Avalon Series – MM_STATUS Log Decoder

Open a terminal on any host on the same LAN as the miner. Pull the full `estats` reply: `echo -n '{"command":"estats"}' | nc <miner-ip> 4028`. Save the output to a timestamped text file named like `1246-$(date +%F).txt`. On Windows without `nc`, use PuTTY in raw mode to `<miner-ip>:4028` and paste `{"command":"estats"}` + Enter. The full reply is the diagnostic — web UI summaries hide most of the fields this page decodes. If you can't reach port 4028, the fault is network / AUC3 / MM-side — jump to the AUC3 network loss companion.

Read `SYSTEMSTATU` first. `SYSTEMSTATU[0] Work: 3` on a 3-board miner = all boards enumerated. `Work: 2` / `Work: 1` / `Work: 0` = enumeration failure — note which slot is missing and jump to the Avalon hashboard-not-detected companion for reseat procedure. `Hot-Recovery` in this field = firmware retried the chain before soft-shutdown, almost always thermal, move to step 5. `Error` in this field = pair with `ECHU` and `MM_STATUS`.

Scan `ECHU[a b c]` for non-zero entries. `ECHU[0 0 0]` = chains clean, move on. Non-zero on exactly one chain = swap that board to a different slot over the next 30 minutes and re-pull `estats`. Fault follows the board = bad board. Fault stays in the slot = AUC3 / ribbon / MM control path. Non-zero on all three chains simultaneously = AUC3 / MM path, not the boards themselves. This slot-swap test is the fastest hashboard-vs-controller isolation on any Avalon.

Read `ECMM[N]` and `MM_STATUS`. `ECMM[0]` + `MM_STATUS = WORK_MODE NORMAL` or `WORK_MODE IDLE` = MM layer healthy. Non-zero `ECMM` + `WORK_MODE FAULT` = MM control fault, proceed to firmware and AUC3 steps. `WORK_MODE FAULT` with clean `ECMM` but non-zero `ECHU` = chain-level fault, follow step 3. `MM_STATUS = Hot-Recovery` = firmware reduced work on the affected chain before soft-shutdown — thermal path.

Scan `PVT_T` array for any entry > 85 C. Shop-vac intake filter. Wipe grille. Verify intake ambient with an IR thermometer at the grille — not the middle of the room, not the hallway. Target ≤ 30 C for A11 / A12, ≤ 32 C for A13+. Above 35 C ambient, thermal fields will never read clean regardless of pad quality or airflow. Fix the room before you open the chassis. Intake dust alone resolves a meaningful percentage of `PVT_T` over-threshold entries.

Compare `MW0` / `MW1` / `MW2` across chains. Pull a second `estats` 10 minutes after the first. The three MW arrays should track each other within ±5%. One chain lagging 10-20% = that board has a thermal or voltage fault — pair with `PVT_T` / `PVT_V` on the lagging chain for isolation. Over 20% = hashboard approaching end-of-life or active throttle, move to Tier 3 pad refresh. All three chains lagging proportionally = pool / stratum issue or MM-wide throttle.

Query `{"command":"version"}` on port 4028 and compare your MM firmware build string against the last-known-good at `avalonminer.org/firmware-document/` (A1166 Pro: 20220926 family; A1246: 20230424 family; A13+: most recent stable release). If you're on a newer or older build, flash last-known-good over the web UI and soak-test 24 hours. Re-pull `estats` — `ECHU` and `ECMM` should clear if the fault was firmware-side. Do NOT interrupt mid-flash; bricked MM on A13+ is a bench recovery.

On A11 / A12 generations, reseat the AUC3 controller USB cable. Power off at the breaker first. Inspect pins for corrosion or blackening before reconnecting. A dab of dielectric grease on oxidised pins has cleared persistent non-zero `ECMM` + `ECHU[N N N]` patterns on multiple three-year-old A1246s in D-Central's repair queue. AUC3 handshake faults masquerade as chain-level faults until you prove them otherwise by reseating the controller.

Measure mains voltage under load at the PSU input with a multimeter. Standard Avalon PSUs expect 200-240 V and tolerate down to 195 V briefly. Sag below 195 V during hash bursts trips `PS` bitmap bit 1 (`Input_UV`) and cascades into `ECHU` faults across all three chains simultaneously. Dedicated 240 V circuit strongly preferred over 120 V for Avalon-class power draw — 120 V sags at peak demand and produces multi-chain `ECHU` patterns that look like hashboard faults.

Decode the `PS` error bitmap bit by bit. Sum of set bits: 1 (`Input_UV`), 2 (`OT1`), 4 (`OT2`), 8 (`OT3`), 16 (`OC_Pri`), 32 (`UV_out`), 64 (`OC_out`), 128 (`CS_error`), 256 (`OC_IOSA`), 512 (`OC_IOSB`), 1024 (`OC_IOSC`), 2048 (`FAN_error`). Work every set bit — `PS error 2049` = bits 1 + 2048 = mains sag + PSU fan failure, which is a different repair path than either bit alone. A single-bit interpretation misses the real fault pattern.

If `PS[0..2]` all read 0, reseat the PSU comm cable (A11 / A12 external PSU) or the internal control-to-PSU harness (A13+ integrated designs). Watch `estats` for 30 minutes — intermittent zero = loose connector, permanent zero = dead AUC3 or dead PSU comm silicon. Cross-brand PSU (e.g., Bitmain APW wired to a 1246) will show permanent `PS[0..2] = 0` because the comm protocol is Canaan-proprietary — use a Canaan PSU from the A1056..A1266 family.

Set DNS to `1.1.1.1` + `8.8.8.8` in the miner's network config. Drop Canaan's default `114.114.114.114` Chinese resolver — it hits intermittent latency from North American ISPs and produces `GHSavg` lag cadences that look like chain faults but clear instantly with a DNS swap. Swap your primary stratum pool for 30 minutes to a known-good mirror. If `GHSavg` recovers on the new DNS + pool, the original lag was pool-side, not silicon.

Refresh thermal pads on the hashboard whose `PVT_T` array reads highest. Arctic TP-3 or equivalent, 1.5 mm pad thickness matched to original. IPA 99% to clean old residue. An 8-12 C drop in `PVT_T` is typical on a three-year-old A11 / A12 and directly resolves thermal-adjacent `ECHU` and `MM_STATUS = Hot-Recovery` patterns. Newer A13+ hashboards use tighter pad specs — match exactly, don't over-compress, match the specific hashboard revision not just the model.

Walk the `PVT_V` array for the problem chain. Chips on the same board should track within ±20 mV. Drift beyond ±40 mV = domain voltage fault. Cross-reference `PVT_V` with `PVT_T` at the same chip position — hot chip + drifted voltage at the same chip position is chip degradation, not a general PSU fault. Mark those chip positions for reflow or replacement. `PVT_V` drift is the chip-level early-warning; log it weekly and you catch degradation six months before thermal failure.

Swap hashboards between slots to isolate fault source. Label slots 0/1/2 with tape. Move the suspect board to a known-good slot. Re-pull `estats`. `ECHU` fault follows the board = bad board, reflow or chip-level. Fault stays in the slot = AUC3 / ribbon / MM control path. This 30-minute physical test separates hashboard faults from controller faults faster than any software diagnostic, and saves you the shipping fee on a board that was never the problem.

Inspect MM control board and AUC3 controller under a loupe for scorched traces, bulging electrolytics near hashboard LDOs, corroded USB pins (A11 / A12), or cracked MLCCs near the PMIC. Canadian-garage operation through two winters of thermal cycling fatigues solder joints and dries electrolytics; visible damage = Tier 4, not DIY. Photograph the damage for the D-Central repair ticket. Don't attempt reflow on MM or AUC3 without test-fixture capability.

Stop DIY when: firmware rollback + AUC3 swap + PSU test all fail, `PVT_T > 85 C` persists after full pad refresh, `ECMM` stays non-zero across two different MM builds, or visible damage on MM / AUC3. You're in test-fixture territory. Book a D-Central Avalon repair slot at https://d-central.tech/services/asic-repair/ — 5-10 business day turnaround, Canadian workshop, ships Canada / US / international. Include the pre-fix `estats` snapshot in the ticket.

Ship with the pre-fix `estats` snapshot included. Anti-static bags, double-box with ≥5 cm foam on every side. Include a note listing: every observed non-zero field and its value, reboot cadence if relevant, MM firmware version string, ambient temp, and your contact info. D-Central bench process uses the pre-fix snapshot to skip the first 30-60 minutes of diagnostic, which saves you repair dollars. Pack for drop: the courier will drop your box at least once — assume it.