v0.4.2

Changelog

• 3b66dbf fix: make CI green — the per-extra matrix and the dashboard lint
• 94bf57a chore: bump to 0.4.2
• c7f22f5 fix(qpi-driver): pin RB's asymptote and bound the echo window, so both measure again
• d87327d fix(qpi-driver): flag the two numbers the 2026-08-16 run reported without support
• 17e33d3 fix(qpi-driver): keep the measured f12 when a sweep cannot resolve the line
• 6e9f38b fix(qpi-driver): report an imprecise measurement instead of refusing it
• 376e049 fix(qpi-driver): stop the readout window at the pulse, not at the instrument
• 8ab161d fix(qpi-driver): let fine_amplitude_12 shorten its own sweep
• 93b10ee fix(qpi-driver): sweep the readout window to the hardware ceiling, not a few factors out
• 3d7f581 fix(qpi-driver): grid the echo delays, and close the gap a degenerate RB fit slipped through
• 27a2782 fix(qpi-driver): judge fine amplitude's overrun from the data, not from its own fit
• 1770f9d feat(qpi-driver): calibrate the readout integration window
• 06eee09 fix(qpi-driver): let a config widen the anharmonicity range f12 is searched in
• 8ad43d8 refactor(qpi-driver): drop the ef ladder's factor-of-two case, pin nr_sigma to the pulse
• 6bd78fe fix(qpi-driver): refuse a DRAG root read off a line through noise
• 9470cc1 fix(qpi-driver): size the Hahn echo window from T1 instead of a fixed 100 us
• 21a2244 fix(qpi-driver): score RB survival against references, not the sweep's own extremes
• 02a5598 fix(qpi-driver): the ef ladder constant read nr_sigma per pulse, not per side
• 1d9ee82 fix(qpi-driver): judge an off-ladder 1-2 pi by population swing, not by the ladder
• 98448f7 fix(qpi-driver): break the EF deadlock by amplifying on the 0-1 readout
• 6fc1561 fix(qpi-driver): sweep the 1-2 drive downward only, never up
• 83d0228 fix(qpi-driver): let the spectroscopy drive ladder climb when a chip cannot be seen
• 2269585 fix(qpi-driver): sweep the 1-2 spectroscopy drive instead of fixing it at 0.10
• 9a542cd fix(qpi-driver): refuse a 1-2 pi that is a factor of two off the ladder
• 9742a44 feat(qpi-driver): let a poor measurement through, marked, instead of refusing it
• 6d16483 fix(qpi-driver): size the three-state readout sweep from the linewidth it measured
• c144a61 fix(qpi-driver): a flat trace is a signal budget, not a broken chain
• 4f8974c test(qpi-driver): assert the Ramsey roots by arithmetic, not against another tuner
• 9d6c611 feat(qpi-driver): report what the ef sweep and the RB decay actually showed
• decce64 fix(qpi-driver): refuse a fine-amplitude sweep no line passes through
• b0f0d0a fix(qpi-driver): measure which root of the Ramsey fringe is the chip's
• 8cef895 feat(qpi-driver): measure the magnitude contrast nothing was choosing for
• d7d48bf fix(qpi-driver): give the echo the T1 ceiling it cannot fit without
• 7e5f834 fix(qpi-driver): a pi amplitude past the top of the sweep was never measured
• e3aa871 fix(qpi-driver): give the pi/2 shortening somewhere to go, and a floor that fits
• 438946f fix(qpi-driver): chunk a 2-D grid by rows, and stop the timeout warning contradicting itself
• 33ed12e feat(qpi-driver): measure the ef ladder, and split an RB sweep that outgrows a schedule
• 7746831 fix(qpi-driver): make four escalatable axes actually escalatable
• caac158 fix(qpi-driver): stop losing calibration results on the last mile
• 8371973 fix(qpi-driver): rb averages harder, bounded by the assembler this time
• 242c41a fix(qpi-driver): keep an embedded Q1ASM program out of the error payload
• 86ea70f fix(qpi-driver): let t1 and t2_echo widen their delays instead of refusing
• cf4aa1e fix(qpi-driver): re-land the routine and fit fixes, none of them on the wire
• 8bf1e67 fix(qpi-driver): cancel the CZ's single-qubit phase instead of doubling it
• 6e6af5b Revert everything after 4686684, to a driver that reports its results
• 13d7729 fix(qpi-driver): stop losing calibration results on the last mile
• c35fb2a fix(qpi-driver): bound the depths escalation at an instruction budget
• 0b3ca0b fix(qpi-driver): cancel the CZ's single-qubit phase instead of doubling it
• 7ded3bb fix(qpi-driver): record a benchmark that runs its own measurement loop
• 2391ccf fix(qpi-driver): escalate the last two nodes that refused instead of resizing
• ddaeaba fix(qpi-driver): resize the sweeps that refused a fit they had already found
• 45539b8 fix(qpi-driver): refuse an RB fidelity fitted off a straight line
• 1061d64 fix(qpi-driver): the ef ladder refuses a partial rotation, not a resolved one
• 4686684 feat(qpi-driver): a refusal carries the fit it refused
• 9b50c07 fix(qpi-driver): free the fine-amplitude intercept, and bound the linear regime
• ee01e7d feat(qpi-driver): report the ef sweep's contrast when the ladder refuses
• d6901d1 feat(qpi-driver): report allxy_check's response, not just its rms
• bae0116 fix: size the ef pulse off rxy.duration, and show the gate fidelity
• e738f3d docs: close RFC 0007 11.5 and document fine_amplitude_90
• 35e6133 feat(qpi-driver): calibrate the pi/2 amplitude with fine_amplitude_90
• 0598838 refactor(qpi-driver): let the fine-amplitude fit amplify any nominal angle
• d5eeca5 feat(qpi-driver): give CalibratedTransmon a measured pi/2 amplitude
• 2d60222 docs(rfc-0007): record that 11.5's equator error survived the detuning refinement
• 60cf28e fix(qpi-driver): centre the ef ladder bound on the pulse it actually plays
• 6e5055e test(qpi-driver): simulate the 1-2 transition, and measure what the map-back buys
• 7ed29dd fix(qpi-driver): make the 1-2 Rabi oscillation visible to a 0-1 readout
• 83d9a30 feat(qpi-driver): refine f01 until the residual detuning is unresolvable
• 0fa0046 fix(qpi-driver): stop a diagnostic score outvoting a measured gate fidelity
• fba85de fix(qpi-driver): hold the 1-2 pi amplitude to the transmon ladder
• 8c2848f docs(rfcs): a broad spectroscopy line is not a wrong one — §11.3 closed
• 703363b feat(qpi-driver): let a routine set its own timeout
• dec4a8c docs(rfcs): record that nothing refuses a line filling its own window
• f8128a0 fix(qpi-driver): stop a widened amplitude sweep at full scale
• b2354f5 fix(qpi-driver): derive the search drive power from the powers configured
• d94defd fix(qpi-driver): read a config number as a number
• e520c37 fix(qpi-driver): hold the confirming sweep's step, not its point count
• 3e88939 fix(qpi-driver): cap an escalating sweep at what the sequencer really holds
• baa03c6 fix(qpi-driver): refuse a readout at chance and an AllXY that is only noise
• 6f0af0a feat(qpi-driver): let resonator_spectroscopy re-centre its own sweep
• 141f51a fix(qpi-driver): skip a two-qubit routine whose endpoint never calibrated
• 93c8d2a fix(qpi-driver): sweep resonator_punchout to full readout scale
• 650f2f9 fix(qpi-driver): declare the gate parameters 22 routines actually read
• f6aac1b docs(rfcs): record that reads is under-declared, so RFC 0007 §11 did not fire
• 8f592bc docs(rfcs): RFC 0007 overclaimed that phase 3 fixed resonator_punchout
• cf4d503 feat(qpi-driver): consume provenance, and correct what it should change
• f046085 feat(qpi-driver): report which inputs nothing has ever measured
• fefd532 feat(qpi-driver): record provenance for what a calibration measured
• 833caf2 feat(qpi-driver): add the provenance sidecar, unused
• 215f874 test(qpi-driver): benchmark against the injected gate error, not a threshold
• 835f28c docs(rfcs): RFC 0007 is implemented
• 78c8676 feat(qpi-driver): calibrate a chip known only from its design document
• a6f2d01 feat(qpi-driver): widen a window the fit says was too short, instead of failing
• 5af1618 fix(qpi-driver): derive the last of the physics-bounded ranges
• 087fb5c fix(qpi-driver): size the readout operating point from the measured linewidth
• ae83b49 feat(qpi-driver): keep the resonator linewidth that was measured
• 9be641b fix(qpi-driver): sweep the drive amplitude to full scale, and find out where that is
• 98c7a38 docs(rfcs): RFC 0007 — phases 1 and 2 are done and merged
• 8d0cfbd fix(qpi-driver): size the confirming sweep from the width the search measured
• 883a65e fix(qpi-driver): only a power that shows a line may judge the others
• b6cc138 feat(qpi-driver): refuse a spectroscopy line drawn through noise, and sweep only what the port can reach
• 5a42706 docs(rfcs): RFC 0007 — §14 was overbroad, and invented a blocker
• ccf2fd0 Fix typo
• d35147a fix(qpi-driver): the simulator's gate paths carry the drive detuning
• 3597c58 docs(rfcs): RFC 0007 — the simulator cannot fail a wrong f01
• 3a570f4 docs(rfcs): RFC 0008 — settle the expiry question, and explain its own vocabulary
• bb78c71 docs(rfcs): RFC 0008 — parameter provenance, lifted out of 0007 §10
• 1149682 docs(rfcs): RFC 0007 — retitle §10, which stopped being a list of gaps
• b16f329 docs(rfcs): RFC 0007 — record what phase 0 changed
• 7132c9e feat(qpi-driver): skip a routine whose input this run failed to produce
• 0a52120 feat(qpi-driver): a routine declares the device parameters it reads
• c4b7e7b refactor(qpi-driver): read the device before the acquisition, not after it
• 70baf36 docs(rfcs): RFC 0007 — resolve the third round of review FIXMEs
• 2916343 docs(rfcs): RFC 0007 — resolve the second round of review FIXMEs
• 1da607b docs(rfcs): RFC 0007 — resolve the review FIXMEs
• 997463d docs(rfcs): RFC 0007 — skip nodes whose prerequisite was never produced
• 7350a2a docs(rfcs): RFC 0007 — calibration without priors
• 7bb91d4 fix(qpi-driver): sum the allowances of a routine that runs several schedules
• b865a08 fix(qpi-driver): give the simulated backend the allowance the DAG reads
• d77cbdb feat(qpi-driver): let qubit_spectroscopy find a qubit that is not where the config says
• df7b866 fix(qpi-driver): refuse a dispersive shift no readout could resolve
• 4892671 fix(qpi-driver): refuse a coherence or Ramsey fit drawn through flat noise
• b190676 fix(qpi-driver): refuse a Rabi oscillation no taller than its noise
• 1a11771 fix(qpi-driver): reset the cluster when opening it
• ddc3f34 fix(qpi-driver): refuse a fine-amplitude slope taken through no contrast
• e504d78 fix(qpi-driver): make the dispersive shift auditable, and guard the sweeps it comes from
• d8b4f0b fix(qpi-driver): refuse a spectroscopy line that is not above the noise
• 88e0d76 fix(qpi-driver): refuse an RB fidelity taken from noise
• e7e0fc6 fix(qpi-driver): decline the flux routines on a flux-tunable-coupler chip
• 5e1da6f fix(qpi-driver): round ramsey's delays to the 1 ns grid
• ecaa5a8 fix(qpi-driver): release the cluster's sync network after every quantify run
• 8570e77 perf(qpi-driver): optimize test_calibration_loop sweep grids and transmon eigenvalue caching
• b3a0395 Make some simulator CI tests run in parallel
• b8b5d26 WIP: Make simulated tests parallel in CI
• 564e0d1 ci: run the invariant checks once instead of per matrix leg