Bitdeer now mines with most of what it builds, so the largest SealMiner customer is Bitdeer's own fleet. Every hour a unit is down is lost hashrate on the company's books, which makes after-sales quality a mining-margin question. Returns surface at benches in the US and Greece; design sits in Singapore and Penang; assembly runs through outsourced partners. The QE's job is to turn that scattered return data into ranked issues, proven root causes and closed CAPAs, and to keep a failure-mode knowledge base everyone uses.
The Field Quality Desk runs the whole loop: escalation queue priced at live hashprice, Pareto and p-chart, Weibull with suspensions, IS / IS NOT, 8D and bilingual stakeholder updates, log triage on the public SealMiner codes, and a failure-mode KB. Fleet data is synthetic. Built for this application.
Who the quality issues come from, where they surface, and what each means for the role.
| Piece | State (Oct 2026, public sources) | What it means for the QE |
|---|---|---|
| Customers | Self-mining 73.0 EH/s on 243,000 rigs and co-mining 15.9 EH/s on 56,000 rigs (June 2026). External buyers hold about 8.2 EH/s of A2s. Wafer inventory is designated for internal use. | The loudest escalations come from Bitdeer's own sites. Downtime cost is measurable in hashprice, so priority can be set in dollars. |
| Products | A2 and A2 Pro (SEAL02), A3 family (SEAL03, air and hydro), A4 (SEAL04) in production since April 2026. Air and hydro variants of each. | Three chip generations and two cooling systems in the field at once. Failure modes need model, cooling and lot on every record. |
| Repair network | Official centres in Massillon OH, Rockdale TX and Knoxville TN; authorised in Carrollton TX and Athens. Bhutan listed as coming soon. Published repair turnaround: 7 working days. | Return data arrives from benches 12 to 13 hours behind Penang. Bhutan returns travel furthest; the demo prices that gap. |
| Build | Single foundry for chips. Outsourced assembly coordinated from Asia. A US factory in Sparks NV is due end-2026 (10,000 units a month). | Every CAPA has a supplier or factory owner. A second assembly site in 2027 needs a clean failure-mode baseline before it starts. |
| Penang | Opened 1 Oct 2025 with seven test labs and two environmental-aging chambers; hardware, firmware and supply chain teams. | Failure analysis and reliability tests can run next door to the QE. Aging chambers can reproduce field failures. |
| Customer terms | 365-day warranty. DOA window 30 days. Sea freight prohibited for returns. Scrap criteria include corrosion, burned boards, mixed boards and third-party firmware. | Warranty denials need consistent evidence. Clear documents prevent disputes before they start (§06). |
After-sales is part of the product in ASIC mining. Buyers compare repair reach and turnaround as much as J/TH.
| Maker | After-sales model (public) | Bitdeer gap and angle |
|---|---|---|
| Bitmain (Antminer) | 12-month warranty on miner and PSU, 6 months on control boards. Sells extended warranty. Authorised overseas repair centres and on-site repair at mining sites. | Widest network and a paid warranty extension. Bitdeer's advantage is owning the fleet: field data from 243,000 rigs feeds design directly. |
| MicroBT (Whatsminer) | One-year warranty. US service centres and repair hubs promising under seven business days; Hut 8 certified to do in-warranty repairs on site. Service centre in Kazakhstan. | Matches Bitdeer's 7-working-day promise and adds on-site certified partners. Bitdeer has no centre in Asia yet. |
| Bitdeer (SealMiner) | 365-day warranty, five live centres in the US and Greece, Bhutan coming soon, public error-code manual and firmware, on-site services for large customers. | A published error taxonomy is a strong base for a field knowledge base. Closing the Asia repair gap and extending codes to A3 and A4 are the next steps. |
Takeaway. Rivals compete on reach. Bitdeer can compete on learning speed: it sees its own field failures first.
The public Error Code Manual V2.0 grouped into failure modes, each with the likely root-cause owner. This is the skeleton of the knowledge base the JD asks for.
| Failure mode | Codes | Usual root cause to test first | Owner | Severity |
|---|---|---|---|---|
| Hashboard not detected / init fail | 420-442, 480-482 | Connector seating, then BGA or connector joint opens after thermal cycling. | Manufacturing | High |
| Chip count, bad chips, low hashrate | 270-282, 450-452, 470-472 | Single ASIC or voltage domain fault; chip bin or wafer lot. | R&D, IC Lab | Medium |
| Board EEPROM, SN and BIN data | 290-302, 310-336 | Data not written or wrong at final test; EEPROM part. | Manufacturing test | Medium |
| Temperature sensor | 200-222 | Sensor or I2C line; connector contamination. | R&D | Low |
| Over-temperature, derate | 240-242, 600-604 | Site inlet temperature and airflow before the unit. | Aftersales, site | Medium |
| Fan | 100-105, 125-126 | Bearing wear from heat and dust; fan date code. | Supplier quality | Low |
| PSU | 800-806 | Input surge or site power quality; capacitor and fuse. | Supplier quality | High |
| Control board, firmware, network | 1-4, 700-702, web 100-109 | Firmware and model mismatch; interrupted upgrade; network or pool config. | R&D firmware | Low |
| Hydro loop | not in V2.0 | Galvanic corrosion from copper in an aluminium loop, coolant chemistry outside pH 7.0 to 8.7, pressure above 350 kPa, filtration. | Supplier quality, site | Highest |
| Transit and handling | inspection | Packaging, route, moisture (the reason sea freight is banned). | Logistics | Medium |
Takeaway. The codes already say what failed. The knowledge base adds why it failed, who owns it, and what was learned last time.
One flow, end to end. Unit data uses the public SealMiner API schema and error codes; repair centres and firmware are live from Bitdeer's public API; the return history is a synthetic fleet with two planted defects.
Severity × frequency × impact. Impact = lost TH/s × live hashprice × days down per site.
Reads a SealMiner API stats reply: boards, chips, PSU, fans, firmware vs latest public build, spec tolerance. Disposition and RMA intake.
Pareto and p-chart; Weibull MLE with suspensions and 90% bounds by lot; lift; IS / IS NOT.
8D draft plus updates for R&D, Manufacturing, Aftersales, Management and customers, in EN and 中文.
SealMiner codes mapped to modes and owners; closed 8Ds write lessons back.
| What the demo finds | How |
|---|---|
| Board drops on A3 Pro Hydro, P1 | p-chart above UCL in the last 6 weeks. Weibull MLE β about 0.6 with 90% bounds below 1 (early life confirmed). Lift concentrates on line L2 builds from July and is absent on L1. Points to a build or screening escape before any design change. |
| Fan wear on A2 Pro Air | β above 1.5 (wear-out) and higher rates at hot, dusty sites. A supplier and maintenance-interval question. |
| Fleet sweep | A read-only collector (Python, port 4028, stats only) sweeps a site; every unit is checked and the flags ranked, so field techs get a worklist before anything ships to a repair centre. |
| Bhutan repair gap | Toggling "Bhutan repair centre open" shows the monthly hashrate saved on current returns. |
| JD duty | How I would do it | Proof here |
|---|---|---|
| Own all escalated quality issues. | One register: every issue has an owner, status, next date and a CAPA number. Nothing is tracked in chat threads. | Demo, Escalations. |
| Investigate performance and failure data; find root causes. | Pareto, then p-chart for timing, Weibull for life shape, lift for concentration, then failure analysis on returned parts to confirm. | Demo, Investigate. |
| Preventive measures: design guidelines, validation plans, process controls. | Every closed CAPA leaves one control: a test limit, a screen, a change-control rule or a design guideline. | Demo, 8D D7. |
| Triage by criticality, frequency, business impact. | Severity from the KB, frequency per 1,000 installed, impact in lost hashrate dollars. Weights reviewed monthly. | Demo, priority formula. |
| Bridge R&D, Aftersales, Manufacturing. | One weekly quality review with the same data for all three; each audience gets the update it acts on. | Demo, stakeholder updates. |
| Centralised knowledge base of failure modes. | Start from the published codes (§03), add cause, checks, owner and lessons. Bench technicians write to it. | §03, demo KB. |
| Prompt, accurate stakeholder updates. | Fixed cadence per priority: P1 every 48 hours until contained, P2 weekly. | Demo, 8D tab. |
Metrics first, then the moves that improve them.
| Metric | Definition | Why it matters |
|---|---|---|
| Returns per 1,000 installed | Monthly, by model, mode and lot. | Separates real trends from fleet growth. |
| DOA and second-return rate | Within 30 days of dispatch; within 15 days of a repair. | Second returns measure repair quality. |
| No-fault-found rate | Bench finds no fault. | High NFF means wasted freight and a diagnosis gap at site. |
| Days down per return | Site failure to unit back hashing. | The cost driver. Logistics often outweighs bench time. |
| CAPA closure and recurrence | Days to close; repeat returns after closure. | Proves the fix worked. |
One register, one taxonomy
Meet bench leads at each repair centre, R&D and manufacturing. Agree one failure-mode list from the error codes. Pull 12 months of returns into one register and rank the open issues.
Close the top three
Full 8D on the three highest-priority issues, with failure analysis in the Penang labs. Start the weekly quality review and the update cadence.
Prevent, then scale
Turn lessons into controls: early-life watch on new lots, screening limits, hydro coolant checks. Hand a baseline to the new US factory before its first build.
Noticed in public after-sales documents. Each is small, and each removes a source of customer disputes.
| Where | What I saw (1 Oct 2026) | Fix |
|---|---|---|
| DOA window | After-Sales Policy: defective within 30 calendar days "from the date of dispatch". Help-centre DOA article: "within 30 days from the date of delivery". | Use one start date in both. |
| Air Operations Manual V2.3 | Covers A2, A3 and A4 Pro Air, but the warranty paragraph names "A2-Air Cooling Series products". | Name all covered series. |
| Error Code Manual V2.0 | Listed under the A2 series. The July 2026 A3 hydro firmware notes "Added error codes", with no public list for A3 or A4. | Publish an A3 and A4 code list. |
| Help-centre repair locations | "Official Repair Center Locations" shows Rockdale only. The live repair-ticket list has Massillon, Rockdale, Knoxville, Carrollton, Athens and Bhutan (coming soon). | Point the article at the live list. |
Independent job-application homework from public information, collected 1 Oct 2026. The demo is my own prototype; its fleet data is synthetic and it contains no Bitdeer code or internal data.