A Braiins support ticket is usually one of four things: a network that cannot reach a pool or the licence server, a hashboard or PSU fault the firmware already named with a code, a tuning or thermal state the customer reads as a fault, or Braiins Manager losing sight of miners. Each one is cheaper to close when the first reply already contains the right command output. The ICT seat covers the hours when Asian and Oceanian farms are awake and Prague is not, so a clean hand-off matters as much as the fix.
The Miner Triage Desk reads grpcurl or BOSminer API output, checks it against the published model specs and error codes, and drafts the next commands, a customer reply in EN or 中文, an escalation packet and a KB entry. Seven sample tickets. Built for this application.
The products a ticket can land on, and what each means for the person answering it.
| Product | State (Oct 2026, public sources) | What it means for support |
|---|---|---|
| Braiins OS | Autotuning firmware for Antminer S19, S21 and T21 families plus Braiins Mini Miners. Control boards: Zynq, Amlogic, BeagleBone, Cvitek, BCB100. About monthly releases; 26.09.1 is current. Dev fee 2 to 2.5% by model. | Most tickets start here. The answer depends on model, control board and build, so those three go in every first reply. |
| Public API | gRPC on port 50051 (REST wrapper since 25.07), proto files public on GitHub. BOSminer API on port 4028 kept for compatibility and marked for future deprecation. | Customers integrate their own tooling. Support reads and writes grpcurl fluently and moves customers off 4028 over time. |
| Licensing | Since 23.04 each miner fetches a time-limited licence from *.braiins.com every 60 minutes, valid 12 hours offline. Port 3336 must be open, plus 443 and 3338 behind Braiins Proxy. | A locked-down firewall shows up as a fee or hashrate question. The fix is an egress rule. |
| Braiins Manager | Cloud fleet management with a local Agent (Rust since 4.0). Antminer, Whatsminer, Avalon and IceRiver on stock, Braiins OS or MARA firmware. Curtailment with Enel, CPower, Voltus, Scioto; strike-price for ERCOT, PJM, Nord Pool. | Tickets come from operators with hundreds of miners at once. Connectivity errors (ECONNREFUSED, TIMEOUT_ERROR) are mostly LAN and firmware state. |
| Toolbox, Proxy, Farm Monitor | Toolbox for batch install and settings; Braiins Proxy aggregates farm connections; Farm Monitor restores Braiins OS when a Cvitek miner reboots into stock. | Batch fixes go through Toolbox. A miner "back on Bitmain firmware" may need Farm Monitor coverage. |
| Pool, Hashpower, Solo | FPPS pool with a 0% effective fee for Braiins OS users; hashrate spot market and guaranteed contracts; solo mining. | Pool and payout questions sit with a separate operating entity and its own terms. Route them, and keep firmware answers to firmware. |
Aftermarket firmware is a support-heavy product. Farms switch when a lockout, a bricked board or a slow answer costs them hashrate.
| Rival | Position (public) | Where Braiins support can win |
|---|---|---|
| Bitmain stock | Free, no dev fee. Recent stock releases may restrict third-party firmware installs (Braiins advisory, 18 Sep 2026). | Clear, fast guidance on which stock builds are safe to flash. A lockout ticket answered well keeps the customer. |
| LuxOS (Luxor) | Antminer and, since 2026, Whatsminer. Bundled with Luxor's pool and hashrate products. | Braiins OS has per-chip tuning, Stratum V2 and Manager. Support makes those features usable at 3 a.m. site time. |
| VNish | Large aftermarket base on Antminer and Whatsminer, sold through resellers. | Direct vendor support with a public error taxonomy and an open API schema. |
| ePIC, MARAFW | Hardware route: replacement control boards, firmware sold to large operators. MARAFW runs inside Braiins Manager. | Manager supports mixed fleets, so one support desk covers the whole site. |
| Foreman, Awesome Miner | Multi-vendor fleet tools. | Manager bundles energy-market features; support can show the curtailment flow working end to end. |
Takeaway. The firmware features are close across vendors. Response time and first-reply accuracy are what operators remember.
Ticket families built from the error codes, miner states, Academy FAQs and 2026 release notes. Each has a first check that settles most cases and a clear owner when it does not.
| Family | Signal | First check | Escalate to | Fix lives in |
|---|---|---|---|---|
| All pools dead | State deadPools | GetNetworkInfo: gateway inside the subnet, DNS reachable. Then user.worker spelling, ports 3333/3336/3337, a second pool host. | Rarely | Customer network |
| Fee or hashrate "missing" | GetLicenseState none, limited or expired | Outbound 3336 to *.braiins.com; behind Proxy also 443 and 3338. | Contract or licence team if a key is involved | Customer firewall |
| Hashboard short of chips | I3, I5, I6; chips below model spec | Compare chipsCount per board with the model table; reseat data and power; ground the chassis; lower the power target. | Repair path with the support archive | Hardware |
| PSU and voltage | P1 to P7, V1 to V4 | PSU model against the supported list; ribbon cable to the PSU; swap PSU. | Repair path | Hardware |
| Thermal pause | T1 to T5, F1 to F3, thermalPause | Fan RPM, inlet temperature, hydro inlet vs outlet (T3 means reversed flow), recent upgrade. | Dev team if it follows an upgrade | Site, or firmware |
| Upgrade regression | New errors right after a release | Release notes for the next patch (26.09.1 fixed T5 stops after 26.09). Count affected sites. | Dev team with a site count | Firmware |
| Install or lockout | Stock UI still showing; install fails | Stock firmware version, control board, boot medium; browser cache. | Dev team for new stock builds | Firmware, Bitmain |
| Manager blind spot | ECONNREFUSED, TIMEOUT_ERROR, Agent Offline | Agent service status; ports 50051, 4028, 80 from the Agent host; did the miner fall back to stock? | Manager team if the Agent is healthy | LAN, firmware state |
| Tuning expectations | Tuner TUNING, PREHEAT; DPS cooldown | Explain the state and its expected end time (shown since 26.08). Judge hashrate after Stable. | No | Expectation |
Takeaway. Five of nine families end at the customer's own network or site. A first reply that asks for the right output closes them in one round.
One flow, end to end, on public data only. The API cookbook parses the proto files from GitHub at load; the release list and hashprice are live; the seven tickets are written from documented failure modes.
Paste grpcurl, BOSminer API, log lines or a Manager export. Ranked findings with sources; downtime priced at live hashprice.
Next commands (read-only first, writes flagged), customer reply in EN or 中文, escalation packet, KB entry.
All 64 methods with a ready grpcurl call and a request body built from the message definition.
Checks a miner's IP, mask, gateway and DNS; plans VLANs and DHCP pools; egress rules with sources.
Read-only script for customers: one paste with every Get* call, password never written out.
| Sample ticket | What the desk finds |
|---|---|
| S21 "all pools dead" after a rack move | Static IP moved to 10.20.5.0/24 with the old gateway 10.20.4.1: no route off the LAN. Also a space in the worker name and no failover host. Reply tells the customer the one setting to change. |
| S19j Pro down a third | Board 2 answers with 84 of 126 chips (I5) and hashes nothing. Commands to disable the board so the other two keep mining, and a repair-path escalation with the archive requested. |
| S21 Hyd. keeps pausing | Water inlet hotter than outlet on all three boards: hoses swapped on a new install (T3). No hardware fault. |
| T5 after the 26.09 upgrade | Matches the issue fixed in 26.09.1. Escalation carries the site count; reply says upgrade, then watch for a day. |
| Manager: 48 workers ECONNREFUSED | Up to 4.6 PH/s Manager cannot see after a power cut. Checks for miners that fell back to stock firmware. |
| JD duty | How I would do it | Proof here |
|---|---|---|
| Braiins OS and Braiins Manager as the primary focus. | Learn both from the error codes, miner states and API outward. Keep a lab miner on the latest build and one on the previous build to reproduce regressions. | §03, demo. |
| cmd.exe, Ubuntu and Linux CLI, API setups, remote access. | Every reply that needs data names the exact command for the customer's OS. Remote sessions follow one checklist: read-only first, every change agreed in the ticket. | Demo, next commands and API cookbook. |
| Install, configure, troubleshoot; educate customers. | Answer the ticket, then link the Academy page that would have answered it, so the next one is self-served. | Demo, reply drafts with sources. |
| Contribute to public and private knowledge bases. | Every resolved ticket with a new cause becomes a KB entry the same day: symptom, confirm, cause, fix, escalate-when. | Demo, KB entry tab; §06. |
| Relay customer feedback to product and development. | Escalations carry the build, model, control board, site count and the exact output. A weekly digest groups tickets by family and release. | Demo, escalation packet. |
| Daily contact with large mining operators. | One named contact per large account on the ICT shift, with their topology on file (VLANs, Agent host, Proxy) so no call starts from zero. | Demo, farm network tab. |
| Travel to data centres for onboarding. | Willing. Onboarding checklist: egress rules, Agent sizing, Toolbox batch install, Farm Monitor ranges for Cvitek boards. | Demo, egress table. |
Metrics first, then the moves that improve them.
| Metric | Definition | Why it matters |
|---|---|---|
| First-reply completeness | Share of first replies that ask for, or already contain, the output needed to decide. | Each missing detail costs a full day across time zones. |
| Resolved in one exchange | Tickets closed after the first customer reply. | The families in §03 should mostly close here. |
| Hashrate-hours restored | Missing TH/s × hours from ticket open to fix confirmed. | Puts a farm's cost on the queue, so priority follows money. |
| Escalation quality | Escalations the dev team accepts without asking a question back. | Developer time is the scarcest resource in the loop. |
| Ticket to KB | New causes written up within one working day. | The next customer finds it in the Academy. |
Learn the floor
Shadow the Prague and Americas shifts. Read the last quarter's tickets by family. Install Braiins OS on a lab miner from stock, break it on purpose, recover it. Agree the hand-off note format between shifts.
Own the ICT shift
Take the Asia and Oceania queue and the ZH Telegram group. Saved replies for the top families with the exact commands. Start the weekly feedback digest to product.
Shrink the queue
Ship the KB entries the tickets asked for. Pre-release checklist with the dev team for the families that follow upgrades. First on-site onboarding if one comes up.
Noticed while building the demo from the public docs. Each is small; each is a place where a customer copying from the docs would get stuck.
| Where | What I saw (4 Oct 2026) | Suggested fix |
|---|---|---|
| Advanced FAQ, power-limit schedule | In the two introductory examples, 0 2 * * * is labelled "at 9:00PM" (it runs at 02:00), and the sed rewrites psu_power_limit = … to psu_target = …, so after the first run the other job's ^psu_power_limit pattern has nothing left to match. The combined example further down gets both right. | Align the introductory examples with the combined one (0 21, same key on both sides). |
| BOSminer API command list | asccount is described as "the details of a single ASC" and asc|N as "the number of ASCs". Upstream CGMiner's API-README has them the other way round. | Swap the two descriptions. |
| Academy top navigation | The Hashpower and Solo links point to braiins.com/hashpower/ and braiins.com/solo/, which redirect once and end on a 404 page. The products live at hashpower.braiins.com. | Point both links at hashpower.braiins.com. |
| bos-plus-api README | The reflection example lists braiins.bos.v1.TunerService, the pre-1.0 name. The changelog renamed it PerformanceService in 1.0.0-beta.2, and the current protos use the new name. | Refresh the example output. |
| Technical Resources tables | The S21 Pro+ (added in 26.07) and S21++ (26.08) are in the supported-devices and power tables but have no row in Default Temperatures. | Add the two rows. |
| Restricted mode wording | The API notes (1.15.0) describe restricted mode as full hashrate with the standard 2.5% dev fee. Licence terms §3.10 and the Licence FAQ describe reduced hashrate while the licence server is unreachable. The proto comments were corrected on 16 Sep 2026, so they are likely the current intent; a customer reading the FAQ still asks which applies. | One sentence in the Licence FAQ naming which applies when. |
Independent job-application homework from public information, collected 4 Oct 2026. The demo is my own prototype; it talks to no miner, contains no Braiins code beyond the public proto files it reads, and its tickets are written from documented failure modes.