, reset_length, measure_length. Mostly constant during the observation window.

Working with the data

from datasets import load_dataset
ds = load_dataset("phanerozoic/qiskit-calibration-drift", split="train")

For drift analysis, filter to is_new_measurement == True to drop heartbeat replicas. For honest aggregations, also exclude rows where is_failure_ceiling == True. Property families are calibrated on independent schedules — cz_error and the single-qubit basis gates cycle together on a chip-wide cadence (look for chipwide_recal_event_id), readout_error refreshes more often, T1 / T2 less often, so the calibrated_time for two different families on the same qubit will generally not coincide.

Sources

• Calibration data: IBM Quantum backend.properties() via Qiskit Runtime.
• Kp (definitive), ap, Ap, F10.7, SN: GFZ Helmholtz Centre for Geosciences (Matzka et al. 2021, Space Weather, doi:10.1029/2020SW002641); SN from WDC-SILSO, Royal Observatory of Belgium.
• Kp (realtime fallback), Bz, Dst, 10cm flux summary: NOAA SWPC.
• Solar wind Bz: NOAA SWPC real-time solar wind (RTSW) 1-minute magnetometer data from the operational L1 spacecraft.
• Cosmic-ray neutron monitor: NMDB Newark station.
• Ambient weather: api.weather.gov (KHPN, Westchester County).
• Archive fills (below): Iowa Environmental Mesonet ASOS/METAR archive for KHPN; Wind MFI and ACE MFI magnetometer data via NASA CDAWeb; OMNI2 hourly Dst (WDC Kyoto; NASA SPDF OMNIWeb, King & Papitashvili 2005, doi:10.1029/2004JA010649); DRAO Penticton 10.7 cm flux (Natural Resources Canada).

Each covariate is the sample nearest to calibrated_time among the samples its source had published by observed_time, within 10 minutes for Bz, 6 hours for the NOAA Kp fallback and 3 hours otherwise; cells the poller wrote before 2026-10-23 allowed up to 48 hours. solar_flux_sfu is the flux in the SWPC summary at observed_time, which takes up each day's 20:00 UT Penticton flux at 21:00 UT. Cells that were empty on 2026-10-23 were filled by the same rule from archives: weather from the Iowa Environmental Mesonet KHPN archive (pressure from the altimeter setting, as api.weather.gov reports it); bz_gsm_nt from Wind MFI one-minute data, with Wind key parameters and then ACE MFI inside Wind's data gaps, a different L1 spacecraft from the SWPC operational one; dst_nt from OMNI2 hourly Dst and the SWPC Dst feed; the GFZ columns from GFZ as published on 2026-10-23; solar_flux_sfu from the Penticton 20:00 UT observed flux rounded to an integer; and solar_zenith_deg by the poller's formula at calibrated_time. Rows observed in the last 35 days get empty cells filled on later polls as sources publish. bz_gsm_nt stays empty where none of those spacecraft has data within 10 minutes, and neutron_flux where NMDB had not served data, which includes the rows recovered from IBM's historical properties on 2026-10-23.

License

CC-BY-4.0, except for the Sunspot Number column (SN) which is CC-BY-NC-4.0 per WDC-SILSO licensing.