Repository navigation
Conversation
|
Changes requested — not ready for Class-A qualification [P1] Correct the MP2307 power stage. The source uses R2 = 10 kΩ and R9 = 3.01 kΩ, which sets approximately 4.0 V with the MP2307’s 0.925 V feedback reference, not 3.3 V. That exceeds the STM32G030’s 3.6 V maximum and risks the connected sensors. The COMP network is also incorrect: C4 is connected between FB and COMP, with no COMP-to-GND compensation network. Please correct the divider and compensation topology against the MP2307 datasheet, then rerun voltage and ERC checks. [P1] Use correctly rated input protection. D2 is the registry part [P1] Apply the declared netclasses. POWER, VALVE, and I2C are declared but assigned to no nets, so the valve power paths and other critical signals are routed with default geometry rather than the stated widths and clearances. Apply the classes explicitly and rerun routing and DRC. [P1] Address the return-path failure. The README records 105 layer changes without a ground-return via within 1.53 mm on a two-layer board carrying switching valve currents, analog probe inputs, and I2C. This is a signal-integrity and EMI design problem, not just a limitation to document. Add ground stitching/return paths or justify a different stackup and rerun the board review. [P2] Bring the schematic and BOM to qualification quality. Twelve decoupling placements are reported far from their devices, a junction exceeds the readable wire limit, and the functional region boxes overlap. Correct the schematic layout rather than accepting these as limitations. Twenty parts also lack distributor identity, so the BOM is not orderable as presented. [P2] Complete reproducibility evidence. The README documents only one local export/router run. Add repeated artifact comparisons and independent PCB review evidence before treating this as Class-A qualification material. |
|
@SammyOina re your review: fixed divider, COMP, Q1/D1, TVS, netclasses, schematic. Still open: return-path vias (#280), BOM identities, independent review. Pushed in 5faa6c9. |
|
@SammyOina re BOM identities: those 14 registry parts are generic placeholders, so nobody picked a real orderable product for them. I didn't invent part numbers. |
|
@SammyOina re BOM identities: fixed. Every part now names a manufacturer number, LCSC where found; APDS-9301 has none. Per-value fuse and inductor added. Pushed in e9b3a4d. |
e9b3a4d to
74011e1
Compare
|
@SammyOina re your review: agent-made now. Return-path warnings 105→1, classes applied, regions disjoint, exports hash-match. Open: reset button, via sizes, BOM, TVS clamp. Pushed in 74011e1. |
JeffMboya
changed the title
JeffMboya
changed the title
82437dd to
b3e7348
Compare
|
PCB, TOP PCB, ANGLED PCB, BOTTOM SCHEMATIC |
SammyOina
left a comment
SammyOina
left a comment
There was a problem hiding this comment.
Changes requested — not ready for Class-A qualification
The latest revision improves the board and makes the routed result reproducible, but the remaining items are still qualification blockers:
-
Correct the 5 V protection on 3.3 V signals.
D_N3is a 5 V TVS onFLOW_PULSE, which feeds STM32 PA6; a 5.1 V clamp is above the STM32G030 input maximum.D_N4is also a 5 V TVS across the 3.3 V rail, so it will not protect the rail before the MCU and sensors are already overstressed. Use protection with a verified worst-case clamp below the relevant absolute maximum and document the current calculation. -
Resolve the input-TVS/buck-rating conflict. The SMAJ15A is listed with a 24.4 V clamp on the same protected rail as the MP2307, whose maximum input is 23 V. Select a protection arrangement whose worst-case clamp stays below the regulator limit, or isolate/protect the buck input separately.
-
Verify the SHT30 footprint and escape. The registry entry is imported without reviewer or datasheet provenance, and the README says the two-layer router cannot escape the 0.5 mm DFN. Verify the exact manufacturer land pattern and produce a manufacturable escape; a routed result that the router itself reports as unescapable is not sufficient.
-
Make the BOM and fabrication output qualification-ready. The README reports only 7 LCSC numbers and 10 manufacturer numbers for 112 rows, while the discussion says the identities were fixed. Reconcile this, provide exact manufacturer/package/rating/datasheet data for every fitted part, and include or reproducibly generate the final Gerber, drill, BOM, and pick-and-place package. The APDS-9301 and generic passives still need orderable identities.
-
Run a valid manufacturing and DRC check. The current DRC has 112 footprint-library warnings because the server lacks the KiCad footprint library table, plus one dangling-track warning. Rerun with the real libraries configured and resolve or justify every warning.
Please also upload the final Gerbers, drill files, BOM, and pick-and-place data to JLCPCB’s PCB/order quotation flow and run its manufacturability/DFM checks, as was done for the recently merged Class-B board. Do not place an order. Record the returned warnings, stackup/clearance constraints, assembly/package issues, and whether every BOM/CPL item is accepted. This is needed to demonstrate that the qualification output is actually fabricatable, not merely routable in Synth.
The remaining six signal-via return-path advisories, the 43.8 mm SWD distance, 16.7 mm reset distance, remote decouplers, unresolved crystal load, and unresolved valve-current assumption should also be corrected or explicitly justified before claiming the Class-A case is passed. The schematic’s OTHER region and crowded PCB references should be cleaned to make the result human-reviewable.
Please update the README with the exact JLCPCB DFM result and regenerated artifact hashes after these changes.
|
@SammyOina Pushed 6d1ae15: clamps, TVS, SHT30, crystal, sourced BOM, library DRC fixed; JLCPCB DFM pending login, listed in README. |
|
@SammyOina 5df49af: final exports, hashes and fab package regenerated; datasheet BOM ratings; justified distances. JLCPCB DFM needs a re-run. |
5df49af to
624326e
Compare
|
@SammyOina 624326e: all five items fixed, distances justified; C_VDD2 is #377, DFM warnings #378; DFM re-run pending. |
|
@SammyOina d7314b4: DFM result recorded; every open item explained in the README or tracked in #377-#380. |
SammyOina
left a comment
SammyOina
left a comment
There was a problem hiding this comment.
Changes requested — remaining Class-A qualification blockers\n\nThe latest revision resolves the earlier power-stage, protection, netclass, BOM, and return-path concerns, and CI is now fully green. The following qualification issues remain:\n\n### [P1] Clear the JLCPCB DFM findings and rerun the final package\n\nThe README records 18 silkscreen-to-hole dangers, one through-hole-to-SMD danger, 50 undersized silkscreen-line warnings, 32 silkscreen-to-pad warnings, and additional mask and annular-ring warnings. The DFM was not rerun after the proposed outline change, and the report itself is not committed.\n\nPlease correct the applicable exporter/layout issues, regenerate the final Gerber/drill package, rerun JLCPCB PCB DFM on that exact package, and commit reproducible evidence. A Class-A qualification board should not be released with known fabrication dangers and below-capability silkscreen geometry.\n\n### [P1] Regenerate the stale schematic evidence\n\nThe attached schematic render still contains an OTHER region, while the latest source and README claim that all parts are grouped and the OTHER region was removed.\n\nRegenerate the schematic and PCB renders from the final source/export and verify that the screenshots, hashes, README measurements, and fabrication package all describe the same revision.\n\n### [P1] Resolve the input protection rating\n\nD_TVS is an SMAJ13A with a 13 V standoff, while the documented normal input range reaches 13.2 V. It is also placed after Q1, leaving the input MOSFET exposed to an upstream transient before the TVS can clamp it.\n\nVerify normal-operation leakage at the maximum specified input, worst-case surge/clamp behavior, Q1 stress, and the full protection chain against the manufacturer datasheets. Select a suitable protection arrangement if the current one cannot satisfy those limits.\n\n### [P2] Resolve or formally disposition the MCU decoupling distance\n\nC_VDD2 remains 11 mm from the MCU supply pin, and the README explicitly states that this is not electrically justified. The fact that closer placements did not release is a Synth limitation, not electrical qualification evidence.\n\nEither produce a released layout with valid per-pin decoupling placement or document an engineering disposition with an independent reviewer before accepting this as Class-A output.\n\n### [P2] Complete the JLCPCB assembly review\n\nOnly PCB DFM was run. SMT DFM and the assembly/quote flow were not run for this 126-part board.\n\nSubmit the final BOM, CPL, and Gerbers to JLCPCB without placing an order, record the assembly/package/hand-fit findings, and reconcile any exceptions with the final fabrication package.\n\n### [P2] Close the remaining component-validation assumptions\n\nThe README still identifies unverified LED/bleeder values, a 5 pF crystal stray-capacitance assumption, and datasheet checks that were not completed. Verify these values against the selected manufacturer parts and update the qualification documentation.\n\nPlease update the README and artifact hashes after the final DFM/export pass. The green CI result does not replace the required fabrication and qualification evidence.
…hored The final run of the Class A irrigation controller imported four parts from KiCad's symbols and authored four. Each carries the footprint_dimensions the registry requires, taken from its footprint's courtyard in the installed KiCad library: the importer and synth_author_part write none for four of them. header_1x5 is byte for byte the part the I2C sensor node's PR adds.
One prompt, one agent driving synth mcp, an empty workspace and an empty user registry. After its release export the agent compacted the layout with synth_compact_layout: seven attempts, each exported through the release gates, kept the tightest that routed (83.9 x 67.0 mm). The source and sidecar are the agent's; the README records what was measured on the exported files, KiCad 10.0.6 DRC (0 errors), every connector flush with the edge, and three exports into empty directories that match the agent's own.
… and placement advice The source and sidecar are the 112-component board that fits 75.1 x 65.4 mm: the agent's design, gap compaction, two fill passes that add 26 parts, two passes that apply the placement advice, and the top-right hole moved to the corner. The README gives the measurements and the three-export reproducibility check, and says what is still open: the reset button and SWD header are farther from the MCU than before, and the added parts' values are not checked against datasheets.
Review fixes: the 5 V clamps on 3.3 V nets are gone and the flow pulse has a 3.3 V zener, the input TVS is an SMAJ13A under the buck's 23 V limit, Q1 is a 30 V AO3401A, the SHT30 is removed, the crystal load capacitors are 6.2 pF for the 8 pF NDK crystal, the valve current is stated, and every part is in a named group. Eighteen parts were added to the voids (126 parts, 31.1 % fill); C_OUT2 sits 11.6 mm from the regulator. Three exports match, KiCad with the library table reports 0 errors, the BOM is sourced for all 126 lines and a fabrication package is generated. The orchestrator's position and group edits are listed in the README; the scripts behind them, the agent reports and prompts are under agent-run/.
…package and measured numbers Three fresh exports of the committed layout with the post-routing label pass give one routed board, one Gerber content hash and one BOM and placement file; KiCad DRC with the real libraries is 0 errors, 0 unconnected and 4 silkscreen warnings. The README records the label results, the reproducibility table, the C_VDD2 search and the part-removal trials that did not release a board, and that the JLCPCB DFM check has to be re-run on the regenerated Gerbers.
…ad of the MOSFET, 0.2 mm via drills and measured evidence
d7314b4 to
e64767f
Compare
|
@SammyOina e64767f: TVS before Q1, 0.20 mm via rings, LEDs, silkscreen fixed; decouplers open #377; main CI red #391. |
What type of PR is this?
This is a feature.
What does this do?
Adds
qualification/class-a/irrigation-controller/: 126 parts, 75.1 x 65.4 mm, released; TVS ahead ofQ1; 0.6/0.2 mm vias.Which issue(s) does this PR fix/relate to?
Resolves #263, #260. Related #39, #280, #377, #378, #379, #380, #388.
Have you included tests for your changes?
No. The board is data; the README compares three fresh exports by hash. Label-pass tests are in #370.
Did you document any new/modified feature?
Yes. The README gives measurements, review answers, BOM ratings, the DFM result and reproducibility;
agent-run/holds prompts and scripts.Notes
Stacked on #326 and #388; needs #310, #345, #348, #368, #370. JLCPCB DFM not yet run on this package.