Automate Test Comparison with Sift Families

Excited to share what I’ve been designing for the past few months here at Sift — Family Data is officially live! 🚀 It's been truly rewarding working with such a talented team here. A big shoutout to the family squad: Bill Raj-Derouin, Alexa Agnew, Adrienne Chang, Alex Luck! Static pass/fail limits in hardware testing don't always cut it when analyzing test run telemetry. Seeing statistical sigma bounds derived from known nominal runs, time-aligned to real mission events, and serving as a dynamic baseline for Family-based, automated rules is just 🤌

A customer schedules an audit. The first question in the room is never "did the parts pass." It's "can we actually find the proof." And that's when somebody grabs a USB stick and starts walking the floor. I've watched that afternoon happen more than once, and it always traces back to the same quiet decision nobody remembers making: the test stand was built to answer good part or bad part, and the data it produced got treated as exhaust. Fine on day one. Rough the day a customer wants a specific serial number's record from eight months ago. It gets worse with scale, not better. We once worked a lab that had grown to more than 45 stations, every one logging to its own drive on its own clock with its own naming. That's not a lab. That's 45 islands sharing a roof, and no way to see yield across any of it. The fix isn't ripping out what works. It's stamping traceability onto every record at the source, leaving the fast local log alone, and adding a path that gets results into somewhere people can actually query. I wrote up how we think about it on the Dynamic Engineering blog. Link's in the comments. Be honest with yourself for a second: if a customer asked tomorrow for one unit's full test record, could you have it in five minutes? #DataAcquisition #TestEngineering #Manufacturing #LabVIEW #QualityAssurance

It was a great pleasure to share more about the Bosch Mobility solutions for SDV - and particularly about the SDV.Suite - during our Live Webninar last week Thu, 16 July 2026! Thanks Olga Grischtschenko & Florian Merkert for sharing the stage 🎉 You can find the recording in the link posted by Bosch Mobility below. #InventedForLife #LetsMakeItReal #SDV #SDM #Automate #Collaborate

A software bug found late in physical integration is 10x more expensive to fix than catching it early. This is the "Rule of Ten" in automotive software development. By mastering the "shift-left" approach and moving up to 70% of testing into virtual environments, the business impact is immediate: 📉 10x Cost Savings by catching defects early in software verification. 💸 30% Less Hardware Costs through smart virtual validation. 🤖 80% Less Manual Effort by combining CI/CT, automation, and AI. Watch the 30-second snippet below to see the math in action! Find the full recording here: https://lnkd.in/e_thN3q3

Marc Hagmeister at dSPACE put the software-defined vehicle's dirty secret into one line this week: "You build software in two-week sprints. Then you wait months for the hardware to test it." He's right. All that agility, parked in the test-bench queue. His fix is Software-in-the-Loop — start testing while the software is still being written, in the cloud, before a single HIL rig frees up. Shift the testing left. Feed the benches better software instead of fighting them for a slot. I'd only add the part nobody puts on the slide. When you shift testing left, you don't run fewer tests. You run 10× more, continuously, from day one. And every one of those runs is now a piece of evidence. Which quietly moves the bottleneck. It stops being "when does the rig free up" and becomes: 1. Can you tie each of those thousands of continuous test results back to the requirement it proves — automatically? 2. When the assessor asks "show me the verification for this requirement," can you answer in seconds, or do you go digging? 3. Does your safety case keep up with a test cadence that just went from monthly to hourly — or does the paperwork become the new test-bench queue? SIL removes the hardware wait. What it exposes is the traceability wait underneath it. More tests, run earlier, is only a win if the evidence stays linked to the requirements as fast as you generate it. Shifting testing left is half the move. Shifting the evidence trail left with it is the other half — and it's the half that shows up in the audit, not the demo. Come see how deep the traceability side of this goes →

I remember a testing session where we spent hours troubleshooting a CAN issue, only to find we were analyzing the bus in isolation. It's easy to get tunnel vision when you're deep in CANoe or CANalyzer, but that isolated view often misses the bigger picture. We learned the hard way that a "green light" on the bus doesn't mean much if the ECU isn't processing those messages correctly. You really need to see the whole system at play. #automotive #embeddedtesting #canprotocol #uds #ecutesting #manualtesting #bms #mobilitysolutions #automotiveelectronics AVOID ISOLATED CAN TESTING ▸ THE COMMON MISTAKE • Focusing only on CAN bus frames. • Assuming bus integrity equals system function. • Over-relying on CANalyzer/CANoe logs alone. ▸ WHY IT'S A MISTAKE • Ignores ECU internal processing. • Misses application layer issues. • Leads to misdiagnosed "no-fault-found" scenarios. ▸ BETTER APPROACHES • Integrate ECU diagnostics (UDS). • Monitor internal ECU states and outputs. • Simulate real-world scenarios, not just bus traffic. ▸ KEY TOOLS & TECHNIQUES • CANoe/CANalyzer for bus, but also for UDS. • Debuggers for ECU code tracing. • JIRA for linking bus behavior to defect tracking. 💡 A healthy CAN bus doesn't guarantee a healthy system; test the ECU's response, not just the bus traffic.

Just yesterday, I caught a tricky edge case in an ECU update that could've easily slipped through a standard regression. It made me think about how often we rely on familiar paths. When designing test cases, especially for complex embedded systems, it’s easy to get comfortable with what worked before. But what if we intentionally shook things up? Spending a little extra time thinking outside the usual "happy path" can uncover those unexpected bugs. It's not about adding hours, but about adding a different kind of thought to your test strategy. #automotiveqa #embeddedtesting #testcasedesign #manualtesting #regressiontesting #ecutesting #canprotocol #udsdiagnostics #bmsquality DIVERSIFY YOUR TEST CASES ▸ BEYOND THE HAPPY PATH • Don't just test ideal scenarios. • Explore unexpected inputs. • Simulate real-world chaos. ▸ THINK LIKE A FAILURE • What if a sensor glitches? • What if a message is delayed? • What if power cycles unexpectedly? ▸ COLLABORATE & BRAINSTORM • Discuss edge cases with developers. • Get fresh perspectives from teammates. • Use defect tracking data for inspiration. ▸ EXPLORE PROTOCOL LIMITS • Test CAN/UDS message overflows. • Validate Ethernet error handling. • Check boundary conditions for BMS. 💡 INTENTIONAL UNORTHODOX TESTING UNCOVERS HIDDEN BUGS.

Everyone tells me they have evals. Almost nobody does. Ask one follow up and it falls apart. How many cases are in your eval set. Who wrote them. When did it last grow. The honest answer from most teams is twelve examples someone wrote the week before launch, and nothing has touched them since. That is not an eval set. That is a screenshot of what you hoped the model would do in March. The engineer you want answers differently. Their set grew every time production surprised them. The weird customer input that broke parsing. The prompt change that quietly made summaries worse and nobody noticed for a week. The tool call that started failing when the provider changed a default. Every one of those became a case. That is the whole tell. Not whether they have evals. Whether the set gets bigger when something breaks. A test suite that never changes is not measuring your system. It is measuring your memory of it.

Does everyone build evals for their personal or team workflows? Aside from regression testing, it seems a very good use case is performance testing on new models like Kimi K3 (or Fable 5 when it came out) Benchmarks are always to be taken with a bucket of salt after all

Hi connections 👋 Continuing from yesterday — once we know how a single bit travels on the wire, the next question is: how do a bunch of these bits become a structured message that every ECU on the bus can understand the same way? That's the CAN frame, and honestly, every field in it exists to solve a real problem engineers ran into. It starts with SOF — a single dominant bit that basically means "everyone, listen up, I'm about to talk." Then comes the part I think is the cleverest bit of the whole protocol: the 11-bit arbitration field. Instead of a bus master deciding who gets to talk, every ECU just watches the ID being transmitted. Lower ID wins. No collisions, no retries, no wasted bandwidth — priority is built into the identifier itself. After that: RTR tells you whether this is a data frame or a request for data. IDE tells you whether you're dealing with a standard 11-bit ID or an extended 29-bit one (useful once a network has too many message types for 11 bits to cover). DLC tells the receiver exactly how many of the payload bytes are actually meaningful — so if you're only sending 3 useful bytes like 22 F1 A7, DLC just says 3. Then the data itself, a CRC to catch corruption, and 7 recessive bits marking EOF so the bus knows it's free again. One detail I really liked learning: bit stuffing. Any time there are 5 identical bits in a row, the controller inserts an opposite bit. This stops receivers from mistaking a long run of bits for an end-of-frame, and it catches line errors almost immediately instead of waiting for the whole frame to fail. And why cap payload at just 8 bytes? Two reasons that still hold up today — the hardware to process larger frames fast was (and still is) expensive, and in safety-critical systems, a higher-priority message can't afford to wait behind a long one. Small frames keep the bus responsive when it matters most. Next post: what happens when 8 bytes genuinely isn't enough — how CAN-TP lets us send full diagnostic payloads over this same bus. 🔜 #CAN #CANbus #CANProtocol #AUTOSAR #EmbeddedSystems #EmbeddedEngineering #ECU #VehicleNetworking #CANFD #CANXL #InVehicleNetworking #VehicleElectronics