duplicate-ref-des

Remedy ¶

Re-annotate the schematic so each physical part holds its own designator, then re-check the BOM, since whichever part was silently merged is the one to look at first.

What it means ¶

One reference designator (U1, R5) assigned to two genuinely distinct physical parts, as opposed to the legitimate case of one multi-unit part whose gates share a ref-des.

Why engineers want it ¶

A duplicated ref-des is a common annotation slip. The two parts merge into one BOM line (one gets built, the other silently omitted) and their connections collide on the same net-join key, so the netlist is quietly wrong. KiCad’s ERC flags it, as do most.

Why this is a reader diagnostic, not an IR check ¶

The IR keys components by ref-des and folds a multi-unit part into one component with sections (WS1-001), so by the time the netlist exists the collision is gone and a duplicate is indistinguishable from a multi-unit part. Only the reader, at ingestion, can tell them apart using its format’s own rule (in KiCad, the same unit claimed twice). It records the verdict in InputDiagnostics.RefDesCollisions; this rule only reports it (architecture/rules-and-checks.md, “Where a rule runs”).

What it means when this rule does not run ¶

This rule only reports a reader’s verdict. When the reader never reached one, the rule has nothing to say, and a clean design also gives it nothing to say. So a reader states whether it computed the diagnostic (InputDiagnostics.supplied). Where the reader did not, this rule is gated to not-applicable, rather than reporting a pass that nobody earned (agni issue 309).

Format	Detects	Rule reads
KiCad	the same unit claimed twice	a real result
gEDA	the same slot claimed twice, or two unslotted placements sharing a designator	a real result
xschem	a repeated instance name, which the format declares unique	a real result
IPC-2581	a repeated refDes, with no gate construct to group placements	a real result
EDIF	nothing: a multi-gate part is instances sharing a designator with no unit to tell them apart	not applicable, with that reason

On an EDIF design the gap is the format’s, not the reader’s, because detecting a collision there would mean reporting every multi-gate part as a duplicate. Not-applicable is the accurate answer there, because the reader never looked, and a report now shows it.

Query structure ¶

report each collision the reader recorded.

select C in ref_des_collisions

Reads: reader.ref_des_collision. Tier P. Site: diagnostic (reader-detected).