What 0.8 shipped, and what is still open

01 / CandidateThe 0.8 implementation is released

0.8 adds two pieces of syntax and a faster router. trunk draws a multi-terminal net as a tree: a branch leaves the nearest copper its net already has, with a junction dot, instead of running back to its own terminal. subcircuit blocks are written once and placed many times, with parameters, exposed terminals and nesting, and are lowered to ordinary declarations before anything else reads the document. And the sparse router prices each grid edge once per route, which compiles a contended bus about one and a half times faster with byte-identical output.

0.7 before it added bundle nudging, which carries a full reversal bus from twelve wires to thirty-two and corrected a limit 0.5 had published; 0.6 added snapshotSchematic and parseSchematicSvg behind their own subpaths; and 0.5 added relative placement and bounded rip-up routing on top of 0.4's removal of the global component and connection ceilings. Quarter-turn geometry, typed primitives, net semantics, collision validation, netlist inspection, structural design rules, accessible descriptions, deterministic fuzzing, and source maps remain part of the same compiler.

source:DC8 "DC" at (90, 120) #purple [type=voltage-dc]
inductor:LX8 "L" at (280, 120) #cyan
adder:HA8 "HA" at (500, 120) #amber [type=half]
qgate:QY8 "R_y" at (710, 120) #blue [parameter="\chi"]
component:BENCH8 "bench" at (870, 290) #slate [width=150 height=80]

02 / GatePublish core before the website consumes it

The release gate runs type checking, 100% statement/branch/function/line coverage, bounded property fuzzing, sixty-one targeted mutants with a 100% kill score, a 270-document byte-identity corpus, six Chromium pixel goldens, installed package-boundary tests, two gzip budgets, and latency/scaling ceilings. Those gates prove the properties they assert. They do not prove a circuit is electrically safe, temporally correct, standards-certified, or manufacturable.

Publication order is non-negotiable, and it is now automated rather than remembered. A tag push publishes @schemd/core to npm and cuts the GitHub release from that version's CHANGELOG.md section, so the prose is written once. A scheduled job on the website then adopts the release — dependency and lockfile together, release snapshot regenerated from npm's own packument — runs the complete web gate against it, and opens a pull request. Nothing merges itself, because a compiler release is allowed to change rendered geometry and that is the part CI cannot judge.

The ordering matters for a reason worth stating plainly: a linked dist/ proves local integration, never that the published tarball and its export map work. Only an install from the registry does that, which is why the sync job installs rather than links.

logic:GO "1" at (100, 120) #purple [type=high]
testpoint:GATE8 "publish gate" at (340, 120) #cyan
load:BELL "bell" at (590, 120) #emerald [type=buzzer]

GO.out -> GATE8.node #purple [digital line]
GATE8.node -> BELL.in #emerald [digital line]

Known boundaries remain deliberate: rotation is limited to quarter turns; text measurement is deterministic rather than font-engine exact; sub-circuits flatten into one model rather than becoming sub-sheets; design rules are structural rather than SPICE, timing, or formal verification; a trunk is the sequential approximation to a Steiner tree rather than the shortest one; routing is deterministic but bounded; and scalar/bus coercion is rejected instead of guessed. See Limitations and Resource budgets before treating untrusted source as a server workload.