0.4.0

See CHANGELOG.md for what changed.

0.3.27

See CHANGELOG.md for what changed.

0.3.26

See CHANGELOG.md for what changed.

0.3.25

See CHANGELOG.md for what changed.

0.3.24

See CHANGELOG.md for what changed.

0.3.23

See CHANGELOG.md for what changed.

0.3.22

See CHANGELOG.md for what changed.

0.3.21

See CHANGELOG.md for what changed.

0.3.20

See CHANGELOG.md for what changed.

0.3.19

Added

• Component names are placed instead of pinned below the symbol. A name now tries
  the four sides of its component in turn — below, above, right, left — and takes the
  first that lands on nothing, so it clears the other symbols, the wires and the names
  already placed. The failure this fixes is silent: a name under one symbol lands
  inside the symbol beneath it, and a wire routed under a row of components crosses
  every name in that row, so the reader reads the ink and not the name. The search is
  per frame over screen-space rectangles with a cell grid, because a hundred-component
  model has a couple of hundred wires and a quadratic test would stutter a drag.

• Settings → Diagram → "Move names out of the way", on by default. Off, every name
  sits centred below its own symbol as it always did. Both surfaces follow it: the
  Studio and the diagrams embedded in notes.

• The README now opens with a whole-studio screenshot in both themes, and carries a
  Mermaid flowchart of what actually happens between the plugin and OpenModelica —
  serialize, translate, compile, run, read the result — with the error path drawn in.
  Two more worked examples (RLC and ResistorSelfHeating) join the mechanical
  one, each with its model, the algebra in LaTeX, and a table of closed-form versus
  simulated values.

• Disclosures in the README for the two things the plugin directory requires to be
  stated: the files it touches outside the vault (the Modelica libraries of your
  OpenModelica installation, read to index classes; the system temporary folder, where
  omc translates and compiles) and its network use (none unless AI assistance is
  switched on, and then only the provider you configure). It also states plainly that
  there is no telemetry, no analytics and no update check.

• scripts/readme-icons.json: the Lucide shapes the toolbar and menus use,
  vendored (ISC, credited in the third-party notices) so the README's screenshots can
  draw the same icons the app does. A test fails if a new icon has no shape.

Changed

• Eleven examples are shipped in the arrangement their author drew: Rectifier, RLC,
  SineAC, BatteryDischarge, DCMotor, HalfWaveRectifier, NonlinearOrifice, PipeFriction,
  Thermal, HeatConduction and ResistorSelfHeating. Several also moved their reference node
  from the source's n terminal to its p; where that reversed a branch the audit
  expectations were re-derived (a battery terminal voltage is now checked as a difference,
  and resistor.i is negative because resistor.p is the grounded end).

• ResistorSelfHeating: ambient is a named allowance in the diagram-box test: the
  electrical-to-thermal chain runs to x 120, and pulling the still-air source back inside
  ±100 would put it on top of the conductor it is wired to.

• The Edit and View buttons are icons alone — Undo, Redo, Copy, Paste, Delete,
  Rotate, Zoom in, Zoom out, Fit to view. They are the eight every editor's toolbar
  has, and with the words gone the whole toolbar fits on one row at the width the
  studio is usually given. The Model row keeps its labels, and every icon-only button
  keeps its NAME in aria-label — which is also the tooltip Obsidian draws — with
  each hint beginning with the action rather than the consequence.

Fixed

• scripts/readme-images.mjs never finished. The Help capture lost its await
  when cropToScene became async, so every run ended with
  image.toPNG is not a function and a non-zero exit — after the Help images had
  stopped being regenerated at all. The README's images are a build artifact, so this
  was a stale picture waiting to be noticed.

• Loading a built-in example can no longer write itself over a file of the same name.
  The name-to-path record is keyed by model name, so a vault holding Modelica/RLC.mo
  kept the shipped RLC associated with it: the header named the user's file while the
  canvas held the bundled model, and the next switch saved the bundled text into that
  file. Two files in the test vault were overwritten exactly that way. The Examples picker
  now loads with fromExample, which drops the record after the outgoing model has been
  flushed — an example is not a document and holds no path until it is saved somewhere.

• The RLC note's damping ratio was wrong by a factor of ten, and the note explained the
  discrepancy with an invented rule. ζ = (R/2)·√(C/L) was 0.158 for the C the note then
  had, not 1.58; the circuit was underdamped all along, so the overshoot the simulation
  showed was simply correct. The note now derives ω₀ = 100 rad/s, ω_d = 86.6 rad/s and
  ζ = 0.5 for C = 1e-3, gives the 16% overshoot as the expected result, and the "a series
  RLC rings at any ζ" claim (with its non-existent complex zeros) is gone from the note,
  the generator and the test header. The account in docs/testing-findings.md is corrected
  rather than deleted, because a wrong note that teaches a false rule is worth remembering.

• The Rectifier note documented a load of 1000 Ω that the model does not have. The
  model's resistor is R=100, so the discharge time constant is RC = 10 ms — half the
  20 ms cycle, not 0.1 s — and the capacitor loses 63% of its charge between peaks. The
  prose, the τ line and the resistor's name now match the model.

• The Rectifier's capacitor is wired across the load again. In the re-arranged model it
  had been drawn between the ground rail and the source's live terminal, which put it
  across the source: capacitor.v simply followed the sine and the load was left
  unsmoothed. One endpoint changed (capacitor.p to diode.n); the verification numbers
  the note has always carried — 9.54057 V peak, 0.367868 over 10 ms — are unchanged.

• Electrical's KVL row reads v_C − v_R = 10 V. The arrangement now references the
  circuit to the source's p terminal, so the resistor's voltage is negative and the two
  add with a sign. The capacitor still charges to +10 V.

• A command id no longer repeats the plugin id. open-modelica-studio reached the
  command palette as "Modelica Studio: Open Modelica Studio", because Obsidian prefixes
  every command id with the plugin's. Now open-view, and a test asserts that no
  registered command id carries the plugin id.

• Saving a model writes through Vault.process rather than Vault.modify, so a
  file another writer touched since it was read is not clobbered by a save.

• The plugin no longer prints to the console on every load. The one line that
  advertised the modelicaStudio debug handle is now behind the debug-log setting; the
  rest of the console output is a diagnostic level (error/warn always, info with
  the setting, debug only with the console's own verbose flag).