<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Hash-Hut Blog]]></title><description><![CDATA[Hash-Hut Blog]]></description><link>https://hash-hut.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Sun, 22 Nov 2026 16:22:38 GMT</lastBuildDate><atom:link href="https://hash-hut.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Rejected and Stale Shares: Measuring What They Cost Your Farm]]></title><description><![CDATA[How stale, invalid and duplicate shares arise in Stratum, how to measure them per worker, and how to turn a reject rate into dollars per day.]]></description><link>https://hash-hut.hashnode.dev/rejected-and-stale-shares-measuring-what-they-cost-your-farm</link><guid isPermaLink="true">https://hash-hut.hashnode.dev/rejected-and-stale-shares-measuring-what-they-cost-your-farm</guid><category><![CDATA[Bitcoin]]></category><category><![CDATA[Cryptocurrency]]></category><category><![CDATA[networking]]></category><category><![CDATA[monitoring]]></category><category><![CDATA[Devops]]></category><dc:creator><![CDATA[hashhut]]></dc:creator><pubDate>Fri, 20 Nov 2026 19:57:26 GMT</pubDate><content:encoded><![CDATA[<p>Here is an example. A worker shows 3.1% rejected shares. Its neighbour on the same switch shows 0.04%. Both draw the same power, and both report the same local hashrate. The first one may be losing revenue, but a reject counter alone does not tell you how much. Depending on the reject types and the pool's credit rules, the loss can be smaller than the percentage, or even zero.</p>
<p>This article covers what the reject codes mean, how to measure the loss, and how to test a fix without fooling yourself.</p>
<h2>Learn the four outcomes of a share</h2>
<p>A share is a proof that the miner found a hash below the pool's easier target. The pool counts accepted shares to estimate your work and split rewards. A submitted share ends in one of these states:</p>
<ul>
<li><strong>Accepted:</strong> valid, on a current job, credited.</li>
<li><strong>Stale:</strong> valid work, but for a job the pool has already retired. Usually a new block arrived first. Whether a stale share is credited depends on the pool and the coin.</li>
<li><strong>Invalid (low difficulty):</strong> the hash does not meet the assigned share difficulty. This points at the machine, such as hardware errors or an unstable profile.</li>
<li><strong>Duplicate:</strong> the same nonce for the same job was submitted twice. This often points at a proxy or a retry. Duplicates are diagnostic, not a loss: they resubmit work that may already be credited.</li>
</ul>
<p>Many dashboards merge these into one "rejected" number. Split them first, because each has a different cause. On the wire, a reject looks like this (Stratum v1, error 21 is "job not found", the usual stale code):</p>
<pre><code class="language-json">{"id": 412, "result": null, "error": [21, "Job not found", null]}
</code></pre>
<h2>Why stale shares happen</h2>
<p>When a new block arrives, the pool sends a job with <code>clean_jobs</code> set, and older work is no longer valid. Any share your ASIC already sent, or was about to send, for the old job arrives late.</p>
<p>A simple model gives the scale. Assume shares are submitted at random times. The fraction that goes stale per block change is roughly the delay divided by the interval between clean-job events:</p>
<pre><code class="language-plaintext">stale fraction ≈ delay / interval between block changes
</code></pre>
<p>Suppose the delay (network latency plus the time the ASIC takes to switch jobs) is 0.3 s. With 600 s between blocks, that is 0.3 / 600 = 0.05%. On a coin with 60 s blocks, the same delay gives 0.3 / 60 = 0.5%. This is a simplified illustration, not a measurement. Your real figure depends on the pool, the coin and the firmware. The model covers only stale shares caused by block changes, so compare it with the stale-only counter and not with the total reject rate.</p>
<p>Two conclusions follow. Faster networks tolerate less latency, and delay variation matters as much as the average. A clean ping test can hide short stratum disconnects, so use reconnect counts and share timestamps instead.</p>
<h2>Match the pattern to the cause</h2>
<p><strong>Stales on a whole rack.</strong> Look at what the rack shares: a switch, an uplink, a route to the pool. Compare reject timestamps with reconnect events and packet loss. Test a second regional endpoint on a small group.</p>
<p><strong>Invalids on one machine.</strong> Check for hardware errors, a missing hashboard and a recent frequency or voltage change. Save the log and note the time of the last change. Then return that one worker to the stock profile and measure again.</p>
<p><strong>Duplicates after a change in the path.</strong> Compare a worker connected directly with one behind a proxy. Keep firmware, endpoint and profile identical, so the proxy is the only variable.</p>
<h2>Put a dollar figure on the reject rate</h2>
<p>The formula:</p>
<pre><code class="language-plaintext">daily loss = gross revenue at 100% credit × (unpaid, non-repeated, difficulty-weighted work ÷ total submitted work)
</code></pre>
<p>Only work the pool does not credit counts as loss. Duplicates do not count, and some pools credit certain stales. A dashboard percentage is a count of shares, not a share of paid work, so a 3.1% counter can mean a smaller loss, or none. The result is an estimate, and block luck swings real revenue around it.</p>
<p>Say a machine would earn $12/day gross at 100% credited work. This is an illustrative input, not a current market figure.</p>
<ul>
<li>At 3% uncredited work: 12 × 0.03 = $0.36/day, or $10.80 over 30 days per machine.</li>
<li>At 0.05%: 12 × 0.0005 = $0.006/day. Across 100 machines that is $0.60/day.</li>
</ul>
<p>Power does not change. Say the machine draws 3.5 kW. That is 84 kWh/day, which at $0.07/kWh costs $5.88/day. Margin before pool fee, cooling, hosting and other costs:</p>
<ul>
<li>0% rejects: 12 − 5.88 = $6.12/day</li>
<li>3% rejects: 11.64 − 5.88 = $5.76/day</li>
</ul>
<p>Now the unfavourable case. Say the same machine only earns $6/day gross because the market is weaker. At full credit the margin is 6 − 5.88 = $0.12/day. At 3% rejects it is 5.82 − 5.88 = −$0.06/day. A reject rate that looks small can flip a thin margin to negative.</p>
<p>Weight by difficulty, not by count. If the pool uses variable difficulty, ten rejected shares can carry far more work than ten accepted ones. Use the sum of assigned share difficulty when your pool exposes it.</p>
<h2>Test a fix without fooling yourself</h2>
<p>Block-based rewards swing with luck, so daily revenue is a poor test of a network fix. The reject ratio is counted in shares and settles quickly. Use two different measures:</p>
<ol>
<li><strong>Reject ratio:</strong> write down the criteria first, for example "difficulty-weighted rejects below 0.1% over 24 hours". Check that the window is long enough: expected rejects = shares × rate, and you should aim for at least dozens of rejects in the window. With 50,000 shares per day, 0.05% is 25 rejects. With far fewer shares, one or two rejects are noise. Compute your own share count per worker and lengthen the test if it is below that.</li>
<li><strong>Revenue:</strong> if you want to see the effect in credits, run the test for at least a week. Compare the same UTC windows, and look at the weak days, not only the average.</li>
</ol>
<p>Change one variable at a time. With two matched workers, run one on the old route and one on the new, then swap them after a day, because units differ. With a single ASIC, alternate old, new, old in comparable windows. If the gap follows the route, work on the network. If it follows the machine, start at the machine.</p>
<p>Roll out any automation (failover, auto-reboot, profile changes) on one worker or rack first. Set limits and cooldowns, and know how you will roll back.</p>
<h2>Where Hash-Hut fits</h2>
<p>The Hash-Hut dashboard gives a pool-side view of accepted and rejected work per worker. The split by reject type, the difficulty weighting and the reconnect log must come from the ASIC logs or your own export. Check which of these the dashboard shows before you plan a test around it.</p>
<p>Hash-Hut describes rejects on its own-node coins as low with healthy hardware and a stable connection. At the time of writing, compare this on your own workers rather than taking it on trust. The statement does not cover BTC and BCH, which go through partner pools, nor LTC and the coins merged with it.</p>
<p>The pool sees accepted work. It does not see temperatures, fans or hashboards, and it does not reboot or tune machines. For the machine side you need firmware tools or a farm monitor.</p>
<p>Hash-Hut Monitoring is self-hosted monitoring that runs on a server inside your farm network. It shows chip temperatures, missing hashboards, fans and PSU, and alerts when hashrate drops below a machine's own norm. That lets you put a hardware cause next to a reject spike. It does not measure reject codes. Auto-reboot is off until you enable it, works only on compatible machines, and is not a cure for rejects. Renewal is free while the farm mines on Hash-Hut with at least 5 PH/s SHA-256 or 250 GH/s Scrypt; check the page for current terms. <a href="https://monitoring.hash-hut.net/?utm_source=hashnode&amp;utm_medium=article&amp;utm_campaign=rejected-shares-and-stale-shares">Start the free 14-day trial of Hash-Hut Monitoring</a> on one rack and compare its hardware alerts with your reject spikes.</p>
<p>Sources: Bitcoin Wiki – Stratum mining protocol · Braiins – Stratum V1 documentation · Foreman – stale and rejected shares</p>
<hr />
<p><em>Disclosure: this article was written by the Hash-Hut team. We run the Hash-Hut mining pool and Hash-Hut Monitoring mentioned above.</em></p>
]]></content:encoded></item><item><title><![CDATA[Moving ASICs off Binance Pool: A Migration Checklist for BTC and LTC+DOGE]]></title><description><![CDATA[A runbook for repointing SHA-256 and Scrypt ASICs after the 10 November 2026 shutdown, from fleet audit to a hashrate check in your own worker statistics.]]></description><link>https://hash-hut.hashnode.dev/moving-asics-off-binance-pool-a-migration-checklist-for-btc-and-ltc-doge</link><guid isPermaLink="true">https://hash-hut.hashnode.dev/moving-asics-off-binance-pool-a-migration-checklist-for-btc-and-ltc-doge</guid><category><![CDATA[Bitcoin]]></category><category><![CDATA[litecoin]]></category><category><![CDATA[Cryptocurrency]]></category><category><![CDATA[Devops]]></category><category><![CDATA[monitoring]]></category><dc:creator><![CDATA[hashhut]]></dc:creator><pubDate>Fri, 20 Nov 2026 19:46:27 GMT</pubDate><content:encoded><![CDATA[<p>Binance Pool stopped working on 10 November 2026 at 07:00 UTC. ASICs still pointed at it have earned nothing for ten days. The fix is the same for 5 machines or 500: audit, repoint, name, verify. Do it in that order and do it in stages.</p>
<h2>Step 1: Audit the fleet before touching a machine</h2>
<p>Build one inventory file. It is the source of truth for every later step, including the recovery check.</p>
<pre><code class="language-csv">ip,model,algo,rack,slot,expected_hashrate,unit
10.0.3.12,Antminer S21,sha256,3,12,200,TH/s
10.0.4.7,Antminer L9,scrypt,4,7,16,GH/s
</code></pre>
<p>Both rows are illustrative. In the example, 200 TH/s per S21 is an assumed measured value. Replace it with your own measured figure for each miner, not the sticker value. Sum the expected hashrate per algorithm. Suppose a small farm has five S21-class machines at that assumed 200 TH/s each. The expected SHA-256 total is 5 × 200 = 1,000 TH/s, or 1 PH/s. You will compare this number with the pool dashboard later.</p>
<p>Mark every machine that was offline or in repair before 10 November. Otherwise you will chase "missing" hashrate that was never there. A network scan also helps find machines you forgot.</p>
<h2>Step 2: One stratum address per algorithm</h2>
<p>You need one address for SHA-256 and one for Scrypt. Pick the region closest to the farm. At the time of writing, Hash-Hut offers eu, asia, ru and by, and the full list is on the pool page.</p>
<pre><code class="language-text">SHA-256 pool URL: stratum+tcp://eu.pool.hash-hut.net:8888
Scrypt pool URL:  stratum+tcp://eu.pool.hash-hut.net:9999
Worker:           mylogin.r03-s12
Password:         (empty)
</code></pre>
<p>Replace <code>mylogin</code> with your Hash-Hut login. TLS is available on the same host and port with <code>stratum+ssl://</code>. Most ASICs have three pool slots. Remove the dead Binance entries from all three. Put your chosen region in slot 1. You can add a second region as backup with the same worker name, but it is not an independent pool. It will not help in a pool-wide, DNS or ISP failure.</p>
<p>Slot order, timeout and failback differ between firmware versions. On the pilot machine, check on your firmware which slot is primary, how long the timeout is and whether the miner returns to slot 1. Test failover on one worker only, with the old config saved.</p>
<p>Autoswitch is a pool-side feature set in the dashboard, not on the ASIC. For the migration, pin SHA-256 workers to BTC and Scrypt workers to LTC. Test autoswitch later, on one rack.</p>
<h2>Step 3: One wallet address per coin</h2>
<p>Hash-Hut does not convert coins at the moment. Each coin is paid to its own address. For this migration you need:</p>
<ul>
<li>A BTC address. Minimum payout at the time of writing: 0.0001 BTC.</li>
<li>An LTC address. Minimum: 0.01 LTC.</li>
<li>An address for each merged coin. At the time of writing, LTC is merged with DOGE, PEP, BELLS, LKY and DINGO. Each has its own balance, address and minimum. Check the LTC coin page for the current list.</li>
</ul>
<p>Enter an address for every coin you leave enabled. If you send payouts to an exchange deposit address, confirm that the exchange supports that network. Exchange and conversion costs are yours, and the pool pays the network transfer fee.</p>
<p>Keep three money states apart in your records: accrued, withdrawable, and paid to the wallet. A balance below the minimum is a delay, not a loss. Payouts run daily at 12:00 UTC once the minimum is reached.</p>
<h2>Step 4: Name workers so you can find them</h2>
<p>The worker name is <code>login.name</code>. Encode the physical location in <code>name</code>: rack and slot. When a machine drops, you should know which shelf to walk to. Stick to lowercase letters, digits and hyphens. That is a safe convention, not a pool requirement.</p>
<p>Generate the names from the inventory instead of typing them:</p>
<pre><code class="language-sh">awk -F, 'NR&gt;1 { printf "%s,%s,mylogin.r%02d-s%02d\n", $1, $3, $4, $5 }' inventory.csv
</code></pre>
<p>The output is <code>ip,algo,worker</code>. Keep it as <code>expected.csv</code>. One name per machine, never reused. Two machines sharing a name merge into one line in the statistics and hide a failure.</p>
<h2>Step 5: Roll out in three stages</h2>
<ol>
<li>One machine per algorithm. Leave it for a full UTC day.</li>
<li>One rack.</li>
<li>The rest.</li>
</ol>
<p>Keep the old config notes for every machine, so a rollback is a lookup. The pilot also confirms the first payout path, from accrual to a transaction in your wallet. That check needs a balance above the minimum. Estimate the days to the minimum from your own accruals. For example, if a single Scrypt pilot (assume one L9 at 16 GH/s) accrues 0.002 LTC a day, 0.01 / 0.002 = 5 days. Your figure will differ. Until the first transaction arrives, keep the balance recorded as accrued.</p>
<h2>Step 6: Check that hashrate has recovered</h2>
<p>A reading taken ten minutes after reconnecting is noisy. Pool-side hashrate is estimated from accepted shares, so use the 24-hour average from the first full UTC day. Three checks:</p>
<p><strong>Worker count.</strong> Copy the worker names from the dashboard worker list into <code>seen.txt</code>, one per line, and diff them against your list:</p>
<pre><code class="language-sh">cut -d, -f3 expected.csv | sort &gt; expected.txt
sort seen.txt &gt; seen_sorted.txt
comm -23 expected.txt seen_sorted.txt
</code></pre>
<p>Each name printed is a machine that is in your inventory but not reporting. Check power, cable and pool slot, in that order.</p>
<p><strong>Hashrate ratio.</strong></p>
<pre><code class="language-text">recovery % = pool 24 h average hashrate / sum of expected (measured) hashrate of online machines × 100
</code></pre>
<p>In the five-machine example, a dashboard average of 960 TH/s gives 960 / 1,000 = 96%. Set your tolerance before you look, for instance 95% for this farm. If a machine is far below the others, compare it with its rack neighbours first. A rack-wide gap points to network or power. A single-machine gap points to the machine.</p>
<p><strong>Reject ratio.</strong></p>
<pre><code class="language-text">reject % = rejected shares / (accepted + rejected shares) × 100
</code></pre>
<p>Hash-Hut publishes no reject guarantee. Measure your own fleet median per worker and compare routes on matched workers over the same UTC window. One worker with several times the fleet median has a hardware or connection problem.</p>
<p>The pool sees accepted work. It does not see chip temperatures, fans or hashboards. If a worker is online but weak, you need machine-side data. Hash-Hut Monitoring is one self-hosted option for that, and it is separate from the pool. If you enable auto-reboot or any other automation there, start on one rack, with a cooldown and a way to switch it off.</p>
<h2>Step 7: Compare earnings fairly</h2>
<p>Compare the same UTC days, not "yesterday versus today". BTC income through the pool is paid on its partner's scheme (FPPS), and LTC on PPS. Merged coins are paid on PPLNS through the partner. Compare each coin's accruals separately, and note that Binance Pool's own reward basis was different. Credits still vary with network conditions. Judge on a week or more, and look at the weakest day as well as the average.</p>
<p>The fee changes the result. In this example, 1 PH/s falls in the Worker grade, 2.5% at the time of writing. That grade ramps in over active days, so the first week may not be at 2.5%. The dashboard shows your current fee. Either exclude the ramp-up days from the comparison, or compare gross credits before the fee:</p>
<pre><code class="language-text">net = gross × (1 − fee)
</code></pre>
<p>On coins through partner pools, the 2.9% includes the partner's fee, so the total at higher grades sits slightly above the grade table.</p>
<h2>Run it</h2>
<p>Your inventory, expected names and tolerance are ready. Open <a href="https://hash-hut.net/binance-pool?utm_source=hashnode&amp;utm_medium=article&amp;utm_campaign=moving-asics-off-binance-pool-a-migration-checklist-for-btc">Hash-Hut's guide for Binance Pool miners</a>, check the current addresses and minimums, and repoint your pilot machines.</p>
<p>Sources: Hash-Hut – pool news and coin pages · Binance – Binance Pool closure notice</p>
<hr />
<p><em>Disclosure: this article was written by the Hash-Hut team. We run the Hash-Hut mining pool and Hash-Hut Monitoring mentioned above.</em></p>
]]></content:encoded></item></channel></rss>