GitHub

Version 0.11.1 - 2026-09-10

  • Fix. Cosmetic changes in the Web UI: file pickers speak the page's language, run history and result pages name the case instead of its path, and only power-flow runs offer the comparison checkbox.

Version 0.11.0 - 2026-09-10

Two runs side by side, and a configuration trap in the solver call.

Highlights

  • Compare two runs. Run history has a selection column and a Compare selected runs button. The comparison page shows where the two runs differ: configuration keys with both values, buses where Q limits engaged, total active and reactive losses, and the largest voltage deviation between the results. Until now this took two browser tabs and a good memory.
  • A grid is solved under the configuration it was imported with. runpf! without an explicit configuration used the globally active one, not the grid's. Importing a grid with its own configuration and then solving it returned a result under foreign settings, without a warning and without a visible difference in the call. A passed configuration still wins.
  • Non-physical generator states appear in the result. When a machine sits at its Q limit while the voltage is on the wrong side of its setpoint, the classic result print now names bus, side, voltage, setpoint, Q and limit. On case118 this is the one line that separates the enforcement methods.

Usability

  • Q limit handling is one control. Checkbox and mode sat in different places of the form. A mode selected with handling switched off did nothing and looked like it did something. Both are together now, and the mode accepts off.
  • Diagnostics report a diagnosis. The self test deliberately runs one step from the stored voltages; the residual is its result. It was rendered as a failed power flow, so every diagnosis looked like a crash. History and result page now say "diagnosed".
  • Case input format names SCF and power-grid-model. The same importer reads both formats; now they can be named that way.

Version 0.10.0 - 2026-09-06

New

  • State estimation: current and PMU phasor measurements, tap estimation, shunt estimation, island-wise solving, bad-data localization, topology validation, Web UI page.
  • Scenarios and N-1: patch model on component ids, engine with reusable working copies, Woodbury screening, Web UI editor.
  • Sparlectra Case Format (SCF): self-describing case file, data section is a valid power-grid-model dataset, byte-identical round trip, study blocks executed.
  • Automatic power-flow mode (power_flow.mode: auto) with escalation ladder and decision log.
  • buildSysimage() and buildApp(): fast-start sysimage and standalone executable, one call each; the Web UI refreshes the sysimage in the background while it keeps running on the old one.
  • Takahashi selected inverse, shared by short-circuit sweeps and state-estimation diagnostics (34x to 264x).
  • MATPOWER busbar couplers and tap-changer nameplates, CGMES export of regulated tap groups.

Changed

  • Every format has its own importer; SCF is the preferred working format, not a mandatory way station.
  • AnalyticLoadFlow.jl is a required dependency, so the APSLF solver is always available.
  • Chi-square band test uses Wilson-Hilferty; multi-island nets are estimated per island.

Fixed

  • DC power flow did not contract closed busbar couplers (sections of one node split by up to 13.6 degrees).
  • Generated measurement sets were ignored, added instead of replacing, and foreign sets were preselected.
  • Case files lost voltage limits, Q-limit hysteresis, phase-tap bands and machine operating points.
  • Q-limit handling scanned its whole event log per bus and iteration, which dominated the power flow on large networks (13659 buses: 60 s, now 13.5 s, same result).
  • Test suite and ensure_casefile share the Web UI case directory; SPARLECTRA_LARGE_CASES_DIR remains as override.
  • Diagnose runs took the form's iteration limit and start settings instead of their own, so the button reported an ordinary converged run instead of the residual of the imported state.

Version 0.9.19 - 2026-08-25

  • Branches carry their physical equipment impedance (r_base_pu/x_base_pu) alongside the live operating point; short circuit and the MATPOWER export read the base, and the CGMES export guard no longer blocks after a series-FACTS run (#329).
  • New: restoreBaseImpedances!, clearUpfcFullControllers!; clearSeriesReactanceControllers! restores the base impedance.

Version 0.9.18 - 2026-08-25

Housekeeping

Version 0.9.17 - 2026-08-25

Improvements

  • N-1 contingency: per-case start-value ladder (contingency.rescue_ladder), case filters, weights, generator outages, overload reporting, Web UI button and weight editor (issue #331).

Breaking

  • Breaking: ContingencyResult.overloaded_branches replaced by overloads::Vector{OverloadRecord}.

Fixes

  • Deprecated retry_flat_start: appends :flat to the ladder, kept one minor cycle.

Version 0.9.16 - 2026-08-23

CGMES deliveries that arrive without their topology file, and a UPFC that finally steers a line's active and reactive flow at the same time.

Highlights

Node-breaker CGMES imports without a TP profile. EMS and substation exports often ship only EQ and SSH, leaving the bus partition implicit in the connectivity nodes and switch states. Sparlectra now works it out itself: connectivity nodes merge across closed switches (an open switch in the SSH wins over the equipment default, out-of-service counts as open), retained switches stay as bus couplers, and boundary nodes hook into the boundary set. Deliveries that do ship a TP are untouched, and on the conformity sets the derived partition matches the sender's bus for bus. A delivery with neither a TP nor connectivity nodes now stops with the import analysis rather than a puzzling error further down. See CGMES Import.

The UPFC, both ways. addUpfcControl! gives you the unified power flow controller. The default model = :quadrature pairs an SSSC and a STATCOM as one named device; model = :full (#326) lifts the quadrature restriction so the series converter injects a voltage of arbitrary phase and the line holds independent P and Q at once, with the active part balanced across the DC link. Forcing the series phase back to quadrature collapses the full model onto the composite. Honest about the first cut: the shunt follows a reactive setpoint (closed-loop voltage regulation is still to come) and convergence is dependable for realistic, moderate targets. See FACTS Devices.

A CGMES workshop tour. A new expert-level Colab tour walks the official ENTSO-E conformity sets end to end: what a delivery is, reading a broken one with the import analysis, bus-branch import with SV validation, node-breaker with and without a TP, and the export round trip.

The rest

  • The classical result knows about FACTS now. Every outer-loop controller (STATCOM, SVC, TCSC/SSSC, HVDC pair, UPFC) shows up in the power-flow result: counted in the header, marked on its branch or bus, and summarised in the control footer. The FACTS tour prints these tables for each device.
  • Ferranti voltage in the bus table. An isolated open-end bus shows the open-end voltage of its one-sided-open branch (flagged open-end) instead of a meaningless start value; the basic tour contrasts the correct pi-stub model against a full disconnect, which quietly drops the charging draw.
  • Smaller things. analyzeCGMES now loads with using Sparlectra; print(summarizeCGMES(...)) renders the readable multi-line report instead of a one-line dump; and every workshop example carries a unique number, a line diagram for each network it introduces, and a note on its AI-assisted origin.

Version 0.9.15 - 2026-08-23

Fix and polish, and observability theory you can execute.

  • Dark states, named. A not-observable set now reports unobservable_state_columns: the states the null space of the measurement Jacobian touches, i.e. exactly what no meter pins down. The column-restricted local check documents its limitation (necessary, not sufficient); the workshop shows the two disagreeing verdicts on the smallest possible counterexample.

  • The state-estimation workshop teaches the theory. WLS foundation and normal equations, the null-space criterion with observable islands, the spanning-tree counting identity, criticality as a zero residual, J/dof as the after-run health number, matrices small enough to read, and a zero-injection section with the honest sigma trade-off.

  • The estimation problem shows its matrix. New measurement_jacobian(net) returns the labeled Jacobian H (described measurement rows, named state columns); the state-estimation example suite writes a measurement-matrix page with a stability verdict (rank, redundancy, cond(H)) on every run. The workshop's observability deep dive now explains what global and local observability ARE and draws the meters into the line diagram.

  • Docs sweep. Stale claims corrected across the pages (RealGrid has 6209 buses, control.controllers is implemented, hvdc_pair is a supported type, chapter pointers follow the tour split, two release dates fixed); the drifted hand-written config key index now points at the canonical configuration.yaml.example; the transformer control theory lives once, on the Control Framework page.

  • Code sweep. Dead code removed, including four dead public bindings (addBusShuntPower!, a deprecated no-op, plus zero_row!, print_jacobian, recalc_trafo_model_data) and six unused private helpers; the Y-bus assembly no longer rescans the isolated-bus list per branch, result tables build without quadratic string growth, and the reverse bus-name lookups are unified.

Version 0.9.14 - 2026-08-22

Intentionally identical to 0.9.13: the registry needs the gapless number, the project prefers not to stay on it. Install this one.

Version 0.9.13 - 2026-08-22

Coordinated transformers, switched shunt banks, and observability you can see.

  • Parallel transformers regulate as a group (#322). Two independent tap controllers on one busbar fight each other through circulating reactive power, and CGMES deliveries produce exactly that setup whenever several tap changers share one RegulatingControl. Now the group is the model: the master runs the voltage loop, followers mirror its steps (followers on addPowerTransformerControl!), and the importer turns shared controls into one group instead of fighting controllers. Part 3 of the workshop tour walks the whole story, from the circulating-current trap to the CGMES data.
  • MSC/MSR switched shunt banks (#324). step_mvar turns the shunt voltage controller into a mechanically switched bank: whole blocks only, never overshooting the target, parking on the reached step instead of hunting between two.
  • Observability you can see. The state-estimation workshop now shows what the observability numbers mean on a drawn network, where measurements must sit (and how a missing one darkens a corner), and how a PMU's time-base offset is estimated. Every workshop notebook states its level.

Version 0.9.12 - 2026-08-22

The workshop splits in two, and every solver reads the same network.

Behavior change

  • The DC power flow now reads the same specification as the AC solver (issue #323). Both solvers build their injections from the prosumer objects; the node-level sums (addBusLoadPower!/addBusGenPower!) are a report layer that NO solver reads anymore. Before, a node-sum edit changed a DC solve but not an AC solve, so the two could silently disagree about what the network is. Affected: only workflows that edited node sums and then ran the DC solver; the fix is one line, add the load as a prosumer. Imported cases fill both layers identically and are not affected.

Highlights

  • The workshop tour splits into basic and advanced. The basic tour carries Newcomer to Advanced (first network to Q(U) control); the new advanced tour carries Expert and Beyond with room to breathe: remote voltage control, HVDC pairing, state estimation, the FACTS limit workshop (now with the in-range view and the generic element table), the N-1 workshop (now with worst-case ranking and CSV export), and the threaded sweeps.
  • Half-open branches show themselves in the classical result. The summary prints one status line per one-sided open branch with the open-end voltage and its Ferranti rise over the feeding bus, the bus table marks the open-terminal bus (open-end in the Control column), and the branch row carries the open-end voltage next to its open@to/open@from marker.

Improvements

  • The Web UI start scripts launch Julia with JULIA_NUM_THREADS=auto (unless already set), so the threaded surfaces actually use the machine's cores; runtime.parallel.* was already on by default and sizes itself from the thread count.

Version 0.9.11 - 2026-08-22

FACTS limit modes, one workshop notebook, and better discoverability.

Highlights

  • STATCOM and SSSC limit modes. A compensator shows its character at its limit, and that is usually during the voltage sag it was installed for. The machine voltage controller can now act as a STATCOM (s_max_mva): the bound is the converter current, so the deliverable Q follows the terminal voltage linearly instead of freezing at a fixed box, and instead of collapsing with V² like the SVC. The series controller can act as an SSSC (v_inj_max_pu): its reactance window is set by the injectable series voltage and narrows as the branch current grows. Both bounds follow the solved operating point from iteration to iteration. Why a UPFC is not on this list, and what to use instead, is explained on the new FACTS Devices page. (#297)
  • The whole workshop in one notebook. The introduction notebook and the workshop page are gone as separate things; the tour now carries everything in five tiers from Newcomer to Beyond, including two new Expert chapters on the FACTS limits and on N-1 contingency batches.

Documentation

  • Search engines and AI agents get proper signposts: robots.txt, sitemap.xml, llms.txt, meta description and preview image, and a page title that no longer collides with a bulk carrier of similar name.
  • The feature matrix gives each analysis domain its own table. State estimation is no longer graded against power-flow features it never claimed to have, and every warning marker now says why it is there.

Bugfixes

  • The docs landing page showed broken image references where the README logo and the Web UI screenshots belong; the README embed now rewrites the image paths for the site.
  • Re-running a configuration with a declared series_reactance controller crashed instead of recognizing the controller was already there.
  • addBusLoadPower! and addBusGenPower! change sums the AC solver never reads, which made them silent no-ops for AC studies. Their docstrings now say so and point to the prosumer-based way; the underlying AC/DC inconsistency is #323.

Version 0.9.10 - 2026-08-21

Multi-core execution, N-1 contingency analysis, and one-sided open branches.

Highlights

  • Multi-core execution. Independent work now runs on Julia threads (start with julia --threads=auto): AC islands (power_flow.islands.mode: solve_parallel), all-bus short-circuit sweeps, and contingency batches, gated by the new runtime.parallel.* keys (enabled, default true). Results stay identical to the serial runs, island for island and row for row. Measured on 16 cores: 4.8x on an 8000-bus fault sweep, 4.1x on a full N-1 over case1354pegase. The demos live in examples/run_parallel_suite.jl.
  • N-1 contingency analysis. New runContingencies! batch API: branch-outage cases from generateN1Branches or imported FOR001 lists, solved warm-started on copies of the base case (never mutating it) and checked against the voltage band and sn_MVA loadings. Islanding and non-convergence are reported per case instead of thrown, parallel circuits get their own cases, and results print as a table or CSV. The solves run without the rescue ladder; retry_flat_start = true grants one flat retry per failed case. See N-1 Contingency Analysis.
  • One-sided open branches. Branches carry from_status/to_status next to the aggregate status: open at exactly one terminal, a branch stays in the model as its exact pi reduction, draws its full charging, and reports the open-end voltage (Ferranti rise) as a branch result. CGMES Terminal.connected maps to the flags on import and export (replacing the half-charging substitute shunt); MATPOWER export represents the state as the exact Y_in bus shunt with an open_terminal= marker.

Performance

  • Takahashi sparse inverse for all-bus fault sweeps, opt-in via runShortCircuit!(net; sweep_method = :takahashi): the whole Thevenin diagonal of an island from one pass over the LU factors instead of one solve per bus, measured 34x to 264x over the serial sweep. Values agree with the default to machine precision; inapplicable islands fall back automatically. Theory in the Short-Circuit Compendium.
  • The solver status moved from global registries onto the Net, and per-island performance entries no longer overwrite each other. Two subtle serial-path details changed with the shared island solve: a failing island raises its error from the result handler (different backtrace origin, same error), and a nonfinite-voltage failure no longer counts its iterations into the aggregate.

Notebooks

  • New: distributed slack (where the participation weights come from and what happens without a valid participant) and transformer taps (OLTC vs PST device math, the phase-tap control loop, the 3WT star equivalent). Every notebook now opens with a topic warmup cell; the tour gained chapters on bus links, one-sided open branches, and parallel sweeps.

Removed

  • The klu linear-solver backend and the KLU.jl dependency: measured slower than UMFPACK on power-flow Jacobians, and a shared KLU factorization is unsafe under threads. Migration: replace linear_solver: klu with linear_solver: umfpack_reuse.

Version 0.9.9 - 2026-08-20

HVDC link results and one-reference-per-island validation.

  • HVDC links are first-class results: every link (Stage-0 or paired, from MATPOWER, CGMES, or the API) is tracked on the net and reported in a new HVDC Link Flows table, a Link row in the branch table (marked "HVDC, not a branch"), the result header, the generator-table terminal markers, a converter-loss line, ACPFlowReport.hvdc_links, and hvdc_links.csv.
  • A synchronous island with more than one angle reference now fails with a message naming the island and its reference buses (both solver paths) instead of the generic unsupported-bus-type error; an island_feed pair whose grid-forming reference was demoted reports invalid_topology. See HVDC Back-to-Back, example exp_hvdc_meshed_ac_tie.jl, and the meshed-operation tour subsection.
  • Branch Type column and report rows agree again: the printed label consults the import metadata first (PST heuristic pending the typed branch kind).

Version 0.9.8 - 2026-08-19

Steerable HVDC links. addHvdcPairControl! couples the two converter injections of a back-to-back link with exact loss accounting, per-terminal Q or voltage targets, and honest at_limit. Both importers attach pairs opt-in (paired_control); Stage-0 fixed injections stay the default. A grid-forming mode = :island_feed feeds an island whose only source is the receiving converter. See HVDC Back-to-Back. (#297)

  • The one-line install detects an older existing copy and offers the update (SPARLECTRA_UPDATE=1/0); the previous copy is kept as Sparlectra.old.
  • Leaner, faster installs: two unused hard dependencies dropped (76 to 54 packages), and the workshop notebooks now install into a fresh temporary environment from GitHub main, keeping Colab's preinstalled stack out of precompilation.
  • The result header now counts every island reference and keeps the reference models apart (Slack: n Source: m) instead of always printing 1; the workshop notebooks carry a network sketch per model.
  • Each GitHub release now ships an SPDX SBOM (Sparlectra-<tag>.spdx.json) as a release asset, generated by CI via tools/generate_sbom.jl (PkgToSoftwareBOM, kept out of Project.toml).

Version 0.9.7 - 2026-08-20

  • Install scripts offer a desktop shortcut (Windows .lnk) or launcher (.desktop / symlink) for starting the Web UI; SPARLECTRA_CREATE_SHORTCUT=1/0 for unattended installs.
  • The Jacobian condition estimate is always part of the result log and the Web UI run overview (with verdict); the output.condition_number option is removed, a leftover YAML key is ignored.

Version 0.9.6 - 2026-08-19

  • #309 Fix Sysimage for windows v0.9.5
  • #308 One-line Web UI install without a GitHub checkout; the install scripts now offer the optional fast-start sysimage build (y/N prompt, SPARLECTRA_BUILD_SYSIMAGE=1/0 for unattended installs).
  • Condition estimate now comes from the solver (lazy, computed only when reports ask); fixed condestJacobian(net) on nets with de-energized buses (DimensionMismatch), the estimate describes the active subsystem.

Version 0.9.5 - 2026-08-18

Highlights

Fast start for the Web UI. Optional PackageCompiler sysimage, built via tools/build_sysimage.jl or the new Fast start page; the start scripts use it automatically and skip the warm-up. See Sysimage.

Controllers in configuration. Outer-loop controllers can be declared under control.controllers, one named entry per controller. See Control Framework. (#305)

Improvements

  • Faster PowerFlow form renders (case-cache scan memoized) and an explicit list of saved case settings a newer configuration file replaced.
  • MATPOWER case downloads retry transient failures; run paths tolerate a symlinked user root.

Version 0.9.4 - 2026-08-14

Highlights

Series compensation (TCSC). addSeriesReactanceControl! steers a line branch onto an active-power target by moving x_pu within its range, with honest at_limit when the target is out of reach. Shows up in the result tables like the tap controllers. Theory in Series Compensation (TCSC), example exp_tcsc_series_reactance_control.jl, Colab notebook TCSC Flow Steering. (#297)

Improvements

  • Distributed slack participation moved into the bus table (dSl alpha, Pg eff MW) instead of a separate block.

Version 0.9.3 — 2026-08-12

Improvements

Distributed slack participation in the classical result print. With power_flow.distributed_slack, printACPFlowResults now lists each participating generator with bus, alpha share, correction dP, and scheduled vs. effective output. Also available in the solver status as distributed_slack_participation, see Power-Flow Configuration.

Version 0.9.2 — 2026-08-11

Highlights

(Source) External grid element. addExternalGrid! models the grid connection as an IEC 60909-0 network feeder — ideal slack by default, non-ideal source with internal_impedance = true. Its declared short-circuit power feeds runShortCircuit! directly, so hand-built and MATPOWER networks need no CGMES delivery for short-circuit studies. See Slack Bus and External Grid Sources. (#299)

Improvements / Documentation

Version 0.9.1 — 2026-08-01

Fixes and refinements after the 0.9.0 release.

Improvements

  • Saved case settings can be reset.

  • Rescue loader handles Q-limit-driven divergence.

  • CGMES deliveries start from their own state by default.

  • DC-line cases run again on old installations.

Version 0.9.0 — 2026-07-31

IEC 60909 short-circuit stage (#277), CGMES export.

Breaking Changes

  • Short-circuit calculation (IEC 60909-0). runShortCircuit! computes Ik''max/min, Sk'' and ip per fault bus from the CGMES short-circuit data. Where attributes were missing and defaults substituted (e.g. motors), the affected rows are flagged — a flagged Ik''max is a lower bound. Web UI gets a Short circuit button for CGMES cases that carry the data; writes `shortcircuitmax.csv/shortcircuit_min.csv`. (#277)
  • CGMES export. writeCGMESFiles writes a complete 2.4.15 delivery (EQ/TP/SSH/SV, optionally zipped). Export → re-import solves to the same power flow and keeps the original mRIDs; new objects get deterministic uuid5 ids, renaming never changes an mRID. Export checkbox in the Web UI run form.
  • Auto slack (power_flow.auto_slack, off by default): promotes the strongest injection to slack when the case registers none, logged.
  • Rescue ladder + DC fallback (power_flow.rescue, power_flow.dc.fallback, both off by default): failed AC solves are retried with alternate start / autodamp / DC-seeded start; if all fails, the DC fallback at least leaves angles and branch P in the net. AC status stays non-converged, honestly.
  • Import analysis. Deliveries that can't be imported now explain themselves: missing boundary sets are named by model id, unresolved references get a histogram, verdict in plain language. Runs automatically on Web UI upload (<case>.import_analysis.txt), also via analyzeCGMES.
  • BaseVoltage reconstruction (cgmes_import.infer_base_voltages, off by default): deliveries missing their BaseVoltage catalog get levels seeded from SV voltages and transformer rated voltages, then propagated. One summary warning.

Improved

  • compareWithSV removes the median angle offset before judging deltas — an IGM cut from a CGM keeps the global angle reference (a real 50Hertz delivery: ~27° on every bus, meaning nothing). Offset is reported separately.
  • Start scripts start_webui.sh/.bat and install_webui.sh/.bat in the repo root.
  • SVC-style voltage control via addShuntVoltageControl!; at the MVAr limit the device clamps to constant B (Q then follows V²). controllableElements(net) lists all controllers in one vocabulary.
  • PSTs update X(α) on tap moves when the winding carries reactance data. Fixed along the way: DTF importer dropped the phase-tap model, and the phase-probe compared stale flows against themselves.
  • Config edits show up on form refresh; newest edit wins between YAML and saved case settings.
  • cgmes_import.placeholder_guards: strict aborts on filler values instead of skipping them.
  • Diagnose / self-check evaluates CGMES cases at their own SV state — all start-value machinery forced off, including a flatstart: true that previously wiped the SV start silently. New artifacts self_check.log, self_check_residuals.csv.
  • CGMES runs: start values selectable (flat/sv), SV comparison now mandatory on every run (sv_compare.csv, sv_compare_flows.csv).
  • Wrong-branch detection only judges the highest voltage level — healthy 45 kV feeders no longer flag SUSPECT while the 380 kV level is clean.

Fixed

  • no slack bus found now says what's actually wrong: all buses isolated (branch-less delivery), slack on an isolated bus, or genuinely no slack — and how to fix each.
  • Failed multi-island runs no longer report iterations=0 / before_nr when the island did a full Newton solve, and never-attempted islands no longer copy the failed island's statistics into ac_island_solver_summary.csv.

Version 0.8.16 — 2026-07-30

CGMES Import

New Features

CGMES import. Sparlectra reads ENTSO-E CGMES 2.4.15 and 3.0 deliveries — folders, ZIPs, nested ZIPs, boundary sets — and builds a bus-branch Net from EQ+SSH+TP+SV: lines, 2- and 3-winding transformers, tap controllers, switches, shunts, loads, machines and injections, with snapshot-faithful handling of open terminals and out-of-service equipment. Each electrical island gets its SV-declared angle reference; assembled multi-area deliveries are supported, with HVDC border crossings kept as separate islands. summarizeCGMES inspects a delivery, compareWithSV validates the solved result against its SV profile. Configuration under cgmes_import.*; details in CGMES Import.

Distributed active-power slack. An island's power imbalance can be shared over the generators by participation factors (one lambda_P per island in the Newton state) instead of landing entirely on the reference bus. Enable with power_flow.distributed_slack.enabled (default false; disabled runs stay bit-identical). Weight modes include the imported MATPOWER APF / CGMES normalPF factors. See the Solver Guide.

PMU angles in state estimation. New VaMeas measurement type and addPmuPhasorMeasurement! for the Vm+Va pair; the estimator solves the PMU time-base offset as an extra state (state_estimation.pmu_ref_offset) and reports it in SEResult.vaRefOffsetDeg.

Tabular phase shifters land in the solved branch. CGMES PhaseTapChangerTabular was read but never applied. The table row at the tap position now folds into the imported branch (ratio and angle; an end-2 angle enters negated per end-referral). On RealGrid this reduces the deviation from the delivered SV state by roughly 40×.

Remote voltage control by machines. A machine can regulate the voltage at a different bus via its reactive output — the counterpart of a CGMES voltage RegulatingControl at a foreign terminal, previously held PV at the machine's own bus. addMachineVoltageControl! programmatically, or cgmes_import.machine_control (default false) from a delivery; parks honestly at_limit. Theory in Remote Voltage Control.

Nonlinear shunt compensators are mapped. Per-section points are summed up to the active section count (an interpretation choice, documented — the place to look if a delivery disagrees). Previously their reactive contribution was missing from the solve.

Improvements

  • The rectangular Newton solver refreshes its Jacobian in place; a PV↔PQ switch triggers a structural rebuild automatically.

Bugfixes

  • Active-set Q-limits now switch on the converged iterate. The start gating could outlast convergence, rejecting runs with remaining PV Q-limit violations instead of enforcing them; a converged iterate now always counts as ready.

Version 0.8.15 — 2026-07-22

DC power flow, faster linear solves, Web UI overhaul

Breaking Changes

  • Standalone DC power flow (rundcpf!, power_flow.solver = :dc) with per-island handling, optional DC-seeded AC starts, and a calculation-model selector in the Web UI.
  • Factorization-reuse linear solvers for the rectangular Newton step (power_flow.linear_solver: umfpack | klu | umfpack_reuse, default unchanged). Symbolic analysis runs once per sparsity pattern, only numeric refactorization per iteration. On case_SyntheticUSA, umfpack_reuse cuts the linear-solve phase from 1.66s to 1.31s; KLU is honestly slower on power-flow Jacobians. New dependency: KLU.jl.
  • Diagnostic report instead of a flat key/value dump, plus a fixed-reference self-check — both available via a dedicated "Diagnose" action in the Web UI.

Improvements

  • Cleaner PowerFlow form: collapsible option boxes, unified solver selection (AC / APSLF / DC), all texts in English, contextual help via docs-backed "?" links.
  • Case selector is now a single editable combobox: type to filter, Enter downloads unknown cases, right-click deletes cached ones.
  • Warm-up solve is deferred and the form render path pre-compiled — the first browser tab no longer stays blank.

Bugfixes

  • Tap controller: probe solves now follow the configured start strategy. control_propose_update! ran its ratio/phase probe power flows with the hard-coded default opt_flatstart = true instead of pf_config.start_mode.flatstart. Imported networks with off-nominal ratio branches (MATPOWER, DTF, CGMES) routinely fail from a flat start while converging from their stored voltage profile, so any outer-loop tap control on such a network aborted with pf_failed — with no hint that the probe start values were the cause.
  • MATPOWER parser accepts newline-separated matrix rows (RTS-GMLC style); validated against the MATPOWER 8.0 RTS-GMLC reference.
  • New Monte-Carlo examples: probabilistic power flow (case14) and a WLS state-estimation error study (7-bus workshop net).
  • Fixed two silent Web UI form bugs where disabled submit elements dropped their values (Diagnose button and DC selection ran normal AC solves).
  • Test suite: 1444 fast-profile and 1835 Web UI extended tests green.

Version 0.8.14 — 2026-07-22

Comment

  • intentionally the same as 0.8.13

Version 0.8.13 — 2026-07-22

Breaking News

  • AnalyticLoadFlow.jl (APSLF) is now integrated as an optional extension.
  • APSLF solver: can run standalone, as the framework solver, or as a start-value generator ahead of the rectangular NR solve. Doesn't yet cover OLTC/PST/Q(U)/P(U), only simple Q-limit switching — details in docs/src/external_solvers.md, demo in examples/powerflow/apslf_demo.jl.
  • Merit-function line search in the rectangular solver: optional Armijo line search inside the autodamp backtracking loop, off by default. Requires autodamp = true. Adds diagnostic fields and a merit_linesearch.log. Background in docs/src/solver.md.
  • Trust-region step control: alternative to autodamp, caps the Newton step adaptively instead of relying on the mismatch criterion. Two modes (scaled/dogleg); dogleg helps when the Newton direction weakens as the radius shrinks, but it's not a fix for bad starting values. Mutually exclusive with autodamp.
  • Web UI: solver dropdown including APSLF, status view shows which solver actually produced the result.

Improvements

  • Wrong-branch detection results are now visible everywhere (report, CSV, log, Web UI, API) instead of only showing up as console warnings. An automatic rescue-retry is deliberately out of scope — detection plus APSLF as a fallback are considered enough for the hard cases.

Bug fixes

  • The classical Q-limit outer loop was losing data per round because the inner solver treats every call as a standalone solve: the Q-limit log got reset each round (undercounting PV/Slack→PQ switches, often showing 0 even when switching actually happened), and diagnostic files from one round overwrote the previous round's. Fixed by keeping the log across rounds and giving each round its own file names.
  • Independently solved AC islands were overwriting each other's diagnostic files (merit, trust-region, iteration-start logs) since they all shared one output directory. Each island now gets its own prefix.

Version 0.8.12 — 2026-07-20

New Features

  • Added an AbstractTapChangerModel supertype and an explicit PowerTransformerTaps.convention field (currently only :neutral_relative), documenting the ratio-tap correction convention on the struct itself instead of leaving it implicit. See docs/src/branchmodel.md for the tap-changer/PST modeling layering (Issue #261 Stage 2).
  • Added a CGMES-style PhaseTapChangerModel <: AbstractTapChangerModel (:symmetrical/:asymmetrical, :tabular staged for later) and three pure formula functions in equicircuit.jl: calcPhaseTapFraction, calcPhaseTapAngleRatio (effective ratio/shift/regulating vector per CGMES v2.4 ch. 4.2/6.2), and calcPhaseTapReactance (CGMES ch. 3 X(α) interpolation, standalone for now — not yet wired into branch x_pu or the outer control loop). PowerTransformerWinding gained a parallel phase_taps field (Issue #261 Stage 3).
  • Added kind = :tabular support to PhaseTapChangerModel, backed by a new TapTablePoint struct (step, ratio, angle_deg, optional x_pu) and a calcPhaseTapTable exact lookup in equicircuit.jl. A table overrides formula-based reconstruction whenever present, integrated into calcPhaseTapAngleRatio/calcPhaseTapFraction/calcPhaseTapReactance. No interpolation between table steps; no importer produces tabular data yet (Issue #261 Stage 4).
  • Added phase_tap_side::Int / phase_taps::Union{Nothing,PhaseTapChangerModel} keywords to create3WTWindings!, letting one winding of a three-winding transformer (MVA-method, star/AUX-bus model) carry a PhaseTapChangerModel — i.e. a Schrägregler on a single 3WT winding, optionally combined with a ratio tap on the same winding. Validated: phase_tap_side ∈ 0:3, and phase_tap_side/phase_taps must be set together. Resolving the attached model into an effective branch ratio/shift, and addressing a single 3WT winding from the outer-loop PowerTransformerControl framework, are analysed but intentionally not implemented (Issue #261).

Improvements

  • Consolidated the ratio-tap correction and tap-range formulas into equicircuit.jl (calcRatioTapCorrection, calcRatioTapRange), the single source of truth used by both calcTransformerRatio (transformer.jl) and the Branch tap-limit derivation (branch.jl). Removed the duplicated inline formulas from both call sites; no change in computed values.
  • The native DTF importer's skew-angle tap computation (_dtf_effective_transformer_tap) now constructs a PhaseTapChangerModel and calls calcPhaseTapAngleRatio instead of computing tap_fraction and calling calcSkewAngleTap inline. Reproduces prior behavior exactly, including the pure-longitudinal case (Issue #261 Stage 3).

Bug Fixes

  • create3WTWindings! previously raised a MethodError for every call, including its own docstring example: the PowerTransformerWinding(...) positional call had drifted out of sync with the struct's field order (a ratio field was inserted ahead of shift_degree without updating this call site), so every argument from shift_degree onward landed one field too early. Fixed by inserting the missing ratio slot; the pre-existing tap_side side-selection logic itself (documented as 1-based [1,2,3], 0 = no tap) was left unchanged. The remaining discrepancy between that documentation and the current implementation is tracked with Issue #261.

Version 0.8.11 — 2026-07-17

New Features

  • writeMatpowerCasefile now has a matpower_export.write_solution option (default true). It writes the solved power flow back into the MATPOWER case — bus VM/VA get the solved values, branches gain result columns 14–17 (PF, QF, PT, QT), pulled from the existing report/loss path, no extra recomputation. A mpc.sparlectra.solution_written = 1 marker flags that the case has a solution. Turned off, you get the old pure model file: 13 branch columns, VM = 1.0, VA = 0.0. Ask for a solution before solving, and it falls back to the 13-column output with a warning. Configurable via YAML, the config API, or the new Web UI option.

Improvements

  • The tap-impedance correction factor from calcTapCorrectedRX (MATPOWER and native DTF import) is now saved on the Branch/transformer metadata. Reimporting a Sparlectra-exported case with the tap_changer_model = 'impedance_correction' marker won't stack a second correction on top. Cases without the marker — including third-party ones — work as before.
  • Shortened the "Tolerance" and "Export Solution" labels so the help icon isn't crammed against the text. The tolerance exponent stepper now looks and behaves like a real number spinbox.
  • Removed the inline explanation text next to "Export Solution" and "Tap-changer model" — it was overlapping the help icon. Full explanations are still one click away; docs were expanded to cover it.
  • Fixed the "Autodamping minimum" spinner step: was 1.0 (way too coarse for a 0–1 range), now 0.01.
  • Removed the remaining inline hints across the PowerFlow form (Tolerance, CSV format, benchmarks, etc.) — help text belongs in the docs, not hardcoded in the views. Docs updated accordingly.
  • Combined one-off Web UI messages (errors, import results, settings-loaded notices) into a single popup dialog instead of scattering them inline. The configuration notice stays inline since it has a link you need to see. Error history page unchanged.

Bug Fixes

  • Fixed wrong Case name in run.log/result files for MATPOWER cases with bus_name metadata — a loop variable was overwriting the case name with the last bus's name.
  • Fixed misaligned PV→PQ header lines in the report (labels too long for the column).
  • Fixed misaligned Solver time/Total time headers (3 characters too narrow).

Version 0.8.10 — 2026-07-17

New Features

  • Added a configurable transformer tap-changer model (transformer.tap_changer_model, allowed values ideal/impedance_correction), applied centrally to all transformers of an imported case. ideal (default) preserves prior Sparlectra behavior where the tap changer only changes the complex winding ratio. impedance_correction re-refers the transformer series impedance through the tapped winding, scaling R and X with the squared magnitude of the regulating vector, |1 + f·e^(jφ)|². The option is read by both the MATPOWER importer and the native DTF importer, but the correction math lives centrally in calcTapCorrectedRX/calcTapImpedanceCorrectionFactor (src/equicircuit.jl) so importers stay free of duplicated tap-impedance mathematics. Configurable via YAML, the GUI-editable configuration API, and a new Web UI expert option.

Version 0.8.9 — 2026-07-16

New Features

  • Added multi-file import for MATPOWER .m and DTF .DAT cases in the Web UI. Imported runnable cases are added to the case selector.

Improvements

  • Enabled MATPOWER DC-line handling through terminal P/Q injections by default.
  • Enabled independent AC-island solving and continued per-island diagnostics by default.
  • Improved Web UI file-import validation, filename handling, conflict reporting, and per-file status output.

Bug Fixes

  • Improved test-run diagnostics, bounded failure output, and excluded generated files from repository-hygiene checks.
  • Made DTF .DAT classification content-aware so only runnable cases and valid FOR002 references appear in the corresponding selectors.
  • Improved the tolerance spinner behavior in the Web UI.
  • Hidden unavailable commit information instead of displaying commit unknown.

Version 0.8.8 — 2026-07-14

New Features

  • Added native DTF import with support for transformer ratio conventions, Schrägregler/skew-angle controls, trailing branch records, and FOR001 contingency metadata.
  • Added an experimental DTF input path for the PowerFlow API and Web UI, including optional MATPOWER export, outage selection, import diagnostics, and explicit handling of unsupported DC-line data.
  • Extended MATPOWER import and export with Sparlectra metadata for bus and branch names, branch types, FOR001 contingencies, and transformer loss data.
  • Added support for the existing mpc.sparlectra.transformer_losses extension so transformer no-load conductance can be preserved across Sparlectra–MATPOWER roundtrips.
  • Added optional MATPOWER DC-line handling through fixed terminal power injections while keeping rejection of active DC-line records as the default.
  • Added user-facing documentation for the DTF format, transformer conventions, Schrägregler controls, trailing records, and FOR002 validation.

Improvements

  • Improved AC-island diagnostics and failure reporting, including per-island status, reference-bus selection, solver settings, mismatch information, Q-limit state, and downloadable diagnostic artifacts.
  • Improved API failure handling so non-converged runs retain their generated artifacts and remain available for ZIP download.
  • Improved Linux Web UI browser startup by using Chromium app windows where available and falling back cleanly to standard desktop browser launchers.
  • Added validation workflows for FOR002 cases A–E, including transformer ratios, branch identity, no-load losses, shunt sensitivity, outage handling, and MATPOWER roundtrip checks.
  • Simplified the DTF/FOR002 validation examples so normal use prints a concise summary while detailed diagnostics remain available on demand.

Bug Fixes

  • Preserved transformer conductance through the existing PowerTransformerWinding.g → getTrafoRXBG / getTrafoRXBG_pu → Branch.g_pu path without creating synthetic terminal bus shunts.
  • Fixed MATPOWER reimport so mpc.sparlectra.transformer_losses restores transformer conductance correctly and does not duplicate loss contributions.
  • Fixed transformer loss reporting so total branch losses can be separated into longitudinal copper losses and voltage-dependent no-load losses.
  • Fixed the Web UI case selector so internal warm-up cases are no longer shown as selectable user cases.

Version 0.8.7 – 2026-06-23

New Features

  • Added a guarded current-injection start pre-solve for large MATPOWER workflows, including YAML/API/Web UI configuration, sidecar persistence, and current_iteration_start.log diagnostics.
  • Added conservative MATPOWER auto-profile handling with safer default recommendations, explicit apply/skip logging, and no silent solver-start or Q-limit overrides.
  • Added explicit fail-fast detection for unsupported active MATPOWER mpc.dcline data in API/Web UI runs.
  • Added classical Q-limit enforcement modes (classic_simultaneous, classic_one_at_a_time) for diagnostic and large-case analysis.
  • Added Web UI case-sidecar settings and compact operation-log retention for local support workflows.

Improvements

  • Improved power-flow diagnostics with DC-start quality metrics, compact/full mismatch summaries, clearer final-mismatch status, and Q-limit validation output with MVAr units and non-converged validity labels.
  • Improved Web UI/API transparency by recording runtime case metadata, MATPOWER import decisions, Q-limit settings, output options, and partial artifact status in logs and effective configuration artifacts.
  • Improved the PowerFlow Web UI layout, Advanced/Expert option grouping, dismissible validation errors, and collapsed Last Errors handling.
  • Improved detailed CSV handling for large and non-converged runs, including partial exports, structured skip reasons, and artifact rediscovery.
  • Restored compact default test-run output while keeping verbose diagnostics available as an explicit opt-in.

Experimental / Developer Tooling

  • Moved the large-case Q-limit comparison utility to examples/experimental/qlimit_large_case_comparison.jl and removed it from the stable API and normal fast test profile.

Bugfixes

  • Fixed MATPOWER import option propagation for Web UI/API runs, including auto-profile, transformer ratio, phase shift, shunt, and voltage-reference options.
  • Fixed Web UI case-sidecar save/load handling with type-safe YAML values and case-local persistence.
  • Fixed Q-limit diagnostics so stale/intermediate violations are not reported as final violations when the final PV/REF check is OK.
  • Fixed effective configuration artifacts so they include the actual resolved runtime casefile and case name.
  • Fixed CSV artifact handling for non-converged runs with available network state.

Version 0.8.6 – 2026-06-18

Improvements

  • Added configurable detailed CSV artifact writing with auto, buffered, and streaming modes so large output artifacts can stream directly to disk while normal cases keep buffered writes.
  • Improved direct detailed CSV export with lower-allocation cell writing and per-file timing metadata for large Web UI/API runs.
  • Removed the remaining per-bus prosumer scan from direct detailed CSV bus control labels, added operation-log CSV progress events, and bounded Q-limit run-log details while preserving full Q-limit CSV artifacts.

Bugfixes

  • Fixed Web UI Q-limit checkbox handling so an unchecked form submission explicitly disables power_flow.qlimits.enabled in the effective configuration.

Version 0.8.5 – 2026-06-17

Improvements

  • Cleaned up single PowerFlow Web UI/API timing output so unavailable solver time is omitted and the measured run duration is shown as wall time.

Bugfixes

  • Fixed MATPOWER .m import for very large cases such as case_ACTIVSg70k.m by replacing regex-based matrix block extraction with deterministic string scanning.

Version 0.8.4 – 2026-06-10

New features

  • A loopback‑only PowerFlow Web UI was added with run history, artifact downloads, logs, MATPOWER case management, help, and writable runtime directories. Optional artifacts now include additional logs, configuration output, and detailed CSV exports for voltages and branch flows.

Improvements

  • Added configurable detailed CSV formatting for technical, German Excel, and US Excel notation.

Bugfixes

  • Fixed rectangular PV/slack voltage initialization so replacing a voltage magnitude preserves the existing phasor angle.

Version 0.8.3 – 2026-05-30

Breaking changes

  • Replaced the former keyword-heavy high-level runner surface with the configuration-driven run_sparlectra framework entry point.
  • Kept run_acpflow only as a thin alias for run_sparlectra; it now accepts the same minimal framework arguments.
  • High-level import, solver, control, benchmark, and output behavior is now controlled through SparlectraConfig or YAML configuration.
  • Framework runs now consistently return SparlectraRunResult for both casefile and net workflows.

New features

  • Added config-driven multi-case MATPOWER execution via matpower_import.cases and the dedicated run_sparlectra_cases helper, while keeping run_sparlectra as a single-case workflow.

Improvements

  • Refactored the high-level ACP/MATPOWER workflow into a clearer framework path with separated import, execution, status, and output handling.
  • Refined SparlectraRunResult status semantics so numerical convergence, solution availability, control-loop status, limit validation, and final framework acceptance are reported separately.
  • Preserved detailed rectangular diagnostics in MATPOWER runner status rows, including Q-limit active-set information, final PV voltage residuals, and wrong-branch metrics.

Bugfixes

  • Fixed file-based MATPOWER start handling so projected/imported voltage and angle starts are actually honored by the rectangular solver instead of being overwritten by an effective flat start.
  • Fixed framework and MATPOWER status handling for rejected numerical solutions, including wrong-branch rejection, active-set rejection, controlled-run PF failures, disabled controls, and synthetic benchmark summaries.
  • Fixed file-based MATPOWER Q-limit lock handling so power_flow.qlimits.lock_pv_to_pq_buses is resolved from original MATPOWER bus IDs to internal Sparlectra bus indices.
  • Fixed the public ensure_casefile binding and updated runnable documentation snippets so fresh checkouts and package installations can load example MATPOWER cases correctly.
  • Fixed the tap-control demo classic-output toggle so the example no longer fails late with an undefined show_classic binding.
  • #228 Remove the old compatibility surface from the high-level runner

Version 0.8.2 – 2026-05-29

New Features

  • Added configurable wrong-branch diagnostics for rectangular power-flow results, including voltage, angle-spread, and branch-angle plausibility checks.

Improvements

  • Hardened matpower_import.auto_profile into a MATPOWER pre-run that logs recommendation evidence, preserves recommend mode without changing the active configuration, applies safe import-convention, comparison-reference, start-mode, and Q-limit guard changes in apply mode, and prints final effective options without rewriting YAML files.
  • Refactored the rectangular complex-state power-flow implementation into focused modules under src/powerflow_rectangular/, with runpf_rectangular! as the network-integrated entry point and run_complex_nr_rectangular as the standalone array-level solver.

Bugfixes

  • Fixed run_acpflow(casefile=...) configuration forwarding so MATPOWER/file-based rectangular solves honor configured power_flow.wrong_branch_* options instead of falling back to default diagnostics.
  • Aligned rectangular autodamping defaults so direct solver calls and configuration-driven runs use the same autodamp_min = 0.05 default.
  • #193 Harden MATPOWER auto-profile recommendation and application
  • #219 Detect wrong/false low-voltage branch convergence and retry safely
  • #220 Mechanically split rectangular power-flow code into focused modules

Version 0.8.1 – 2026-05-26

Breaking Changes

  • Removed the old run_net_acpflow public wrapper. Use run_acpflow(net=...) for already constructed networks and run_acpflow(casefile=..., path=...) for file-based workflows.

Highlights

  • Added a generic outer-loop control framework above runpf!. Transformer tap/phase control now uses this framework as the first concrete controller implementation.

Improvements

  • Added structured ControlRunResult output and latest_control_result(net) for inspecting controller status, outer iterations, PF solve count, controller rows, and trace rows.
  • Added machine-readable control trace rows for transformer control.
  • Added run_acpflow(net=...) as the preferred high-level entry point for already constructed networks, and made it the single public in-memory entry path.
  • Simplified examples/tap_control_demo_grid.jl into a lightweight demo of the generic control framework.
  • Documented the control configuration section, including that control.controllers is reserved for future YAML-based controller instantiation.

#179 Introduce a generic control framework above the existing power-flow solver.

Version 0.8.0 – 2026-05-25

Breaking Changes

  • The public AC power-flow path now supports only the sparse rectangular Newton-Raphson solver. Legacy polar/classic methods, dense PF matrices, and finite-difference PF Jacobian options are no longer supported as user-facing runtime choices.
  • Power-flow configuration has moved to structured YAML sections and typed configuration objects. The old flat keyword-style configuration path is deprecated/removed for the cleaned rectangular workflow.
  • Obsolete sparse switches such as power_flow.sparse, opt_sparse, and state_estimation.sparse are no longer valid configuration keys. Sparse matrix handling is mandatory for the production PF core.

Highlights

  • Added a central configuration workflow:
    • default template: src/configuration.yaml.example
    • optional user override: examples/configuration.yaml
    • typed config objects for power flow, MATPOWER import, state estimation, diagnostics, output, benchmarking, runtime, and performance profiling
    • early validation of unknown or obsolete keys
  • Simplified the production power-flow path to the sparse rectangular AC solver with sparse Y-bus assembly, sparse analytic rectangular Jacobian, and sparse linear solves.
  • Improved MATPOWER import and runner workflows:
    • central YAML-driven execution
    • configurable MATPOWER import options
    • cleaner compact summaries
    • better separation of numerical convergence, Q-limit validation, and solution availability
  • Added configurable performance and timing output for MATPOWER and rectangular PF runs:
    • representative wall time
    • solver time
    • result-output time
    • timing coverage
    • optional allocation information
    • rectangular workspace metadata
  • Improved large-case output handling:
    • configurable result-table row limits
    • summary/compact/full result-output modes
    • safer default behavior for large MATPOWER cases
  • Added rectangular workspace reuse/preallocation controls:
    • power_flow.rectangular_workspace_reuse
    • power_flow.rectangular_preallocate_workspace
    • power_flow.rectangular_workspace_min_buses
  • Reworked example scripts under top-level examples/:
    • matpower_import.jl is now a thin YAML-driven entry script
    • tap_control_demo_grid.jl uses central configuration helpers
    • export_solution.jl writes deterministic export artifacts under examples/_out/export_solution/<case>_<timestamp>/
  • Added a clearer test profile structure:
    • default fast profile for normal development
    • extended profile for MATPOWER/output/documentation-heavy checks
    • all currently aliases extended

Configuration Notes

  • The main configuration template is now:

    src/configuration.yaml.example

  • A local user/example override can be placed at:

    examples/configuration.yaml

  • MATPOWER benchmarking moved to the top-level benchmark section:

    benchmark:
  enabled: true

    The old matpower_import.benchmark key is rejected with a migration message.

Output and Examples

  • examples/export_solution.jl now produces files instead of only printing to the console. Typical output files are:

    summary.txt
internal_solution.csv
external_solution.csv
comparison.csv
*_export.m

  • examples/matpower_import.jl uses the central configuration and writes logs under:

    examples/_out/

Documentation

  • Added/updated documentation for:
    • central configuration
    • power-flow configuration
    • MATPOWER import configuration
    • state-estimation configuration
    • performance profiling
    • test profiles
    • examples overview
  • Issue #199: Central configuration, PF solver simplification, sparse-only PF core, and test-framework cleanup
  • Issue #201: YAML redesign

Version 0.7.8 – 2026-05-16

Highlights

  • Improved Q-limit handling for large MATPOWER imports, especially cases with many generators that have zero or very narrow reactive-power ranges.
  • Added compact console reporting for large MATPOWER example runs while keeping the full diagnostics in the logfile.

Fixes

  • Fixed rectangular power-flow status caching to use weak network keys so repeated benchmark/example solves can release imported Net objects after callers consume the status.
  • Fixed direct run_acpflow(net=...) and rectangular solver defaults so the Q-limit guard remains opt-in unless a caller or config explicitly enables it.
  • Fixed direct run_tap_controllers_outer! defaults so the Q-limit guard remains opt-in when the exported tap-control API calls runpf!.
  • Fixed MATPOWER example config forwarding so Q-limit guard options from YAML are preserved in the effective config and passed through all run_acpflow paths.
  • Fixed MATPOWER example console row limiting so console_max_rows from YAML is forwarded to Q-limit event and final active-set row caps in the run_acpflow paths.
  • Fixed rectangular NR status reporting: numerical convergence, Q-limit active-set consistency, final convergence, comparison status, and rejection reasons are now reported separately.
  • Fixed Q-limit guard behavior so strongly violating active PV buses can be locked to PQ during eligible Q-limit checks, reducing final active-PV limit violations.

Improvements

  • Reduced console noise for large Q-limit active-set cases by replacing long PV→PQ and violation tables with compact summaries.
  • Improved MATPOWER import example logging: the console now shows the essential run status, while detailed auto-profile evidence, diagnostics, and solver traces remain available in the logfile.

Version 0.7.7 – 2026-05-13

Highlights

  • Improved MATPOWER import diagnostics for large and difficult cases:

    • clearer YAML-backed logging and terminal summaries
    • VM/VA reference checks
    • branch-shift and transformer convention scans
    • bus-shunt and fixed-reference residual diagnostics

  • Added configurable MATPOWER transformer and phase-shifter import conventions:

    • matpower_ratio
    • matpower_shift_sign
    • matpower_shift_unit

  • Improved MATPOWER voltage reference handling:

    • configurable BUS.VM vs. GEN.VG handling
    • hybrid comparison mode
    • better diagnostics for PV/REF buses without online generators
    • correct handling of buses switched from PV to PQ by Q-limits

  • Improved rectangular NR robustness for large MATPOWER cases:

    • preserved slack and PV setpoints during flat start
    • added DC-angle and blended-voltage start support
    • added wrong-branch diagnostics for suspicious low-voltage solutions

  • Improved PV→PQ Q-limit handling:

    • clearer per-unit and MVAr diagnostics
    • hysteresis/deadband support
    • compact post-solve active-set consistency checks
    • separate handling of PV and REF/slack Q-limit violations

Fixes

  • Fixed MATPOWER slack/reference voltage import so VM/VA values are preserved unless explicitly overridden.
  • Fixed nominal-tap transformer handling so explicit TAP = 1 branches remain transformer models.
  • Fixed repository-local MATPOWER .jl case loading and example regression-test handling.
  • Fixed several Julia 1.12 / Revise world-age issues in MATPOWER diagnostic and example paths.
  • Fixed flat-start setpoint extraction for networks with isolated buses.
  • Improved warning/error capture in matpower_import.jl with compact logfile summaries.

Diagnostics

  • Added MATPOWER auto-profile pre-run mode (matpower_import.auto_profile = recommend|apply) to summarize or apply robust import, flat-start, PV/REF voltage-source, and Q-limit settings while preserving explicit YAML overrides.
  • Added branch-neighborhood reports for selected high-residual buses.
  • Added residual-cluster diagnostics for PEGASE-style mismatch regions.
  • Added negative branch impedance scans while preserving signed MATPOWER BR_R / BR_X values.
  • Added MATPOWER reference-data consistency diagnostics, including documentation of the known case300.m fixed-reference mismatch around BUS_I 196 / 2040.
  • Issue #186: Singular Jacobian / world-age issue in rectangular NR solver.## Version 0.7.6 – 2026-05-11

Bugfix

  • Fixed rectangular MATPOWER runs so a singular Newton Jacobian is reported as non-convergence instead of aborting the example, and tightened Julia 1.12 / Revise entry-point calls to avoid world-age binding warnings.

Version 0.7.5 – 2026-05-10

New Features

  • Added PV→PQ Q-limit switching start controls (qlimit_start_iter, qlimit_start_mode, qlimit_auto_q_delta_pu) for rectangular power-flow runs.
  • Added configurable bus-shunt modeling with classic Y-bus admittance stamping and a rectangular-solver voltage-dependent injection mode for keeping shunt effects in nonlinear mismatch terms.
  • Added automatic rectangular Newton damping (autodamp) for difficult flat-start power-flow cases; the solver backtracks the Newton step from damp down to autodamp_min and accepts the first residual-reducing trial.
  • Added start projection (start_projection) for difficult power-flow seeds, including optional DC-angle starts and raw/DC blend scans for both the internal rectangular solver and external-solver PFModel starts.

Bugfix

  • Replaced the singular sparse linear-solve fallback with a rank-revealing QR path before dense SVD fallback to avoid large pinv/LAPACK failures in ill-conditioned rectangular Newton steps.
  • Wrapped MATPOWER voltage-angle comparison differences into the minimal ±180° range before applying angle tolerances.
  • Synthetic tiled-grid benchmark example now falls back from .yaml to .yaml.example and reports when built-in defaults are used.

Improvements

  • Improved MATPOWER case parsing and network construction performance for large cases by reducing parser allocations, pre-sizing network containers, and avoiding repeated bus-name lookups while adding MATPOWER branches.

Version 0.7.4 – 2026-05-08

New Features

Bugfix

  • Fixed YAML subset parser cleanup on parse errors so temporary config files are closed before mktemp cleanup on Windows.

Improvements

  • Improved large-network MATPOWER and rectangular-solver performance by aggregating prosumer-derived bus types and specified power injections in linear time.

Version 0.7.3 – 2026-05-04

New Features

  • Added transformer tap control in AC power flow with an outer-loop controller workflow.
  • Added addTapController! support for:
    • voltage control via tap ratio (mode = :voltage)
    • branch active-power control via phase shift (mode = :branch_active_power)
    • coupled ratio + phase control (mode = :voltage_and_branch_active_power)
    • tap-control reporting in classic and structured ACP flow outputs

Improvements

  • Updated branch-model documentation with practical controller-direction probing guidance for phase-shift control.

Tests

  • Added regression tests for transformer tap controller behavior.

Version 0.7.2 – 2026-04-15

Bug Fixes

  • Fixed the increased loading time for large test cases

Version 0.7.1 – 2026-04-15

Improvements

  • Refactored decision logic for qlimit_mode = :adjust_vset:
    • consolidated Q-limit event handling via shared active-set flow
    • extracted voltage-step handling into dedicated helper logic
  • Added typed VoltageAdjustConfig support for prosumers and integrated it into voltage-regulation detection.
  • Simplified rectangular mismatch API by removing unused derivative keyword arguments from mismatch_rectangular(...).

Solver Robustness

  • Rectangular solver now handles reduced Ybus matrices (caused by internal isolated buses) by expanding them back to full network dimension for mismatch/Jacobian processing.
  • For rectangular runs with active-link merges and internal isolated buses, solver now uses a rectangular FD fallback path instead of switching to :polar_full.

Version 0.7.0 – 2026-04-15

New Features

  • Added support for P(U) and Q(U) controller models in power flow calculations
  • Characteristic curves can be defined via discrete points, with selectable interpolation methods:
    • linear
    • piecewise constant
    • cubic
    • polynomial
  • Added support for controllers on non-PV generators
  • Added documentation and examples for the new controller features

Version 0.6.4 – 2026-04-12

New Features

  • Marked legacy Jacobian solvers as deprecated:
    • runpf_full! / method = :polar_full
    • runpf_classic! / method = :classic
    Both now emit a deprecation warning when used.
  • Set rectangular complex Jacobian (method = :rectangular) as the default for:
    • run_acpflow
    • run_acpflow(; net = net, ...)
  • Updated examples and user documentation to use :rectangular as the recommended/default solver method.

Bugfixes

  • For backwards compatibility, if vm is set, the coresponding bus type is set to PV

Version 0.6.3 – 2026-04-11

New Features

  • Added :adjust_vset controller-based Q-limit handling at PV buses (adaptive Vset steps before optional PV→PQ fallback).
  • Power-flow bus typing is now derived from attached prosumers (Slack > PV > PQ); addBus!(busType=...) is legacy-only and no longer defines operational PF type.

Version 0.6.2 – 2026-04-02

New Features

  • adding sign validation and optional autocorrection of Q-limits before running power flows.
  • Provide an option to lock selected PV buses from being switched to PQ
  • Added pre-run PV Q-limit preview logging in MVAr for easier diagnostics before the PF iteration loop.

Version 0.6.1 – 2026-03-24

New Features

  • Bad Data Detection (BDD) and Statistical Diagnostics for State Estimation (SE)

Bug Fixes

  • Fix Issue 139

Notes

  • Bad-data detection and statistical diagnostics are available via validate_measurements, runse_diagnostics, summarize_se_diagnostics, and print_se_diagnostics.

Version 0.6.0 – 2026-03-17

New Features

  • Added initial State Estimation (SE, WLS), documentation and examples.
  • Added zero-injection-bus (ZIB) support in SE, documentation, and examples.
  • Improved results reporting and logging.
  • Refactor testsuite for clarity.

Notes

  • State Estimation is currently marked as experimental.
  • Passive buses / ZIB are currently represented through zero-injection pseudo measurements in the WLS workflow.

Version 0.5.0 – 2026-03-11

Highlights

  • Create and use machine-readable ACPFlowReport output
  • Introduced bus-links + documentation
  • Consolidated recent documentation

Version 0.4.35 – 2026-02-02

Fixes

  • Stabilized MATPOWER case import and bus indexing in the presence of isolated buses.
  • Fixed bus classification and ordering to ensure consistent solver input for flatstart and non-flatstart runs.
  • Improved internal consistency checks for network topology before solver execution.

Improvements

  • Refined MATPOWER comparison tooling (Vm/Va diff diagnostics and angle alignment).
  • Minor internal cleanups in MatpowerIO and network initialization to improve robustness.

Version 0.4.34 – 2026-06-02

Fixes

  • Closed Issue 110:
  • Corrected per‑unit conversion issues discovered during the refactoring of the MATPOWER import interface, including proper handling of line, transformer, and shunt parameters.
  • Fixed PU conversion inconsistencies for shunts and aligned all shunt handling with the unified Y‑model.
  • Harmonized Q‑limit checking across all solvers and ensured consistent PV/PQ switching behavior.

New Features

  • Updated shunt modeling: addShunts now exclusively supports the Y‑model; voltage‑dependent shunt power must be represented via loads or generators.
  • Unified solver logic for reactive power limits and PV/PQ switching
  • added new testcase matpower vs manual network

Version 0.4.33 – 2026-02-02

New Features

  • Added on-demand MATPOWER case handling via FetchMatpowerCase.ensure_casefile (automatic download of .m cases and optional generation of .jl cases)
  • Added option to control flat start behaviour in AC power flow (flatstart)

Internal Changes

  • Refactored MATPOWER case import logic and removed redundant example-based loaders

Version 0.4.32 – 2026-30-01

New Features

  • Added function to export calculated network

Version 0.4.31 – 2026-28-01

Title

  • Change license from BSD-3-Clause to Apache License 2.0

Version 0.4.30 – 2026-28-01

Bug Fixes

  • Importing Matpower files, wrong conversion to per unit system

New Features

  • Option to choose flatstart

Version 0.4.29 – 2025-12-13

New Features

  • Support native 3-winding transformers in Sparlectra without AUX bus generation and allow non-PU parameterization #90
  • Support 2-winding transformers as well in non-PU parameterization #90

Version 0.4.28 – 2025-12-12

Bug Fixes

  • Close Issue #85: test totalBusPower vs. TotalLosses failed for solver != rectangular

Version 0.4.27 – 2025-12-11

New Features

  • Added 3Bus Testcase with PV Generator
  • Added createTest5BusNet with multiple generators at one bus and zero injection generator
  • Issue #81 Fix Calculatation of Generators with Zero Injektion
  • Mark PV->PQ switching in printACPFlowResults Log

Bug Fixes

  • fixes per unit calculation for line shunt admittance
  • fixes calculation of losses

Version 0.4.26 – 2025-12-04

  • Issue #74: Multiple Generators at One Bus Not Handled Correctly
  • Documentation Updates

Version 0.4.25 – 2025-11-29

Added

  • Rectangular (Complex-State) Newton–Raphson Solver

Version 0.4.24 – 2025-11-20

Added

  • Q-limit handling with automatic PV→PQ switching when a generator hits its reactive power limits.

Version 0.4.23 (2025-11-11)

New Features

  • Full-state Newton–Raphson solver

Version 0.4.22 (2025-08-27)

Bug Fixes

  • small fixes

Version 0.4.21 (2025-03-14)

New Features

  • adding functions to remove elements from a net

Version 0.4.20 (2025-03-11)

  • internal reorganization and small bugfixes

Version 0.4.19 (2024-10-14)

Bug Fixes

  • pu calculation for transformer impedance

Version 0.4.18 (2024-04-14)

Bug Fixes

  • closes issue#48 "printout jacobian runs into error"

Version 0.4.17 (2024-04-14)

New Features

  • added testcase for importing Matpower files

Bug Fixes

  • bugfix wrong function call in createnet_powermat

Version 0.4.16 (2024-04-13)

Bug Fixes

  • bugfix shunt index for isolated buses, closes issue #38

Version 0.4.15 (2024-04-12)

New Features

  • Implemented a function to detect isolated buses and incorporate them into the network calculation (Issue #38)

Version 0.4.14 (2024-04-12)

Bug Fixes

  • bugfix addShuntPower, closes issue #36

Version 0.4.13 (2024-04-12)

New Features

  • added attribute for Lineparameters based on length
  • added update parameter function for network
  • added workshop documentation

Bug Fixes

  • taking line length not (always) into account for line parameters
  • parsing emtpy lines of Matpowerfiles
  • documentation rendering

Version 0.4.12 (2024-04-08)

New Features

  • added functions to facilitate the modification of networks.
  • documentation available at https://welthulk.github.io/Sparlectra.jl/.

Bug Fixes

  • print prosumers

Version 0.4.11 (2024-04-05)

New Features

  • make changes to imported Matpower networks after import.
  • added functions to facilitate the creation of networks.

Enhancements

  • added documentation make file

Bug Fixes

  • import and parser for Matpower .m files

Version 0.4.10 (2024-03-30)

New Features

  • removed numerous redundant functions, partially restructured classes
  • removed support for CGMES due to the availability of numerous alternative libraries
  • removed support for the legacy custom JSON data format (potentially subject to reintroduction at a later stage)
  • added functions to facilitate the creation of networks
  • better performance

Bug Fixes

  • calculation of branch flow and losses
  • branches out of service

Version 0.4.8 (2024-03-26)

  • first package release registered in the Julia registry

Version 0.4.1 (2023-12-19)

  • Initial release of Sparlectra

Version 0.4.0 (2023-11-30)

  • Initial public commit of Sparlectra