Exclude ETH_MDIO/MDC from IEEE MDI 100 Ω. Deduplicate RD+/RX and TD+/TX
pair aliases to one cert per channel; inherit LAN8720A 10/100 when the
magnetics-side nets lack the PHY. USB D+/D− skip AF tr/f (90 Ω cert owns
them); RMII stays on the English PE-AF-003 AN-133 line.
Describe PCB analysis, protocol cert (USB/Ethernet/MAC–PHY), findings
tree, top-bar UI, and Markdown export. Gitea is the source of truth.
No predecessor-product mentions.
Replace the left sidebar with a thin top toolbar and tab/sheet navigation.
Light is default; dark mode lives in Settings and persists. Report export
is Save as Markdown (.md) with the findings-tree hierarchy — not Excel.
USB D+/D− is one English certified/not-certified line vs the packed 90 Ω
cite. RMII ETH_RXD/TXD/TXEN is identified as PHY–MAC, not RJ45 100 Ω.
J2 missing-field L1/L2/L3 noise is silent. PCB AI failures go to
review_errors; not_reviewed count matches one designator per row.
Certifications, Protocols, domains, empty states, and the protocol empty message are English. Error / Warning / Info stay. Sidebar version is stamped 2.84.0. A PCB exam records extract, pcb_checks, protocol, af, and total seconds.
PE-BOM-011 treats EP/EPAD/THERMAL PAD and the footprint exposed land as one pad when both sides have one. PE-ESD-001 quotes a datasheet only with part and section, and does not warn on a 3V3 jack rail, uncitable VBUS, or optical S/PDIF. CC Rd is the packed 5.1 kΩ on the Type-C CC pin net. Ethernet MDI pair Z is PASS, FAIL, or ERROR against the cited 10/100 or gigabit clause.
Alta Frequenza / af-allegati only hold TI SI (SNLA027B, SDAA499), which do not state vSafe5V/vSafe0V. CabCon R2.5 §1.6 defers volts to USB PD. HDMI 1.4 TMDS stays UNKNOWN.
LS/FS rates, pull-up/down, and VBUS come from the base spec. Cable Z0 90 ohm is CABLE; PCB should 90 ohm is RECOMMENDED. L2 differential_impedance stays UNKNOWN. Rp/Rd/Ra from CabCon Tables 4-27/28/29. IEEE Section One/Three and HDMI still missing. PDFs not in git.
NUMERIC rows cite document title, revision, section/table, and page. USB Zdiff, Type-C Rp/Rd, 10BASE-T, and 1000BASE-T stay UNKNOWN. PDFs are not in git.
USB 2.0 is protocol, Type-C is connector, USB 2.0 over Type-C is the
implementation. Ethernet/HDMI skeletons stay UNKNOWN with needed_document.
No invented ohms; Mini HDMI is connector-only; RJ45 is not 1000BASE-T.
PCIe/CXL keep logical vs physical ids. I2C, CAN, RGMII and missing
high-speed processor interfaces get required_checks; numbers stay
UNKNOWN/MISSING_SOURCE until an official cite. No invented ohms.
L3 CHANNEL structure for timing budget, insertion loss, return loss,
crosstalk, and complete channel. Numeric PASS/FAIL only from existing
channel FACT; otherwise a visible skip. No OpenEMS, no invented S-parameters,
no M6 packs.
Instances, L0–L3 max level, measured/limit/margin/source/method, traceability chain. FAIL listed first, not under Warning. Visible skips. No pack numbers, no L3 solver.
Presence and connection only. FAIL solely on missing MANDATORY structure.
AXI4 internal is NOT_APPLICABLE, not FAIL. RECOMMENDED never FAIL. No L1–L3
and no invented ohms/mm/ps.
physical_bus_instance from declaration, silicon/part, pin/net names, and
pairs. USB and DDR fixtures emit groups; AXI4 internal attaches no PCB
nets. Ambiguity is REVIEW, not FAIL. No L0–L3 certifier and no invented Z.
Versioned JSON skeletons without invented ohms/mm/ps. RECOMMENDED is not
FAIL-mandatory; typical-as-standard is rejected. Report Protocolli is empty
until recognition (M1+). Native src overlays inherited periscopex on the
package path.
PCB tree folders are Error / Warning / Info. U18 with PE-PLC-002
ERROR is ERROR, not WARNING. J2 and other connectors list with ICs.
PE-PLC stays in the PCB exam bucket. No deploy.
WROOM lands are not the ESP32 die table (U11). QFN-48 pads 25–48 are
not extras when extraction stopped at 1–24 (U12). VQFN is the QFN
family (U19 24-QFN 4×4). Real extra signal pads still ERROR.
USB-C A5/B5 are contacts, not EP. Datasheet pintable, KiCad
.kicad_mod, and embedded PCB pads join by pin identity. Missing
pintable skips; scalar pin_count is never an ERROR authority.
POST /api/library/datasheets now writes an IC inbox card or a
passive/discrete model. Empty pintables never enter library/extracted.
AF Board+AI runs after unchanged run_pcb_checks (PE-SI / HF line stay).
Add /api/library datasheet import and component GET/PUT with no exam.
Reorganize pytest into datasheet, library, schematic, PCB, and AF+AI.
Document Rust criteria (none chosen; no rustup) and coding conformity.
Phase B: stackup vs fab spec, via current with datasheet I (no IPC via
chart), ESD pad distance, USB-PD, PoE isolation voltage, antenna, and
PDN Z(f). Skip without evidence. Not DRC or FEM.
Phase A: stub, return-split, and BOM termination on buses present on the
graph; DDR/CPU/FPGA class certifiers only when that device exists. Skip
without evidence. Via, pad, track, and zone stay distinct. Not DRC or FEM.
Replace the schematic-only DeepSeek scaffold with what this tree actually is: full board analysis for fab, gated interface certifiers, evidence rules, and the run/Docker/update commands that exist on disk.
USB-C/RJ45/PoE/DDR3 are the method, not a catalog. No connector means
no finding. Bare RJ45 is Ethernet only; PoE needs PoE evidence. HubAudio
has USB-C and PoE MagJack, not DDR. Skip-I unchanged.
Dedicated deterministic engines certify class and connection integrity
(CC1/CC2, Rd 5.1 kΩ, VBUS/GND, SuperSpeed if USB3; 10/100 vs GbE
magnetics; PoE only with evidence). Missing evidence is INSUFFICIENT
or N/A, never invented PoE, SuperSpeed, Z, or I. No FEM, no via IPC.
KiCad 9/10 name-only nets fill the index. PE-LAY-004 counts tracks, vias,
and pours and skips NC nets. Missing I_load is INSUFFICIENT. EP pads are
not PE-BOM-011; abs-max binds the named pin. Kelvin/EMI/PI ignore NC and
IN±. LQW decodes nH; BLM/FB is a bead, not DCR as Z.
michele