Skip to content

v0.24.0

Choose a tag to compare

@rjwalters rjwalters released this 20 Oct 15:33
· 3894 commits to main since this release

Upgrade notes / behaviour changes

  • --edge-clearance 0 no longer permits copper across Edge.Cuts
    (Issue #6043). edge_clearance=0 (and an unresolved None) used to turn the
    board-edge keepout off. It now means "no margin beyond not touching the
    cut": the router keeps every trace and via off the outline and interior
    cutouts by a small physical floor (half the widest trace or via plus half a
    grid cell). There is no opt-out. Pads that sit on or very near the outline
    (edge connectors, castellations) now get a thin keepout ring when no clearance
    is configured; set a positive --edge-clearance if you need a specific value.

  • Crashed kicad-cli DRC runs now fail gates that used to pass (Issue #6724).
    These results used to read as clean and now fail. A kct route whose
    post-route kicad-cli DRC crashed or was killed fails the DRC gate even
    without --strict-drc. A route whose first pour-oracle DRC did not
    complete exits 10 (unverified) instead of 0. That includes a first
    oracle DRC that times out (the default per-call budget is 120 s), not only
    a crash. The recipe pipeline gate fails the DRC leg even with
    require_drc=False, which still excuses only a missing kicad-cli. If you
    start seeing these, read the reason: a signal kill (SIGKILL (out of memory?)) usually means too many Docker DRC containers at once, so run
    DRC one board at a time; an "unsupported option" (kicad-cli printed
    Unknown argument: <opt>) means your kicad-cli is too old for an option
    kct passed; exit 3 ("could not load the board") usually means the board
    was saved by a newer KiCad than your kicad-cli, so upgrade KiCad. Any
    other exit status is reported as-is, with kicad-cli's stderr.

  • KiCad 8/9: geometric DRC now actually runs, without zone refill
    (Issue #6724). Before, kct passed KiCad 10's --refill-zones to every
    kicad-cli; KiCad 8/9 rejected it and the rejected run read as a clean
    DRC. Now KiCad 8/9 run the check without --refill-zones, so kct route,
    kct audit, the recipe gate and benchmark metrics may report geometric
    violations that were previously hidden. KiCad 10+ behaves as before
    (zones are refilled before checking).

  • diffpair_length_skew and match_group_length_skew numbers shift (Issue #6725). Via length now follows the board stackup and the layers each via connects on, and pad/via entry geometry is trimmed, so reported skews change by a few hundredths of a millimetre per via (more where a via declared F.Cu/B.Cu only carries tracks on an inner layer). Pairs and groups can newly fail or newly pass; boards routed with an earlier kct may need kct route re-run to be retuned against the new measure.

  • A via connected on only one layer no longer adds length in kct check (Issue #6725). kicad-cli gives a via no height when the tracks ending in it (and any pad overlapping it: a surface pad's own layers, a through-hole pad's footprint side) use only one copper layer. diffpair_length_skew and match_group_length_skew now do the same on the written board, and no longer charge the via's declared span. A pair or group whose skew came from such a stub via can newly pass, or newly fail. The router's tuners still charge the declared span, because one routed fragment of a net may not show the via's other layer.

  • Large boards routed with a fine --grid on the Python backend now use
    that grid
    (Issue #6741). A run that was silently coarsened to ~500k cells
    now allocates the full requested grid (e.g. 6.05M cells for 195 x 155 mm at
    0.1 mm) and is correspondingly slower and larger in memory. Build the C++
    backend (kct build-native), or pass a coarser --grid that still satisfies
    your clearance, if the run is too slow.

  • New kicad-cli DRC failure reasons (Issue #6775). Scripts that match
    GeometricDRCResult.reason should know that exit 3 (board failed to load)
    is now kicad_cli_load_failed. Every other non-zero, non-signal exit is
    now kicad_cli_exit_error. Both used to be kicad_cli_crash, which now
    means only a signal kill or an exception. All three are still failures
    (GeometricDRCResult.failed), so gates behave as before.

  • kct route holds coupled P/N traces to the pair gap, not the class
    clearance
    (Issue #6810). For a coupled class whose pair gap is larger
    than its clearance, the pre-save check can now fail on P/N traces closer
    than the pair gap -- the copper kct check (diffpair_clearance_intra) and
    kicad-cli (diff_pair_gap) already reject.

  • kct route --differential-pairs can spend longer on pairs it cannot
    couple
    (Issue #6835). Without --diffpair-per-pair-timeout, the coupled
    search's only bound is its iteration budget, which is now 1,400,000 per pair
    instead of 40,000 (PCIE_TX, PCIE_RX and LVDS<n> pairs keep 40,000). A pair that
    fails both phases can take on the order of 15-20 s instead of about 0.5 s.
    Set KCT_COUPLED_FLAGOFF_MAX_ITERS=40000 to restore the old cap for every
    pair, or pass --diffpair-per-pair-timeout to bound each pair by wall clock.

  • kct route --grid <value> now refuses grids past a per-backend cell
    ceiling
    (Issue #6875). Since #6741 an explicit grid is never coarsened,
    so a very fine grid on a large board could allocate until the machine ran
    out of memory. An explicit grid needing more than 20,000,000 cells on the
    Python backend, or 100,000,000 on the C++ backend (counted at the largest
    layer count the run would try), now exits 1 before routing with an error
    naming the cell count and suggesting kct build-native, the finest
    --grid that fits, or --grid auto. Pass --force to route at the
    requested pitch anyway. --grid auto is unaffected, and no committed
    board recipe comes near either limit (the largest, board 09's
    --grid 0.05 investigation, is 15.6M cells at 6 layers).

  • MCP clients must not send undeclared arguments (Issue #6914). tools/call
    requests carrying keys that are not in a tool's inputSchema now return a
    validation error; remove the extra keys.

  • kct pcb import-ses can now fail on an edited board (Issue #6915). If
    the board changed after export-dsn, re-export and re-route, or pass
    --allow-stale. Boards with no sidecar still import, with a warning.

  • Evidence producers refuse an unqualified KiCad environment (Issue #6962).
    kct readiness and the scripts/boards/ producers now exit 2 and write no
    evidence when kicad-cli is not the pinned image's release (10.0.6) -- e.g.
    Homebrew KiCad 10.0.7 -- or when native DRC reports any
    lib_footprint_issues finding; both used to be recorded silently. Re-run
    with KCT_KICAD_DOCKER=1 to route kicad-cli through CI's pinned image.
    scripts/ci/init_kicad_libraries.py no longer writes into your KiCad
    config: it populates an isolated config home and exports it (to
    $GITHUB_ENV in CI, or via eval "$(python3 scripts/ci/init_kicad_libraries.py --shell)").

  • Code that called a removed router method must switch to the remaining
    API
    (Issue #6984).

    • ViaRules.standard_6layer(): build the rules by hand with
      ViaRules(through_via=ViaDefinition(ViaType.THROUGH, drill_mm=0.3, annular_ring_mm=0.15, start_layer=0, end_layer=5, name="Through")).
    • Corridor helpers: use the remaining RoutingGrid corridor methods,
      set_corridor_preference, clear_all_corridor_preferences and
      get_corridor_cost. There is no per-net clear any more: clear them all
      and re-apply the corridors you want to keep. If you need to know which
      nets have a corridor, record that yourself when you call
      set_corridor_preference.
    • LayerStack.get_layers_adjacent_to_plane and
      get_layers_with_reference_plane: iterate LayerStack.layers and
      filter on layer_type == LayerType.PLANE, plane_net, is_routable
      and reference_plane.
    • Router.set_via_cache_enabled: the via-validity cache is now always
      on, with no off switch. Router.clear_via_cache() still clears it.
  • Rebuild the native router; re-routes may differ (Issue #6994). The C++
    router's BUILD_VERSION is now 60, so run kct build-native after
    upgrading (a v59 extension is rejected and routing falls back to the slower
    Python path). Native A* search radii shrink by one cell where float noise
    used to inflate them (e.g. Default rules on 0.1 / 0.05 / 0.025 mm grids) and
    rip-up considers one more ring of blocking nets, so re-routing an existing
    board can produce different copper.

  • KiCadPCBExporter.generate(), .write() and .generate_trace_segments()
    are removed
    (Issue #7022). They wrote a whole board from a string template
    with footprints that had no pads, so the result could not be routed, checked
    or manufactured. Create the board from the schematic instead (kct create-pcb or schematic-to-PCB sync), then apply the layout to it with
    layout.apply_to_pcb("board.kicad_pcb", "board_placed.kicad_pcb", net_map={...}). update_pcb_placements keeps its signature, now logs a
    warning naming layout refs that are not on the board, and reports the number
    of footprints it actually moved.

  • LEDBlock places its LED at rotation 180 inside the block (Issue #7022),
    so export_placements() reports the LED footprint's rotation 180 degrees
    different from before. Boards laid out with the old orientation shorted the
    LED's anode to its cathode trace.

  • PCB.export_gerbers(..., create_zip=False) on a preset now returns a
    directory
    (Issue #7023). With no create_zip argument the jlcpcb, pcbway
    and oshpark presets still return a ready-to-upload zip; code that passed
    create_zip=False and relied on the zip it got anyway should drop the
    argument. include_drill=False and layers=[...] passed to a preset now
    take effect.

  • kct bom and kct sch summary|validate|preflight|hierarchy exit non-zero on a missing/unparseable schematic (Issue #7028).
    Scripts that relied on exit 0 for a bad schematic path will now fail; fix the
    path or handle the error.

  • Project.create boards differ on disk (Issue #7047). The outline is now
    four gr_line segments instead of one gr_rect; the Cmts.User,
    Eco1/2.User, Margin and User.1-User.9 layers and the explicit
    stackup are no longer pre-declared; the root board uuid is gone; a
    title_block is added; and the file is tab-indented. Scripts that place
    items on those layers must add the layer first, and scripts that look for a
    gr_rect outline should use PCB.get_board_outline().

  • --batch-routing is deprecated (Issue #7051). Scripts passing it now get a warning on stderr (and a DeprecationWarning); remove the flag, routing output is identical. Autorouter.route_all_negotiated(batch_routing=...) is accepted but ignored. The BatchPathfinder, BatchRouteRequest, BatchRouteResult, batch_heuristic_gpu and compute_batch_costs_gpu import paths from kicad_tools.acceleration.kernels are gone. GPU/batched routing was evaluated in #7053 (docs/research/gpu-routing-design.md) and is not worth building now.

  • kct placement align --axis row --reference left|right (and --axis column --reference top|bottom) now exits non-zero with an error instead of
    silently centering; use row: center/top/bottom, column: center/left/right.
    Reported move counts may be lower than before. Library callers:
    align_components and ROW/COLUMN constraints raise ValueError for
    mismatched references (Issue #7056).

  • Derived HV clearances rise for working voltages above about 50 V
    (Issue #7057). kct creepage --standard, kct audit --hv-standard and the
    kct readiness HV isolation gate now derive clearance from the rated impulse
    voltage, so for example 120 V needs 0.5 mm and 230 V needs 1.5 mm at the
    default overvoltage category II (both were 0.2 mm). Layouts that passed
    before can newly fail. Circuits that are not directly mains-connected, or
    that sit behind transient protection, can pass
    --overvoltage-category I / --hv-overvoltage-category I; fixed-installation
    equipment should pass III or IV. 48 V and below is unchanged.

  • Creepage around slots now reads shorter (Issue #7058). Both former
    over-estimates are fixed, so kct creepage / kct audit HV checks can newly
    fail on boards that previously passed. Slots narrower than dimension X
    (PD2: 1.0 mm, used when no pollution degree is given) no longer add creepage.
    Re-run HV checks and review any newly failing pair.

  • Regenerate previously exported CPL files (Issue #7059). CPL coordinates now use
    the Gerbers' frame (aux origin, Y up), so every row changes, including the sign of Y
    on boards without an aux origin. Remove any manual offsets applied on the assembler
    side to compensate for the old frame.

  • Differential-pair impedance numbers change (Issue #7060). The
    edge-coupled model now reports the physically solved Zdiff, which is lower
    than before for tightly coupled pairs (about 5 % to 20 % at gaps near the
    trace width). Expect:

    • impedance-targeted diff pairs to route at different widths and gaps
      (tight pairs: narrower traces to reach the true target);
    • kct check impedance findings and kct impedance diffpair output to
      change;
    • previously routed impedance-controlled boards to be flagged out of
      tolerance where they were sized with the old, optimistic model.
      Re-run kct impedance diffpair for your geometry and re-size if needed.
  • kct parts import -i/--interactive is deprecated (Issue #7065). It was
    never implemented; it is now ignored with a one-line stderr warning. Remove
    it from scripts.

  • kct impedance CPWG results change (Issue #7075). CPWG Z0 and
    geometries synthesised by cpwg_geometry_for_impedance differ from earlier
    releases, and microstrip/CPWG loss figures are 2-3x lower. Re-run any CPWG
    impedance targets; out-of-range geometry now raises ValueError.

  • Single-ended microstrip impedance, delay and loss numbers change
    (Issue #7076). Re-check anything pinned to the old values:

    • Router widths auto-resolved from target_single_impedance grow about 4-5 %
      (e.g. 50 Ω on Stackup.jlcpcb_4layer() F.Cu: 0.387 → 0.403 mm).
    • The impedance validator reports 2-5 % higher Z0 on existing copper, so a
      tight tolerance can newly fail (e.g. 0.18 mm traces on board 07 go from
      51.72 to 53.64 Ω).
    • εeff falls 5-7 %, so microstrip delay drops 2.5-3.7 % per mm (board 07:
      5.934 → 5.716 ps/mm) and the loss estimate drops too (16.9 → 15.4 dB/m).
      Delay-to-length conversions for length matching change accordingly.
      Diff pairs with an explicit gap use the coupled solver and are unaffected.
  • kct drc can newly exit 1 or 2 on schematic parity (Issue #7077).
    Scripts that gate on its exit code should check what changed:

    • NOT RUN exits 2. A board with no schematic next to it, or one that is
      not fully annotated, now exits 2 (incomplete, as kct check does). So
      does kct drc <report>.json/.rpt when the report has no parity
      findings: an empty schematic_parity array cannot tell "clean" from
      "never ran". Pass --allow-incomplete (same name and meaning as
      kct check --allow-incomplete) to get exit 0 back.
    • Parity findings gate by severity. Error-severity findings exit 1;
      warning-severity findings exit 2 under --strict. Waive reviewed ones in
      .kct_waivers.json.
    • --mfr mode uses the same gate (Issue #7170), and kicad-drc-summary
      keeps its exit code (only its verdict text changes).
  • estimate_pcb / kct estimate now raise ValueError for layer counts above the pricing table's maximum (Issue #7080), instead of silently pricing them as 2-layer. ENIG and other finish totals drop for quantities above 1.

  • kct audit verdict on hosts without kicad-cli (Issue #7083). A board
    whose ERC could not run (kicad-cli not installed, or ERC timed out after
    60 s) or whose schematic <-> PCB sync check crashed now gets
    REVIEW WARNINGS / "verdict": "warning" instead of
    READY FOR MANUFACTURING, so kct audit --strict now exits 2 in those
    cases (without --strict it still exits 0). Install kicad-cli, or pass
    --skip-erc to opt out of ERC explicitly; --skip-erc and boards with no
    schematic are unchanged.

  • kct export now refuses boards whose DRC report lists unconnected items
    (Issue #7091). The connectivity safety floor blocks on error-severity opens
    as well as net shorts, even with --skip-preflight; route the board, or pass
    --skip-drc-floor for a known-safe workaround. A DRC (or ERC) report
    generated for a different PCB (or schematic) no longer counts as a clean
    check: preflight shows WARN and the floor only warns, so re-run DRC on the
    board you are exporting (or pass --drc-report) to get a real verdict.
    Under --strict-preflight such a mismatched report is a hard FAIL: the
    export is blocked and nothing is written until DRC/ERC is re-run for the
    board being exported.

  • kct validate --lvs --min-confidence, filtered kct erc / kct erc explain, and kct sch validate without a working kicad-cli can now exit
    non-zero
    (Issue #7094). These used to exit 0 while real problems were
    hidden or unchecked. --min-confidence and the ERC --errors-only /
    --type / --sheet filters are display-only now, so a hidden error still
    fails the run; use --filter-config ignore rules for deliberate
    suppression. kct sch validate exits 1 with passed: false when ERC did
    not run; pass --skip erc to opt out of ERC explicitly.

  • kct drc can newly fail (Issue #7096). Boards with unrouted nets that
    previously printed DRC PASSED now exit 1. Pass --ignore-unconnected to keep
    the old clearance-only verdict.

  • Re-check boards placed with kct placement optimize since #3812
    (2026-06-20)
    (Issue #7106). On a board whose Edge.Cuts outline does not
    start at (0, 0), kct placement optimize, refine, suggest, snap and the
    MCP placement sessions could save most parts outside the board while
    reporting success. Run kct placement check (or kct check) on any such
    board. If parts lie off the board, restore the pre-optimization file, or
    re-run kct placement optimize with this release.

  • MCP route_net no longer overwrites the input board by default (Issue #7111).
    Omitting output_path now writes <stem>_routed.kicad_pcb next to the input.
    Pass output_path equal to pcb_path to restore the old in-place behaviour.

  • Regenerate .kicad_dru files written by kct init (Issue #7116). They began
    with ; comment lines, which KiCad does not treat as comments: KiCad silently
    ignored the entire file and DRC fell back to board-setup defaults. Re-run
    kct init (or delete the .kicad_dru and re-run) to get a file KiCad loads.

  • create_checkpoint no longer accepts score or drc_violation_count;
    context tools reject unknown session IDs
    (Issue #7121). Both values are
    now taken from the live session. The seven context tools return
    Session not found for any ID that start_session did not issue (or that
    has been committed, rolled back or expired) instead of auto-creating a
    context. Callers that passed those parameters should drop them.

  • kct audit / kct report on a .kicad_pcb now run ERC, and the sign-off
    skills use new gate commands
    (Issue #7136).

    • kct audit and kct report on a .kicad_pcb now run kicad-cli sch erc
      against the resolved schematic. That is slower and needs kicad-cli (a
      missing kicad-cli is ERC NOT RUN, WARNING).
    • An audit of a .kicad_pcb with no schematic beside it is now WARNING
      (ERC NOT RUN) instead of READY, so kct audit --strict exits 2. Pass
      --skip-erc if a PCB-only audit is intended.
    • Agents following /kct:manufacturing-readiness or /kct:tapeout run new
      gate commands: kct check --output <dir>/check-report.json judged by a
      jq rule, kicad-cli pcb drc --refill-zones --severity-error --exit-code-violations --format json into <pcb-dir>/drc_report.json,
      and kct export --drc-report <pcb-dir>/drc_report.json. The gates need
      jq. Never use kct check --allow-incomplete for sign-off.
  • kct validate --sync can newly exit 1 on a value mismatch (Issue #7150).
    Fix the schematic or PCB value so they agree, or run kct sync --apply to
    copy the schematic value onto the PCB.

  • JLCPCB bundles gain paste Gerbers; strict exports check the DRC
    report's manufacturer
    (Issue #7153). kct export --mfr jlcpcb now puts
    <board>-F_Paste.gtp (and <board>-B_Paste.gbp when the board has a
    B.Paste layer) in gerbers/gerbers.zip; scripts that assert an exact
    Gerber file list need updating. --strict-preflight exports now FAIL when
    the auto-discovered or --drc-report DRC report was generated for a
    different manufacturer or capability tier than the run. Re-run
    kct check --mfr <run's mfr> to refresh the report, or export with the
    --mfr the report was built for.

  • kct erc / kct sch validate can newly report isolated_pin_label /
    single_global_label findings on hierarchical designs
    (Issue #7161).
    They are exactly the findings kicad-cli sch erc reports for the same
    schematic, which the removed name-count filter used to hide. They are
    warnings by default (exit 0); a project that raises isolated_pin_label to
    error in its .kicad_pro now fails, as kicad-cli does. Fix the label (or
    waive it with --filter-config, which is counted as Filtered:). The
    Python helpers filter_cross_sheet_global_labels,
    filter_cross_sheet_global_labels_objs and
    filter_cross_sheet_power_violations are removed from kicad_tools.erc.

  • kct readiness can newly fail or NOT RUN the bom gate (Issue #7166).
    The gate fails when the bundle's bom_lcsc_values preflight result is FAIL.
    It reports NOT RUN when the bundle manifest has no such result, for example
    a bundle exported before this release; regenerate the bundle with
    kct readiness --generate. kct export --strict-preflight can newly fail
    on IC package-family or pin-count mismatches between an assigned LCSC part
    and its footprint. bom_lcsc_values now reports WARN instead of OK when an
    assigned part is in neither the parts cache nor the offline catalog. That
    WARN never blocks export, not even under --strict-preflight, and never
    blocks readiness.

  • kct drc --mfr can newly exit 1 or 2 (Issue #7170). It now uses the
    same exit gate as plain kct drc, reversing #7077's separate 0/1 rule for
    --mfr. It exits 1 on any fab-limit breach (including the newly judged
    hole_to_hole, via_diameter and too-large drill_out_of_range) and on
    any error-severity violation kct cannot judge against the fab's limits
    (not checked). It exits 2 on warnings under --strict, and when schematic
    parity did NOT RUN unless --allow-incomplete is passed, so
    kct drc report.json --mfr <fab> on a saved report now needs
    --allow-incomplete to exit 0, as plain kct drc report.json already did.
    Project.check_drc() / check_manufacturer_rules() now also return
    CheckResult.UNKNOWN entries (rule_name == "not_checked"), which are not
    is_compatible. kicad-cli's hole_clearance now parses as
    drill_clearance (it was clearance), so --type/--category filters and
    kct fix-drc see it as a drill finding. --mfr --format json output is new.
    The Python helper filter_phantom_wire_violations is removed from
    kicad_tools.erc.cross_sheet.

  • bom_lcsc_values can newly FAIL (Issue #7180). Wrong-package LCSC numbers assigned in a .kct bom_entries block, and coarse-vs-fine SO pitch mismatches (SOIC/SOP/SO at 1.27 mm against TSSOP/MSOP/VSSOP at 0.65 mm or finer), are now caught by preflight and block the readiness bom gate. LCSC package strings never state pitch, so when the pitch comes only from the family name the check compares the 1.27 mm vs fine class only; fine-vs-fine (0.65 vs 0.5 mm) never fails. Fix the assignment; do not suppress the check.

  • DRC gates that can newly fail (Issue #7188).

    • kicad-drc-summary can newly exit 1: on unrouted boards (error-severity
      unconnected_items), and for any kicad-cli error-severity type the
      severity table did not mark blocking -- e.g. hole_clearance, or
      solder_mask_bridge / silk_* when the project's severities say error
      and --fab does not reclassify them.
    • kct mfr validate can newly exit 1 (clearance and connectivity are now
      checked), and its --format json violations come from DRCChecker:
      each entry keeps type / detail and adds severity and rule_id; a
      new warnings list has the same shape; ok is true exactly when there
      are no errors. Warnings never fail. The "design meets manufacturer
      requirements" line is gone.
    • kct audit / kct readiness: an audit whose kicad-cli run reports
      error-severity opens after the zone refill now fails its DRC row.
      Readiness no longer counts warning-severity opens toward
      native_drc.error_count; unconnected_count still counts every open.
    • kct mfr validate --layers now defaults to auto-detection from the
      board's copper layers (as kct check does) instead of 2; an explicit
      --layers N still overrides. Boards that previously got 2-layer rules
      (boards 03, 05 and 06: 4-layer; board 07: 6-layer) are now judged against their real
      layer count. Keepout zones (pour-fragment cutouts) are no longer
      reported for their placeholder zone clearance of 0.
  • Auto-discovered drc_report.json / erc_report.json must be
    content-bound to count as a pass
    (Issue #7190). A report beside the board
    that records no pcb_sha256/schematic_sha256 (and is not pinned with the
    board in output/readiness.json) now shows as a preflight WARN instead of
    OK/DRC passed; one whose recorded hash no longer matches the file now
    FAILs kct export --strict-preflight. Re-run kct check --output drc_report.json (or kct build) after changing the board, or pass the
    report explicitly with --drc-report/--erc-report.

  • .kicad_dru keepout rules now depend on the kicad-cli release (Issue #7198).
    On KiCad >= 10.0.6, kct stops writing the # BEGIN kct keepout rules block
    and removes an existing one, so committed sidecars change when regenerated.
    Keepout findings on vias, pads, footprints and zone fills now appear only
    under KiCad's native type, not the kct rule name: update any
    .kct_waivers.json entry or --type filter keyed on the old name.
    A sidecar written on 10.0.6 and checked by an older kicad-cli no longer
    enforces keepouts. Regenerate it on that release or upgrade to 10.0.6.

  • Double-sided JLCPCB exports now WARN, and an impossible package pitch
    now FAILs --strict-preflight
    (Issue #7204). Boards with bottom-side
    parts get a bottom_side_in_cpl WARN; this never blocks an export. Parts
    that need JLCPCB Standard PCBA (0201/01005 chips, 0.35 mm IC pitch, BGA
    pitch under 0.5 mm) get an assembly_package_limits WARN under the default
    --assembly-tier auto. Pass --assembly-tier standard to judge against
    Standard instead. A part beyond every tier the fab offers (for example a
    0.25 mm-pitch BGA at JLCPCB) is a FAIL and blocks --strict-preflight.
    The JLCPCB assembly profile values also changed to match the published
    limits: min_bga_pitch_mm is 0.3 (was 0.4) on profile.assembly, which is
    the Standard tier.

  • Boards whose footprints lack courtyards no longer pass kct readiness
    (Issue #7204). kct check now emits a courtyard_missing warning for each
    such footprint, and courtyard_missing is an assembly-affecting rule, so
    kct readiness --assembly needs --ack-warnings courtyard_missing. The
    courtyard_overlap coverage behind the warnings is a blocking unknown,
    which --ack-warnings cannot clear, so readiness and /kct:tapeout refuse
    sign-off until the gap is fixed. Fix: give each footprint an F.CrtYd
    outline (the stock KiCad library one, or pad extent + 0.25 mm), or, for a
    part with no body, set (attr ... allow_missing_courtyard).

  • Protection patterns require wider current-carrying traces (Issue #7206).
    OvercurrentProtection / ReversePolarityProtection minimum widths now
    follow IPC-2221 (3 A: 0.90 -> 1.37 mm; 10 A: 3.0 -> 7.19 mm on 1 oz), so
    existing copper sized by the old rule can violate the pattern's routing
    constraint. Widen the traces, or pass the board's real copper_oz=.

  • A project.kct that declares a current now needs target_ampacity (Issue #7206).
    Add target_ampacity (amperes, continuous) to the net classes that carry it in
    your net-class map (net_class_map.json); otherwise kct check and
    kct readiness report a blocking ampacity: unknown. kct never infers which
    net carries a spec current.

  • Pours on nets with target_ampacity are now measured and can fail (Issue #7206).
    A copper pour whose narrowest neck between the net's pads is under the
    IPC-2221 width for the declared current is an ampacity error. That includes
    a thermal spoke narrower than that width into any one pad or via, whatever
    the length of its gap. Parallel spokes are not added together. Widen the
    spokes or the neck, or connect the pad solidly. The rating is conservative
    about pour copper, but it does not rate via barrels: a pour that hands its
    current to another layer through a small via is not flagged for it. The
    pour reads ampacity: unknown:pour not measurable (<net>: <reason>) when:

    • a zone of the net has no committed fill (refill the zones and save);
    • the net has fewer than two copper pads;
    • a pad's copper outline cannot be modelled;
    • the fills lie on no board copper layer;
    • an internal layer has no copper weight (inner_copper_oz <= 0);
    • the copper does not join all the net's pads;
    • shapely is not installed.
  • New stackup_copper preflight FAIL blocks kct export --strict-preflight
    (Issue #7209). A board whose stackup declares copper the fab profile does not
    offer (e.g. 3 oz inner on JLCPCB) now fails stackup_copper. Default exports
    still write the package and list the failure as a warning. Align the stackup
    with an orderable build.

  • Pour-only rails with a declared current now block sign-off (Issue #7210). A net with a declared target_ampacity that is carried only by a copper pour (no track segments) used to read ampacity: checked; it now reports ampacity: unknown:zone-only, not measured. ampacity is a blocking category, so kct readiness and tapeout refuse to sign off while it is unknown. Route a track that carries the current, or remove the current declaration for that net, to clear it.

  • kct validate --sync can newly fail on footprint lib-ID drift (Issue #7213). A placed footprint whose package or Library: nickname differs from its symbol's Footprint field is now a footprint_mismatch error (exit 1). Fix the PCB (or schematic) so both name the same Library:Footprint. A footprint missing the nickname on one side only is reported as a warning and does not fail the check.
    kct sync uses the same comparison: a placed footprint whose nickname
    differs from the schematic's (e.g. Capacitor_SMD:X vs Resistor_SMD:X)
    now produces an update_footprint action, so review the plan before --apply.

  • kct check reports new warnings on many existing boards (Issue #7220).
    thermal_asymmetry and via_at_pad_edge run by default. On the committed
    fleet they report: 00: 1, 01: 1, 03: 6 (C10, C11, and J1 A1/A12/B1/B12) and
    04: 9 thermal_asymmetry; 02: 2, 05: 3 and 06: 17 via_at_pad_edge. Plain
    kct check exit codes and kct readiness verdicts are unchanged: both are
    warnings in the advisory bucket, and neither is an assembly-affecting
    warning-review rule. A pipeline that fails on any warning (--strict, or
    a script that checks summary.warnings == 0) will now fail. That includes
    board 05's redesign/build.py, which refuses a rebuild until its three
    pad-edge vias are moved, and board 04's reviewed
    check_manufacturing.py, whose DRC leg now reads "0 errors, 9 warnings"
    until the fill repair stops forcing those terminals solid. Fix the copper.
    Use --skip thermal_asymmetry or a .kct_waivers.json entry only after a
    review. Direct ViaInPadRule() callers are unaffected: the edge warning is
    opt-in there (report_pad_edge=True), and kct check turns it on.

  • kct readiness can now block on silkscreen text height, text thickness and
    obscured reference designators
    (Issue #7228). silkscreen_text_height,
    silkscreen_text_thickness and silk_ref_obscured join the
    assembly-affecting warnings, so a board that has any of them needs
    --ack-warnings <rule> or a fix (kct place-silk-refs for obscured
    references). The committed example boards report none of them.

  • Filtered kct drc / kct validate --sync runs can newly fail (Issue #7230).
    kct drc --net X (and --type/--category/--errors-only) can now exit 1,
    and kct validate --sync --errors-only --strict can now exit 2, where they
    previously exited 0. CI that relied on a filtered run passing must filter
    the output itself or fix the findings.

  • Read verdict, not the exit number, across sign-off commands (Issue #7230).
    The two exit ladders stay: kct drc/erc/validate --sync/sch validate
    exit 1 for a board with errors, kct check/audit exit 2 and keep 1 for
    "the tool could not run". A pipeline that compares numbers across commands
    should read the JSON verdict (gate_verdict for kct audit) instead.
    JSON consumers that reject unknown keys must accept the new top-level
    verdict, exit_code and gate_verdict keys.

  • Regenerate JLCPCB packages for boards with through-hole parts (Issue #7239).
    The JLCPCB CPL and BOM content changes for every board that has THT parts
    (headers, connectors, DIP/TO packages): those rows are now assembled by the
    fab. Re-run kct export --mfr jlcpcb / kct readiness, and send the new
    BOM/CPL through procurement review before ordering.
    kct readiness can now block such a package: a THT BOM line with no LCSC
    part number fails the bom gate, as does a THT part still listed in a
    manual-assembly-bom.csv. Assign a JLCPCB-sourceable part or drop it from
    the manual BOM.

  • A declared target_diff_impedance can now fail kct check (Issues #7250, #7293). A coupled segment more than 20 % off its declared differential target is an impedance error. A matched diff pair with no judged segment (unrouted partner, or no run inside the coupling window) now reports impedance: unknown:diff_pair_unmeasurable in the coverage map instead of checked.

  • kct check can now fail on a project.kct mismatch (Issue #7292). A board whose outline differs from
    requirements.mechanical.dimensions, or whose copper-layer count differs from
    requirements.manufacturing.layers.preferred (or exceeds layers.max), now reports Overall: FAILED.
    Fix the spec or the board; run kct check --skip spec_conformance to defer.

  • MCP get_drc_violations and export_assembly can newly fail (Issue #7260).
    get_drc_violations can now return passed: false on boards it used to
    pass: 2-layer boards are checked against 2-layer fab limits, and passed
    is kct check's verdict, so a board whose ERC/LVS cannot run (no
    kicad-cli, no schematic) is verdict: "incomplete" (pass
    allow_incomplete: true for exit_code: 0). Its layers parameter is
    removed (the count comes from the board), and an omitted manufacturer
    resolves like kct check (fab_profile.json > project.kct > jlcpcb).
    export_assembly now runs preflight and refuses on any FAIL by default;
    pass strict_preflight: false to downgrade non-blocking FAILs to warnings.
    Like kct export, it now writes sibling <board>.kicad_pro /
    <board>.kicad_dru next to the source PCB.

  • An open known defect now makes a board not ready (Issue #7261). Board 03
    (USB joystick) records defect #7250 (USB pair 71.6 ohm vs the 90 ohm spec)
    and reads blocked / "Needs work" until that is fixed, so kct fleet status
    reports one fewer ship-ready board. read_readiness() and the gallery now
    reject a ready report that lists a defect, or a known_issues list that
    differs from its hashed known-issues.json evidence. kct readiness carries
    a board's recorded issues forward; pass --resolve-known-issue N to drop one.

  • Re-run kct readiness / kct check --output for reports bound only by readiness
    co-pinning
    (Issue #7277). Such reports now read as unbound (a WARN, never a pass) until
    readiness records report_sources for them or they carry their own pcb_sha256.
    ERC reports of multi-sheet schematics must be regenerated; flat ones are unaffected.

  • Re-running the zone fill can change generated copper (Issue #7295). Pads that were previously forced solid may now carry a per-pad thermal_width/thermal_gap override instead. Re-fill boards on the pinned KiCad toolchain (KCT_KICAD_DOCKER=1) and re-pin readiness evidence; committed boards are unchanged until regenerated.

  • Rebuild the native router extension (Issue #7296): run uv run kct build-native. Routing with a manufacturer set now
    keeps vias at least min_solder_mask_dam_mm (0.1 mm on JLCPCB) off SMD pad copper, so some routes differ;
    pass DesignRules(min_via_pad_edge_gap=0.0) to restore the old behaviour.

  • --length-match-diffpairs is no longer a silent no-op under
    --auto-layers
    (Issue #7377). Runs that pass --differential-pairs --length-match-diffpairs without --layers N / --no-auto-layers, or
    --route-engine lattice --differential-pairs on the auto-layers,
    --adaptive-rules, --auto-mfr-tier or --auto-pcb-size paths, now get
    diff-pair skew tuning: serpentine meanders on the shorter leg, so the routed
    copper differs from earlier releases. This is what --no-auto-layers /
    --layers N runs already did. Drop --length-match-diffpairs to keep the
    old grid output; the lattice auto-tune has no separate off switch.

  • Rule writers no longer lower stricter authored rules (Issue #7385).
    kct mfr apply-rules <x.kicad_pro>, kicad-fab-rules apply and kct init
    on an existing project now keep any authored minimum or netclass value
    that is stricter than the manufacturer profile. Before, these commands
    could be used to relax a project to the profile; they now keep the
    stricter value and list what they kept. To replace a project's rules with
    the profile on purpose, set the project text variable
    KCT_PRESERVE_BOARD_RULES=0. This is the same escape hatch kct route has
    used since #6191. Any relaxation the policy still permits (a
    template-signature Default netclass relaxed to the profile, or anything
    under KCT_PRESERVE_BOARD_RULES=0) is printed and reported as
    relaxed_by_policy in the --format json rule_merge payload. These
    writers still leave rule_severities alone: an authored
    isolated_copper: error or lib_footprint_mismatch: error is never
    downgraded by them (only kct route applies the #6191 non-blocking
    severities). They do write the profile's micro-via floors
    (min_via_annular_width 0.05 mm when the project has none); the
    apply-rules header now names that floor next to the standard-via
    annular ring.

  • known_issues warnings: {rule: count} is rejected (Issue #7389). Record the
    exact findings as {rule, ref, pad} identities (--known-issue N:summary:kind:rule@pad,...) and regenerate the hashed known-issues.json.

  • Schematic block behaviour changes from Issue #7399. Re-check any
    schematic generated with these blocks:

    • ReversePolarityBlock(topology="pfet"), BarrelJackInput and
      BatteryInput now connect the input to the P-FET drain and the output
      to its source, so a regenerated netlist swaps those two pins.
    • ESDProtectionBlock(channels=N) with N >= 2 raises ValueError when N is
      more than the ESD array has (2 for the default USBLC6-2SC6). Before,
      N was quietly capped at 4 and the extra channels got ports on no pin.
    • VoltageDividerSense's y is now the top of R1, as in VoltageDivider.
      Before, it was the R1/R2 junction. A ratio of 1 or less now raises
      ValueError; before, it built a 0-ohm or negative top resistor.
  • --strict-preflight fails on header parts whose pin count is not verifiable
    (Issue #7446). Under strict preflight, a BOM row on a PinHeader_RxC /
    PinSocket_RxC footprint whose LCSC part record is known but states no pin count,
    or states conflicting ones, now fails bom_lcsc_values ("pin count not
    verifiable"). Without --strict-preflight it stays a warning. Parts missing from
    the parts cache and offline catalog still only warn, even under strict. To clear
    the failure, assign a part whose record states its pin count.

  • PCB.save() can now raise ValueError (Issue #7479). Directly assigning Pad.number/type/net_name/uuid/pintype/pinfunction, Footprint.name/reference/value/description/tags/uuid, mutating Footprint.properties, or appending to Footprint.pads on a loaded board never reached the saved file; save() now refuses (and writes nothing) instead of dropping the edit, naming each ref/pad.field. Use update_footprint_reference(), update_footprint_value() or assign_net_to_footprint_pad(). Edits that stay in memory and are never saved are unaffected.

  • USBConnector ESD protection now places Power_Protection:USBLC6-2SC6 (was
    Device:D_TVS); custom esd_tvs_symbol values must use USBLC6-2SC6 pin numbering.
    UCC27211GateDriver's default driver_symbol is now Driver_FET:LM5109AMA.
    The default is a schematic-level stand-in with different pin numbers than the UCC27211
    and emits a UserWarning; pass driver_symbol= for any board that will be fabricated.

Fixed

  • The trace optimizer no longer joins Kelvin sense branches (Issue
    #5398). The router builds a current-sense net as a star of separate
    branches that meet only on the shunt pad, but the post-route optimizer
    checked collisions against other nets only. A chamfered corner or a
    straightened staircase could run one branch over a sibling of the same net
    away from the shunt, which KiCad DRC cannot see. A rewrite that touches
    same-net copper the route did not already touch (a sibling branch or
    another terminal's pad) is now rejected, and that route keeps its routed
    geometry. The log reports Kelvin branch isolation: N route(s) kept pre-optimization geometry. On the board 05 DRV8301 stress fixture this had
    merged ISENSE_A+'s U3.32 branch into the U3.25 branch about 6 mm from the
    R12 shunt pad.

  • kct retune-lengths tunes boards saved by KiCad 10 (Issue #5798). On a
    board whose copper names its net ((net "DQ0"), KiCad 10's format) the
    routing loader filed every track under net 0. The tuner then saw every
    member of every selected group as unrouted, changed nothing, and reported
    the group as unattainable. Board 07's routed artifact is such a board.
    Loaded copper is now matched to its net by name before tuning.

  • Match-group tuning keeps a member's earlier meanders when a later attempt
    fails
    (Issue #5798). When a later attempt found no host segment, or hit
    the group's insert ceiling partway through a member, the tuner reset that
    member to its original route. That discarded meanders that had already
    passed DRC and been counted against the group budget. On board 07,
    SDRAM_ADDRESS's two shortest members lost their meanders this way, so the
    group spread stayed at 52.44 mm and the whole group was rolled back. It now
    improves to 41.69 mm and the meanders are kept (the group is still over its
    5.0 mm limit, so kct retune-lengths still writes nothing for it).

  • kct route no longer reports shorts against input copper it does not
    write
    (Issue #5798). With --preserve-existing, a net the run re-routes
    drops its old input copper from the output. The post-route short check
    still tested routed copper against that dropped copper, so a board that
    kicad-cli passes with 0 errors could exit with ROUTING FAILED: cross-net copper shorts. Board 07 with --length-match-groups reported 19 such
    shorts, all at dropped SDRAM escape-stub vias. The check now uses only the
    input copper the output carries.

  • Match-group tuning spends its insert ceiling on the shortest members
    first
    (Issue #5798). The tuner visited group members in declaration order,
    so on a large group the 16-insert ceiling could run out before the members
    that needed the most length were tried (board 07's SDRAM_ADDRESS BA0/BA1).
    Members are now visited by descending length deficit. The limits themselves
    are unchanged, and results are still reported in declaration order.

  • Match-group tuning lets a member at its insert cap use the group's unused
    ceiling
    (Issue #5798). In groups of more than four members each member
    gets at most 2 meanders, even when the group's 16-insert ceiling is far from
    spent. On board 07's complete route SDRAM_BYTE0 stopped at 5.80 mm (limit
    5.0) with DQ6 at its cap 0.8 mm short and 4 ceiling inserts unused. After
    the first pass, a member that stopped at its cap now gets one more insert
    from the leftover ceiling, largest remaining deficit first. kct retune-lengths --group SDRAM_BYTE0 now reaches 4.89 mm and writes the board
    (kicad-cli 10.0.6: 0 errors, as for the input). An explicit
    max_inserts_per_member is still a hard cap, and a group whose ceiling is
    spent, such as SDRAM_ADDRESS, is unchanged.

  • Demo boards 04-07 carry schematic pad pin types (Issue #6025). The
    board 04, 05, 06 and 07 generators now call annotate_pcb_file_pintypes
    right after writing the unrouted PCB, so every committed unrouted and
    routed .kicad_pcb (and board 07's readiness/checked.kicad_pcb) carries
    (pinfunction ...)/(pintype ...) and kct detect-mistakes can use
    pin-type rail evidence on them. Copper is unchanged; readiness reports
    were re-pinned by each board's producer.

  • The router no longer lays copper across interior Edge.Cuts cuts when no
    edge clearance is set
    (Issue #6043). With edge_clearance=None and no
    known manufacturer, or with edge_clearance=0 / --edge-clearance 0, the
    grid used to block nothing, so a trace could run straight through a board
    window, cutout or round hole. The grid now always gets a physical keepout
    floor in those cases: the widest trace or via half-width on the board plus
    half a diagonal grid cell. It is reinstalled on trial resets, parallel
    workers and both fine-grid retries, and works on the Python and native
    backends. validate_routes now also reports copper overlapping a cut as an
    edge violation when no clearance is set. Positive clearances are
    unchanged.

  • Board 05 MOSFET symbols now match the TO-220 footprint pad names (Issue #6290).
    Q1-Q6 use a project-local board05_custom:Q_NMOS_GDS symbol (G=1, D=2, S=3) so
    schematic pin numbers equal the footprint's pad names; LVS reports 0 pad-name
    mismatches (previously 36) without any relaxation of the comparator.

  • Board 03's committed copper meets its own router net classes (Issue #6394). routing-plan.json and usb_joystick_routed.kicad_pcb are re-derived from a fresh lattice route (kct route --route-engine lattice --differential-pairs --native-zone-clearance, committed net-class map, routed under jlcpcb and verified under jlcpcb-tier1). This replaces the hand-repaired plan whose 13 Power 0.2 mm violations both DRC engines reported. Board 03 now has 0 errors in kct check --mfr jlcpcb-tier1 and in kicad-cli 10.0.6 DRC (--all-track-errors --refill-zones), passes usb_geometry() (common In2.Cu run 4.10 mm, D+/D- 15.72 mm, skew 0, MCU launches 3.80 / 3.65 mm), and its readiness receipt is ready. Its .github/routed-drc-tolerance.yml allowance is removed. The recipe now declares the router's net classes in the routed sidecars itself (router_net_classes=), and route_demo.py stitches, applies the tier-1 sidecars and fills before checking, as the recipe does, so its native DRC no longer reports starved thermals the shipped fill does not have. One warning remains: USB_D+ / USB_D- have 2 / 3 vias (#7093).

  • Board 01's committed readiness evidence is fresh again (Issue #6404).
    boards/release_package.py now runs the package_release.py producers on the
    pinned KiCad toolchain (they bypassed it because they pass custom engines),
    and board 01's package and readiness.json were regenerated so
    read_readiness() validates every board.

  • Lattice diff pairs no longer seal in a routed pad between their own start pads (Issue #6697).
    When a coupled segment starts at two pads of one part that straddle a foreign pad (board 03:
    VBUS U2.5 between USB_D+ U2.4 and USB_D- U2.6), the relocated pair anchor started the coupled
    run right on that pad's only exit. Pour nets used to get a #6401 via site there; since #6599
    VBUS is routed, so U2.5 was stranded (1 kicad-cli unconnected item, negotiation stuck at the
    12-round cap). The router now reserves an escape via site for such a routed, single-layer pad
    on the nearest legal lattice node off its outer edge (log: Pair escape via site (#6697)).
    The pair routes around it and the pad's net drops its via there. Board 03 under the lattice
    protocol (seeds 42 and 43) now converges in 2 rounds, 27/27 nets, with 0 kct check errors and
    0 kicad-cli errors or unconnected items, and both USB_D U2.4->R4.1 and USB_MCU_D stay coupled.

  • Lattice diff-pair ties no longer tie one net of a pair alone (Issue #6697).
    The #6698 same-part far-row tie was all or nothing only when both nets had a
    tie candidate: if just one net had one, that net was tied and its partner
    routed single-ended, attaching P and N by different mechanisms (board 06rf
    USB2_D uncoupled length 8.677 -> 12.951 mm). Now neither net is tied and the
    router logs Diff-pair ties P/N: not used (partner net has no tie candidate).

  • Lattice uncoupled diff-pair fallback legs now route partner-aware
    (Issue #6703). When a pair's coupled search declines, its two single-ended
    legs no longer route as unrelated nets: each leg order is tried, the second
    leg is pulled toward the first leg's corridor, and each foreign pad is kept
    at its own class while the legs are searched (so no class-pad shortfall gets
    a pair net demoted). The partner-aware legs are kept only when they cut the
    flush's summed skew by at least 0.25 mm and the remaining fallback pairs
    still route as well behind them; otherwise the previous legs stay. Board
    07rf TMDS_D0/D1 raw skew drops from 7.213/8.393 mm to 0.642/1.824 mm and all
    three TMDS pairs pass kct check's 0.075 mm skew limit with one meander
    each. Boards with fallback pairs route slower (07rf: about 2x; 06rf: about
    1.35x); at most four fallback pairs per pass are weighed, and the routing
    deadline stops the extra searches.

  • A killed or failed kicad-cli no longer reads as a clean DRC (Issue #6724).
    run_geometric_drc used to pre-create its report and ignore kicad-cli's exit
    status. A kicad-cli killed by the OOM killer left an empty file that parsed
    as 0 violations, so kct route's pour oracle printed 0 -> 0 (converged) on
    boards KiCad calls unconnected. Now a signal kill or an unexpected exit
    status gives reason=kicad_cli_crash, and a missing, empty or truncated
    report gives kicad_cli_no_report. Neither result counts as a clean run,
    and the note names the signal (SIGKILL (out of memory?)). The same fix
    covers kct drc / run_drc, the zone refill/fill-via-DRC paths (a killed
    --save-board restores the board), kct audit's ERC, kct sch validate's
    ERC, netlist export, kct readiness exports and kct export preflight
    (an empty DRC/ERC report is a WARN).

  • Geometric DRC now actually runs on KiCad 8/9 (Issue #6724).
    run_geometric_drc always passed --refill-zones, which only KiCad 10's
    kicad-cli pcb drc has. KiCad 8/9 rejected the command line (exit 1, no
    report), and that read as a clean DRC, so the geometric DRC never ran there.
    The flag is now passed only when the installed kicad-cli supports it (a
    memoised --help probe), in run_geometric_drc, run_drc / kct drc
    and the connectivity refill cross-check. A command line kicad-cli does
    reject is reported as reason=kicad_cli_unsupported_option, naming the
    option and saying a newer KiCad is needed, instead of as a crash.

  • The --refill-zones capability probe only caches a definitive answer
    (Issue #6724). A killed, failed or timed-out kicad-cli pcb drc --help
    (30 s cap) is no longer remembered as "no refill support"; the geometric
    DRC then passes --refill-zones and lets kicad-cli's own rejection
    decide, so a KiCad 10 run never silently skips the zone refill.

  • kct check and the skew tuners measure net length exactly as kicad-cli does (Issue #6725). diffpair_length_skew, match_group_length_skew and the router's pair (#6664) and match-group (#6675) tuners now share one function, router/kicad_length.py. A via is charged KiCad's stackup height between the layers it actually connects (outer copper full, inner end copper half, solder mask excluded; from the board's (setup (stackup)), else KiCad's default for (general (thickness)) and the copper layer count) instead of the manufacturer profile's full board thickness. Track chains entering a pad or via are clipped at the copper boundary and the in-copper part is replaced by a straight line to the centre. On the test fixtures, per-net lengths agree with kicad-cli to about a micron, so the tuners now measure skew the way kicad-cli measures it. board_thickness_mm and blind_buried_supported no longer enter the measurement.

  • Net length now matches kicad-cli on track junctions, pass-through pads and off-centre via ends (Issue #6725). A bench comparison against kicad-cli 10.0.6 on boards 03, 06b, 06rf and 07rf found up to 0.57 mm of disagreement on a single net that the first kicad-length model missed. kct now groups track segments into unbranched runs the way kicad-cli does, so a track that passes through a pad centre is no longer clipped at that pad. A run that ends anywhere inside a via's copper is now clipped at the via and run straight to its centre. A via's span now comes from the layers where a track ends inside it, or where a pad overlaps it. A surface pad adds its own layers, and a through-hole pad adds only its footprint's side. A track that only passes through the via adds nothing. Where a via sits inside a pad that a run ends in, the via trim comes first and the in-via stretch is no longer removed twice. A via connected on only one layer gets zero height on the written board, as in kicad-cli. Every diff-pair net on the four bench boards now agrees with kicad-cli to within 0.1 µm.

  • Board 07 ADDR_BUS skew brought inside 0.5 mm (Issue #6730). The committed
    board 07 regression fixture was re-tuned in place with kct retune-lengths
    (ADDR_BUS 1.0998 -> 0.0169 mm; kicad-cli agrees) and the relaxed test pins
    were tightened. Board 03 USB_D is unchanged: its 1.32 mm skew is inside the
    2.5 mm gate, and reaching 0.5 mm would break the usb_geometry() bounds.

  • Board 07 pour repair holds its copper off foreign pads at the net-class
    clearance
    (Issue #6736). The step-7 pour-connectivity repair checked its
    offset vias, pad stubs and bridge legs against a flat 0.15 mm and measured
    foreign pads by their circumscribed circle, so a 45-degree +1V2 bridge leg
    shipped 0.1913 mm from U1.24 (GND) against KiCad's 0.2 mm Power rule. It now
    resolves max(0.15, class(P), class(N)) per neighbour net and measures pads
    with their true shape through the shared clearance kernel. Board 07's
    kicad-cli clearance count is back to 0.

  • kct route no longer crashes on Windows (Issue #6739). The
    KCT_RSS_TRACE peak-RSS tracer imported the Unix-only resource module
    at load time, so every kct route run on Windows failed with
    ModuleNotFoundError: No module named 'resource'. The import is now
    optional; without it the tracer reports 0 bytes instead of crashing. The
    native-DRC failure advice also no longer assumes signal.SIGKILL exists.

  • kct route no longer mistakes (pad inside a footprint description for a
    pad
    (Issue #6740). The router's pad-block scan now skips quoted strings, so
    a footprint such as (descr "Heatsink pins (pad 4, net HS) are 25.4 mm apart.") loads with only its real pads instead of failing with
    "Unsupported routing geometry 'HS' for pad J1.4,".

  • kct route --grid is no longer silently coarsened on large boards
    (Issue #6741). The Autorouter's internal adaptive cell cap (Issue #972)
    replaced the requested pitch with a coarser one once a board exceeded
    ~500k cells on the Python backend, without a message and after the
    Grid resolution ... exceeds clearance check had passed the requested
    value -- a 195 x 155 mm board at --grid 0.1 routed at ~0.35 mm
    (Grid: 561x446x2 = 500,412 cells). Boards loaded through kct route
    (load_pcb_for_routing) now route on exactly the pitch that was checked,
    explicit or auto-selected, matching --max-cells's documented "no effect
    when --grid is an explicit value". Direct Autorouter(...) library callers
    keep the adaptive cap, which now logs a warning naming both pitches and
    can be disabled with adaptive_grid=False.

  • Killed kicad-cli no longer reads as a successful render/SVG export (Issue #6742). run_pcb_export_svg, run_pcb_render, the MCP PCB/schematic screenshot exports and the report schematic figures now fail on any non-zero exit status (including signal kills such as OOM -9) even when a truncated file was left behind. The export safety floor also warns when the DRC report is zero bytes instead of silently treating it as clean.

  • Board 06 pour repair holds its copper off foreign copper at the net-class
    clearance
    (Issue #6748). The step-7 pour-connectivity repair checked its
    offset vias, pad stubs and bridge legs against a flat 0.15 mm and measured
    foreign pads by their circumscribed circle, so repair copper on a
    Power-class pour net (GND, +3V3, +1V8, +1V2, VBUS_USB) shipped
    inside the 0.2 mm Power clearance (three sites at 0.162-0.193 mm on a fresh
    regen). It now resolves max(0.15, class(P), class(N)) per neighbour net,
    measures pads with their true shape through the shared clearance kernel,
    widens the segment prefilter margin to the largest pair clearance, and
    accepts an off-45-degree chord only when both dogleg variants step 12's
    quantizer may ship also clear. Port of board 07's #6736 fix.

  • Fleet-router comparability gate no longer counts fab-only DRU rules as DSN losses (Issue #6756). Silk, text, annular, edge and hole rules are recorded under not_routing_rules, and a class-scoped clearance is covered by its emitted DSN class block.

  • A wide net no longer seals its fine-pitch neighbours' exits (Issue #6761).
    On a pad column whose pitch is narrower than a via of its own nets (board
    05's U2: 0.65 mm pitch, 0.6/0.8 mm vias), a connection that starts or ends
    on that column now keeps its trace out of every other-net pad lane of the
    column and drops no via into one, so it leaves the column straight (or
    changes layer in its own lane) before it turns. Before, PHASE_C's 45-degree
    F.Cu diagonal and 0.8 mm via across U2.10/U2.11 left one exit for two pads,
    and GND_SENSE U2.10 or VM U2.11 stranded. The fences are hard but never
    lose a connection: a search they make impossible is retried once without
    them. Board 05 (grid, 4 layers, seed 42 and 7) routes 32/33 signal nets
    instead of 31/33, with U2.9..U2.12 all connected and no kicad-cli DRC
    violations. KCT_DISABLE_ESCAPE_FENCES=1 restores the previous search.
    Native backend build version 57.

  • Board 06 routed .kicad_pro now declares the recipe's net classes (Issue #6763). The
    diffpair-test recipe passes build_net_class_map() to write_drc_constraints after routing
    (as kct route does), so kicad-cli pcb drc enforces the Power 0.2 mm clearance the
    router used.

  • A rejected --refill-zones is retried, and a skipped refill is a WARNING
    (Issue #6775). When the kicad-cli pcb drc --help probe was killed or
    timed out, kct passed --refill-zones anyway, and KiCad 8/9 rejected it
    with Unknown argument: --refill-zones. The DRC then failed as
    kicad_cli_unsupported_option. Now the DRC is re-run once without the
    flag, and that answer is remembered for the process. Whenever a requested
    refill is not passed, kct logs a WARNING that zone fills may be stale;
    before, this was logged only at INFO. One shared run_kicad_drc_command()
    now drives run_geometric_drc, run_drc / kct drc and the
    connectivity refill cross-check, so kct drc and the connectivity check
    no longer drop the flag when the probe's answer is unknown.

  • kicad-cli DRC failures are classified by cause (Issue #6775). Exit 3
    is reason=kicad_cli_load_failed ("could not load the board", possibly a
    newer KiCad file format). Any other non-zero exit, including a plain
    exit 1 such as cannot open display, is reason=kicad_cli_exit_error, and
    the note quotes kicad-cli's stderr. kct route prints separate advice
    for each cause: upgrade KiCad for an unsupported option or a load failure,
    check memory only after a SIGKILL, and otherwise read kicad-cli's own
    output. kct drc also names a rejected option instead of reporting a bare
    "status 1".

  • The skew tuners now measure each net in the order kct check and kicad-cli read it (Issue #6778). Where three or more track ends meet, kicad-cli's line merge depends on the order the tracks are listed in the file. The router's pair (#6664) and match-group (#6675) tuners measured in route order, while the routed file's order came from content-derived UUIDs (#6052), which is effectively random per net. On 07rf the group tuner logged DDR_DATA_BYTE_0 at 0.000 mm while kct check failed it at 0.284 mm: one 0.2 x 0.2 mm stub inside a via on DM0. kct route still gives router copper content-derived UUIDs, but each net now takes its UUIDs in a fixed geometric order, so the net appears in that order in the routed file and after any KiCad save. The tuners sort a route the same way and measure its coordinates as written (4 decimals). Routed boards get the same UUIDs as before, assigned to different tracks, and their copper order changes; the copper itself does not.

  • Length tuners measure vias in back-side through-hole pads like kct check
    (Issue #6780). Router pads now carry their footprint's copper side
    (Pad.side), so the per-Route length model applies kicad-cli's
    footprint-side rule for through-hole pads, and a via inside a through-hole
    pad is measured as a stub in both forms. Previously the pair and group
    tuners charged such a via the full F.Cu-B.Cu height, and could converge on
    a skew target that kct check and kicad-cli reject.

  • DSN export carries min_copper_edge_clearance; fleet audit recognises implied diff_pair_gap (Issue #6784). The exported (boundary ...) is inset by the project's copper-to-edge clearance, and pads already in the band are listed as omissions. dsn_audit counts a diff_pair_gap no greater than the pair nets' class clearance as carried, and records Native zone clearance / SMD Pad Clearance as not-routing.

  • Diff-pair skew tuner no longer overshoots or gives up on the first blocked placement (Issue #6788). tune_diff_pair_skew sizes multi-loop trombones exactly (a 5.3 mm skew used to add 6 mm and overshoot), never counts a zero-length serpentine as an insert, and on a clearance rejection retries a centred meander, the reverse side, a smaller amplitude and up to 3 ranked hosts before rolling back.

  • Diff-pair and match-group skew tuners no longer double-count a shared via (Issue #6792).
    The tuners measure a net held as several Route fragments by merging them,
    and every fragment that met at a via carried it, so its span was charged once
    per copy. Route-form length now counts each via once: two fragments on
    In1.Cu/In2.Cu sharing a via inside a through-hole pad (1.6 mm, 4 layers)
    measure 9.0475 mm, the whole-net kicad-cli value, instead of 10.095 mm
    (+1.05 mm), and the tuners converge on the length kct check reports.

  • Diff-pair skew tuner no longer trades a skew error for an uncoupled-length error (Issue #6795).
    On the lattice checker-limit path a meander that would push the pair over its
    max_uncoupled_total_mm limit is rejected; overshoot now has its own count in the
    kct route Summary: lines, failure messages are printed, and an overshoot revert reports
    inserts_applied == 0.

  • Kelvin nets tie neighbouring duplicate pins before the star (Issue
    #6799). Two same-net pins of one package with no room for a trace between
    them (board 05's DRV8313 PGND pins U2.6 and U2.7, 0.65 mm pitch) are now
    joined pad-to-pad and routed to the shunt as one branch. Before, each pin
    got its own isolated branch, U2.7's branch sealed U2.6's exit, and U2.6
    stayed unconnected. Resistor pads and pins with room for a trace between
    them keep their own branch. Board 05 now routes 33/33 signal nets.
    The rule looks only at geometry, so an adjacent force pin and sense input
    of one package on the same sense-named net are tied too; give such a
    sense input its own net or a net tie at the shunt.

  • Fleet-router dsn_audit no longer counts class-scoped clearances on empty classes as losses (Issue #6800). A clearance rule whose named net classes have no member nets under the frozen project is now reported under not_routing, since kicad-cli applies it to no net.

  • kct route no longer fails its pre-save clearance check on coupled
    diff-pair copper laid at its own pair gap
    (Issue #6810). The router's
    self-check held the P and N traces of a coupled pair to the net-class
    clearance instead of the pair gap, so board 06rf (class MIPI: clearance
    0.15 mm, pair gap 0.1016 mm) exited 3 on 53 false segment violations while
    kct check and kicad-cli both passed. P-to-N traces of a pair the router
    coupled-routes are now held to the same floored pair gap the exported
    diff_pair_gap rule and kct check use. Pads, vias, other nets and
    +/- pairs the router does not couple (e.g. Kelvin ISENSE+/ISENSE-)
    keep the full clearance, and P/N copper below the gap is still reported.

  • kct fleet status no longer reports false "0 nets in PCB" schematic drift (Issue #6815). The drift check now reads pad nets through the schema parser shared with kct validate --sync, so KiCad 10 name-only (net "VCC") boards (02, 04-07) pass. Boards that declare their PCB under project.kct artifacts (e.g. 09) are now discovered, and boards with no output (e.g. 08) are listed in a skipped footer and JSON array with a reason.

  • A hand-written net-class-map length_constraint without net_id no
    longer crashes kct check
    (Issue #6820). Every net-class-map consumer
    (kct check, kct route, kct creepage, kct zones, kct audit) now
    reports a clean error naming the map file and the entry instead of a
    KeyError traceback. Also, routing_policy now checks the via cap of a
    half-routed differential pair itself; it was handed to
    diffpair_uncoupled_length, which skips pairs without track on both nets,
    so the cap was silently never evaluated.

  • Mesh engine and parallel routing no longer cross interior Edge.Cuts
    cutouts
    (Issue #6822). kct route --route-engine mesh built its navmesh
    from the outline's bounding box only, so board windows, cutouts and round
    holes were never holes and a net could be routed straight across one, at any
    edge clearance. Each interior Edge.Cuts loop is now a navmesh hole, and every
    mesh leg and via keeps the edge clearance from it, or the #6043 physical
    floor when no clearance is set. The thread-safe grid that parallel routing
    rebuilds now reinstalls the board-edge keepout, so parallel routing can no
    longer cross the board edge or a cutout. Grid-engine output is unchanged.

  • Stitch pad-to-via stubs are always 45-degree-legal (Issue #6823).
    run_stitch -- behind kct stitch and kct route's pour-link closer --
    joined a pad to a via site a few hundredths of a mm off its axis with one
    straight stub 1-2 degrees off the 0/45/90/135 set, which
    routing_plan.save_plan() rejects (board 03's VCC and GND welds). Such a
    site is now reached through a dogleg whose legs are clearance-checked
    (both dogleg variants are tried), the extended-escape polar fallback
    doglegs its 22.5-degree bearings, and the shared writer
    (add_trace_to_pcb / trace_to_track_segments) doglegs any remaining
    off-angle leg.

  • --matrix layer assignment no longer overwrites an authored
    preferred_layers
    (Issue #6824). A net class that names its own preferred
    layers keeps them; the automatic matrix hint only fills classes with none.

  • Board 03 route harnesses and the determinism smoke now load the committed net-class map (Issue #6829).
    The research route harnesses and board_route_determinism_smoke.sh pass the board's declared
    --net-class-map instead of silently falling back to the generic 0.5 mm skew default; a declared
    map that is missing on disk now raises FileNotFoundError.

  • Lattice fan-via gateway follow-ups (Issue #6834). A discarded tap-chain retry or rejected pad-growth attempt no longer leaves its gateway diagnostics in the route log, and a chained pair's gateway via limit now counts vias already used on the pair's other segments.

  • Lattice diff pairs couple under a dual-row SMD connector's facing row
    (Issue #6836). When a pair's matching pins sit behind the opposite-polarity
    row of an SMD USB-C receptacle, the fan-via gateway retry now also tries
    "pass-under" combinations: the coupled body reaches the gateway vias from
    the pad side, under both rows, on a layer those pads do not own. They are
    tried only after every existing gateway combination has declined, so pairs
    that already couple keep their copper. On bench board 06rf, USB2_D now
    couples on B.Cu instead of falling back to uncoupled legs.

  • Lattice diff pairs retry a centerline turn that sits inside a partner pad row (Issue #6839).
    When the emitted offset leg broke the partner-pad clearance floor, the pair used to fall back
    to uncoupled with no retry. The search is now re-run (bounded) without the stubs through the
    offending turn; board 06b LVDS1 now couples 2/2 on seeds 42 and 43.

  • Coupled diff-pair search keeps clear of the partner's vias (Issue #6840).
    The joint P/N search measured a partner via as an ordinary trace cell at
    trace-to-trace spacing on one layer, so a rail could run past or into its
    partner's barrel; the commit gate then rejected the whole coupled route
    (Board07 MIPI_DAT0/1 and TMDS_D0/1/2). Each rail now keeps the trace-to-via
    clearance from every partner via on every layer, and a layer change keeps
    both new barrels clear of the partner's copper. The rejection log no longer
    prints worst=+nanmm under the wrong check name: it reports the real
    overlap, e.g. N via vs P segment on B.Cu, gap -0.0315mm. kct route also
    no longer lets the trace optimizer reshape a committed coupled pair, which
    had moved 06rf MIPI_D0's N rail onto another row (20.1 mm uncoupled), and
    a coupled via drop at the pair's own SMD pads now honours the
    manufacturer's via-in-pad capability (it used to bypass it, e.g. two
    via_in_pad errors on 06rf MIPI_CLK under --mfr jlcpcb).
    Native router build version 59.

  • Polarity-swap via move honours via-in-pad and tracks each rail's via
    (Issue #6847). CoupledPathfinder(allow_swap_via=True) (library-only; kct route keeps it off) could still drill a swap via into an SMD pad under a
    manufacturer without a via-in-pad process, e.g. --mfr jlcpcb; it now takes
    the same gate as the plain coupled layer change. The search also filed each
    rail's swap via at the partner's cell, so the #6840 partner-via clearance
    measured each rail against its own barrel; vias are now recorded where each
    rail drills them.

  • Coupled diff-pair routing no longer puts signal on reserved plane
    layers
    (Issue #6857). The joint P/N search (C++ and Python) now honours
    DesignRules.allowed_layers (--reserve-plane-layers, on by default) and
    per-net hard avoid_layers (ampacity classes, --strict-layers), as the
    single-ended router already did. Board07's MIPI_DAT0 at a 700k coupled
    budget previously committed 4 P and 6 N segments on the In1.Cu GND plane;
    it now commits coupled on F.Cu/B.Cu.

  • Coupled diff-pair routing keeps the better of the joint-search and pose-centreline candidates (Issue #6858). A successful joint search no longer displaces a better pose-centreline trunk (06rf MIPI_CLK/MIPI_D0 at a raised coupled budget); both candidates pass the same copper gates and the lower uncoupled length, then skew, then via count wins, with ties kept on the joint result. A [coupled-select] log line records each comparison.

  • Coupled diff-pair search declines a pair whose SMD pad sits on a
    forbidden layer
    (Issue #6860). When allowed_layers or a net class's
    hard avoid_layers (ampacity classes, --strict-layers) forbid the layer
    of an SMD start or end pad, or leave no layer at all, the coupled search
    now declines the pair the way the single-ended router declines such a net,
    instead of routing it on the forbidden pad layer. The pose-centreline
    trunk is skipped for such a pair and itself refuses a forbidden pad layer.
    The pair takes the usual declined-pair fallback, and a [coupled-declined]
    line gives the reason.
    Default settings are unaffected.

  • Diff-pair ties no longer put copper on reserved plane layers (Issue
    #6866). The lattice tie step that joins a connector's equivalent pins
    (#6698) now plans ties only on layers the pair may use: those in
    DesignRules.allowed_layers (--reserve-plane-layers, on by default) and
    outside either net's hard avoid_layers (ampacity classes,
    --strict-layers). It shares one filter with the coupled search (#6857).
    When no permitted layer fits, the pair keeps its usual connections and the
    log says has no permitted tie layer. On board 03 the 1.70 mm USB_D- tie
    moves from the In1.Cu GND plane to B.Cu.

  • Footprint text such as (footprint X) or (zone ...) in a quoted
    descr or tags no longer breaks routing or zone removal
    (Issue #6873).
    kct route split boards into footprints with a regex, so a descr or
    tags string mentioning (footprint ...) or (module ...) crashed
    load_pcb_for_routing with zip() argument 2 is shorter than argument 1.
    It now slices each footprint at the parser's own offsets. Auto-pour zone
    removal, the zone-fill presence checks, the post-route connectivity check,
    the placement fixer and the .kicad_mod pad reader also skip tokens inside
    quoted strings, honouring escaped quotes. Before, a string such as
    "pour (zone keepout, see note" left every real zone after it in place.

  • Routing paths report the grid pitch they actually used (Issue #6875).
    kct route --format json gains a grid object (requested_mm, used_mm,
    cols/rows/layers/cells, selection explicit-or-auto, coarsened
    and coarsened_reason), and the text Grid: line now names the pitch and
    how it was chosen (Grid: 0.1mm pitch (explicit), 251x201x2 = 100,902 cells). The routing paths that keep the #972 adaptive cell cap no longer
    coarsen with only a log warning: route_pcb returns the grid in
    stats["grid"], AdaptiveAutorouter in RoutingResult.grid /
    grid_warning (copied into kct route-auto's per-net warnings), and the
    MCP route_net tool in a new grid field plus a warnings entry. Both
    route_pcb and AdaptiveAutorouter accept adaptive_grid=False to route
    on exactly the requested pitch.

  • kct export --stackup-id is accepted again (Issue #6881). The unified
    kct parser never declared the flag that docs/guides/stackup-presets.md
    documents, so kct export board.kicad_pcb --stackup-id JLC04161H-3313 was
    rejected before reaching the stackup verification step. The flag is now
    declared and forwarded, with its choices and help read from the same
    constant as the standalone export parser. --no-spec, --no-merge-lcsc,
    --allow-unbuildable-bom and --bom-source were missing the same way and
    are now accepted too. A parity test fails if the two parsers drift again.

  • Lattice diff pairs honour a non-pad preferred_layers entry (Issue #6883). A coupled pair whose class prefers a legal layer none of its pads share (e.g. an allow_plane_layers inner layer) now tries its gateway before a planar pad-layer run commits; a declined gateway falls back to the planar run as before.

  • kct optimize-traces no longer joins Kelvin sense branches (Issue
    #6889). The #5398 gate covered only the optimizer inside kct route;
    kct optimize-traces rewrote an already-routed board with no Kelvin check,
    so a chamfer could still run one ISENSE branch over a sibling, or onto a
    sibling terminal's pad, away from the shunt. It now reads the board's
    Kelvin nets from its pads, optimizes each branch on its own and keeps any
    branch whose rewrite would touch another, printing the same Kelvin branch isolation line. Both paths also reject a rewrite that pushes contact the
    router already made further from the shunt pad. Boards with no sense-named
    net are processed exactly as before.

  • kct runs on Windows consoles (Issue #6894). Help output no longer
    crashes with UnicodeEncodeError under cp1252 stdout, and kct interactive
    works without readline. A new advisory Windows Smoke CI job imports
    every router/cli module on windows-latest.

  • kct check --only/--skip help lists every valid category (Issue #6896). The top-level kct check --help previously named 4 of ~39; it is now generated from CHECK_CATEGORIES.

  • kct drc accepts the run-DRC options kicad-drc has (Issue #6898).
    --strict, --category/-c, --verbose/-v, --suggest/-s, --filter-config,
    --keep-report, --output/-o and the -t/-m/-l aliases are now accepted and
    forwarded, so kct drc board.kicad_pcb --strict exits 2 on warnings. The
    standalone kicad-drc --rules / --compare print a deprecation pointer to
    kct mfr rules / kct mfr compare.

  • Board bundles no longer ship broken relative markdown links (Issue #6899). A shared helper, scripts/boards/_bundle_links.py, rebases links in markdown copied into output/manufacturing/ when the target is in the bundle and demotes them to plain text when it is not. The board 03/04 design-source/README.md copies are fixed, the board 07 procurement-review.md link now points at the shipped firmware/source/README.md, and a test walks every bundle for broken links.

  • MCP tools no longer report success when nothing was checked (Issue #6913).
    validate_assembly_bom reports assembly_ready: false for an empty BOM,
    validate_pattern fails a decoupling pattern with no capacitors, and
    export_gerbers / export_assembly fail when no copper Gerber was produced
    and export_gerbers no longer lists stale files from earlier runs.

  • Lattice route log keeps a gateway-first decline reason (Issue #6932). A
    diff pair whose gateway attempt declined and then routed planar now prints its
    Diff-pair gateway (lattice) ... gateway declined (...) line instead of
    suppressing it.

  • Fleet-router benchmark drift check scoped to the selected boards (Issue #6941).
    benchmarks/external/fleet_routers/run.py run --boards ... now only fails the
    frozen-manifest drift gate for the boards it was asked to run; drift on
    unselected boards no longer blocks the run. verify still checks every board.

  • Board 03's committed readiness evidence is current again (Issue #6945). readiness.json
    and readiness-verification/ were regenerated with kct readiness --verify after #6876
    changed net_class_map.json and the routed .kicad_dru, and the bundled kicad_project.zip
    was rebuilt. A new test checks every board's recorded input hashes against the tree.

  • kct drc/check/audit/reason/repair-clearance --mfr accept the advertised aliases (Issue #6946). The standalone parsers now take the same alias set as the unified kct parser (jlc, osh, seeedstudio, ...) and normalize it to the canonical manufacturer ID, instead of failing with invalid choice.

  • kct route --grid ceiling error gives only advice that works (Issue #6948).
    It no longer suggests building the C++ backend when the grid is over the C++
    limit as well. It also no longer suggests a coarser --grid that the
    grid-coarser-than-clearance check would then refuse.

  • Lattice diff pairs can take their gateway through their own reservations
    (Issue #6956). A pour-pad via site (#6401) inside a declared pair's gateway
    envelope at the pair's own footprint, and every pair escape via site the
    pair causes (#6697), now exempt the pair's two nets and still block every
    other net. Board 03's USB_D[1] (U2.4 -> R4.1) now couples through In2.Cu,
    and usb_geometry() passes the common run, length and skew checks. A
    gateway decline line now counts every combination's reason per group and
    names the pad-attach blockers, not just the last combination's.

  • Lattice router no longer drills a route's own vias under the hole-to-hole
    floor
    (Issue #6957). The lattice A* checked a new via only against
    committed copper, so one path could hop F.Cu -> B.Cu -> F.Cu with its two
    vias 0.566 mm apart (board 03 VBUS: 0.266 mm hole gap against 0.5 mm). The
    search now rejects a via closer than via_drill + min_hole_to_hole to any
    earlier via on its own path (the grid engines' #6464 / #6473 check), and the
    per-net self-DRC gate declines such a route instead of shipping it.

  • Lattice diff-pair search no longer declines pair-offset-degenerate on an end overshoot (Issue #6958). A fat centreline that overshoots the goal anchor and doubles back along its own last (or first) edge is now trimmed to the anchor before offsetting; mid-path reversals still decline.

  • Lattice coupled-pair legs end at their pad's copper, not the pad centre (Issue #6959). Board 03's USB_MCU_D+/USB_MCU_D- launches drop from 4.328 mm to 3.804 / 3.649 mm (with #6977's pair gateway reservation exemptions), under the 4.0 mm usb_geometry() bound. Only coupled-pair copper changes; gateway via-pads are not clipped.

  • Diff-pair shadow constructor pre-screens partner vias at the real trace/via floor (Issue #6963).
    The opt-in coupled shadow constructor (enable_shadow_construction, e.g.
    board 06's KCT_BOARD06_SHADOW=1) sized its via-vs-partner keep-outs,
    escape-channel seal penalty and lateral shadow-via offset with
    trace_clearance, while the exact gate behind them enforces
    max(trace_clearance, via_clearance). When via_clearance is the larger
    value, candidates the pre-screen admitted were rejected later, sometimes
    declining a whole shadow side. All four sites now share one helper with the
    gate's floor. The default routing path is unchanged.

  • kct route --length-match-groups runs on the default layer-escalation
    path
    (Issue #6968). The default --auto-layers flow, and the rule-relaxation
    and combined escalation flows, now run the match-group tuner. Before this fix
    only --no-auto-layers runs reached it, and the flag did nothing silently.
    When every detected group's net class has length_critical: false, the run
    now says so on stderr, even under --quiet, and still prints each group's
    measured skew. Under --preserve-existing, the tuner counts a member's
    preserved copper that is written to the output (for example input escape
    stubs) toward that net's length. That copper is never edited.

  • kct route --preserve-existing keeps net ownership of KiCad-10 name-only copper (Issue #6970). Existing segments and vias written as (net "NAME") with no numeric header were loaded under net 0, foreign to every real net; they now load under their real net id.

  • Lattice diff-pair search no longer overshoots an end anchor and returns along the end stub (Issue #6975). The trim from #6958 turned those centrelines into guaranteed pair-polarity-twist declines, because the trimmed end runs the opposite way to the one its stub was picked for. The parity-filtered pair search now rejects a goal arrival that doubles back onto its stub, a start node that leaves back along its stub, and a goal arrival that loops through its own goal node. On boards 03, 06rf and 07rf no centreline needs trimming any more, and the 34 trim-induced twists are gone. Board 06rf's PCIE_RX now couples on its pad layer instead of through a B.Cu gateway, which removes two vias per net.

  • Match-group meanders are checked pairwise against each neighbour's net class (Issue #6979).
    The tuner used to clamp a group's relaxed clearance up to the strictest class on the whole board,
    so one 0.2 mm net anywhere cancelled the relaxation. Each neighbouring track, via and pad is now
    held to max(group floor, that net's class clearance), pads on unrouted nets included. Net-0 and
    unclassified copper is held to the Default class clearance (or the board clearance rule when no
    Default class is declared), so a meander can no longer sit 0.17 mm from a no-net pad under a
    0.2 mm Default class. The diff-pair partner stays held to the intra-pair gap.

  • kct route runs pour auto-skip, auto-pour and the plane-via reservation on
    KiCad-10 name-only boards
    (Issue #6989). The pour stage read nets only
    from (net N "NAME"), so on a board written as (net "NAME") with no net
    table it found no nets. Zoned pour nets were not auto-skipped, missing pours
    were neither created nor reported, and the #6401 plane-via reservation was
    skipped. Routed copper then differed from the same board in the numbered
    dialect. The stage now resolves nets with the routing loader's mapping, new
    pours on such a board are written as (net "NAME"), and the post-fill check
    no longer warns "Net format corruption detected" for a board that is
    name-only by design. Output on numbered boards is unchanged.

  • A routing-cache hit no longer returns different copper than the fresh
    route
    (Issue #7003). The cache dropped which differential pairs the
    coupled pre-pass had committed, so the trace optimizer rewrote pair rails on
    a hit that it leaves untouched on a fresh route (board 06b: 284 vs 242
    items). The set is now stored in the entry and replayed; CACHE_VERSION is
    bumped so older entries are re-routed once.

  • kct sch wire commands agree on endpoint coincidence (Issue #7009).
    remove-wire, remove-component and disconnect now bucket endpoints at
    micron resolution by rounding, like cleanup-wires, instead of truncating
    at 0.1 mm, so 12.69999 and 12.7 are no longer treated as different
    points. --tolerance still means the same distance in mm.

  • The routing cache no longer replays copper from a different C++ router
    build
    (Issue #7014). Cache entries are now keyed on the loaded native
    router BUILD_VERSION (the value kct build-native --check reports, or
    none when routing falls back to Python), so a cache hit after a C++
    rebuild no longer silently skips the new router's behaviour. Existing
    cache entries are invalidated once. PYTHONHASHSEED is not part of the
    key: routing boards 01 and 06b under two hash seeds gave identical copper.

  • In-process kct route keeps explicit --clearance / --manufacturer
    defaults
    (Issue #7017). kct route used to parse its flags with a
    separate outer parser and rebuild the router's argv flag by flag, dropping
    any value equal to the default. To keep an explicit --clearance 0.15 or
    --manufacturer jlcpcb (#4875) it sniffed sys.argv, which an in-process
    kicad_tools.cli.main([...]) call never fills, so both values were lost and
    the router took them from the board's own rules. kct route now hands
    everything after route to the router unchanged, so what you type is what
    runs. kct route --help now prints the router's own help, so it lists every
    flag the router accepts, including diagnostics and experimental switches
    that were reachable only through python -m kicad_tools.cli.route_cmd.
    Global --verbose, -q/--quiet and --units before route work as
    before.

  • PCB block pad geometry: the SOT-23-5 / SOT-23 fallback pad tables and
    LEDBlock's cathode trace
    (Issue #7022). The built-in fallback pad tables
    in kicad_tools.pcb.placement (used when kicad_footprint_reader is not
    installed) were rotated 90 degrees from the KiCad library, so LDOBlock
    traces ended on the wrong pins. They now match KiCad 10's
    Package_TO_SOT_SMD. LEDBlock now turns the LED 180 degrees inside the
    block so its cathode faces the resistor. Before, the cathode-to-resistor
    trace ran straight through the anode pad.

  • PCB.export_gerbers honours layers, include_drill and create_zip for
    the jlcpcb, pcbway and oshpark presets
    (Issue #7023). The preset path used
    to drop all three: create_zip=False still returned a zip, include_drill=False
    still exported drill files, and an explicit layer list never reached
    kicad-cli. The preset now supplies the defaults and explicit arguments
    override it; a layer list without Edge.Cuts logs a warning.
    GerberExporter.export_for_manufacturer gains matching keyword-only
    layers, generate_drill and create_zip overrides. PCB.export_gerbers_zip
    no longer wraps a nested gerbers.zip inside its archive and now honours
    include_drill.

  • Board-03 experiment row routes under jlcpcb-tier1 (Issue #7026). The
    expand_route_experiment.py 03 row now passes --manufacturer jlcpcb-tier1,
    matching the sidecars the board-03 recipe fills with, so a captured plan's
    pour links are judged against the recipe's fill.

  • kct bom and kct sch summary|validate|preflight|hierarchy fail on a missing or unparseable schematic (Issue #7028).
    A nonexistent or empty root .kicad_sch used to exit 0 (an empty BOM, a
    fallback summary, or per-check warnings); it now exits 1 with an error on
    stderr. Failing sub-sheets are still skipped.

  • Board-03 save_plan() stores net names, not per-board net ids (Issue #7033). Copper
    (net N) refs are rewritten to (net "NAME") on capture, unknown ids raise ValueError,
    and apply_plan() rejects a plan holding numeric net ids instead of failing with a bare
    KeyError or binding copper to the wrong net.

  • PCBBlock placed at 90/270 degrees now lands its traces on the pads
    (Issue #7046). Block ports, component positions, internal traces, block
    vias and ComponentPlacement.pad_position(..., block_rotation=...) now
    rotate in KiCad's footprint convention (the same transform as the exported
    footprint angle), instead of the opposite sign, which put block traces up to
    a pad pitch off their pads. Rotations 0 and 180 are unchanged.

  • kct placement align --tolerance is now honoured (Issue #7056). Parts
    already within tolerance of the target line are left untouched, and
    aligned / "Aligned N components" report parts actually moved;
    distribute and reference alignment counts likewise.

  • IEC 60664-1 clearance is now keyed on the rated impulse voltage, not the
    working-voltage peak
    (Issue #7057). kct creepage --standard and
    kct audit --hv-standard looked the steady-state peak working voltage up in
    Table F.2, which is keyed on impulse withstand voltage, so 230 V mains got
    0.2 mm where basic insulation at overvoltage category II needs 1.5 mm. The
    required clearance is now the larger of Table F.2 at the Table F.1 rated
    impulse voltage (from the working voltage and the new
    --overvoltage-category / --hv-overvoltage-category, default II) and
    Table F.8 at the steady-state peak, floored by pollution degree. Table F.2's
    non-standard 3200 V and 6300 V rows are gone (now 3000 V and 6000 V), the
    PD2 / group II / 80 V creepage cell is 0.95 mm (was 0.9), and the creepage
    table is cited as IEC 60664-1:2020 Table F.5. --standard iec62368 shares
    the same tables and is corrected too.

  • kct creepage no longer overstates creepage around board slots (Issue #7058).
    The detour search is no longer pinned to the straight-line nearest-point pair
    (it is now seeded from both shapes, so it is the true shortest surface path),
    and cutouts narrower than IEC 60664-1 dimension X (PD1 0.25 / PD2 1.0 /
    PD3 1.5 mm) are bridged instead of followed. The X table is unverified against
    the paywalled standard text. The report records dimension_x_mm.

  • CPL coordinates now match the package's Gerbers (Issue #7059). Pick-and-place
    files were written in the board-relative, Y-down frame (and pcb_path subtracted a
    sheet-absolute aux origin from them), so bottom-side and off-centre parts landed in
    the wrong place. kct export and PCB.export_placement now emit the frame of
    kicad-cli pcb export pos --use-drill-file-origin (aux origin, Y up); OSH Park uses
    the origin-less frame like its Gerbers.

  • Edge-coupled differential impedance now comes from a field solve, not an
    empirical coupling factor
    (Issue #7060). CoupledLines took the
    single-ended Z0 and applied a fitted kc = exp(-a*s/h)*(1-exp(-b*w/h))
    with silent clamps, which overstated Zdiff of tight pairs by 12 % to 24 %
    with real copper (+11.7 % at board 06's USB2 geometry). Edge-coupled
    stripline and microstrip now solve the two-conductor quasi-static problem
    by boundary elements (even and odd mode, finite copper thickness, offset
    stripline, dielectric interface for microstrip), the kc factors and the
    [20, 200] / [15, 180] ohm clamps are gone, and
    TransmissionLine.differential_microstrip delegates to the same model so
    only one coupled-microstrip model remains. Results agree with Cohn
    (stripline) and Kirschning & Jansen (microstrip) within 1 % at zero
    thickness and with independent finite-difference solutions within 2 % at
    35 um copper.

  • Stock LDOBlock now applies cleanly to a real board (Issue #7062). Its
    caps are turned 270 degrees so pad 1 faces VIN/VOUT, GND leaves pin 2 as a
    dogleg under the package, and the VIN/VOUT/GND ports sit on block copper.
    PCBLayout.apply_to_pcb no longer refuses (or skips) the block's own copper.

  • Board 04 schematic is ERC-clean under KiCad 10.0.7 (Issue #7063). Power
    nets (+5V, +3.3V, GND) are now named with global labels instead of
    local labels that shared a name with the power symbols
    (same_local_global_label); schematic and readiness evidence re-pinned on
    the pinned 10.0.6 toolchain.

  • Four CLI flags that were documented but dead now work (Issue #7065).
    kct parts import is reachable from the top-level kct parts (it was only
    available via python -m kicad_tools.cli.parts_cmd). kct sch validate
    gains --lib-path DIR and --skip CHECK (both repeatable); --lib-path
    now actually resolves symbols missing from the embedded lib_symbols for the
    pin-aware orphan_label, wire_stub and pin/pad-count checks (embedded
    symbols stay authoritative). kct optimize-traces --verbose prints one row
    per net, and --format json adds stats.per_net. The
    optim.fom.check_hard_constraints docstring now says manufacturer is
    reserved and unused.

  • docs/reference/cli.md matches the real parser for optimize-traces, parts and sch validate (Issue #7070). The never-implemented optimize-traces --length-match is gone, the parts import/suggest/cache subcommands and the sch validate options are documented, stats.per_net is documented in machine-output.md, optimize-traces -v now has help text, and tests/test_cli_docs_drift.py guards these sections against parser drift.

  • kct impedance CPWG impedance and dielectric loss corrected (Issue #7075).
    Coplanar-waveguide-with-ground Z0 now uses the conductor-backed closed form
    (Ghione & Naldi) instead of a sinh modulus that over-stated it by 50-230 %;
    microstrip and CPWG dielectric attenuation drops by a factor of eps_eff
    (Pozar eq. 3.198). Degenerate geometry raises ValueError instead of
    returning 50 Ω, and the balanced CPWG solver no longer returns micron-scale
    geometries.

  • Microstrip Z0 and εeff now include the Hammerstad-Jensen 1980
    finite-thickness terms
    (Issue #7076). TransmissionLine.microstrip read
    0.4-5.4 % low in Z0 for real copper weights and over-read the effective
    permittivity by 5-7 %; it now agrees with a 2-D field solve to within 0.35 %.
    Computed impedances, width_for_impedance suggestions, propagation delay and
    the loss estimate all shift (see the upgrade note).

  • kct drc reports schematic parity as PASSED, FAILED or NOT RUN
    (Issue #7077). kicad-cli's schematic_parity findings used to be dropped,
    and a board whose schematic was missing or not fully annotated (kicad-cli
    skips parity, exits 0 and writes an empty list) read as parity-clean.
    DRCReport now parses the findings into DRCReport.schematic_parity with
    their kicad-cli severities, and kct drc (table, summary, --mfr) and
    kicad-drc-summary print one Schematic parity: PASSED | FAILED (N issue(s)) | NOT RUN (<reason>) line from one shared helper. A NOT RUN
    parity never prints DRC PASSED; the verdict says DRC INCOMPLETE.
    --format json gains a schematic_parity object (status, checked,
    reason, error_count, warning_count, issues), a derived
    schematic_parity_checked alias, and parity_errors/parity_warnings in
    summary. .kct_waivers.json entries also match parity findings.

  • Silkscreen width/height DRC rules now cover B.Silkscreen (Issue #7078).
    The generated Silkscreen Width and Silkscreen Height custom rules only
    matched F.Silkscreen, so undersized bottom-side text passed against the
    fabricator minima. All shipped manufacturers/rules/*.kicad_dru files are
    regenerated.

  • ImpedanceRule judges diff pairs at the routed gap (Issue #7079). The pair
    Zdiff now uses the measured edge-to-edge spacing to the partner net instead of a
    hard-coded 0.15 mm. Traces with no measurable partner gap, or whose solver call
    fails, are no longer judged against silent 100 / 50 Ω fallbacks, and pair
    violations no longer carry a bogus "requires 0.000mm" hint.

  • Cost estimator no longer prices unlisted layer counts as 2-layer or multiplies the per-order finish fee by quantity (Issue #7080). Layer counts missing from the pricing table (e.g. 3) are priced at the next listed count up; the surface finish fee is charged once per order.

  • kct audit / kct report no longer say READY when ERC or the sync check never ran (Issue #7083).
    ERC now finds kicad-cli the same way DRC does (find_kicad_cli()), so a
    standard macOS install that is not on PATH actually runs ERC instead of
    reporting PASS / kicad-cli not found (skipped). When kicad-cli really is
    unavailable, ERC times out, or the schematic/PCB sync check crashes, the
    check is shown as NOT RUN (table, --format json via a new
    could_not_verify field, and kct report's ERC status), an action item
    explains why, and the verdict is capped at REVIEW WARNINGS. The
    zone-fill action items now point at kct zones fill <pcb> instead of a
    manual KiCad step.

  • kct export no longer ships unrouted boards (Issue #7091). The
    connectivity safety floor and the DRC preflight now count KiCad's
    unconnected_items (opens), which since #4498 sit outside the report's
    errors and were never read. A board with error-severity opens is refused
    by the floor (N connectivity error(s) (S net short(s), O unconnected item(s))) and FAILs preflight (DRC has 0 error(s), 36 unconnected item(s)), so --strict-preflight blocks it; waived and warning-severity
    opens still pass. A DRC report recorded for a different PCB (such as the
    *_routed sibling's drc_report.json next to the unrouted board) is no
    longer read as a clean DRC: preflight reports WARN DRC report is for <x>, not <y> and the floor warns instead of passing silently. ERC preflight
    applies the same check against the schematic.

  • Lattice fallback diff pairs prefer equal P/N via counts (Issue #7093). The
    uncoupled-fallback leg-pair selection and flush score now rank via-count
    parity after routed legs and class-pad shortfalls, before skew (board 06rf's
    USB3_RX1 goes from 2/4 to 2/2 vias, skew 2.557 to 0.507 mm). A segment that
    still commits with unequal vias now prints a per-segment WARNING in the
    lattice diff-pair report, including on partially-coupled pairs. When every
    candidate is asymmetric, the fallback now also re-searches each leg on its
    partner's layer sequence (a matching layer change, or none), so a via-parity
    candidate exists to rank: a fresh board 03 route gives USB_D 3/3 vias instead
    of 2/3. A second mirrored round re-routes each parity hit's partner under
    the mirrored leg; it replaces the first-round pick only when no worse on both
    skew and length, so neither is traded for the other. Board 03's
    committed copper is not re-derived yet: USB_D+
    (16.07 mm) is still over the board's 16 mm bound, and the MCU launch is over
    its 4 mm bound (#7256).

  • Crossed lattice fallback diff-pair legs get equal via counts
    (Issue #7093). A fallback pair whose pads swap sides (board 03's USB_D
    J1.A6->U2.3) has to cross itself, so one leg changes layer. Two things
    kept it at unequal vias. First, a via-equal candidate that the #7284 gate
    rejects for its skew or length no longer stops the mirrored partner-leg
    search from running. Second, when the exact mirror still finds no
    acceptable via-equal pair, a partner that changes layer once is also
    copied from or to another copper layer (for example B.Cu -> F.Cu instead
    of In2.Cu -> F.Cu), so the copy can cross under its partner instead of
    going round it. Board 03 under --manufacturer jlcpcb-tier1 routes USB_D
    with 3/3 vias instead of 3/2, at 15.780 / 15.851 mm (skew 0.071 mm, was
    0.195 mm). Under --mfr jlcpcb it gives 3/3 vias at 15.432 / 15.432 mm,
    which is now inside the 16 mm bound (it was 16.022 mm). Routed copper on
    06b, 06rf and 07rf is unchanged.

  • Lattice diff pairs now try to match via counts on extra-pad attachments
    (Issue #7093). A pair's extra same-net pads, such as a USB-C receptacle's
    second A/B row, are attached by single-ended connections. Until now nothing
    kept P and N on the same layer changes there, so a pair whose coupled run
    had equal vias could still commit unequal totals. After the board is
    routed, a pair with unequal totals and extras on both nets has one extra
    re-searched against the finished copper, copying the other extra's layer
    changes. The new copy is kept only if it makes the totals equal and is no
    longer than the longer of the two extras plus 0.5 mm. No other copper
    moves, and the pass stops (keeping the copper) once the run's deadline
    passes. The outcome is logged as a Diff-pair extra parity (lattice) line.
    To copy an a -> b -> a hop onto a planar extra, the lattice search gains
    a crossing layer mirror: it lays copper off the hop layer only within
    1.5 mm of either pad, so the copy cannot come back planar. On the current
    benches (03, 06b, 06rf, 07rf) no committed copper changes: board 06rf's
    USB2_D (2 / 4) is not fixed, because its two J1 extras cross each other,
    and the log line now says that no attachment fit.

  • Unequal diff-pair via counts are always reported as a pair-rule failure
    (Issue #7093). Every route_all_* run now checks each engaged pair's routed
    copper and prints a WARNING: diff pair P/N committed with unequal via counts ... pair-rule failure line, also kept in
    Autorouter.pair_via_parity_failures. Before this, the grid engine's
    independent fallback and pairs left to the single-ended main loop reported
    nothing. The lattice's per-segment warning now counts the legs that were
    actually committed, not the fallback's own record. This adds hybrid segments,
    and a discarded tap-chain, gateway-offer or gateway-rollback retry can no
    longer hide or invent a mismatch.

  • Three validators no longer read clean when they didn't check (Issue #7094).
    kct validate --lvs --min-confidence now filters only what is displayed:
    a below-threshold mismatch (e.g. R1 renamed to R9 on the PCB) still
    makes the result non-clean and the exit code non-zero, and the output says
    N match(es) below --min-confidence X hidden (JSON:
    hidden_below_min_confidence). kct erc and kct erc explain compute the
    exit code before the --errors-only / --type / --sheet view filters
    (after exclusions and --filter-config), and print No violations match the filter (N hidden: E error(s), W warning(s)) instead of ERC PASSED - No violations found (JSON: summary.hidden_by_view_filter). kct sch validate
    reports an ERC that could not run (kicad-cli missing, timed out, killed, or no
    report) as INCOMPLETE: erc not run (<reason>) under a new checks_not_run
    list, sets passed: false, and no longer lists erc under "Checks run".

  • kct drc no longer passes an unrouted board (Issue #7096). Error-severity
    unconnected items (missing connections) from kicad-cli reports now count toward
    the verdict and exit code in table, summary, --mfr and --format json modes;
    the JSON summary (and --mfr --format json) carries an explicit unconnected
    count, and .kct_waivers.json unconnected_items entries waive them for
    KiCad JSON reports as they already did for .rpt text reports.

  • PlacementSession.query_move now reports real deltas (Issue #7099). The
    "before" snapshot was taken after the hypothetical move, so score_delta was
    always 0.0 and new_violations / resolved_violations always empty. This
    made MCP query_move / apply_move, find_best_position (always []),
    query_swap, query_alignment and kct placement refine query return false
    answers. query_swap also no longer crashes on unhashable violations once a
    swap produces any.

  • kct placement optimize no longer saves parts off the board on boards
    with a non-zero origin
    (Issue #7106). Since #3812 the placement optimizer
    compared origin-relative part positions against an absolute board outline.
    On any board whose Edge.Cuts does not start at (0, 0), kct placement optimize, refine, suggest, snap and the MCP placement sessions
    clamped parts into the wrong rectangle and wrote them outside the real
    board edge while still reporting success
    (16 of 17 parts on the
    04-stm32-devboard example, 44 of 47 on 09-usbc-pd-power). In-board parts
    were also flagged as boundary violations, which skewed session scores. The
    outline from both the Shapely and the legacy Edge.Cuts paths, KiCad keepout
    zones (--auto-keepout), the no-Shapely board-edge keepout and
    get_board_edges now all use the origin-relative frame that part positions
    use.

  • query_move routing impact and export_assembly fields now report computed values (Issue #7107).
    routing_impact.crossing_changes is the real before/after inter-net crossing delta and
    estimated_length_change_mm is the change in per-net pin-MST length (it previously summed
    centre distances over star springs, with wrong sign/magnitude). pnp.rotation_corrections
    now counts the CPL rows actually rotated by the manufacturer table instead of always 0.

  • MCP docs match the server; get_drc_violations is now a registered tool (Issue #7111).
    docs/mcp/tools.md documented a DRC tool the server never registered
    (Unknown tool); it is now registered (manufacturer maps to rules=,
    plus severity_filter). analyze_board is documented under its real name
    board_summary, board_inspect is documented, and the route_net,
    get_unrouted_nets, measure_clearance, placement_suggestions and DRC
    parameters/responses now match the server. A new tests/test_mcp_tools_doc.py
    keeps tool names and parameter tables in sync.

  • MCP route_net now saves the routed net (Issue #7114). Since the tool
    was added it called merge_routes_into_pcb with a signature that never
    existed, so every successful route ended in TypeError and
    success: false. It now merges the router's S-expression into the board
    text and writes it to the resolved output path (<stem>_routed.kicad_pcb
    by default), with the trace width and via sizes the router used.

  • Three MCP tools no longer fail or mislead (Issue #7115). placement_suggestions
    now works (it raised ImportError on every call) and returns ranked suggestions
    from the same engine as kct placement suggest; its never-implemented strategy
    parameter is removed. export_gerbers and export_assembly count copper layers
    inside gerbers.zip, so the default zip_output=true no longer reports a false
    "no copper-layer Gerber files" failure. board_view wraps its caption to
    max_size_px (400 px now draws the board) and returns success: false telling the
    caller to raise max_size_px instead of a caption-only image.

  • kct init writes the same .kicad_dru rules as generate_dru (Issue #7116).
    The hand-written template mapped silkscreen width to silk_clearance, hole-to-edge
    to hole_to_hole and solder-mask dam to physical_clearance; it now uses
    text_thickness/text_height and physical_hole_clearance, with via rules
    scoped to vias, and drops the unsupported mask-dam rule.
    The ; header comments (invalid in KiCad DRU; KiCad silently ignored the whole file)
    are now #. The old hole_size (max ...) rule (6.3 mm for JLCPCB) is not emitted by
    generate_dru and is not restored here.

  • MCP route_net merges collinear segments before saving (Issue #7120).
    A straight run is now one segment instead of one per grid step, matching
    kct route; an optimizer failure falls back to the raw routes.

  • MCP session tools now share one session store, and restore_checkpoint
    really restores
    (Issue #7121). start_session and the seven context tools
    (record_decision, get_decision_history, annotate_decision,
    get_session_context, create_checkpoint, restore_checkpoint,
    get_session_summary) used to keep separate stores, so
    get_session_context failed right after start_session, pcb_path and
    intents were empty, and unknown session IDs were silently accepted. The
    context now lives on the placement session and closes with it.
    create_checkpoint captures the live positions, score, violation counts and
    move depth; restore_checkpoint undoes or re-applies moves and reports
    success only when positions match the snapshot. PlacementSession.undo()
    also re-syncs the incremental DRC engine, so DRC counts no longer go stale
    after an undo.

  • Hierarchical placement fallback no longer mixes coordinate frames (Issue #7124). place_hierarchical_from_pcb now shifts the raw Edge.Cuts outline into the origin-relative frame when the Shapely path is unavailable, so offset boards keep parts inside the real outline.

  • kct -q footprint generate and kct --verbose erc <report> no longer fail (Issue #7126). The
    footprint generate and erc <file> argv shims now skip leading global options
    (--verbose, -q, --units) before looking for the command, as kct route does.

  • kct audit runs ERC on a .kicad_pcb, and the sign-off skills refuse
    bad boards
    (Issue #7136). kct audit <board>.kicad_pcb used to force an
    ERC skip and print [OK] ERC: PASS -- skipped by user request without
    running anything. It now resolves the schematic the way kct check does
    (board_routed.kicad_pcb finds board.kicad_sch) and runs
    kicad-cli sch erc; with no schematic it reports ERC NOT RUN and the
    verdict caps at WARNING. "Skipped by user request" now appears only with
    --skip-erc. The /kct:manufacturing-readiness and /kct:tapeout gates are
    now exact command blocks whose exit code is the verdict: Gate 1 judges
    kct check's JSON report (any NOT RUN other than Manifest refuses),
    Gate 2 runs kicad-cli pcb drc --exit-code-violations and requires
    0 error violations and 0 unconnected items, and Gate 3 passes that report
    to kct export --drc-report. Unrouted, clearance-violating and
    schematic-less boards are refused instead of signed off.

  • Project.route() saves merged copper (Issue #7138). Collinear
    per-grid-cell segments are now merged into straight runs before the board
    is written, matching kct route and the MCP route_net path. The merge is
    one shared helper, merge_collinear_routes; RoutingResult segment and
    length totals reflect the merged copper.

  • kct drc waiver application tolerates report-like objects without schematic_parity (Issue #7141). Fixes an AttributeError regression from #7077 in apply_waivers_to_report.

  • kct drc schematic parity now runs on <base>_routed.kicad_pcb boards
    (Issue #7147). kicad-cli only reads <pcb-stem>.kicad_sch, so parity was
    NOT RUN (exit 2) on every fleet board. kct drc now resolves the board's
    schematic (project.kct, then <base>.kicad_sch with _routed etc.
    stripped) and runs kicad-cli on a temporary copy of the board folder with
    that schematic staged under the board's name; the board folder is never
    written (no stray .kicad_prl). New --schematic PATH override, the JSON
    schematic_parity.schematic field records the schematic used, and NOT RUN
    now names the real cause (no schematic found, several candidates, or a
    schematic kicad-cli could not netlist).

  • kct optimize-placement no longer ignores gr_rect board outlines without Shapely (Issue #7148). The no-Shapely fallback now uses the shared Edge.Cuts reader (gr_rect, gr_poly, gr_line, gr_arc) instead of a footprints+10 mm box that could place parts off-board (board 04). A board with no outline at all now warns loudly on stderr, and the new --require-outline flag makes it an error.

  • Board 04 J1 value is consistent (Issue #7150). The PCB footprint value is
    now SWD-6, matching the schematic, instead of SWD.

  • kct export bundles agree with themselves (Issue #7153). JLCPCB
    Gerbers now include the F_Paste/B_Paste stencil layers that JLCPCB's
    KiCad 7/8 Gerber guides list, so an SMT-assembly bundle no longer ships a
    CPL with nothing to cut a stencil from (a paste layer missing from the
    board's layer table is still not plotted). Preflight now reads the
    manufacturer/tier a kct check DRC report was generated for and WARNs
    when it differs from the run's --mfr (jlcpcb-tier1 vs jlcpcb counts
    as different), instead of reporting "DRC OK" beside a report that says
    FAIL. The design report's DRC Status names the --mfr rule set and shows
    the kct and kicad-cli error counts on separate rows, noting that one
    defect can be reported by both engines. README.txt and the console now
    name the Gerber archive by its real path, gerbers/gerbers.zip.

  • Export report ERC runs against the export's schematic (Issue #7153).
    ReportDataCollector and ManufacturingAudit accept an explicit
    schematic_path, and kct export passes the schematic it already holds, so
    the report's ERC section no longer reads SKIPPED when the schematic is not
    auto-discoverable from the PCB path. An explicit skip_erc still skips.

  • extract_board_outline no longer returns an Edge.Cuts cutout as the board outline (Issue #7155). When several closed contours exist (gr_rect, gr_poly, chained gr_line/gr_arc), the largest-area one is returned instead of the first by element type.

  • Route-time match-group tuner now clears preserved copper on other nets (Issue #7159). Under --preserve-existing, vias and tracks of preserved nets outside the tuned groups are treated as fixed clearance obstacles, so meanders no longer land on them (board 07 DRC regression). Lengths and published routes are unchanged.


Release notes truncated to fit GitHub's 125,000-character limit (202,401 characters in full). Full v0.24.0 section: https://github.com/rjwalters/kicad-tools/blob/v0.24.0/CHANGELOG.md