Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

• Plot annotations in reports. Waveform, region, FFT and comparison-overlay plots can carry text labels, vertical and horizontal reference lines, and shaded spans; waveform, FFT and comparison-overlay plots can also carry a figure caption beneath the plot (a region zoom inherits its parent waveform's annotations but has no caption of its own). Annotations are set from a new Annotate… editor in the report generator app or from Python (PlotAnnotation), and persist in a <source>.annotations.json sidecar that is restored when the waveform is re-imported. See examples/report_annotations.py, and examples/report_annotations_advanced.py for the surfaces beyond a single plot: all four plot types, per-annotation colour/size/arrow overrides against the PlotStyle defaults, a region zoom inheriting clipped annotations, the sidecar round-trip, and the anchor list behind the Annotate… dialog. The README and the annotation section of the report-generator guide now carry a GIF of the API in use, generated by scripts/media/make_annotation_gif.py, which executes the code it renders and holds the plots that execution produced. Same asset in both places, so there is nothing extra to regenerate.
• Dense binary waveform stream + zoomable live view. The gateway's Time canvas now receives the dense record (up to 100 000 samples per channel, --stream-max-points) as compact binary WebSocket frames (?format=binary; layout in the gateway API reference), draws it as a glitch-true min/max envelope, and lets you scroll/pinch-zoom and drag-pan into it client-side — no extra instrument read, and the window stays put as new acquisitions arrive. The JSON WebSocket frames scripts consume are unchanged (still ≤ 2000 points). scpi_control.server.frames.decode_binary is the Python-side decoder. New --stream-max-fps (default 20) bounds the live-view rate. The per-channel update rate itself is unchanged by this branch: on an SDS824X HD a full-record read costs ~250 ms regardless of point count, so the gateway measured the same ~1.3 frames/s with dense 100k-sample frames that the old 2000-point live view already ran at — reading is per-transfer, not per-point. What changed is resolution (2000 → up to 100 000 samples per frame), not rate.

Security

• Report-text-derived plot filenames could escape the plots directory. A section title or channel label containing .., / or : reached the saved PNG's filename unsanitized — on Windows, a : could silently write to an NTFS alternate data stream instead of the intended file, and a / or .. could crash generation or write outside the report's plots folder. Filenames are now built from an allowlist of the derived text, with a second check confirming the resolved path stays under the plots directory before anything is written.
• PDF report text reached ReportLab's markup parser unescaped at most call sites. Titles, section headings, channel labels, and comparison table titles were passed straight to ReportLab's Paragraph, which parses its input as XML mini-markup — a <, > or & in any of them could corrupt the page or silently drop content. Every text construction in the PDF generator now routes through one escaping helper.
• Markdown report text reached table cells and link/alt text unescaped. Technician/equipment metadata, measurement names and units, comparison-table cells, and manifest entries were interpolated raw into | ... | table rows, where an unescaped | truncated the cell and corrupted the row; waveform, region, and overlay labels were interpolated raw into ![...] alt text, where an unescaped ] closed the link early and corrupted everything after it. Both are now escaped through dedicated helpers before reaching Markdown syntax (the Markdown-text sibling of the PDF escaping fix above). Also fixed in the Markdown generator: a section title equal to a Windows reserved device name (NUL, CON, COM1, ...) could produce a plot filename that Windows treats as a reference to the device itself regardless of the .png extension, and two overlay plots in the same section whose channel labels collapsed to the same sanitized filename (e.g. A/B and A:B) could silently overwrite each other's PNG — the overlay filename now includes a disambiguating index, matching the existing waveform/region plot pattern.

Changed

• Scope sessions no longer sit on a fixed quarter-second poll floor: the floor is 1 / --stream-max-fps and the instrument's own readiness gate and read time set the pace. PSU/AWG sessions keep 0.25 s. :WAVeform:MAXPoint? is asked once per session instead of every tick.
• For embedders subscribing to InstrumentSession.publish directly: waveform and reference messages now carry samples (a numpy array) and seq instead of a points list; scpi_control.server.frames.to_json() converts to the wire shape. The WebSocket JSON contract itself is unchanged.

Fixed

• Translucent fills reached the PDF fully opaque. svglib honours fill-opacity/stroke-opacity but ignores a bare opacity:, which is what matplotlib writes for a translucent patch — so an annotation span drawn at PlotStyle.annotation_span_alpha (0.25 by default) was painted as a solid block over its own gridlines, and a plot legend's frame lost its 80% transparency. Lines were never affected: those get stroke-opacity, which svglib already reads. PDF plots now render both at the alpha they were drawn with. Markdown reports were always correct, so PDF and Markdown output of the same report now match.
• A code span next to bold text could corrupt or crash PDF generation. The internal placeholder used to protect `code` spans while markdown was converted to PDF markup was itself shaped like the bold markdown it was being protected from, so the two collided. An AI-written summary mixing a code span with **bold** text — an ordinary shape for an LLM to produce — now renders both correctly instead of losing the code span or failing generation outright.
• AI-generated summary/findings could ship into a report built from different data than they described. The AI Analysis panel's generated content stayed attached after starting a new report or importing new waveforms, so a report built afterward could silently include narrative text about a previous dataset. It's now cleared whenever the underlying waveforms change.

Removed

• WaveformRegion.markers, an unused field that was serialized but rendered by nothing. Use PlotAnnotation instead.

[7.1.0] - 2026-08-09

Added

• Five examples for features that had none: math_channels.py (waveform arithmetic), reference_waveforms.py (golden-reference comparison), screen_capture_example.py, measurement_badges_example.py (Tektronix badge pooling), and protocol_decoding.py (I2C/SPI/UART).

Changed

• The hero GIF now also appears on the documentation landing page and in examples/README.md, not only the root README. Same asset, so there is nothing extra to regenerate.
• Every network-facing example now takes --host, defaulting to mock, and runs with no instrument attached — the built-in mock supplies the waveforms. Previously all but one required editing a hardcoded SCOPE_IP before they would run at all. function_generator_basic.py's --ip flag is renamed --host for consistency.
• Converted examples now exit non-zero on unexpected failure instead of printing an error and exiting 0. Anyone scripting or automating against these examples should check the exit code again — a blanket except Exception: print(...) around main() previously made several examples report success even when nothing worked (e.g. basic_usage.py pointed at an unreachable host).
• The examples execution guard covers 33 of 36 examples, up from 15. The three it cannot execute — psu_usb_connection.py (USB/VISA transport), psu_gui_test.py (Qt GUI), gateway_rest_client.py (needs a running gateway) — now say so in their docstrings and in the README, so the gap is visible rather than silent.
• README badges regrouped into two rows and extended from six to nine. The first row answers "should I install this?" (version, downloads, release status, Python versions, platform); the second answers "is this looked after?" (CI, coverage, docs, licence). New badges: monthly PyPI downloads, PyPI release status, and supported platforms. The platform badge is static and linked to the CI workflow, since that job's ubuntu/windows/macos matrix is the only thing backing the claim — it needs updating by hand if an OS ever leaves the matrix. Because pyproject.toml sets readme = README.md, these also appear on the PyPI project page from the next release onward.
• The README hero GIF now tours four examples running with no instrument attached — basic_usage.py, math_channels.py, measurement_badges_example.py and screen_capture_example.py — instead of showing a single waveform capture. Every frame's text is captured by running the example as a subprocess at generation time, so the image cannot drift from the code. The file keeps its path (docs/images/mock-demo.gif), so the README's <img> source is unchanged; only its alt text was updated to describe the new content.

Fixed

• Docs: the landing page's Quick Example called scope.measure.frequency() and scope.measure.vpp(), neither of which exists — anyone copying it got an AttributeError on the last two lines. It now uses scope.measurement.measure_frequency(), and runs against the built-in mock with no instrument attached, matching the README's quick start. Verified by executing the block as written.
• Serial protocol decoders (I2C/SPI/UART): decoding could fabricate phantom events — bytes that were never on the wire — when a frame began too close to the end of a captured buffer. _sample_at_time resolved an out-of-range sample time to the nearest captured sample instead of rejecting it, so a truncated frame decoded exactly like a fully-captured one, indistinguishable from a genuine decode at every call site. This affected real hardware captures as much as the mock. Truncated frames are now dropped instead of decoded, so a decoder may report fewer events for the same capture than before — there is no migration path to describe, since anyone relying on the old behavior was relying on fabricated data. examples/protocol_decoding.py's documented UART event summary changes from {'DATA': 1} to {} accordingly.
• Report generation: MarkdownReportGenerator.generate() and PDFReportGenerator.generate() caught Exception around their whole write path, so a programming error (e.g. an AttributeError from a report model change) was indistinguishable from an environmental failure (permission denied, disk full, a bad output path) -- both just returned False, and the real bug was visible only as a log line. They now catch only OSError; every other exception propagates. This is not a signature change -- generate() still returns bool and no caller needs updating -- but a caller that previously saw False for a latent bug will now see the exception instead.
• Oscilloscope (legacy Siglent dialect): channel.unit no longer returns the raw echoed response. Real hardware answers C1:UNIT? with the header still attached (C1:UNIT V, RC01020-E01C p.137); previously that header was returned verbatim, so scope.channel1.unit came back as "C1:UNIT V" instead of "V". Code that worked around the old value should drop the workaround. The mock's C1:UNIT? answer was updated to match the same header-echoed form.
• CI: the Codecov upload step passed file:, an input codecov/codecov-action dropped at v4. It was silently ignored — the run logged Unexpected input(s) 'file', stayed green, and uploaded only because the CLI falls back to searching the tree for a report. Now files:, so the explicit path is honoured.

[7.0.1] - 2026-08-07

Added

• Docs: docs/images/social-preview.png, a purpose-built 1280×640 GitHub social-preview card. Links shared to Reddit/HN/X/Discord previously rendered the default GitHub card. Upload it under Settings → Social preview; BRANDING.md has the steps (it used to point at resources/Test Equipment.png, which is 1024×1024 and gets cropped to 2:1).
• Docs: docs/images/mock-demo.gif, a terminal demo of a hardware-free capture, and docs/images/gui-live-view.png / gui-fft.png, desktop-GUI screenshots showing live traces instead of an empty window.
• scripts/media/ — generators for all of the above, each driving the real library against the built-in mock so the images cannot drift from the code. The demo GIF's printed output is captured by executing the snippet rather than typed by hand.

Changed

• README rewritten as a showcase, 1008 → 269 lines. The GUI manual, the inline API reference, the v1.0.0 migration guide, and the maintainer build/publish steps are now linked to their existing homes on the docs site (docs/gui/, docs/user-guide/, docs/about/changelog.md, docs/development/) rather than duplicated. The four things that lived only in the README — the supported-instrument list, the CLI table, the GUI's Power Supply and Data Logger tabs, and the interactive-tutorial link — are kept.
• README now leads with a runnable hardware-free example. The previous quick start needed an instrument at 192.168.1.100 before it did anything.
• Docs landing page uses the live-trace GUI screenshot instead of the disconnected one.

Fixed

• README no longer claims CAN and LIN protocol decoding. Only I2CDecoder, SPIDecoder, and UARTDecoder exist, and decoding is done in software — the CAN/LIN entries in supported_decode_types describe what the instrument can decode, which the library never drives.
• README no longer presents channel_controls.png as a distinct "Channel Controls" screenshot; it was byte-identical to main_window.png. That file and the orphaned 01_main_window.png (also byte-identical) are removed.

Known issues

• The mock answers :MEASure queries with fixed constants (Peak-to-Peak 2.000 V, RMS 707 mV) that do not track the waveform it synthesizes — a capture of a 3.28 Vpp square wave still measures 2.000 V. Waveform samples are state-coupled; measurement queries are not. No Measurements-tab screenshot is shipped for this reason.

[7.0.0] - 2026-08-04

⚠️ Breaking Changes

• PSU: ovp_level/ocp_level on SPD1305X/SPD1168X now raise NotImplementedError instead of silently sending a command the firmware discards. No SPD1000X programming manual documents a SCPI protection subsystem; set protection on the front panel. (Same honesty gate the SPD3303X received in v5.0.0.)
• Oscilloscope: setting trigger.trigger_type to a type the connected dialect cannot express (everything except EDGE on Tektronix) now raises FeatureNotSupportedError instead of silently leaving the scope on its previous trigger.
• Oscilloscope: setting an invalid trigger type now raises ValueError (consistent with the mode/slope setters) where it previously raised InvalidParameterError; callers catching SiglentError around trigger.trigger_type assignments should catch ValueError too.
• DAQ: an overload reading is no longer returned as the raw 9.9E+37 sentinel. Reading gains an overload: bool field; overload readings carry value=NaN with overload=True. Code that compared readings against 9.9e37 should check reading.overload instead.
• Oscilloscope (legacy Siglent): acquire(..., format="WORD") now raises FeatureNotSupportedError instead of returning fabricated data. The legacy programming guide documents no way to request 16-bit samples; use format="BYTE".
• Connections: read_raw() takes a framing argument (Framing.AUTO by default, so existing calls are unchanged). Custom BaseConnection subclasses must accept it (importable as from scpi_control.connection import Framing).
• Connections: a response that is nothing but a terminator is now a timeout on read() on both transports, and on query() on sockets. read() against a bare \n — or b"\x00\n", a NUL byte then a terminator — used to return ""; it now logs the discarded stray byte, keeps reading for the real answer, and raises SiglentTimeoutError with the session marked out of step so the next send clears it. SocketConnection.query() inherits the fix because it calls read(); VISAConnection.query() does not, because it delegates straight to pyvisa's own query() rather than this module's read(), so a terminator-only answer there still comes back as "". A lone terminator is the leftover of an earlier exchange, not an answer to the question just asked — returning it as "" is how one late newline shifted every later answer by one query. Code that read "" as "the instrument had nothing for me" must catch SiglentTimeoutError instead, except when going through VISAConnection.query(). (A response that is empty after whitespace but not terminator-only — e.g. " \n" — still comes back as "".)
• Sockets: a response cut short by the peer now raises SiglentConnectionError instead of being returned as a successful short read. A truncated measurement or screenshot was previously indistinguishable from a complete one.
• Sockets: a non-ASCII byte in a text response now raises SiglentConnectionError instead of UnicodeDecodeError. Code catching ValueError around read()/query() was silently swallowing it and must catch the transport error instead.
• VISA: is_connected now requires a successful connect(). A connection that failed after opening the resource previously reported itself connected.
• Power supply: unrecognized PSU models no longer advertise OVP/OCP support (has_ovp/has_ocp are now False in the generic fallback). Code that set ovp_level/ocp_level on an unlisted model now gets NotImplementedError instead of a command the instrument silently discards. Migration: if your PSU genuinely has a protection subsystem, add it to PSU_MODEL_REGISTRY with manual citations.
• Oscilloscope: trigger.coupling = "AC" on the tektronix dialect now raises FeatureNotSupportedError — the Tek edge-trigger coupling vocabulary (DC|HFRej|LFRej|NOISErej, TBS1000C PM p.151 / MSO2 PM p.2-661) has no AC token; the old code sent one anyway.
• Power supply: on the SPD3303X/SPD3303X-E, output3.voltage, output3.current, output3.measure_*(), output3.get_mode(), output3.timer_enabled, output3.waveform_enabled and reading output3.enabled now raise FeatureNotSupportedError. CH3 is the DIP-switch-selected fixed rail: its voltage is chosen on the front panel (QS0503X-E01B p.21) and the command set covers only CH1/CH2 for setpoints (p.39), measurement (p.38), the timer (p.41) and the waveform display (p.40), while p.42's status bitmap has no CH3 state bit — which is why reading the state used to hang. Switching CH3 on and off still works and is unchanged (OUTPut {CH1|CH2|CH3},{ON|OFF}, p.40). get_configuration() omits the unsupported fields instead of raising, and reports the per-output capability flags alongside.
• Oscilloscope: trigger.level now raises FeatureNotSupportedError where no level command is documented, instead of returning a fabricated 0.0 and silently discarding writes — a LINE source on the legacy, LeCroy and Tektronix dialects, and the external input on Tektronix (the ⅘/6 MSO command set documents TRIGger:A:LEVel:CH<x> but no AUX form). The modern Siglent dialect is unaffected: its :TRIGger:EDGE:LEVel takes no source argument, so nothing is gated there. Code that read 0.0 as a real trigger level was reading a number the library invented. trigger.get_configuration() reports such a level as None rather than raising, so the rest of the trigger state still comes back.
• Power supply: OutputSpec and PSUCapability (scpi_control/psu_models.py) are now frozen dataclasses. psu.capabilities hands back the shared PSU_MODEL_REGISTRY entry itself, not a copy, so an in-place mutation (e.g. PSU_MODEL_REGISTRY["SPD3303X"].has_ovp = True, a workaround some callers used against the v5.0.0 OVP gate) would change behaviour for every PowerSupply instance in the process; it now raises FrozenInstanceError instead. Constructing a new PSUCapability/OutputSpec still works. Callers that were patching a registry entry should build a replacement with dataclasses.replace(existing, has_ovp=True) (and register it, if it needs to be looked up by model name) instead of mutating in place.

Added

• Frequency response: scpi_control.frequency_response.sweep() steps a function generator across a frequency range, captures a reference and a response channel at each point, and returns gain and phase with the capture geometry behind them. Autoranges the response channel per point, reports an unmeasurable point as None with a stated reason rather than a plausible number, and exports CSV with a metadata header (to_csv) or a Bode plot (plot). See examples/frequency_response_sweep.py. Every accuracy claim is validated against a mock instrument and an analytic RC model — there is no function generator on the development bench, so none of it is hardware-verified.
• Connections: resync() discards anything the instrument left unread and reports the byte count. Called automatically before the next send after a read times out.
• Connections: drain_input() discards bytes the instrument has already queued and reports the count, taking no protocol-level action. Use it for housekeeping after a completed exchange (the terminator behind a screen dump, say); resync() is for recovering a session whose position is unknown, and on VISA it prefers a device clear, which aborts the instrument's pending operation. Both are no-ops on BaseConnection, so existing subclasses keep working unchanged.
• Typed vocabulary: Coupling, TriggerMode, TriggerSlope, TriggerCoupling, TriggerType, TriggerSource, BandwidthLimit, TrackingMode — string-compatible enums accepted by every token-valued setter (plain strings still work; getters still return strings).
• Capability discovery: scope.capabilities (derived from the connected dialect's command tables) and psu.capabilities (the detected model's registry profile) for ask-before-call feature checks.
• Structured errors: InvalidParameterError (now also a ValueError) carries parameter/value/valid_options/dialect/model; CommandError carries the sent command where available.
• Power supply: OutputSpec carries per-output capability flags (programmable, measurable, switchable, state_readable, supports_timer, supports_waveform) reflecting what the model's manual documents for that output, readable through psu.capabilities.
• Oscilloscope: scope.capabilities.trigger_level_sources reports which trigger sources have a documented level command on the connected dialect.

Changed

• Invalid token-valued parameters are rejected locally, before anything reaches the wire, with the connected dialect's valid options in the message.
• The mock scopes now reject undocumented wire tokens (queueing -224 like the real SDS824X HD) instead of accepting any word, with accept-sets derived from the same tables the drivers use.
• scope.capabilities.trigger_sources is documented as what the driver can send for the connected dialect, not what the attached model physically has — SCPI offers no way to interrogate the latter. Measured on an SDS824X HD (2026-08-04), three cases are accepted with no error queued and silently coerced to something else: a channel that is switched off (lands on LINE), EX on a model with no external trigger input (lands on LINE), and EX5 (a no-op from a channel source; lands on the highest enabled channel from LINE). Read scope.trigger.source back when the distinction matters. The dialect-wide set is deliberately not restricted, because other modern-dialect scopes do have a working external input.
• FeatureNotSupportedError now also subclasses NotImplementedError, and the PSU capability gates (tracking, OVP, OCP, timer, waveform) raise it instead of a bare NotImplementedError. Existing except NotImplementedError handlers keep working; the errors are now also catchable as SiglentError.

Fixed

• Oscilloscope (modern Siglent): waveform captures are now scaled by the channel's probe attenuation ratio. Verified against an SDS824X HD: the wire preamble's gain/offset exclude the probe factor, so 10x-probe captures previously read 10x too small.
• Oscilloscope (modern Siglent): trigger.trigger_type = "SLEW"/"GLIT"/"INTV" now sends the dialect's real tokens (SLOPe/PULSE/INTerval) instead of legacy spellings the scope rejects.
• Connection: a timeout during a sized binary read (e.g. slow screenshot start) now raises SiglentTimeoutError and keeps the session usable, instead of misclassifying as a dead connection.
• Mock: the modern mock now answers a trigger-token query with the command table's canonical wire spelling (SLOPe, NORMal, RISing, HFREJect) whatever casing the write used, matching an SDS824X HD measured 2026-08-04 — slope, SLOPE and sLoPe all read back SLOPe there. It previously echoed the sent casing for mode/slope/coupling and force-uppercased the trigger type, so neither shape matched hardware and no mock test could catch code that depended on the response casing.
• Mock: the modern mock now answers :CHANnel<n>:PROBe?, :CHANnel<n>:BWLimit?, :TIMebase:DELay?, and *OPC? with hardware-measured response shapes, rejects invalid probe/trigger-type parameters with -224 like real firmware, and reports probe-free preamble gain — so provenance channel snapshots are exercised in CI. The mock also now seeds the modern dialect's default channel coupling as DC instead of the legacy D1M token.
• DAQ: R? responses are parsed as the IEEE 488.2 definite-length block the instrument actually sends, so the first reading of every batch is no longer silently discarded. Because R? erases readings on the instrument, those readings were previously unrecoverable.
• DAQ: reading tokens that cannot be parsed are now logged at warning level instead of being dropped silently at debug level.
• Oscilloscope (Tektronix): captures now widen the transfer window before measuring it, so a narrow DATa:STOP left behind by another program or a recalled setup no longer truncates every capture indefinitely.
• Mock: the DAQ mock wraps R? in a definite-length block (leaving READ?/FETCh? bare, as the instrument does) and the Tektronix mock models the DATa:STARt/DATa:STOP window, so both regressions are now catchable in CI.
• Sockets: a reply that arrives after its read timed out is no longer returned to the NEXT query — the connection discards anything left unread before it sends again, and reports the discarded byte count. Every value after a single timeout could previously be shifted by one query. (A reply still in flight when the next command is sent cannot be distinguished from that command's answer; call resync() or reconnect if you need certainty after a timeout.)
• Sockets: binary content containing # followed by digits is no longer mistaken for an IEEE 488.2 block header, which truncated the response and left the remainder to poison the next read. A block header is now only recognised after a printable-ASCII prefix.
• Sockets: the terminator following a binary block is consumed by peeking rather than a fixed two-byte read, so a late second newline can no longer become an empty response to the next query.
• Sockets: read() now skips a leading run of stray terminator (\r/\n) and NUL bytes instead of returning it as an empty response. This is the backstop for a terminator that lands after _drain_terminator's peek window has already closed, with no timeout along the way to set _desynced and trigger a resync() drain — without it, that byte alone reproduces the High-7 symptom of every later answer shifting by one query.
• VISA: every exchange now holds the connection lock, so a GUI and the server can no longer interleave mid-transfer; connect() is a no-op when already connected instead of leaking a second resource over the first, and a failed connect() closes and clears both the resource and the ResourceManager; read() skips a leading stray terminator instead of returning it as an empty response, and read()/read_raw() now translate VisaIOError the way write()/query() already did; read_raw(None) frames definite-length blocks like the socket transport and drains the terminator left behind a completed block the same way, with the read termination character disabled during binary reads so a 0x0A inside a waveform cannot end the read; and query_binary() no longer demands exactly max_bytes (which blocked the full timeout and then raised, after the data had arrived). These are contract fixes checked against a stub VISA resource, not against hardware — this repo has none, and pyvisa is not installed locally.
• Oscilloscope: a whole acquisition holds the connection lock, rather than each exchange inside it, so a concurrent source or interval change cannot be assembled into the middle of one waveform. The screenshot path holds it across its write, wait and read.
• Oscilloscope: a screenshot that arrived intact is no longer discarded because the housekeeping behind it failed. The post-capture drain now runs best-effort and logs a warning, where a dropped connection during it used to surface as RuntimeError: Failed to capture screenshot with the complete image already in hand. That drain also no longer issues a VISA device clear — a modern SCDP capture over VISAConnection was issuing a VISA device clear on every single screenshot.
• Oscilloscope: trigger.level now works for the EX and EX5 external inputs on the legacy Siglent and LeCroy dialects, where <trig_source>:TRLV documents them (RC01020-E01C p.128 / MAUI p.7-33). A startswith("C") check had been silently dropping those writes and returning 0.0 for the reads.
• Mock: the PSU mock rejects the undocumented CH3 setpoint, timer and waveform forms with -224 like real firmware while still accepting OUTPut CH3,ON, and the legacy scope mock answers the EX/EX5 TRLV forms — so both gates are exercised in CI rather than merely untested.
• PSU: the data logger and the GUI PSU panel now check an output's capabilities before measuring it, instead of calling and catching. For an output with no documented measurement command (an SPD3303X's fixed CH3 rail), the logger writes empty CSV fields where it previously wrote ERROR, and the panel shows --- without polling. Column count and order are unchanged. Code scanning SPD3303X logs for the ERROR sentinel on CH3 should treat an empty field as "not measurable on this output".

[6.0.0] - 2026-07-30

⚠️ Breaking Changes

• A token name is now an identity rather than a unique credential. scpi-web token add <name> no longer fails when that name already exists — it mints an additional token, so one person can hold several devices, and scpi-web token revoke <name> cuts off every one of them at once. Anything that relied on the duplicate-name error to detect an existing name should call TokenStore.names() instead.
• DuplicateTokenName has been removed from scpi_control.server.auth. It can no longer be raised, so any except DuplicateTokenName clause is now an ImportError waiting to happen.
• TokenStore.names() now returns distinct names, sorted. It previously returned one entry per token, which is one entry per device under the new model.
• A fresh gateway no longer mints an identity called default, and no longer prints a ?token=… bootstrap URL on first start. Instead it opens a host-only admin panel and asks what to call you, so the first identity is a real person rather than an anonymous credential that ends up owning every session it creates. Existing installations are unaffected — this changes only what happens when the token store is empty.
• Revoking an identity now also closes that identity's live streams and releases the sessions they owned. A revoked colleague mid-capture loses their view, and their session becomes immediately claimable.
• Waveform values read from a high-resolution Siglent instrument in the default 8-bit transfer mode change by a factor of 256 — they were that much too small, and are now correct. This is a defect fix rather than an API change, but it is the change in this release most likely to affect you, and it is listed here so it is not missed: waveform data captured from an HD instrument before 6.0.0 should be treated as unusable rather than rescaled, because the 8-bit path also discarded the low byte. The full entry is under Fixed.

Added

• Handing out gateway access no longer means sending someone a permanent secret. scpi-web invite <name> prints a link and a six-digit code for the same ten-minute invitation: send the link, or read the code down the phone. Either one is exchanged in the browser for a real token, so a scientist never sees an scpi_… string and a leaked chat message stops working in minutes. The same command is how you get someone back in after they clear their browser or switch laptops. The sign-in screen now asks for a join code first, with the raw-token field kept for scripts and CI.
• scpi-web token list now shows how many devices each identity holds and when it was last used.
• The SCPI terminal now works for any connected instrument, not just oscilloscopes. It moved out of the oscilloscope's control rail (eight tabs down to seven) into a drawer across the bottom of the window, opened from a Terminal button in the header and closed with Escape, so a power supply session can be driven by raw SCPI for anything its panel doesn't expose. POST /api/sessions/{id}/command is the new kind-agnostic route behind it; the older /scope/command is unchanged and still refuses a non-scope session, and both now delegate to one shared implementation of what a command does.
• A power supply session now has a readout strip like an oscilloscope's: one card per output showing measured voltage, current and power, and whether that rail is live. Unreadable values still show as --.--, and an output whose on/off state the supply will not report still shows as unknown rather than a confident off.
• Internally, which view a session shows is now decided by a per-kind registry instead of a chain of session.kind checks in the app shell. No visible change on its own, but it's what makes the next instrument kind a single registry entry instead of another branch threaded through the shell.
• The gateway can now host a function generator (SDG series) alongside oscilloscopes and power supplies. Connect one discovered on the network from the home screen, set each channel's waveform, frequency, amplitude, offset and phase, switch outputs on and off, and kill every output with one button. Duty cycle appears for a pulse and symmetry for a ramp, only where they apply. The home screen's "Connect manually" panel gained a one-click mock generator button alongside the mock scope and mock power supply, so an AWG session is reachable with no hardware and no direct API call.
• The readout strip shows what the generator reports back rather than what was sent — an AWG clamps amplitude against its load setting and snaps frequency to its resolution — and every channel appears, including disabled ones, so which outputs are live is answerable at a glance. A value the instrument will not report shows as --.--, and an output whose state it will not report shows as unknown rather than off.
• POST /api/sessions accepts kind: "awg", and the new /api/sessions/{id}/awg/... routes read and control the channels.
• A session that fails now says why. A dropped stream or an error from the instrument shows as a banner under the header carrying the detail, instead of only turning the status dot red. Dismissing the banner clears the message but does not change what the status dot reports — the connection state is the instrument's to report, not a side effect of closing a notification.
• Any session can be disconnected from the header. Disconnect moved out of the oscilloscope toolbar, so a power supply or function generator session no longer has to be left by reloading the page.
• A non-owner is now told, in session, that it's read-only and who owns it, with the same Claim action the home screen offers. Previously they found out by having a write rejected.
• The gateway now serves an admin panel on the machine it runs on, at http://127.0.0.1:8766/. It lists who has access with device counts and last-seen, issues an invitation showing the link and code together with a countdown, cancels one sent to the wrong person, and revokes someone with a confirmation that names how many devices it signs out. There is no sign-in, and three independent things stand in for one: the listener binds loopback, so the operating system refuses every non-local connection; requests must carry a Host of 127.0.0.1 or localhost, which is what stops a DNS-rebound page the bind would let through; and any request carrying a foreign Origin is refused, which is what stops an ordinary cross-origin page that satisfies both of the others. --admin-port moves it; --no-admin switches it off. There is deliberately no flag to bind it to another address.
• Cancelling a pending invitation is new — previously the only remedy for one sent to the wrong person was waiting ten minutes for it to expire.
• A Sessions screen in the admin panel lists every open session with its owner, viewer count and idle time. Release clears the owner, making the session immediately claimable by anyone, with no confirmation. Close ends the session after confirming, and warns when that session is recording so an active capture is not lost accidentally.
• The admin panel gained a page header.
• The live-session list in the Sessions screen now displays readable right-aligned numeric columns for Viewers and Idle time.
• A Horizontal panel in the oscilloscope controls, containing a new timebase stepper that steps in a 1-2-5 ladder (1 ms → 2 ms → 5 ms → 10 ms) with engineering-unit labels, making both slow sweeps and sub-microsecond settings easily reachable.
• capture.csv gained an optional max_points parameter for parity with the JSON export, allowing an oversized record to be trimmed to a smaller sample count before export.

Changed

• On modern Siglent instruments, the gateway now asks the instrument to thin the waveform record to the points the display needs, rather than transferring everything and discarding most of it client-side — no change to what you see on screen. Dialects without the interval command continue to transfer the full record and thin on the server.
• The CSV and JSON exports now stream instead of loading into memory; when a waveform is too large to export safely, the endpoints return an error naming the actual point count and showing how to proceed with max_points. On dialects that cannot report record length in advance, the export proceeds unguarded and logs a warning if it oversizes — nothing is ever silently truncated or decimated.

Fixed

• The gateway no longer freezes while a long capture is in progress. The oscilloscope poll previously fetched a waveform on a timer without checking whether the instrument had finished acquiring — at slow sweep rates this meant requesting much faster than the scope could produce frames, blocking the read until the acquisition was complete, and freezing every control in the web UI (since the session's only worker thread also services commands). On modern Siglent instruments, the poll now checks whether a new acquisition has landed before fetching; other dialects use adaptive timing-based backoff to prevent the freeze.
• On modern Siglent instruments, the live view updates once per completed acquisition. At slow sweep rates, a single update every 14 seconds (at 1 s/div, for example) is expected behaviour — it matches the instrument's own display rate — and is not a defect or regression.
• A channel whose display query failed is now logged instead of silently failing. Previously a poll-path failure made a channel appear 'off': no waveform, no error message, a gateway that appeared healthy and did nothing.
• A thinned live view on a modern Siglent instrument no longer reports a time axis that is too short by the thinning factor. When the live view asks the scope for every Nth point, the instrument keeps reporting the acquisition's sample spacing rather than the spacing between the points it actually sends — so a trace thinned 7:1 claimed to cover a seventh of the sweep it really covered, with sample_rate wrong by the same factor. Measurements and exports were never affected (they are never thinned); the live view's x-axis was. Verified against an SDS824X HD.
• A thinned read that comes back short now fails loudly instead of returning a truncated trace scaled onto a time axis that looks correct, and a deep record that fits comfortably once thinned is no longer refused as too large for one transfer.
• Waveforms captured from a high-resolution Siglent instrument in the default 8-bit transfer mode were 256 times too small. These scopes report one code-per-division figure for their native 16-bit samples and reuse it unchanged for an 8-bit transfer, where the instrument sends only the high byte — so a 3.07 V signal was read back as 0.012 V. Every default waveform read on an SDS800X HD was affected: live view, exports, saved records and any analysis derived from them. 16-bit (WORD) captures were always correct. Verified against an SDS824X HD by comparing both transfer widths against the instrument's own measurement of the same channel. Waveform data captured from an HD instrument before this release should be treated as unusable rather than rescaled, since the 8-bit path also discarded the low byte.
• The live view's new-acquisition check never actually ran on a modern Siglent instrument. The scope answers that query with a response header (INR 8193) rather than a bare number, which the driver could not read, so it concluded the instrument had no such capability — silently, on every poll, with nothing in the log. The poll therefore always fell back to timing-based backoff, which is the very thing the check exists to avoid.
• A measurement the instrument cannot compute yet is no longer reported as a failure. Modern Siglent scopes answer **** for an item that has no value at that moment; that surfaced as a parse error and, in the gateway, as a poll query failed warning for an entirely routine condition. It now raises the new MeasurementUnavailableError (a CommandError subclass, so existing handlers keep working) and the gateway reports the reading as unavailable without logging an alarm.
• A mock oscilloscope now reports its own acquisition length, like a real one. :ACQuire:POINts? had no mock answer, so the query failed and the driver concluded the record length was unknown — which meant the live view's frame-thinning was never exercised against a mock at all, on any dialect. A mock session's live view now thins its frames the same way a real instrument's does.
• The timebase control's stepper no longer moves in 0.1 s jumps or hides sub-microsecond sweeps under a six-decimal display. It previously lived in the Trigger panel with 100 ms increments — one click from 1 ms/div landed on 0.101 s/div — and made sub-microsecond sweeps effectively unreachable.
• Revoking a token now takes effect immediately. The gateway reloads its token store when the file changes, so scpi-web token revoke no longer requires restarting the server to lock someone out — which meant the documented remedy for a leaked credential silently did nothing until someone remembered to restart.
• The token store is now written atomically, so a crash or a concurrent read during a write can no longer leave a tokens.json that the gateway refuses to start from.
• An AWG channel whose waveform the instrument will not report now shows the same --.-- marker every other unreadable field uses, instead of a blank dropdown.
• The home screen's no-instruments-found message no longer singles out "start a Mock scope" — it points at the mock button for whichever instrument kind, since a mock power supply and mock function generator are just as available.
• A revoked credential no longer keeps streaming to a client with an active token, and no longer blocks a new user from claiming the session. A periodic liveness check in each stream catches a revocation in a different process (default every 5 seconds), and closes the client's WebSocket with code 4403 to signal that the credential was revoked rather than the session ending (4410).
• The admin panel's Close and Revoke confirmations are now genuinely modal: a backdrop over the page, focus trapped inside, dismissible by Escape or a backdrop click, and focus returned to the triggering button when closed. Previously they appeared as inline panels below the table, so on a long list they could render below the fold and appear not to work.
• Acting on one session (releasing or revoking) no longer disables the action buttons on every other row — each row's actions are now independent.
• The live-session list refreshes every 10 seconds, so the Idle column stays current instead of being a snapshot from page load. Refreshes pause while a Close confirmation is open, so the list cannot reshuffle under a question you are reading, and a failed refresh leaves the previous rows visible rather than blanking the table.
• The gateway serves index.html with Cache-Control: no-store on both the gateway and admin panel, so a rebuilt webapp appears on an ordinary reload instead of requiring a hard refresh and cache clear.

[5.8.0] - 2026-07-27

Added

• The web gateway can now host a power supply (SPD3303X series) alongside oscilloscopes. A power supply found by the network scan shows up under "Power supplies" with its own Connect button, and "Mock power supply" in the Connect-manually rail opens a PSU session with no hardware at all. The panel gives every output a voltage setpoint, a current limit, an on/off switch, and live voltage/current/power readings that update as the instrument does.
• The output switch reflects what the instrument actually reports, never what was just asked for, and a reading the supply will not answer shows as --.-- or "state unknown" rather than as a confident zero or a confident "off". An SPD3303X's CH3 has no output-state query at all, so this is the normal case there, not an edge case.
• POST /api/sessions accepts an optional kind field ("scope" by default, so an existing caller that omits it is unaffected), and the new /api/sessions/{id}/psu/... routes read and control PSU outputs. A session refuses to start if the connected instrument identifies as a different kind than the one asked for. The 26 existing /scope/ routes are unchanged, as is every public name on InstrumentSession.

[5.7.1] - 2026-07-26

Fixed

• A synthesized waveform no longer carries a spurious full-amplitude sample where a cycle begins. synthesize builds its time array as t0 + i / sample_rate, and for an arbitrary t0 — a trigger crossing, a free-run drift offset, a device model's lead-in — that sum cannot land exactly on a whole number of periods, so the cycle count came out an ULP below an integer and the modulo mapped it to 0.9999999999999991 instead of 0. Continuous kinds never showed it, but square read its low level at the instant it should switch high and ramp reset a sample early: one sample at full amplitude, roughly 50 int8 codes of spike in a mock capture. The cycle count is now snapped to a whole cycle when it sits within a few ULP of one, which moves the cycle position by less than any consumer can resolve and leaves every other sample bit-identical. Note this covers the cycle wrap only; a sample landing exactly on a duty or edge threshold is still subject to the same float error, which shifts that edge by one sample rather than inverting a sample, and tests/test_cycle_boundary.py pins that residual so it stays visible.

[5.7.0] - 2026-07-26

Added

• batch_capture gained a circuit breaker: max_consecutive_failures (default 3) stops a run after that many back-to-back capture failures, counted across configurations and reset by any success. This guards the common unattended failure — a trigger level the signal never crosses — which otherwise times out on every single capture: at a 70 s timeout and 100 triggers per configuration, that is hours of waiting to collect nothing, previously surfaced only as one log line per failure. Everything gathered before the breaker trips is still returned, with the failed entries carrying their error field, and an error-level log names the reason and the shortfall. Note this changes the default behaviour of a run whose captures all fail: it now stops early instead of attempting every planned capture. Pass max_consecutive_failures=None to restore the old behaviour; a run with genuinely sparse triggers should raise max_wait instead.
• A mock AWG's output can now be captured by a mock oscilloscope: wire AwgLoopback(awg_connection, awg_channel=1) into the scope's signals=, and a SCPI write to the AWG changes the next capture. Previously the mock AWG was a write-only surface — you could set a waveform and read it back, but nothing in the system responded to its output, which made it untestable end to end without hardware. MockConnection's signals= now accepts a callable returning a SignalSpec as well as a static one, evaluated at every acquisition; this is a general extension point, not AWG-specific, so any dynamic signal source works the same way. scpi_control.dut.RCLowPass is a new first-order RC low-pass usable as a device under test between the two instruments, or on any array on its own, using exact zero-order-hold discretisation; the mock applies it with a lead-in rendered before the capture window so a capture has no settling transient at its head. Two conversions are worth knowing, because getting them wrong produces plausible-looking output: AWG amplitude is peak-to-peak and is halved into SignalSpec's peak amplitude, and phase is degrees and becomes radians — an output that is off reads flat. ARB is captured as a sine with a logged warning, since the mock stores no arbitrary sample data, and extreme duty values are clamped rather than raising. Non-breaking: a static SignalSpec in signals= behaves exactly as before.

Changed

• start_continuous_capture with output_dir set now returns per-capture metadata instead of an empty list. Entries carry timestamp, elapsed_time, capture_num and the files written, omitting the bulky waveform arrays (those are on disk). Previously the function returned [] whether it had written ten thousand files or none, so a caller had no way to tell what happened. In-memory mode (no output_dir) is unchanged and still returns the waveforms.

Fixed

• start_continuous_capture no longer hides save failures. Saving sat inside the capture loop's broad except, which logs and continues, so a rejected file_format or an unwritable output_dir failed identically on every iteration for the run's entire duration while the function returned an empty list — an overnight run producing an empty directory and no signal at all. This is what kept the npz default (fixed in 5.6.0) invisible for so long. Saves now have their own handler: if the first save fails and no file has been written, the run stops immediately with a SiglentError naming the format and chaining the underlying cause, because that is configuration rather than a transient fault and every later attempt would fail the same way. Once one file has landed the configuration is proven, so later failures are counted, logged, and summarised at the end without aborting a long unattended run. The raised error keeps its original type when it is already a library exception, so a caller catching InvalidParameterError still catches it. A run that writes nothing because no save was ever attempted — every capture yielding no waveforms, from disabled channels or a failing acquire() — now ends with a warning too; that is the same empty-directory symptom reached by a different route, and it was equally silent.
• batch_capture no longer discards the whole run when interrupted or when the instrument stops answering. It had no KeyboardInterrupt handler at all, so an operator stopping a run they could see was doomed lost every capture already taken; it now keeps and returns them, matching what start_continuous_capture already did. The guard covers the whole per-configuration body rather than only the capture, because the gap between configurations — two socket writes plus the settle delay — is exactly where an impatient operator tends to press Ctrl-C.
• batch_capture now records any SiglentError as a failed entry, not just SiglentTimeoutError. A dropped link mid-run previously propagated and discarded every capture already taken, which is the precise loss the failed-entry path exists to prevent. Connection failures count toward the circuit breaker as well — an instrument that has stopped answering is exactly what it is for. A failure while applying a configuration stops the run and returns what was collected.

[5.6.0] - 2026-07-26

Fixed

• save_waveform now accepts npz and h5 as aliases for NPY and HDF5. npz is start_continuous_capture's default, so an overnight run using default arguments previously raised on every save and wrote nothing; four shipped examples were broken the same way. The same spellings were already valid as file extensions via auto-detect — that inconsistency is what produced the bug.
• batch_capture now parses the SI unit strings its own docstring documents ('1us', '500mV'), via the new scpi_control.units.parse_si_value. Previously the documented example raised TypeError and discarded every capture already taken. Numeric scales keep working unchanged.
• capture_single now polls acquisition_status() instead of sleeping a fixed 0.5 s, with a timeout derived from the current timebase and a new optional max_wait. On a slow timebase it previously returned the previous acquisition, which batch_capture then recorded under the new config's label — wrong data, correctly formatted, no warning. A timed-out capture now appears in batch_capture's results as an entry with empty waveforms and an error field, rather than aborting the run or holding stale data.
• capture_single no longer clobbers a user-configured NORM trigger mode. It previously always armed a single-shot acquisition regardless of trigger mode; it now leaves NORM alone and waits for the next natural trigger, matching what TriggerWaitCollector.wait_for_trigger already did, and continues to arm a single shot in every other mode as before.
• parse_si_value scales by shifting the decimal exponent rather than multiplying by a float, so '10us' is exactly 1e-05. The multiply-based approach produced 9.999999999999999e-06, which reached the instrument as TDIV 9.999999999999999e-06 instead of TDIV 1e-05.
• The oscilloscope screenshot at the top of the README no longer fails to load on the PyPI project page. It used a repository-relative path, which GitHub resolves against the repo but PyPI cannot — PyPI renders the README as a standalone document with no repo context, so the image silently broke there while looking fine on GitHub. It now uses an absolute raw URL. Note that PyPI project pages are immutable per release, so the fix appears from the next release onward; already-published pages keep the broken image.

[5.5.0] - 2026-07-26

Added

• Four new synthetic signal kinds join sine/square/triangle/ramp/dc/noise: chirp, a repeating frequency sweep from frequency to end_frequency over sweep_time (linear or, with sweep_log, logarithmic), with phase carried across the retrace so the sweep boundary is discontinuity-free — it's also the one kind the mock free-runs rather than trigger-aligns, since a sweep has no stable period; exponential, a square wave through an RC network with time constant tau, evaluated at its periodic steady state so it's settled from the first cycle rather than over the first few, and split by duty; pulse, a trapezoid whose width and edge rate are set independently of the period by pulse_width and edge_time (edge_time may be 0 for an ideal edge) rather than by duty, which pulse ignores — pulse_width is the 50%-to-50% (FWHM) width, matching both instrument convention and the threshold the repo's timing analyzer measures at, so the flat top runs for pulse_width - edge_time; and multitone, a fundamental plus a coherent harmonic series (harmonics gives the relative amplitudes of the 2^nd, 3^rd, ... harmonic), where amplitude sets the fundamental rather than the peak of the sum — deliberately not normalized, since normalizing would make THD depend on the harmonic set. The new kinds give the repo's timing, THD and spectrum analyzers a closed-form answer to check against for the first time, rather than only a pure sine, a square, or noise. Non-breaking: every new SignalSpec field is optional and appended at the end of the dataclass, so existing specs and positional construction are unaffected.

Fixed

• Signal synthesis documented ringing as having "no effect" on kinds without edges. That was wrong: ringing finds its edges with np.diff(samples) != 0 — any nonzero sample-to-sample change — so on a continuous signal every sample qualifies and the impairment acts as a derivative-weighted filter rather than an edge response. At 1 MSa/s, 1 V amplitude and 50 kHz ringing, the measured deviation is 1.04e-1 V on chirp, 5.60e-2 V on exponential, 1.33e-2 V on multitone and 9.85e-3 V on sine; only dc is a genuine no-op. The behaviour is unchanged and defensible as a band-limited edge response — only the documentation was wrong, and it now states the real effect with those measured figures.
• SignalSpec accepted non-finite kind parameters. pulse_width=nan passed validation entirely (both nan <= edge_time and nan > 1/frequency are False) and produced a finite but silently wrong waveform with no nan in the output to give it away; tau=inf produced an all-nan trace and a numpy RuntimeWarning. end_frequency, sweep_time, tau, pulse_width and edge_time are now checked with np.isfinite and raise InvalidParameterError.
• A multitone spec whose harmonics contained a non-numeric element raised a raw TypeError from numpy instead of the InvalidParameterError the module promises for every bad parameter.
• The mock's trigger search now sees the ideal signal. _trigger_crossing synthesizes a short reference trace to locate where the signal crosses the trigger level, but left the impairments enabled, so drift, glitches and ringing each moved the crossing (68.85 us to 66.41, 64.70 and 10.25 us respectively) and ringing added roughly 21 ms per acquisition. It now zeroes them alongside the noise it already stripped, so trigger alignment no longer jitters with the impairment settings.

[5.4.0] - 2026-07-26

Added

• The mock instrument can now misbehave the way real hardware does. It keeps a real SCPI error queue behind SYST:ERR? (drained by *CLS and *RST) and rejects out-of-range parameters instead of storing them — a bad value is accepted by the transport, ignored, and reported as -222,"Data out of range (<parameter>)" on the next query, which is what an instrument actually does. Validation covers the Siglent scope, PSU and AWG personalities, plus the Tektronix and LeCroy scope dialects; DAQ mode gains the SYST:ERR? queue itself but has no numeric parameters to validate (its only two writes, ROUT:SCAN and TRIG:SOUR, are non-numeric). reject_if_invalid gained an optional max_magnitude override, since the shared scope-calibrated bound rejected legitimate AWG frequencies (and tightened it for AWG percentages/phase, which never legitimately need it), plus an optional non_negative bound for parameters whose real-driver validation is >= 0 rather than > 0 (trigger holdoff, AWG phase/ramp symmetry, PSU voltage/current). FunctionGenerator.get_error() and PowerSupply.get_error() go from timing out entirely to answering SYST:ERR? for the first time; DataLogger.get_error() already answered before this work, but only ever the hardcoded +0,"No error" — it can now report a real queued error too. Unimplemented commands still time out under strict=True, unchanged.
• SignalSpec gained optional signal impairments — baseline drift, glitches and edge ringing — so measurement and analysis code can be exercised against imperfect signals instead of mathematically clean ones. All default to off, so existing synthesised output is unchanged. Ringing is continuous across stream() chunks, the same way drift already is, rather than resetting at each chunk boundary; it's an edge impairment, meaningful only on signals that actually have edges ("square", or a pulse-like "ramp").

[5.3.0] - 2026-07-25

Fixed

• Provenance and report output no longer state things the software cannot know. Enhanced CSV headers label the save timestamp as Saved and only emit Captured when provenance carries a real acquisition time; the report manifest shows unknown instead of passing a file's modification time off as a capture time; AI-generated Key Findings and Recommendations are labelled as such in both Markdown and PDF output, so machine-written text is distinguishable from engineering judgement in a signed report; manifest file paths render verbatim instead of being italicised and stripped of underscores by the markdown converter; the shipped comparison and batch examples no longer stamp synthetic waveforms with a real instrument identity; and scpi-extract reports missing provenance without inventing a reason for it.

[5.2.0] - 2026-07-25

Fixed

• Measurements now work on modern-dialect Siglent oscilloscopes (SDS800X HD, SDS5000X and siblings). measure() previously sent the legacy PAVA? command, which does not exist on those instruments, so every measurement failed — including the whole GUI Measurements tab and the web gateway's measurement calls. It now uses the documented :MEASure:SIMPle subsystem. Note that a modern measurement is not side-effect-free: it enables the measurement function, pins the measurement mode to simple (an instrument left in advanced mode is not guaranteed to answer the simple-value query the same way), sets the simple-measurement source, and switches the requested item on, which is visible on the instrument display. These are left in place rather than cleared, because the instrument's clear command is all-or-nothing and would remove measurements you configured yourself — so enabled measurement items accumulate on the instrument display across repeated calls rather than being switched off.

[5.1.0] - 2026-07-25

Changed

• Licensing metadata migrated to PEP 639: the package now declares License-Expression: MIT and ships License-File: LICENSE instead of the deprecated license = { text = "MIT" } table. Building from source now requires setuptools>=77 (raised from 61); installing a prebuilt wheel is unaffected.
• PyPI summary, keywords, and classifiers rewritten to lead with the SCPI/test-automation category, and the README now opens with a short pitch instead of a single dense paragraph.

Fixed

• GUI: the Measurements tab now works for all 15 measurement types. Top, Base, Max, Min, Positive Width, and Negative Width previously showed --- (indistinguishable from an instrument fault) because the panel called core methods that didn't exist; it now routes every type through the instrument's measurement dispatch.
• GUI: the duty-cycle marker divides the pulse width by the signal's actual period instead of the gate span, so a marker spanning several cycles reads the true duty cycle — or N/A when no period is detectable — rather than a value that shrank as the gate widened.
• GUI: the DAQ "Suggest Thresholds" action no longer errors on success; the analysis-result signal now carries the structured suggestion instead of rejecting it.
• scpi_control.gui.widgets now imports its widgets lazily, so importing a Qt-free submodule (such as the measurement-marker math) no longer pulls in PyQt6. Importing those modules previously failed outright wherever the optional gui extra was not installed. from scpi_control.gui.widgets import ChannelControl is unchanged.

[5.0.0] - 2026-07-24

⚠️ Breaking Changes

• Web gateway: every /api/* request now requires a bearer token, and the live-stream WebSocket requires one too. On first run scpi-web mints a token and prints a http://127.0.0.1:8765/?token=… URL; mint more with scpi-web token add <name>. Migration: scripted HTTP clients must send Authorization: Bearer <token>; WebSocket clients must offer the scpi-token.<token> subprotocol (alongside a plain scpi). A query-parameter token is accepted only on the initial web-UI page load and is rejected on /api/*. GET /api/health stays unauthenticated. See the gateway security guide.
• Web gateway: the interactive API docs at /docs and /redoc are removed and the OpenAPI schema moved to /api/openapi.json (token required) — the old locations served the whole control surface unauthenticated.
• Reference waveforms: stored metadata moved from a pickled dict to a JSON string so reference files load without allow_pickle. Files saved by 4.x raise an error naming the file until converted. Migration: run scpi-web references migrate once.
• Power supply (SPD3303X/-E): ovp_level/ocp_level now raise NotImplementedError — the SPD3303X command set has no protection subsystem, so these calls never armed anything on real hardware (they emitted a FutureWarning in 4.1.0). The model's has_ovp/has_ocp capabilities are now False. Migration: stop calling these on an SPD3303X; gate on has_ovp/has_ocp if you support multiple PSU models.
• Testing: MockConnection now defaults to strict=True, so an unmatched PSU/AWG/DAQ query raises TimeoutError like real hardware instead of returning "". Migration: pass strict=False explicitly to restore the old lenient behavior.
• Testing: MockConnection no longer answers the legacy C<n>:WF? waveform read on a modern-dialect instance (SDS800X HD / SDS5000X); it now times out, matching real modern hardware, which documents no such command. Legacy-dialect scopes are unaffected.

Added

• Web gateway: named bearer tokens — scpi-web token add|list|revoke <name>, stored hashed in ~/.siglent/tokens.json (relocate with --config-dir).
• Web gateway: session ownership. The identity that creates an instrument session may write to it; anyone authenticated may read, stream, and export. Non-owner writes return 409. POST /api/sessions/{id}/claim takes over a session whose owner has been idle past --abandon-after (default 300 s) and is not actively watching the stream; POST /api/sessions/{id}/owner hands it over by name (or releases it with ""). The web UI shows a read-only badge and a claim button to non-owners.
• Web gateway: --allow-port to permit instrument ports beyond 5025, --max-sessions (default 8) to bound concurrent sessions, and an unauthenticated GET /api/health endpoint.
• Web gateway: scpi-web references migrate converts pre-5.0 reference files to the new format (atomic and idempotent; --dir to point at a non-default store).

Fixed

• Security: loading a waveform or reference archive no longer unpickles it, so a crafted .npz can no longer execute code via scpi-extract, the report generator's loader, or the gateway. The one remaining place that reads the old pickled format is scpi-web references migrate, which the user runs explicitly on their own files.
• Security: reference lookup no longer honours absolute paths or .. traversal, and listing references no longer deserializes every file in the storage directory.
• Security: session creation validates the target before connecting — hostnames are resolved and every resolved address checked, loopback/link-local/cloud-metadata addresses are refused, ports must be allowlisted, and a failed connect returns a generic message rather than reflecting the peer's bytes (no SSRF port-scanning or banner-grabbing).
• Web gateway: full-resolution CSV/JSON serialization of deep-memory captures runs off the event loop, so one large export no longer freezes the gateway for other users; the session cap holds correctly under concurrent requests.
• Reference storage: rename_reference no longer leaks the source file handle (a Windows-only WinError 32 that made it fail); reference lookups now confine themselves to the storage directory.
• Web UI: file downloads (capture CSV, screenshot, waveform JSON, trend CSV) fetch with the bearer token instead of bare <a href> links, which could not carry an Authorization header and would have returned 401 against the authenticated gateway.

[4.1.1] - 2026-07-24

Fixed

• Packaging: the six screenshots in the README are now absolute URLs, so they render on the PyPI project page. They used repository-relative paths, which PyPI cannot resolve because it renders the README with no repository context — the images showed as broken there while displaying correctly on GitHub.

[4.1.0] - 2026-07-23

Added

• Testing: a vendor-example conformance corpus (tests/wire_forms.py) pins every command in the covered SCPI tables to a request/response pair transcribed from the vendor's programming guide, with document and page citations. A coverage test (tests/test_wire_conformance.py) fails the suite if any command in those tables (the legacy and modern scope dialects, the SPD power-supply overrides, and the SDG function-generator overrides — 159 commands) has no cited corpus entry, so an invented command can no longer be added silently to a covered table. (The IEEE-488.2 base, the generic fallbacks, and the Tektronix/LeCroy/DAQ tables are not yet enforced — see docs/development/wire-form-inventory.md, which records the full audit.)
• Oscilloscope: modern-dialect deep-memory waveform capture is chunked over :WAVeform:MAXPoint windows using :WAVeform:STARt, so records larger than a single transfer are reassembled correctly.
• Oscilloscope: modern-dialect (SDS800X HD / SDS5000X) waveform capture now goes over the documented :WAVeform:SOURce/:WAVeform:PREamble?/:WAVeform:DATA? subsystem instead of the legacy C<n>:WF? DAT2/DESC forms, which have zero occurrences anywhere in the modern programming guide (audit H9). Voltage and time reconstruction use the guide's own documented formulas (p.758/p.759).

Fixed

• Oscilloscope (legacy Siglent): measurements use the documented C<n>:PAVA? <param> form and parse its two-field response, and sample-rate responses carrying an SI magnitude letter (SARA 500.0kSa) now parse. The previous PAVA? <param>,C<n> form and scientific-only sample-rate parsing failed on real hardware (audit H7/H30/H8/H34).
• Power supply (SPD3303X/-E): measurement queries use the documented MEASure:VOLTage? CH<n> form (channel as an argument), tracking mode uses the documented numeric OUTPut:TRACK {0|1|2}, and output state is read by decoding the SYSTem:STATus? word — the instrument documents no output-state query (audit H6/H19/H20).
• Function generator (SDG): parameter readback uses the documented bare C<n>:BSWV? / C<n>:OUTP? queries and parses the returned key-value list; the previous per-field selector grammar (C<n>:BSWV? FRQ) does not exist on the instrument (audit H5).
• USB/GPIB/serial: VISAConnection can now be instantiated — it implemented neither read nor read_raw, so it raised TypeError at construction and every documented VISA example failed (audit H10).
• Testing (mock fidelity): the mock now answers documented legacy probe (C<n>:ATTN?) and bandwidth-limit (BWL?) queries instead of timing out; the DAQ dispatch no longer matches R? as a substring (which hijacked SYST:ERR?/TRIG:SOUR?) and interprets scan-list/trigger writes; and MockConnection.read_raw honors the requested byte count.

Deprecated

• Power supply: setting ovp_level/ocp_level on an SPD3303X/-E now emits a FutureWarning. The SPD3303X command set contains no protection subsystem, so these calls have never armed anything on real hardware. In v5.0.0 the model's has_ovp/has_ocp capabilities become False and the calls raise NotImplementedError.
• Testing: MockConnection gained strict=True, which makes unmatched PSU/AWG/DAQ queries raise TimeoutError like real instruments instead of returning "". Strict becomes the default in v5.0.0 — pass strict=False explicitly to keep the old behavior past that release.
• Testing: MockConnection still answers legacy C<n>:WF? writes on a modern-dialect (SDS8xx HD/ SDS5000X) instance, even though the driver's modern capture path no longer sends it (audit H9, Task 18). This is a backward-compatibility shim for anything still issuing that form by hand; the modern guide documents no such command, and the mock handler is removed in v5.0.0.

[4.0.0] - 2026-07-22

⚠️ Breaking Changes

• Power supply and function-generator readback (measure_voltage/measure_current/measure_power, setpoint/OVP/OCP getters, and AWG frequency/amplitude/offset/phase) now raise CommandError when the instrument returns an unparseable response, instead of silently returning 0.0. Callers that relied on the silent zero must handle the exception.

Added

• Dev tooling: pytest-xdist in the dev extra for parallel local test runs (python -m pytest -n auto); complements --testmon (use one or the other per invocation).
• Report generator: before/after comparison and multi-DUT batch reports — load multiple capture-file runs, overlay their waveforms, and get delta tables (comparison) or per-DUT summary with aggregates and yield (batch). Available from the Python API and the GUI's new Report → Comparison / Batch Report dialog.
• Report generator: optional raw-data appendix (source-file manifest with SHA-256 checksums, capture timestamps, and instrument identity from provenance) and a configurable sign-off block (template-defined roles with signature/date lines). Both also work on ordinary single-run reports.
• Report generator: top_flatness waveform statistic (top-plateau deviation as % of amplitude); the probe-calibration preset now evaluates overshoot/undershoot (pass/fail) and top flatness (warning).

Fixed

• Report generator: the waveform loader no longer drops acquisition provenance when loading capture files.
• Analysis: corrected THD (each harmonic is read at its own bin, robust to an off-bin fundamental), noise/SNR (estimated from a signal-free residual instead of the whole-signal RMS), vamp (Vtop − Vbase amplitude, not the vertical midpoint), overshoot/undershoot (only reported for flat-topped signals), duty cycle (correct when the capture starts on the high level), DC classification (no longer fooled by a large offset over real ripple), FFT peak detection (finds peaks below 0 dB), and SciPy window handling for the power spectrum / spectrogram.
• Reports and the live webapp spectrum now use a single THD engine, so they report the same number for the same capture. Report-path THD now sums the first 5 harmonics (previously 10).
• DAQ AI threshold suggestions now bind each extracted number to its label instead of grabbing the first number in the sentence.
• Comparison/batch reports: pass/fail criteria now match analyzer statistics (case-insensitive plus display-name aliases) — previously the shipped criteria template produced empty verdicts. Criteria that cannot be evaluated are surfaced as warnings (never silently dropped), half-open RANGE criteria are enforced, verdicts are severity-aware (only critical criteria gate) with a distinct INCOMPLETE state, and yield reports an incomplete count instead of inflating the pass rate.

[3.3.1] - 2026-07-21

Changed

• README and docs now cover the full current feature set: multi-vendor scopes, DAQ, PSU/AWG models, the web gateway and usb extras, all four CLI tools, load_waveform()/scpi-extract usage, and API reference pages for signal_synth, waveform_io, provenance, and scpi_extract. Every example now carries a fully spelled-out docstring (what it shows, what it needs, what to expect).

Fixed

• DataCollector.save_data()/save_batch() no longer raise InvalidParameterError with their default format — the format is now auto-detected from the filename extension (pass CSV, CSV_ENHANCED, NPY, MAT, or HDF5 explicitly to override).
• Report-analyzer calibration guidance and example-script output are now ASCII-safe, so they no longer crash on Windows consoles with legacy codepages.
• Documentation sweep: corrected the README's report-generator sample to the real API, fixed the default SCPI port (5025) and GUI frame-rate claims, repaired all broken links and anchors across README and the docs site, synced the docs changelog page through 3.3.0, and excluded internal planning files from the built documentation site.

[3.3.0] - 2026-07-21

Added

• signal_synth.stream(): an infinite or duration-bounded generator of phase-continuous voltage chunks for live-signal simulation — optional wall-clock pacing (realtime=True), correct seeded-noise semantics (seed advances per chunk, so seeded streams are reproducible without repeating noise blocks), and eager parameter validation.
• Dev tooling: pytest-testmon in the dev extra for impact-based local test selection (python -m pytest --testmon); CI and pre-merge checks still run the full suite.

[3.2.0] - 2026-07-21

Added

• Synthetic signals: a public generator API — SignalSpec plus synthesize()/make_waveform() in scpi_control.signal_synth — producing sine, square (with duty cycle), triangle, ramp, DC, and Gaussian-noise waveforms as numpy arrays or ready-to-analyze WaveformData, with seedable reproducibility.
• Mock scopes now synthesize realistic waveforms by default (all vendor personalities): traces are computed from the mock's live timebase, V/div, offset, and trigger state at every acquisition — over-range clips at 8-bit full scale, the trigger level/slope aligns the edge at the window center, and unseeded captures animate with fresh noise. Pass signals={ch: SignalSpec(...)} to choose the signal, or waveform_payloads bytes for the old fixed-payload behavior (unchanged). Web-gateway mock sessions stream synthesized signals out of the box.

Changed

• MockConnection's default (no waveform_payloads given) no longer serves a fixed 4-byte payload — channels synthesize state-coupled signals instead.

[3.1.0] - 2026-07-21

Added

• Report generator: a built-in probe-calibration template preset (ReportTemplate.create_probe_calibration_template(), seedable from the Template Manager via "Create Probe Cal Template") — probe_calibration test type, a compensation procedure, and overshoot/undershoot/ringing/flatness pass/fail limits.
• Examples: four new runnable, no-hardware examples — report_ai_qa.py (local-LLM tool-calling Q&A over a report), network_discovery.py (scan the network for SCPI instruments), report_branding.py (apply company branding/colours to a report), and report_computed_analysis.py (deterministic, LLM-free report analysis).
• Examples: a tests/test_examples_smoke.py guard that executes the no-hardware examples, compile-checks the rest, and blocks known-stale tokens from reappearing.
• Docs: a real screen capture and a real 1 kHz calibration-square-wave plot from a Siglent SDS824X HD (in docs/images/), plus the raw capture committed as a test fixture (tests/fixtures/cal_square_sds824x.npz) that exercises the analyzer against genuine hardware data.
• Acquisition provenance: waveforms captured with acquire()/get_waveform() now record the instrument identity, per-channel settings, trigger configuration, timebase, and UTC timestamp, embedded in every save format (new keys only — existing files and keys are unchanged; pass provenance=False to skip the snapshot on high-rate paths).
• scpi_control.waveform_io.load_waveform(): public parser that reads all five waveform file formats (old and new files) into numpy arrays plus normalized metadata/provenance, with an optional to_dataframe() pandas helper.
• scpi-extract CLI: inspect provenance (--info), dump raw data (--csv), or emit machine-readable metadata (--json) from any saved waveform file.
• Waveform savers now persist the previously dropped timebase, voltage_scale, and voltage_offset fields; plain CSV gains a #-commented provenance header (channel, sample rate, scales, instrument, timestamp) (suppress with save_waveform(..., bare=True)).

Changed

• Report generator AI: every oscilloscope system prompt now carries one shared grounding rule (claim only what the report data or a tool actually returned; say so plainly when a value is missing rather than inventing it).
• Report generator AI: key-findings and recommendations are parsed more tolerantly — multi-digit numbering, markdown-bold list markers, and a leading preamble line no longer corrupt the extracted list.
• Report generator (internal): the oscilloscope and DAQ LLM prompt layers now share one grounding constant and the prompt-lookup / chat-prompt assembly via llm/_prompt_helpers.py, removing duplicated boilerplate.

Fixed

• Report generator analysis: WaveformAnalyzer's noise check no longer computes an O(n²) full autocorrelation — it now reads only the two autocorrelation lags it uses, so analysing large captures (e.g. 1,000,000-sample records) completes in milliseconds instead of hanging. Also removes a divide-by-zero RuntimeWarning on constant signals.
• Examples: repaired broken examples — simple_capture.py and advanced_analysis.py referenced a non-existent waveform.time_interval (now the sample period 1.0 / sample_rate), and the interactive tutorial saved with an invalid format="NPZ" (now "NPY"). The report-generation and branding examples now degrade cleanly when reportlab is absent instead of crashing.
• Examples: updated the stale Siglent-Oscilloscope package name (and dead repo links) to SCPI-Instrument-Control, and corrected the README's hardware/no-hardware annotations.
• Screen capture on modern-dialect scopes (SDS800X HD): ScreenCapture now selects the correct SCDP command per dialect and reads the raw BMP sized by its own header, instead of over-reading a fixed 10 MB — which timed out and dropped the connection. scope.screen_capture.get_screenshot_pil() now works on the modern scopes.
• Report generator analysis: WaveformAnalyzer.detect_signal_type no longer misclassifies clean periodic signals (e.g. a square wave captured over many periods) as noise. The noise check now measures spectral concentration — whether one frequency bin dominates — instead of testing autocorrelation at an arbitrary fixed lag.
• Report generator analysis: WaveformAnalyzer.detect_signal_type now classifies pulse/PWM signals (non-50%-duty square waves) as pulse instead of sawtooth. Two-level (flat-topped) signals are separated from ramps before the harmonic scorers run.
• Report generator DAQ AI: the data-logger analysis prompts now carry the same grounding rule as the oscilloscope prompts (claim only what the data shows; say so when a value is missing), and the session-summary prompt no longer invites a "Here is the summary" preamble.
• Report generator PDF: characters outside the built-in font's range (⚠, ✓/✗, σ, Ω, …) no longer render as blank boxes — they are normalized to readable text, with a catch-all so any unexpected glyph can never box out.

[3.0.0] - 2026-07-17

⚠️ Breaking Changes

• The report generator's WaveformData fields are renamed, and channel moved from the first constructor argument to the third. channel_name, time_data, and voltage_data are now channel, time, and voltage, matching the base class's field order (time, voltage, channel, ...). Because the position changed along with the name, positional construction such as WaveformData("CH1", t, v, ...) does not fail with a recognizable rename error — it silently assigns "CH1" to time and blows up later with AttributeError: 'str' object has no attribute 'shape'. Construct with keywords (channel=, time=, voltage=) to migrate safely. This is a documented public API — see docs/report-generator/api-reference.md.
• WaveformData.source and WaveformData.description are removed, on both the library and report waveform types. WaveformData(..., source="x") now raises TypeError instead of silently accepting the keyword.
• WaveformData.timebase and WaveformData.voltage_scale are None unless an instrument actually reported them, instead of being derived by assuming a 14-division horizontal grid and an 8-division vertical one — Siglent's geometry, previously applied to Tektronix and LeCroy captures too — with a flat trace given a bare 1.0 V/div. Real acquisitions are unaffected: all three vendor backends pass genuine scope values. Code that does arithmetic on timebase/voltage_scale from a synthetic waveform (a math channel, or hand-built test data) — e.g. wf.timebase * 1e6 — now raises TypeError on None instead of silently using an invented number. Guard with a None check, or supply real values.

Added

• Tektronix and LeCroy oscilloscope support (core control, waveform acquisition, measurements): two more wire dialects (tektronix, lecroy) alongside the existing Siglent legacy/modern pair, auto-detected from *IDN? via a manufacturer-first routing step. Covers the Tektronix TBS1000C Series and 2 Series MSO, and the LeCroy WaveSurfer 3000z and WaveRunner 8000 series. See the SCPI Dialects guide for the full per-vendor command tables and known gaps (e.g. LeCroy statistics/cursors/holdoff, Tek 16-bit waveform transfer).
• Tektronix MSO ⅘/6 Series support (MSO44, MSO46, MSO54, MSO56, MSO58, MSO58LP, MSO64), including 6- and 8-channel models.
• Badge-based measurements for the modern Tektronix MSO families, which also enables measure() on the MSO 2-Series.
• Report generator — local-LLM analysis and richer PDFs (pip install "SCPI-Instrument-Control[report-generator]"). Everything here is local-only — no cloud providers, no API keys.
• Local-LLM (Ollama) tool calling: the model can call read-only report tools to ground its answers, behind a capability gate that no-ops when the local model cannot do tool calls.
• Waveform analysis tools for the model — analyze_plateaus, list_edges, and analyze_spectrum — plus WaveformAnalyzer.calculate_spectrum.
• Deterministic no-LLM analysis: a ComputedAnalyzer populates per-waveform statistics and regions on every report, and composes a summary, findings, and recommendations when no LLM wrote them. Reports now attribute their summary by source (manual / AI / computed) rather than a bare AI flag.
• Vector PDF plots: waveform, FFT, and region plots render as scalable vector graphics instead of rasterized images.
• Page framework: a running header/footer and page numbers on every page, with section headings kept from stranding at the bottom of a page.
• Template branding: a template's logo, company name, header/footer text, and four brand colours apply to the generated PDF, and a built-in starter template can be seeded from the Template Manager.

Changed

• On the Siglent modern dialect, trigger.holdoff, measurement statistics, cursors, and channel.unit now raise FeatureNotSupportedError immediately instead of writing an unsupported command and timing out (SiglentTimeoutError). Code that caught the timeout to detect these unsupported operations must now catch FeatureNotSupportedError instead.
• Probe-ratio wire format on the legacy and modern dialects now serializes floats compactly (10.0 is sent as 10), matching how real scopes echo the value back.
• add_measurement now validates and case-normalizes its measurement type (unknown types raise instead of being sent verbatim to the instrument).
• Channel numbers are validated against the connected model's channel count instead of a fixed 1-4 range. Scopes with fewer than four channels now raise InvalidParameterError for a channel they do not have, where they previously queried it.
• The report generator's WaveformData is now a subclass of scpi_control.waveform.WaveformData (see Breaking Changes above for the field rename and reorder this involved). Report waveforms now inherit the library's array-shape validation, and channel is now guaranteed to be a str by the type itself. The loader's explicit str() calls at each construction site are unchanged and now redundant, not removed — they stay as a defensive measure against np.str_/bytes values handed back at the MAT/HDF5 boundary.

[2.0.0] - 2026-07-15

⚠️ Breaking Changes

• Removed the siglent compatibility shim. import siglent now raises ModuleNotFoundError; use import scpi_control (identical API). The shim had emitted a DeprecationWarning since 1.0.0, which announced its removal in v2.0.0. If you cannot migrate yet, pin SCPI-Instrument-Control<2.0.
• Python 3.9+ required (was 3.8+; Python 3.8 reached end of life in October 2024). CI now tests Python 3.9 through 3.14.

Added

• Browser-based lab gateway (pip install "SCPI-Instrument-Control[web]", then scpi-web or python -m scpi_control.server): a FastAPI server that manages named multi-instrument sessions and serves a React UI to any browser on the LAN. Mock-first — every feature works against the built-in mock scope (mock: true).
• Live waveform streaming over WebSocket, with channel, timebase, trigger, and acquisition (run/stop/single/auto) controls
• LAN instrument discovery and a dashboard-style home screen
• Measurements with live values and cross-tab-synchronized selection
• Software math channels (M1/M2) streamed as canvas traces
• SCPI terminal, instrument screenshot PNG, full-resolution waveform export as CSV and JSON
• FFT spectrum view computed server-side from full-resolution data (window selection, peak markers, THD readout)
• Software Butterworth filters (lowpass/highpass/bandpass) streamed as F1/F2 traces
• Reference waveforms: save named snapshots, ghost overlay on the canvas, live correlation and max-deviation statistics
• Measurement trend recording: server-side ring buffer at ~1 Hz, live trend chart, CSV export; the measurement selection locks while recording
• Dual-dialect SCPI support: legacy (e.g. SDS1104X-E) and modern (e.g. SDS800X HD) Siglent command sets behind one API, auto-detected from *IDN? with a manual override

Changed

• Connection layer hardened: thread-safe socket handling, exact-length binary reads, default SCPI port 5025
• wait_for_trigger honors a user-configured NORMAL trigger mode instead of forcing SINGLE
• Black formatting standardized on the 26.x stable style; dev extra now requires black>=26.5,<27 (installed on Python >= 3.10, where Black 26 is available)
• Removed the unused uplot frontend dependency (the UI draws on a hand-rolled canvas)

Fixed

• Mock-fidelity and hardware-behavior fixes from the 2026-07 code audit (trigger vocabulary mapping, waveform preamble parsing, measurement timeout handling, and related issues)

[1.1.0] - 2026-01-12

Added

• Data acquisition / data logger instrument support (DataLogger): Keysight 34970A/DAQ970A-style SCPI units — DMM function configuration (V/I/R/ temperature), scan lists, triggering, alarms, channel scaling, and timed logging helpers (PR #39)

[1.0.1] - 2026-01-07

Packaging and metadata fixes for the 1.0.0 rename release; no functional changes.

[1.0.0] - 2026-01-06

⚠️ MAJOR RELEASE - Package Renamed

This is a major release with a package rename. The project has been renamed from Siglent-Oscilloscope to SCPI-Instrument-Control to better reflect its expanded capabilities beyond just Siglent oscilloscopes.

Breaking Changes

• PyPI Package Name: siglent → SCPI-Instrument-Control
• Python Import Name: import siglent → import scpi_control (recommended)
• Old import still works with deprecation warning (will be removed in v2.0.0)

Added

Backward Compatibility Layer - Created siglent/ compatibility shim package - Old import siglent syntax still works with DeprecationWarning - Re-exports all scpi_control modules for seamless backward compatibility - Warning message guides users to migrate imports - Compatibility layer will be removed in v2.0.0

Migration Guide - Comprehensive migration documentation in README.md - Step-by-step upgrade instructions - Comparison table showing what changed and what stayed the same - Code examples for old vs new import syntax - Explanation of why the rename was necessary

Changed

Package Structure - Renamed siglent/ directory to scpi_control/ (preserving git history) - Updated all internal imports from siglent.* to scpi_control.* - 192 import occurrences updated in scpi_control/ modules - All imports updated in tests/ directory - All imports updated in examples/ directory - Package now reflects multi-instrument, multi-vendor capabilities

PyPI Metadata (pyproject.toml) - Package name: siglent → SCPI-Instrument-Control - Version bump: 0.5.1 → 1.0.0 (major release) - Updated description to emphasize universal SCPI instrument control - Updated keywords to reflect multi-instrument support - Updated all URLs to point to new repository name - Homepage, Repository, Issues, Documentation, Changelog

Documentation - Updated README.md with new project name and branding - Added "Package Renamed" notice at top of README - Updated all badge URLs to reference new repository name - Updated installation commands to use SCPI-Instrument-Control - Changed GUI application overview section title - Updated git clone directory name in examples - Added comprehensive Migration Guide section with: - Installation update instructions - Import statement migration examples - Backward compatibility explanation - CLI command compatibility notice - Comparison table of changes - Rationale for the rename

Command-Line Tools (No Changes) - siglent-gui command unchanged for convenience - siglent-report-generator command unchanged - No changes needed to scripts or automation using these commands

Why the Rename?

This library has evolved significantly beyond its original Siglent oscilloscope focus:

1. Multi-Instrument Support: Now supports oscilloscopes, power supplies, and function generators
2. Multi-Vendor Support: Works with any SCPI-compatible equipment, not just Siglent
3. Universal Protocol: Based on industry-standard SCPI commands (IEEE 488.2)

The new name SCPI-Instrument-Control accurately represents what the library does: control any SCPI-compatible test equipment.

Migration Instructions

For New Users

pip install SCPI-Instrument-Control
pip install "SCPI-Instrument-Control[gui]"
pip install "SCPI-Instrument-Control[all]"

from scpi_control import Oscilloscope, PowerSupply, FunctionGenerator

For Existing Users

Option 1: Update imports (recommended)

pip uninstall siglent
pip install SCPI-Instrument-Control

Change imports:

# Old
from siglent import Oscilloscope
from siglent.gui.app import main

# New
from scpi_control import Oscilloscope
from scpi_control.gui.app import main

Option 2: Use compatibility layer (temporary)

pip install SCPI-Instrument-Control

Keep old imports (will show DeprecationWarning):

from siglent import Oscilloscope  # Still works, but deprecated

Note: The compatibility layer will be removed in v2.0.0. Please migrate when convenient.

Technical Details

Git History Preservation - Used git mv siglent/ scpi_control/ to preserve file history - All commit history for individual files maintained - No loss of attribution or change tracking

Import Updates - Used automated find/replace for import statement updates - 192 occurrences in scpi_control/ directory - All test files updated - All example files updated - No manual edits required for import changes

Compatibility Implementation - siglent/__init__.py re-exports entire scpi_control module - Uses warnings.warn() with DeprecationWarning category - Warning includes migration instructions and timeline - stacklevel=2 ensures warning points to user code, not library

Dependencies

No dependency changes. All existing dependencies remain the same.

Deprecation Timeline

• v1.0.0 (2026-01-06): Old import siglent works with deprecation warning
• v2.0.0 (TBD): Compatibility layer removed, import siglent will fail

Users have until v2.0.0 to migrate their imports.

[0.5.1] - 2026-01-05

Fixed

Documentation Build and Deployment - Fixed GUI API documentation with incorrect module references - Corrected all wildcard references (*) to proper underscores (_) in docs/api/gui.md - Fixed 20+ module reference errors (e.g., siglent.gui.main*window → siglent.gui.main_window) - Documentation now builds successfully without errors - Set up automated GitHub Pages deployment - Created .github/workflows/docs.yml for automatic documentation deployment - Documentation auto-deploys to GitHub Pages on every push to main branch - Integrated with existing make docs-generate automation - Updated PyPI documentation URL - Changed from README-only link to proper documentation site - PyPI now links to https://little-did-I-know.github.io/Siglent-Oscilloscope/ - Users can access complete API documentation and guides from PyPI

MkDocs Build System - Ensured autodoc generation works correctly with mkdocstrings - Verified API reference pages generate from Python docstrings - Confirmed cross-references between documentation pages work properly

Changed

Documentation Infrastructure - Documentation now automatically generated and deployed on every main branch update - API documentation uses mkdocstrings for automatic extraction from code - Examples documentation auto-generated from example scripts - Improved documentation discoverability for users installing from PyPI

[0.5.0] - 2026-01-04

Added

Automated Test Report Generation 📊 - Installation: pip install "Siglent-Oscilloscope[report-generator]" - PDF and Markdown Report Generators - Professional publication-ready reports with waveform plots and analysis - Comprehensive metadata tracking (test ID, operator, timestamp, scope model) - Multiple output formats (PDF via ReportLab, Markdown with embedded plots) - Automatic file organization and asset management

• Signal Type Detection (siglent/report_generator/utils/waveform_analyzer.py)
• Automatic waveform classification using FFT harmonic analysis
• Detects 9 signal types: sine, square, triangle, sawtooth, pulse, DC, noise, complex, unknown
• Confidence scoring for classification accuracy
• THD (Total Harmonic Distortion) calculation for waveform quality assessment
• Pattern matching algorithms for periodic signal identification:
  • Square waves: odd harmonics at 1/n amplitude ratio
  • Triangle waves: odd harmonics at 1/n² amplitude ratio
  • Sawtooth waves: all harmonics at 1/n amplitude ratio
• Signal type displayed in both PDF and Markdown reports with confidence percentage

• Added SignalType constants class for standardized type identification

• Enhanced Waveform Statistics (siglent/report_generator/models/report_data.py)

• Comprehensive signal analysis with 25+ measurement parameters
• Amplitude measurements: Vmax, Vmin, Vpp, VRMS, Vmean, DC offset
• Frequency and timing: frequency, period, rise time, fall time, pulse width, duty cycle
• Quality metrics: SNR (Signal-to-Noise Ratio), THD, noise level, overshoot, undershoot, jitter
• Automatic statistics calculation via WaveformData.analyze() method
• Smart unit formatting with SI prefixes (mV, µs, kHz, etc.)

• Statistics integration in PDF and Markdown report generators

• Plateau Stability Analysis (Optional Feature)

• Measures noise on high and low plateaus for periodic signals
• Uses run-length encoding to identify flat signal regions
• Analyzes middle 60% of each plateau to exclude edge transitions
• Reports three metrics:
  • Plateau High Noise: Standard deviation on high-level plateaus
  • Plateau Low Noise: Standard deviation on low-level plateaus
  • Plateau Stability: Average noise across all plateaus
• User-configurable via Report Options dialog checkbox
• Auto-applies to pulse, square, triangle, sawtooth, and sine waves

• Helpful for power supply ripple, logic level stability, and signal quality testing

• LLM Model Detection Feature (siglent/report_generator/widgets/llm_settings_dialog.py)

• "Detect Models" button in Ollama and LM Studio configuration tabs
• Automatically queries server for available models and populates dropdown
• Changed model input from text field to editable combo box
• Preserves previously selected model after detection
• User-friendly error messages with troubleshooting steps
• Shows model count and lists detected models
• Leverages existing LLMClient.get_available_models() API

• Works with both Ollama Python SDK and OpenAI-compatible endpoints

• Report Options Dialog (siglent/report_generator/widgets/report_options_dialog.py)

• New checkbox: "Plateau Stability Analysis (Advanced)"
• Tooltip explains feature usage and applicability
• Settings persist with report templates
• Integrated with ReportOptions model

Changed - Report Generator Features

• PDF Generation Progress Tracking (siglent/report_generator/generators/pdf_generator.py)
• Enhanced with granular waveform-level progress updates
• Progress now updates for each waveform being processed: 20%, 25%, 30%, ..., 80%
• Prevents freezing appearance during matplotlib plot generation
• Progress callback shows current operation: "Processing section X/Y", "Rendering waveform Z"
• Smooth progression instead of jumping from 20% to 100%

• Integrated with QProgressDialog in GUI for visual feedback

• Unicode Character Normalization (siglent/report_generator/generators/pdf_generator.py)

• Comprehensive character mapping for AI-generated text compatibility
• Handles 30+ Unicode characters that ReportLab can't render:
  • Smart quotes (U+201C, U+201D, U+2018, U+2019) → regular quotes
  • Em-dash (U+2014) and en-dash (U+2013) → hyphens
  • Bullets (U+2022) → asterisks
  • Ellipsis (U+2026) → three dots
  • Math symbols: ≤→<=, ≥→>=, ≠→!=, ×→x, ÷→/
  • Degree symbols: °→deg, ℃→C, ℉→F
  • Non-breaking spaces and special formatting characters → regular space
• Applied to all AI-generated text fields: executive summary, AI insights, AI summary
• Prevents empty boxes, question marks, or missing characters in PDFs

• Maintains readability while ensuring PDF compatibility

• Waveform Statistics Display (PDF and Markdown Generators)

• Statistics tables now include signal type with confidence
• Plateau stability metrics shown when enabled and applicable
• Enhanced formatting with proper units and precision
• Color-coded headers and organized metric grouping in PDFs
• Responsive table layout that adapts to content

Fixed - Report Generator Features

• AI-Generated Text Rendering in PDFs
• Fixed special Unicode characters not rendering (showing as boxes/question marks)
• Root cause: LLMs generate smart quotes, bullets, and math symbols that ReportLab can't handle
• Solution: Comprehensive character normalization before PDF generation

• Now handles text from Claude, GPT, Llama, and other LLMs correctly

• Progress Bar Freezing During Plot Generation

• Fixed progress bar appearing to freeze at 20% during PDF generation
• Root cause: No progress updates during slow matplotlib plot rendering
• Solution: Track waveforms across sections and report per-waveform progress
• Users now see smooth progression throughout the entire generation process

Technical Improvements - Report Generator

• FFT-Based Signal Analysis
• NumPy FFT with harmonic ratio analysis for signal classification
• Autocorrelation for period detection in non-periodic signals
• Robust against noise with configurable confidence thresholds

• Optimized for real-world oscilloscope waveforms

• Extensible Waveform Analyzer

• Static methods for modular analysis capabilities
• Separation of detection, measurement, and formatting logic
• Easy to add new signal types or analysis algorithms

• Comprehensive docstrings with algorithm explanations

• Report Options Architecture

• ReportOptions dataclass for type-safe configuration
• Passed through generator chain to all analysis functions
• Enables feature flags for optional expensive computations

• JSON-serializable for template saving

• Model Detection Integration

• Reuses existing LLMClient infrastructure
• Handles both Ollama native API and OpenAI-compatible endpoints
• Graceful error handling with actionable user feedback
• No duplicate code between Ollama and LM Studio detection

Power Supply - Now Stable ✅ - Power supply support graduated from BETA to stable - API is now considered production-ready - Removed experimental warnings and beta tags - Full support for SPD3303X series power supplies - Installation: Standard package or pip install "Siglent-Oscilloscope[power-supply-beta]" (alias maintained)

Pre-Commit Checks and Code Coverage 🔍 - New make pre-commit-branch target for lightweight branch commit validation - Code formatting checks (Black, Flake8) - Fast parallel test execution - ~1 minute validation for rapid development - Enhanced make pre-pr with codecov integration - Full test suite with coverage reporting - Automatic coverage upload to Codecov - Comprehensive validation before pull requests - Codecov configuration (.codecov.yml) - 70-100% coverage range targets - Project and patch thresholds - Proper ignore patterns for tests, examples, docs - Coverage documentation (docs/development/PRE_COMMIT_CHECKS.md) - Complete guide for pre-commit workflows - Coverage concepts and best practices - Troubleshooting and CI/CD integration

Changed

Documentation Updates - Updated README with comprehensive Automated Report Generation section - Installation instructions for [report-generator] extra - Code examples and feature highlights - Added to Features, Installation, Optional Extras, and Examples sections - Removed BETA designation from power supply in documentation - Enhanced package description with automated report generation features

Project Organization - Reorganized main directory structure for better clarity - Moved test/development scripts to scripts/ directory: - test_llm_model_detection.py - test_pdf_progress.py and test_report_progress.pdf - test_signal_detection.py - test_unicode_rendering.py and test_unicode_rendering.pdf - Moved ICON_SETUP.md to docs/development/ for better organization - Removed duplicate codecov.yml (keeping .codecov.yml) - Cleaned up empty node_modules/ directory - Root directory now contains 22 essential files/directories

Fixed

Development Dependencies - Added pytest-cov>=4.0.0 to dev dependencies - Fixes "unrecognized arguments: --cov" error - Ensures coverage plugin available after pip install -e ".[dev]" - Added codecov>=2.1.0 to dev dependencies - Required by make codecov-report and make pre-pr - Enables coverage uploads in development

Makefile Pytest Integration - Updated all pytest calls to use python -m pytest - Ensures pytest-cov plugin is properly loaded - Fixes coverage generation in nested make calls - Updated targets: test, test-cov, test-fast, test-exceptions

[0.4.0-beta.1] - 2026-01-04

Added (EXPERIMENTAL 🧪)

Power Supply Control - BETA Release - ⚠️ EXPERIMENTAL: Power supply API is unstable and may change without warning - Installation: pip install "Siglent-Oscilloscope[power-supply-beta]" - Target Stable Release: v0.5.0 (pending community feedback) - Feedback: Please report issues at GitHub Issues

Core Power Supply Features: - Main PowerSupply Class (siglent.power_supply.PowerSupply) - SCPI-based communication over Ethernet (port 5024) or USB - Multiple connection types supported: - Ethernet/LAN via SocketConnection (default) - USB via VISAConnection (new - requires [usb] extras) - GPIB via VISAConnection (IEEE-488) - Serial via VISAConnection (RS-232) - Automatic model detection from *IDN? response - Capability-based feature availability - Context manager support for automatic connection management - Support for multiple power supply models via capability registry

• Model Support:
• Siglent SPD3303X / SPD3303X-E (triple output, 30V/3A + 30V/3A + 5V/3A)
  • Full feature support including OVP, OCP, timer, waveform generation, tracking modes
• Siglent SPD1305X (single output, 30V/5A)
• Siglent SPD1168X (single output, 16V/8A)

• Generic SCPI-99 PSUs (fallback with conservative defaults)

• Output Control (siglent.power_supply_output.PowerSupplyOutput)

• Voltage setpoint control with validation against model limits
• Current limit configuration
• Output enable/disable
• Real-time voltage, current, and power measurements
• Operating mode detection (CV/CC)
• Over-voltage protection (OVP) and over-current protection (OCP) settings
• Timer functionality (Siglent SPD specific)

• Waveform generation enable/disable (SPD3303X specific)

• Advanced Features:

• Tracking Modes for multi-output PSUs:
  • Independent mode (default)
  • Series tracking (voltages add)
  • Parallel tracking (currents add)
• Dynamic output creation based on model capabilities
• Model-specific SCPI command variants

• Comprehensive error handling and logging

• Model Registry and Capability System (siglent.psu_models)

• PSUCapability dataclass with complete model specifications
• OutputSpec defining per-output voltage/current/power limits
• Automatic model detection with fuzzy matching
• Generic SCPI fallback for unknown models

• Extensible registry for adding new models

• SCPI Command Management (siglent.psu_scpi_commands)

• Model-specific command variants (Siglent SPD vs generic SCPI)
• Template-based command generation with parameter substitution

• Support for multiple SCPI dialects

• Data Logging (siglent.psu_data_logger)

• PSUDataLogger: Manual data capture with configurable channels
• TimedPSULogger: Automated time-series data collection
  • Configurable sampling intervals
  • Background thread-based acquisition
  • Real-time data access during logging
  • Export to CSV, JSON, or custom formats
• Voltage, current, and power logging for all outputs

• Timestamp-based data organization

• Connection Layer (siglent.connection)

• VISAConnection class for USB/GPIB/Serial support (new)
  • PyVISA-based connection supporting multiple transport protocols
  • USB-TMC protocol for USB connections
  • GPIB (IEEE-488) support
  • Serial (RS-232) support
  • Optional dependency: install with [usb] extras
  • Pure Python backend (pyvisa-py) requires no proprietary drivers
• Resource discovery utilities:
  • list_visa_resources() - List all VISA devices
  • find_siglent_devices() - Automatically find Siglent instruments

• Graceful fallback when PyVISA not installed

• Examples:

• examples/psu_basic_control.py - Basic voltage/current control, measurements (Ethernet)
• examples/psu_usb_connection.py - USB/GPIB/Serial connection examples (new)
• examples/psu_advanced_features.py - Tracking modes, data logging, timer, waveform generation

• examples/psu_gui_test.py - GUI integration test (experimental)

• Testing:

• tests/test_power_supply.py - Unit tests with mock connection
• tests/test_visa_connection.py - Tests for USB/VISA connections (new)
• Hardware tests marked with @pytest.mark.hardware
• Coverage for core functionality, model detection, SCPI commands, VISA connections

Documentation: - Added docs/development/EXPERIMENTAL_FEATURES.md - Comprehensive guide for experimental features - Guidelines for marking features as experimental - Version numbering strategies (alpha/beta/rc) - Installation and discovery patterns - Documentation standards - Testing requirements - Graduation and deprecation processes - Updated docs/development/contributing.md with experimental features section - Module docstrings with experimental warnings and installation instructions

Changed

• Version Bumped: 0.3.2 → 0.4.0-beta.1
• Pre-release version indicates experimental status

• Follows semantic versioning with beta tag

• Package Description Updated:

• PyPI description now mentions power supply support (beta)
• Added SPD series models to package description

• New keywords: "power supply", "SPD", "PSU", "voltage", "current"

• Export Structure:

• siglent.__init__.py clearly separates stable vs experimental exports
• PowerSupply, PSUDataLogger, TimedPSULogger marked as experimental (v0.4.0-beta.1)
• Documentation in docstring warns about experimental status

Technical Details

Experimental Feature Implementation: - Module-level FutureWarning on import with clear experimental notice - Optional dependency groups in pyproject.toml: - [power-supply-beta] - Power supply features (no extra dependencies) - [usb] - USB/GPIB/Serial support via PyVISA (optional) - [experimental] group added for future experimental features - Graceful degradation when optional dependencies not installed

Architecture Highlights: - Connection abstraction layer (BaseConnection) - SocketConnection for TCP/IP Ethernet (default) - VISAConnection for USB/GPIB/Serial (optional via PyVISA) - Consistent SCPI interface across all connection types - Shares exception hierarchy (SiglentConnectionError, SiglentTimeoutError, CommandError) - Capability-based design allows easy addition of new models - SCPI command abstraction supports multiple manufacturers and SCPI dialects - Pluggable transport layer - easy to add new connection types

Known Limitations (Beta Status): - Limited hardware testing (primarily SPD3303X-E) - Some SCPI commands may vary between models - Timer and waveform generation features only tested on SPD3303X - Remote sensing support not yet implemented - No GUI integration for power supply control (gui test script only)

API Stability Notice: - Power supply API may change in future releases without deprecation warnings - Breaking changes possible in any 0.x release - Recommend pinning to specific version for production use: Siglent-Oscilloscope==0.4.0-beta.1 - Feedback welcome to help stabilize API before v0.5.0 stable release

Upgrading from 0.3.x

For Oscilloscope Users (No Changes Required): - All oscilloscope functionality remains stable - No breaking changes to existing APIs - Update to 0.4.0-beta.1 is backward compatible

For Power Supply Users (New Feature):

# Install with power supply support (Ethernet/LAN only)
pip install "Siglent-Oscilloscope[power-supply-beta]==0.4.0-beta.1"

# Install with USB support (includes PyVISA for USB/GPIB/Serial)
pip install "Siglent-Oscilloscope[usb]==0.4.0-beta.1"

# Install both
pip install "Siglent-Oscilloscope[power-supply-beta,usb]==0.4.0-beta.1"

For Developers: - See docs/development/EXPERIMENTAL_FEATURES.md for guidance on experimental features - Power supply modules will show experimental warnings on import - Set PYTHONWARNINGS=default::FutureWarning to see all warnings during development

[0.3.1] - 2026-01-02

Added

• Comprehensive MkDocs Documentation
• Added complete user guide documentation (5 files, ~3,000 lines):
  • basic-usage.md - Foundation for connecting and controlling oscilloscope
  • waveform-capture.md - Advanced capture techniques and data formats
  • measurements.md - Automated measurement capabilities
  • trigger-control.md - Comprehensive trigger configuration
  • advanced-features.md - FFT, math channels, automation, protocol decoding
• Added GUI documentation (7 files, ~4,700 lines):
  • overview.md - GUI introduction and installation
  • interface.md - Complete UI reference with keyboard shortcuts
  • live-view.md - Real-time waveform visualization
  • visual-measurements.md - Interactive measurement markers
  • fft-analysis.md - Frequency domain analysis
  • protocol-decoding.md - I2C/SPI/UART decoding
  • vector-graphics.md - XY mode and waveform generation
• Added connection guide (~960 lines):
  • connection.md - Network setup, troubleshooting, VNC access
• Added development documentation (3 files, ~2,950 lines):
  • building.md - Build system, testing, documentation generation
  • structure.md - Codebase organization and design patterns
  • testing.md - Testing strategy and best practices
• Added API reference:
  • gui.md - Auto-generated GUI API docs using mkdocstrings
• Total: 17 documentation files, ~11,900 lines
• Material for MkDocs theme with admonitions (tip, info, warning)
• Comprehensive examples, troubleshooting sections, and cross-references
• mkdocstrings integration for auto-generating API docs from Python docstrings

Fixed

• Windows Executable Build Workflow
• Fixed 7-Zip archive creation failing with "The system cannot find the file specified" errors
• Updated Windows build workflow to correctly reference README.md and LICENSE from parent directory
• Changed paths from README.md LICENSE to ../README.md ../LICENSE to match macOS/Linux builds
• Resolved workflow exit code 1 error during archive creation

[0.3.0] - Unreleased

⚠️ BREAKING CHANGES

• Exception Class Renaming (Issue #3 from Code Review)
• ConnectionError renamed to SiglentConnectionError to avoid shadowing Python's built-in ConnectionError
• TimeoutError renamed to SiglentTimeoutError to avoid shadowing Python's built-in TimeoutError
• Migration Guide:
  • Update imports: from siglent.exceptions import SiglentConnectionError, SiglentTimeoutError
  • Update exception handling: except (SiglentConnectionError, SiglentTimeoutError) as e:
  • Backward compatibility aliases provided for transition period (will be removed in v1.0.0)
  • If you use from siglent import ConnectionError, update to from siglent import SiglentConnectionError
• Why: Prevents naming conflicts with Python built-ins, improves code clarity, follows best practices
• Impact: All code that imports or catches ConnectionError or TimeoutError from siglent.exceptions needs updating

Added

• Waveform Validation System (siglent/gui/utils/validators.py)
• WaveformValidator class for comprehensive data quality checks
• Validates waveform data before plotting or processing
• Catches common issues that cause blank plots:
  • None/missing waveforms
  • Empty voltage or time arrays
  • Mismatched array lengths between time and voltage
  • All-NaN or excessive NaN values (>50%)
  • Invalid voltage ranges (all zeros, infinite values)
  • Suspiciously large voltages (>1000V)
• validate() method returns (is_valid, list_of_issues)
• validate_multiple() separates valid from invalid waveforms
• get_summary() generates diagnostic strings like "CH1: 50,000 samples, range -2.5V to +2.5V"
• Detailed Error Dialog Widget (siglent/gui/widgets/error_dialog.py)
• DetailedErrorDialog class for user-friendly error reporting
• Two-level error display:
  • User-friendly summary for non-technical users
  • Expandable technical details (stack trace, context) for debugging
• Features:
  • Error icon and timestamp display
  • "Show Details" / "Hide Details" toggle button
  • Read-only text area for stack traces and context
  • "Copy to Clipboard" button for comprehensive error reports
  • Automatic dialog resizing when showing/hiding details
• Structured error info dictionary format:
  • type: Error type name (e.g., 'TimeoutError')
  • message: User-friendly error message
  • details: Additional error details
  • context: Dictionary of context info (operation, settings, etc.)
  • traceback: Full stack trace string
  • timestamp: Error occurrence time
• Convenience function show_error_dialog() for quick usage
• Real-Time Status Updates (LiveViewWorker)
• New status_update signal (pyqtSignal(str)) for user feedback
• Status messages during acquisition cycle:
  • "Checking enabled channels..."
  • "Acquiring CH1...", "Acquiring CH2...", etc.
  • "Validating waveforms..."
  • "Live view: CH1, CH2 (50,000, 100,000 samples)"
  • "No enabled channels", "Not connected"
• Status bar updates reflect worker progress in real-time

Changed

• LiveViewWorker Error Handling Enhanced
• Changed error_occurred signal from pyqtSignal(str) → pyqtSignal(dict)
• Errors now emit structured dictionaries with full context
• Error info includes:
  • Error type, message, details
  • Operation context (update_interval, operation name)
  • Full traceback for debugging
  • Timestamp for error tracking
• Integration with WaveformValidator for data quality checks
• Only emits valid waveforms (invalid ones logged at WARNING level)
• Enhanced logging: acquisition results logged at INFO/WARNING for visibility
• WaveformCaptureWorker Validation Integration
• Validates all captured waveforms before emitting via WaveformValidator.validate_multiple()
• Logs validation failures at WARNING level (visible to users)
• Only emits valid waveforms to prevent blank plots
• Enhanced error messages include validation failure details
• Progress message updated: "Processing waveforms..." → "Validating waveforms..."
• WaveformDisplayPG Pre-Plot Validation
• Validates all waveforms before plotting via WaveformValidator.validate_multiple()
• Invalid waveforms logged at WARNING level with specific issues
• Info label shows "Invalid data - check logs" when all waveforms fail validation
• Enhanced diagnostic logging:
  • Valid waveforms logged at INFO level with summary
  • Runtime validation checks for None and empty arrays
  • Voltage range logging: "[−2.5V to +2.5V]" or "[all NaN]"
• Only stores and plots valid waveforms
• Main Window Error Handling Integration
• Connected to new status_update signal from LiveViewWorker
• New _on_live_view_status() method updates status bar with worker messages
• Enhanced _on_live_view_error() method:
  • Accepts structured error dictionary instead of plain string
  • Shows DetailedErrorDialog for rich error information
  • Brief error message in status bar (60 chars max, 5 second timeout)
  • Fallback to QMessageBox for legacy string errors
• User-friendly error display with expandable technical details

Fixed

• Blank Plot Issue from Invalid Waveforms
• Root cause: Invalid waveforms (None, empty arrays, all NaN) were being plotted
• Solution: Comprehensive validation before plotting in all code paths
• Workers now validate data before emitting to GUI
• Display widget validates again before rendering as safety check
• Cryptic Error Messages
• Users previously saw raw exception strings in status bar
• Now see structured error dialogs with context and debugging info
• Technical details hidden by default but available on demand
• Missing Waveform Quality Diagnostics
• Added comprehensive validation with specific issue reporting
• Users now see exactly why waveforms failed (e.g., "CH1: All voltage values are NaN")
• Validation results logged at WARNING level for visibility
• Bare Exception Handling in Vector Graphics (Issue #2 from Code Review)
• Replaced bare except: clauses with specific exception types in vector_graphics.py
• Now catches CommandError, SiglentConnectionError, SiglentTimeoutError explicitly
• Prevents catching system exceptions like KeyboardInterrupt and SystemExit
• Improves debugging and error handling clarity
• Socket Read Race Condition (Issue #5 from Code Review)
• Added timeout protection in socket.py read loop
• Prevents infinite loop if oscilloscope doesn't send newline-terminated responses
• Raises SiglentTimeoutError with detailed message showing bytes received
• Improves reliability and error diagnostics
• Version Mismatch (Issue #1 from Code Review)
• Fixed version inconsistency between __init__.py (0.1.0) and pyproject.toml (0.2.6)
• Both now correctly report version 0.2.6 (will be bumped to 0.3.0 for this release)

Technical Improvements

• Input Validation for SCPI Commands (Issue #4 from Code Review)
• Added ASCII validation before encoding commands in socket.py
• Raises CommandError with clear message if non-ASCII characters detected
• Prevents UnicodeEncodeError exceptions during command transmission
• Example: CommandError: SCPI command contains non-ASCII characters: "C1:VDIV 1.0V\u2013"
• Magic Number Constants (Issue #6 from Code Review)
• Added named constants for waveform conversion in waveform.py:
  • WAVEFORM_CODE_PER_DIV_8BIT = 25.0 (codes per division for 8-bit ADC)
  • WAVEFORM_CODE_PER_DIV_16BIT = 6400.0 (codes per division for 16-bit ADC)
  • WAVEFORM_CODE_CENTER = 0 (center code for signed integer data)
• Improved code documentation with conversion formula from SCPI manual
• Makes waveform parsing logic easier to understand and maintain
• Centralized waveform validation logic in reusable WaveformValidator class
• Structured error reporting enables better debugging and user support
• Separation of user-facing messages from technical diagnostics
• Thread-safe error propagation from workers to main GUI thread
• Validation happens at multiple checkpoints (capture → emit → display)
• All docstrings updated to reference new exception names
• Backward compatibility aliases ensure gradual migration path

[0.2.6] - 2025-12-31

Added

• Background Waveform Capture Worker (waveform_capture_worker.py)
• Non-blocking waveform acquisition in separate thread
• Progress updates during multi-channel capture
• Cancellable long-running downloads
• Thread-safe signal/slot architecture
• Real-time status updates: "Downloading CH1 data from scope..."
• Progress Dialog for Capture Operations
• Visual progress indication during waveform downloads
• Shows channel-by-channel progress
• Cancel button to abort slow captures
• Auto-closes on completion
• Prevents concurrent capture operations
• Intelligent Waveform Downsampling
• Min-max decimation algorithm for large datasets
• Preserves signal peaks, valleys, and transients
• Configurable threshold (default: 500,000 points)
• Downsamples 5M point waveforms to 500K for display
• Original data fully preserved for export/analysis
• Status indicator shows "(display downsampled)" when active
• Modern Graph Visual Styling (PyQtGraph)
• GitHub-inspired dark theme (#0d1117 background)
• Thicker, smoother waveform lines (2.0px with antialiasing)
• Vibrant channel colors for better visibility:
  • CH1: Bright Yellow (255, 220, 50)
  • CH2: Turquoise/Cyan (64, 224, 208)
  • CH3: Hot Pink (255, 105, 180)
  • CH4: Bright Green (50, 255, 100)
• Subtle dotted grid (20% opacity)
• Modern typography (Segoe UI, 11pt labels)
• Muted axis colors for professional appearance
• Sample count with thousands separators (e.g., "5,000,000 samples")

Changed

• Waveform Capture Architecture Refactored
• Moved from synchronous to asynchronous capture model
• Capture operations now run in WaveformCaptureWorker thread
• Main GUI thread remains responsive during downloads
• Enhanced error handling with user-friendly messages
• Canvas Rendering Optimization (Matplotlib)
• Replaced all blocking canvas.draw() calls with canvas.draw_idle()
• Removed redundant canvas.update(), canvas.repaint() calls
• Deferred rendering prevents GUI thread blocking
• Applied to 17+ instances across waveform_display.py
• Downsampling Threshold Increased
• Changed from 100,000 to 500,000 point threshold
• Provides 5x more waveform detail on display
• Maintains smooth performance even with millions of samples
• Progress Dialog Signal Handling Improved
• Disconnects canceled signal before closing to prevent race conditions
• Only triggers cancellation if worker thread is actually running
• Eliminates spurious "User cancelled capture" messages

Fixed

• Critical: GUI Freezing During Waveform Capture
• Fixed multi-second GUI freeze when capturing large waveforms (5M samples)
• Root cause: Synchronous SCPI queries blocked main thread for 5-10+ seconds
• Solution: Background worker thread handles all network I/O
• GUI remains fully responsive during capture operations
• Progress Dialog Race Condition
• Fixed spurious cancellation events when dialog auto-closed
• Dialog now properly disconnects signals before closing
• Prevents "cancelled" state after successful capture
• Large Waveform Display Performance
• Fixed severe lag when plotting 5+ million point waveforms
• PyQtGraph could still block GUI thread with massive datasets
• Min-max downsampling reduces display points by 10x while preserving signal fidelity
• Incomplete Waveform Display Issue
• Fixed waveforms not showing full captured data
• Increased downsampling threshold from 100K to 500K points
• Users now see 5x more detail in displayed waveforms

Performance

• Waveform Capture: No longer blocks GUI (runs in background thread)
• Display Rendering: 10x faster for large waveforms via intelligent downsampling
• Canvas Updates: Non-blocking deferred rendering throughout
• User Responsiveness: Can interact with GUI during long captures

Technical Improvements

• Thread-safe capture worker with Qt signals for status updates
• Signal/slot disconnect pattern prevents race conditions
• Min-max decimation preserves signal integrity during downsampling
• NumPy-optimized downsampling algorithm for performance
• Clean separation of capture, processing, and display concerns

[0.2.5] - 2025-12-30

Added

• Buy Me a Coffee Badge
• Added support/donation badge to README
• Links to https://buymeacoffee.com/little.did.i.know
• Comprehensive Test Suite
• Added tests/test_channel.py - 50+ test cases for channel control (280 lines)
  • Tests for enable/disable, voltage scale, offset, coupling
  • Probe ratio, bandwidth limit configuration
  • Multi-channel validation
• Added tests/test_trigger_comprehensive.py - 40+ test cases for trigger (320 lines)
  • Mode control (AUTO, NORMAL, SINGLE, STOP)
  • Source, level, slope configuration
  • Edge trigger setup and actions
  • Holdoff and coupling control
• Added tests/test_measurement_comprehensive.py - 45+ test cases for measurements (340 lines)
  • Frequency, period, Vpp, RMS, amplitude measurements
  • Min/max/mean voltage measurements
  • Timing measurements (rise/fall time, duty cycle)
  • Statistical measurements and cursor support
• Added tests/test_waveform_comprehensive.py - 35+ test cases for waveform handling (280 lines)
  • WaveformData creation and properties
  • Binary waveform capture and parsing
  • Multi-format save/load (CSV, NPZ, MAT, HDF5)
  • Waveform analysis and comparison
• Added tests/test_socket_connection.py - 33+ test cases for connection (270 lines)
  • Connection lifecycle (connect, disconnect, reconnect)
  • Command sending and querying
  • Binary data queries
  • Context manager and error handling
• Total: 490+ new test cases across 5 test modules
• Test Coverage and Quality Assurance
• Integrated pytest with coverage reporting in CI workflow
• Added Codecov integration for test coverage tracking and visualization
• Multi-version testing across Python 3.8-3.12
• Coverage badge display on GitHub and PyPI
• Coverage Improvement: Overall coverage increased from 39% to 42%
  • channel.py: Enhanced test coverage with comprehensive scenarios
  • trigger.py: Added extensive mode and configuration tests
  • measurement.py: Added tests for all measurement types
  • waveform.py: Added capture, save/load, and analysis tests
  • connection/socket.py: Added full connection lifecycle tests
• Professional Project Badges
• CI build status badge (GitHub Actions)
• Test coverage badge (Codecov)
• PyPI downloads per month badge
• GitHub issues tracker badge
• GitHub stars badge
• Last commit timestamp badge
• Codecov Configuration
• Added codecov.yml with project-specific settings
• Configured coverage thresholds and reporting
• Pytest Configuration
• Added [tool.pytest.ini_options] to pyproject.toml
• Configured test markers for hardware and GUI tests
• Added strict pytest configuration for better test quality
• Contributing Guide
• Comprehensive CONTRIBUTING.md with development guidelines
• Code style, testing, and PR submission instructions
• Development setup and best practices documentation
• Community Standards
• Added CODE_OF_CONDUCT.md (Contributor Covenant 2.1)
  • Private reporting contacts (email, GitHub, security advisory)
  • Clear enforcement responsibilities (maintainers defined)
  • Anti-retaliation policy
  • Step-by-step "What Happens Next" process
  • Appeals process for disputed decisions
• Added SECURITY.md with vulnerability reporting process
• Added security best practices and safe usage guidelines
• Development Automation
• Added .pre-commit-config.yaml for automated code quality checks
• Configured Black, Flake8, isort, Bandit security scanning
• Added file cleanup and validation hooks
• Makefile for Development
• Added comprehensive Makefile with common development tasks
• Commands for testing, linting, formatting, building, publishing
• Quick setup commands: make dev-setup, make check
• Shortcuts: make test-cov, make format, make gui
• Pre-PR commands: make pre-pr, make pre-pr-fast, make pre-pr-fix
• Pre-PR Validation Scripts
• Added scripts/pre_pr_check.py - Comprehensive Python validation script
• Added scripts/pre_pr_check.sh - Bash version for Unix-like systems
• Automated checks: formatting, linting, security, tests, coverage, build
• Options: --fast (skip slow checks), --fix (auto-fix issues)
• Color-coded output with detailed error reporting
• GitHub Issue Templates
• Structured bug report template (.github/ISSUE_TEMPLATE/bug_report.yml)
• Feature request template (.github/ISSUE_TEMPLATE/feature_request.yml)
• Issue template configuration with links to discussions
• Pull Request Template
• Comprehensive PR template (.github/PULL_REQUEST_TEMPLATE.md)
• Checklists for code quality, testing, and documentation
• Sections for type of change, testing details, and migration guides
• Dependabot Configuration
• Automated dependency updates (.github/dependabot.yml)
• Weekly updates for Python packages and GitHub Actions
• Grouped updates by dependency type (dev, security, GUI, core)
• Interactive Tutorial
• Jupyter notebook tutorial (examples/interactive_tutorial.ipynb)
• Step-by-step guide for oscilloscope control
• Examples of waveform capture, FFT analysis, measurements
• Multi-channel capture and data export demonstrations

Changed

• SEO and PyPI Metadata Improvements
• Enhanced package description with comprehensive feature highlights for better discoverability
• Expanded keywords from 7 to 20 terms covering automation, data acquisition, GUI, protocol decoding, and visualization
• Improved search ranking for oscilloscope automation, SCPI control, and lab equipment software
• CI/CD Enhancements
• Enhanced CI workflow with dedicated test suite job
• Added pytest-cov and pytest-xdist dependencies
• Improved test execution with verbose output and coverage reporting
• Test Organization
• Moved manual test scripts to scripts/ directory
• Reorganized interactive GUI tests as manual scripts
• Ensured automated tests properly handle optional dependencies
• README Improvements
• Added Community and Support section with links to issues, discussions, security
• Added Resources section highlighting tutorial, examples, and guides
• Added Quick Start for Contributors with Makefile commands
• Improved Contributing section with detailed instructions

Fixed

• Test Suite Issues
• Fixed CI test failures due to missing PyQt6 dependencies
• Moved manual test scripts (test_live_view.py, test_pyqtgraph.py, test_dependency_check.py, test_waveform_display.py) to scripts/ directory
• Prevented pytest from collecting non-test GUI demo scripts
• Ensured GUI tests skip gracefully when PyQt6 is not installed
• Fixed hanging test in test_socket_connection.py::TestSocketQueryBinary::test_query_binary
  • Test was using mock_socket.recv.return_value which caused infinite loop in read_raw()
  • Changed to side_effect to return data once then raise timeout to signal end of data
• Python 3.8 Compatibility
• Fixed pyproject.toml license field to use PEP 621 compliant format
• Changed license = "MIT" to license = {text = "MIT"} for Python 3.8 compatibility
• Resolved build errors in older setuptools versions

[0.2.4] - 2025-12-29

Added

• Vector Graphics / XY Mode Features (requires [fun] extras)
• New vector_graphics.py module for generating waveforms for XY mode display
• VectorDisplay class for managing XY mode and waveform generation
• Shape factory with generators for:
  • Basic shapes: circle, rectangle, polygon, star, line
  • Lissajous figures for classic oscilloscope patterns
  • Text rendering (experimental)
• VectorPath class with transformation methods (rotate, scale, translate, flip)
• Waveform export to CSV, NumPy, and binary formats for AWG upload
• GUI Integration: New "Vector Graphics 🎨" tab in GUI application
  • Shape selection with dynamic parameter controls (Circle, Rectangle, Star, Triangle, Lissajous, Line)
  • XY mode enable/disable directly from GUI
  • Waveform generation with sample rate and duration controls
  • Export to CSV, NumPy, or Binary format for AWG upload
  • Works even without scope connection (offline waveform generation)
  • Graceful degradation: shows installation instructions if [fun] extras not installed
• Example script: examples/vector_graphics_xy_mode.py with animations
• Optional dependency group [fun] in pyproject.toml:
  • shapely>=2.0.0 (geometric operations)
  • Pillow>=10.0.0 (text rendering)
  • svgpathtools>=1.6.0 (SVG path support)

Changed

• Updated README.md with comprehensive Vector Graphics / XY Mode section
• Added GUI tab documentation with step-by-step usage instructions
• Added example use cases (calibration, education, pattern testing)
• Updated installation instructions to include [fun] extras option
• Added [fun] to Optional Extras requirements section
• Updated siglent/gui/main_window.py to integrate Vector Graphics tab
• Updated siglent/gui/widgets/__init__.py with import note for optional panel

[0.2.3] - 2025-12-29

Changed

• project.toml version number due to pypi version number conflict

[0.2.2] - 2025-12-29

Changed

• GitHub Workflow Updates
• Simplified PyPI publishing workflow
• Removed TestPyPI publishing step for streamlined releases
• Workflow now publishes directly to PyPI on releases or manual trigger

[0.2.1] - 2025-12-29

Changed

• Project Structure Reorganization to follow Python packaging best practices
• Moved all test files to tests/ directory
• Moved development utilities to scripts/ directory
• Consolidated documentation into docs/ directory
• Created docs/development/ for build and deployment documentation
• Updated MANIFEST.in to properly exclude development files from distribution
• Updated .gitignore to properly exclude build artifacts
• Added docs/development/PROJECT_STRUCTURE.md documenting the project layout

Fixed

• Improved .gitignore to properly exclude dist/, build/, and egg-info directories

[0.2.0] - 2025-12-29

Added

• High-Performance Live View with PyQtGraph
• 100x faster real-time plotting (1000+ fps capability vs 5-10 fps)
• Replaced matplotlib with PyQtGraph for live waveform display
• Non-blocking threaded data acquisition
• Smooth updates at 5-20 fps with responsive GUI
• Supports all 4 channels simultaneously
• Interactive Visual Measurement System
• Click-and-drag measurement markers directly on waveforms
• 15+ measurement types with specialized visual markers:
  • Frequency/Period with auto-detection
  • Voltage measurements (Vpp, Amplitude, Max, Min, RMS, Mean)
  • Timing measurements (Rise Time, Fall Time, Pulse Width, Duty Cycle)
• Real-time measurement updates as you adjust marker gates
• Visual gates and markers with color-coded styling
• Auto-placement with intelligent positioning
• Measurement Configuration Management
• Save measurement setups to JSON files
• Load previously saved configurations
• Configuration browser and management
• Shareable measurement templates
• Measurement Export Functionality
• Export results to CSV format
• Export to JSON with full configuration
• Batch measurement support
• Timestamp and metadata inclusion
• Background Worker Thread (LiveViewWorker)
• Prevents GUI freezing during SCPI queries
• Thread-safe signal/slot communication
• Configurable update intervals (default: 200ms)
• Automatic error handling and reporting
• Visual Measurement Panel (visual_measurement_panel.py)
• Add/remove markers via UI
• Enable/disable individual markers
• Live measurement value display
• Auto-update mode with 1-second refresh
• Marker list with real-time results
• New Measurement Marker Classes
• Base MeasurementMarker abstract class
• FrequencyMarker with period auto-detection
• VoltageMarker with threshold visualization
• TimingMarker with edge detection
• PyQtGraph-based Waveform Display (waveform_display_pg.py)
• Drop-in replacement for matplotlib display
• Preserved all existing features (cursors, zoom, pan)
• Optimized update performance
• Configurable visual styling
• Measurement Data Models (measurement_config.py)
• MeasurementMarkerConfig dataclass
• MeasurementConfigSet for collections
• JSON serialization/deserialization
• Configuration validation

Changed

• Migrated live view from matplotlib to PyQtGraph for dramatic performance improvement
• Replaced timer-based acquisition with threaded worker pattern
• Enhanced waveform display with marker support methods
• Updated main window to integrate visual measurement panel
• Improved channel enabled detection to handle scope response format

Fixed

• Channel Detection Bug: Fixed channel.enabled property failing to detect "C1:TRA ON" response format (channel.py:48)
• GUI Freezing: Moved blocking SCPI queries to background thread to maintain GUI responsiveness
• Live View Startup: Removed non-existent refresh_state() call that caused crashes
• Waveform Update Performance: Eliminated unnecessary canvas redraws during live view

Dependencies

• Added pyqtgraph>=0.13.0 to GUI dependencies
• Updated installation instructions for [gui] extra

Technical Improvements

• Thread-safe Qt signal/slot architecture for live updates
• Abstract base class pattern for extensible measurement markers
• Dataclass-based configuration management
• Separation of rendering and calculation logic
• Optimized matplotlib blitting for fast partial updates (fallback mode)
• Coordinate system handling for zoom/pan compatibility

Performance

• Live view: 100x faster (5-10 fps → 1000+ fps capability)
• SCPI queries: Moved to background thread (GUI unblocked)
• Canvas updates: <1ms per frame vs 100-500ms previously
• Measurement calculations: Real-time NumPy-based processing

0.1.0 - 2025-12-25

Added

• Initial release of Siglent oscilloscope control package
• Complete programmatic API for controlling Siglent SD824x HD oscilloscopes
• TCP/IP socket connection via SCPI protocol (port 5024)
• Full channel control (all 4 channels)
• Voltage scale and offset configuration
• Coupling mode (DC/AC/GND)
• Probe ratio settings
• Bandwidth limiting
• Comprehensive trigger control
• Trigger modes (Auto, Normal, Single, Stop)
• Edge trigger configuration
• Source selection and level control
• Slope and coupling settings
• Waveform acquisition and processing
• Binary waveform download
• Automatic voltage conversion
• NumPy array output
• CSV and NPY export formats
• Automated measurements
• Frequency, period, Vpp, RMS, amplitude
• Rise/fall time, duty cycle
• Min/max/mean calculations
• Cursor support
• PyQt6-based GUI application
• Connection management dialog
• Acquisition controls (Run/Stop/Single)
• Live waveform view with configurable update rate
• Single waveform capture
• Multi-channel matplotlib display with oscilloscope-style theming
• Interactive zoom and pan
• Waveform export to PNG/PDF/SVG
• Console script entry point (siglent-gui)
• Comprehensive example scripts
• Basic usage and configuration
• Waveform capture and export
• Automated measurements
• Live plotting
• Full documentation
• API documentation in README
• PyPI deployment guide
• Build instructions
• Example scripts with explanations

Technical Details

• Python 3.8+ support
• Dependencies: PyQt6, NumPy, Matplotlib
• MIT License
• Type hints throughout codebase
• Context manager support for oscilloscope connections
• Comprehensive error handling with custom exceptions