{"id":"22ba0ef3-87ad-46c7-bbaf-6a420fc4a40e","arxiv_id":"2505.00303","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"Using 2021-2023 surplus solar data and Antminer S21 XP Hyd specifications, the authors estimate that Bitcoin mining on surplus power would have earned KEPCO roughly $294-348 million in 2023.","lead":"This paper asks whether South Korea's surplus solar electricity, left over after net metering, could profitably power Bitcoin miners, and estimates $294-348 million in annual profits for KEPCO. A generalist should care because it tests a concrete and controversial policy idea: using otherwise wasted renewable energy to mine Bitcoin while easing a state utility's debt.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Assumption 2 (24-hour surplus availability) is load-bearing: without storage, solar surplus is daylight-only, so miners run only ~6h/day, cutting revenue by ~4x while fixed depreciation costs stay constant, likely flipping the headline profits negative.","rationale":"The reader's weakest assumption is exactly the one I identify: Assumption 2's 24-hour availability. This is the most load-bearing concern because it directly scales the miner count and daily revenue, and it is physically implausible without storage. Unlike the price-forecast leakage, which affects only the predicted-price scenarios, the 24-hour assumption also undermines the 'Actual Price' baseline scenarios in Table 9, so it cannot be dismissed as a secondary issue. The paper itself notes a legal barrier in the KEPCO Act (Section 5) and lists excluded costs, but the 24-hour assumption is more fundamental: even if the legal issue were resolved and costs added, the profitability would collapse under a realistic solar availability profile. I agree with the reader's REJECT verdict and high confidence, because the optimal strategy under realistic conditions would require either substantial storage investment (not costed) or accepting a much smaller mining operation with far lower revenue. The concrete test I propose would settle the matter quantitatively; until then, the central claim is unsupported.","tokens_in":11281,"tokens_out":2863,"duration_ms":35667,"concrete_test":"Recompute Table 9 using an actual hourly solar generation profile for South Korea rather than Assumption 2. Obtain typical hourly PV output (e.g., from Korea Meteorological Administration solar irradiance or PVWatts for representative regions), scale it so monthly totals match the KEPCO surplus kWh, and set the miner fleet to the number that can run only during surplus hours (or add battery storage costs to enable 24h operation). Then recompute revenue and costs for all four cases in Table 9. If the resulting profit for Simulation 2 with Random Forest drops below zero (or even below $50M), the central claim that Bitcoin mining on surplus electricity yields significant revenue does not survive.","verdict_should_be":"REJECT","load_bearing_attack":"The central profitability claim in Section 5 and Table 9 rests on Assumption 2 in Section 3: \"The availability of surplus power is assumed to be 24 hours per day.\" This assumption enters directly into the miner count and revenue calculation. The monthly surplus electricity (kWh) is divided by 24 hours per day to estimate how many Antminer S21 XP Hyd units can operate; the daily revenue formula in Section 4.4 multiplies by 144 blocks/day, implicitly assuming every miner runs for all 24 hours. In reality, the surplus comes from household solar PV, which generates only during daylight hours (roughly 5–7 hours/day in South Korea, with seasonal and weather variation). The monthly net surplus kWh is an accounting residual after net metering, not a continuous 24/7 power supply. If miners can only operate during solar-producing hours and no storage is added, then daily operating hours fall to roughly 20–30% of the assumed 24 hours. Revenue scales with operating hours, but miner depreciation, networking, cooling, and maintenance costs do not. Recomputing Table 9 with, say, 6 hours/day of operation reduces the revenue in every case by a factor of ~4 (to roughly $75–100M), while the cost column stays at $41–62M; the profitability margin shrinks to near zero or negative before even adding excluded costs such as cooling, labor, networking, and profit sharing. Simulation 2's fixed 30,565 miners, based on a three-year average, would also be idle during low-solar months and at night, or would require drawing grid power at retail rates, which the paper does not price. Because the entire 'surplus electricity' premise is that the energy is otherwise wasted and free, this assumption is not a minor modeling choice; it is the mechanism that makes the multi-hundred-million-dollar profit figures plausible.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes using surplus electricity from household solar net-metering in South Korea to operate Bitcoin miners, with KEPCO as the assumed operator. The authors estimate monthly miner counts from three years of KEPCO surplus data, use the Antminer S21 XP Hyd specifications and network hash rates to compute expected Bitcoin mined per day, and generate 2023 Bitcoin price predictions with Random Forest and LSTM models. They compare 12-month profitability across two simulation fleets and report profits of $294–348M after depreciation, concluding that the strategy could generate significant revenue for KEPCO while alleviating its debt. The paper also acknowledges legal barriers in the KEPCO Act and lists several cost items that are excluded from the analysis.","tokens_in":11580,"tokens_out":11720,"duration_ms":129943,"significance":"If the quantitative analysis were sound, the paper would be one of the first empirical case studies of a national Bitcoin-mining strategy built on solar surplus, with clear policy relevance for KEPCO and for utilities in other countries with net-metering surpluses. The authors deserve credit for using public data, presenting an explicit cost table, including an actual-price baseline, and acknowledging the legal obstacles. However, the central profitability estimates are not currently supported: the model assumes 24-hour availability of a daylight-only resource, the price 'predictions' are based on features that contain the same-day target price, and the cost model excludes major cost categories while being labeled conservative. Because all three issues are load-bearing for the headline numbers, the current manuscript does not establish the feasibility claim it advances.","major_comments":[{"comment":"The assumption that surplus power is available 24 hours per day contradicts the data source described in the paper: the surplus is residual household solar generation after net metering, and the paper itself lists energy storage as a future research item. This assumption is directly load-bearing because the monthly surplus kWh are converted into an all-day miner count, and the daily revenue formula multiplies by 144 blocks per day. Under realistic daylight-only availability (roughly 5–7 hours per day), either the same fleet operates about one quarter of the day, cutting mining revenue in Table 9 by roughly 75% while depreciation costs remain, or the fleet size must be quadrupled to consume the same energy, quadrupling the capital-cost line while leaving revenue approximately unchanged. Neither case reproduces the reported profits, and the difference can plausibly turn the remaining profit negative once excluded operating costs are added. The analysis should use hourly surplus profiles and, if storage is assumed, include storage costs and round-trip losses.","section":"Section 3, Assumption 2; Sections 4.2.2, 4.4; Table 9"},{"comment":"The Bitcoin price 'predictions' are not genuine out-of-sample forecasts. The feature set in Table 4 includes Momentum with N = 1 (P_today minus the previous day's price), K%, D%, and RSI, all of which are computed using the current day's price P_today, which is exactly the target variable. The model can therefore recover the target from its own inputs, which explains the very high in-sample-style R² of 0.91 on the 2023 test set. This is label leakage, not forecasting skill, and it undermines the claim in Section 4.4 that the graphs indicate 'reliable forecasting.' The prediction exercise should be redone with strictly lagged features and a genuine recursive evaluation, or removed from the paper in favor of the actual-price baseline.","section":"Section 4.1.1–4.1.2; Table 4; Figure 5"},{"comment":"The cost model is not conservative, despite the claim in the conclusion. Table 8 excludes networking equipment, cooling systems, labor, maintenance and replacement costs, pool fees, facility costs, and any profit share or payment to the surplus electricity providers, and it also assumes zero electricity cost. Table 9 then reports profit as revenue minus only straight-line miner depreciation. For example, a fleet of 30,565 water-cooled miners would consume roughly 173 MW at full load, so the associated facility, cooling, networking, and staffing costs cannot be assumed to vanish. At minimum, these items should be quantified or, failing that, the results should be presented as upper bounds, not as a conservative estimate. The treatment of the surplus energy as free also needs an economic justification, since the electricity is generated by households under a net-metering arrangement.","section":"Section 4.3; Table 8; Section 5"},{"comment":"No sensitivity or robustness analysis is provided for the two most consequential assumptions, the 24-hour availability factor and the zero-cost electricity. Table 9 reports only point estimates, yet the profit figures scale essentially linearly with the assumed daily operating hours and are highly sensitive to the excluded cost items. A minimal acceptable analysis would vary the availability factor between realistic daylight bounds and add a best- and worst-case operating-cost estimate, because the current results give no indication of how fragile the $294–348M headline profit range is.","section":"Section 4.4 and Table 9"}],"minor_comments":[{"comment":"The text says 'For Simulation 1, using 40,439 miners,' but the reported cost of approximately $62M corresponds to 45,439 miners, not 40,439; the number in Table 7 is 45,439.","section":"Section 4.5, cost calculation"},{"comment":"The word 'tress' in the sentence about decision trees appears to be a typo for 'trees.'","section":"Section 2.1.2"},{"comment":"The labels '45.439' and '30.565' in the workflow figure use a decimal point where a thousands separator is intended; they should read '45,439' and '30,565.'","section":"Section 5 and Figure 12"},{"comment":"The exact formula used to convert monthly surplus kWh into the number of operational miners is not shown in the text, and no data or code availability statement is provided; this makes it difficult to audit the central calculation.","section":"Reproducibility"}],"recommendation":"reject","confidential_remarks":"The authors have identified a genuinely interesting applied problem, and the paper is transparent about many of its assumptions. However, the three principal flaws — the physically inconsistent 24-hour surplus assumption, the label leakage in the price-prediction exercise, and the incomplete cost model — are not local presentation issues; they invalidate the numerical profitability estimates that form the paper's main contribution. I would be willing to consider a substantially rewritten version that uses hourly solar generation profiles, storage costs where relevant, genuinely lagged predictors, and a complete operating-cost model, but the present analysis does not meet the bar for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nQuick take: the paper is an honest, clearly written application of a standard mining-profit calculation to South Korean household-solar surplus data. It is the first to combine KEPCO's net-metering surplus figures with Bitcoin network hash rate and a current miner's specs. That data combination is the real novelty. The authors also deserve credit for stating their assumptions plainly, listing excluded costs, and acknowledging that the KEPCO Act currently blocks this use of surplus power.\n\nBut the headline profitability, $294–348 million per year, rests on an assumption the paper itself states: surplus power is available 24 hours per day. It is not. The surplus comes from household solar, so it is daylight-only without storage. The paper uses monthly kWh, divides by 24 hours, and derives miner counts and daily revenue as if the machines run around the clock. If real operating time is closer to 6 hours/day, revenue falls by a factor of about four, while depreciation stays fixed. The profit numbers in Table 9 shrink to roughly $75–100 million in revenue against $41–62 million in cost, i.e., near zero or negative before accounting for the costs the paper explicitly excluded (cooling, networking, labor, provider compensation). This is not a minor modeling choice; it is the mechanism that makes the headline plausible.\n\nThe price-prediction section has a similar problem. The technical features (Momentum = P_today − P_{t−N}, K%, D%, RSI) are all constructed from the same day's price, so the 'predicted' price is mostly a transformation of the target. The reported R² of 0.91 on the test period does not demonstrate forecasting ability; it demonstrates in-sample smoothing. A true out-of-sample test would use lagged features only.\n\nThere are also small slips: Table 9's cost calculation uses 40,439 miners, while the text and Table 7 say 45,439. That looks like a typo, but it should be caught.\n\nOverall, the framework is clear and the data are real, but the central claim as stated is not supported. The paper is worth a serious referee, because the policy question is live and the dataset is unique, but it needs major revision: a realistic availability factor, explicit storage and grid-backup costs, and a genuine forecast evaluation. I would send it out rather than desk-reject, and the referee report should be specific about the availability assumption.\n\nFor your own work, I wouldn't cite it yet, but it could be a useful classroom example of load-bearing assumptions in energy economics.\n\nYours,","headline":"A transparent but physically unrealistic profitability estimate for Bitcoin mining on South Korean solar surplus; the 24-hour availability assumption alone likely flips the headline numbers.","tokens_in":12168,"tokens_out":2895,"would_cite":false,"duration_ms":28327,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A new study argues that South Korea's leftover solar electricity could mine $294–348 million in Bitcoin profit per year.","keywords":["surplus electricity","Bitcoin mining","net metering","KEPCO","Random Forest","LSTM","solar energy","profitability analysis"],"falsifier":"Use hourly or sub-daily solar surplus output from KEPCO's net-metered households (or the Jeju project) to compute how many hours per day the fleet can actually run; if the result is materially below 24 hours, the 30,565-miner base case and its $294 million profit figure collapse unless storage costs are added.","tokens_in":1495,"feed_emoji":"⚡","tokens_out":2565,"duration_ms":59678,"temperature":0.7,"pith_summary":"This paper tries to establish that surplus electricity from household solar systems, currently left unused after net metering, can be profitably redirected to Bitcoin mining. The authors calculate that operating 30,565 to 45,439 Antminer S21 XP Hyd units on that surplus would generate roughly $294 to $348 million in profit over 12 months after equipment depreciation. They argue this revenue stream could help Korea Electric Power Corporation (KEPCO) reduce its debt, minimize wasted energy, and resolve unsettled payment issues from net-metered solar households. The study is an applied empirical analysis: it uses actual surplus electricity data from 2021 to 2023, Bitcoin network hash rates, and machine-learning price forecasts to arrive at the profit estimates.","feed_headline":"Bitcoin mining on surplus solar could net $294M a year","feed_subtitle":"Simulating 30,565 miners on KEPCO's leftover solar power yields $294M-$348M in annual profit after depreciation.","key_machinery":"The load-bearing calculation is the daily mining revenue formula: revenue = predicted Bitcoin price × block reward (6.25 BTC) × (fleet hash rate / network hash rate) × 144 blocks per day. The fleet size is derived from monthly surplus electricity in kWh, reduced by a 3.59% transmission-loss rate, and divided by the power draw of the Antminer S21 XP Hyd (5,676 W). Bitcoin price enters through two machine-learning forecasts, Random Forest and LSTM, and hardware cost is handled through a 7.5-year straight-line depreciation schedule.","core_discovery":"The central claim is that otherwise-wasted solar surplus in South Korea can be converted into a significant national revenue stream through Bitcoin mining. The paper computes the daily Bitcoin harvest as the product of the block reward (6.25 BTC), the miner fleet's share of the total network hash rate, and 144 blocks per day, then multiplies that by the Bitcoin price predicted by Random Forest or LSTM models. Across four cases (two price models and two fleet sizes: a maximum of 45,439 miners and a fixed 30,565-miner fleet), the lowest 12-month profit is about $294 million and the highest is about $348 million, after straight-line depreciation of the mining hardware. The paper presents this as the first empirical feasibility study connecting South Korean electricity surplus with Bitcoin mining, and concludes that the revenue could materially help KEPCO's finances, provided legal barriers are lifted.","pith_inferences":["If the 24-hour surplus availability assumption is relaxed to daylight-only generation, the miner fleet would need storage or grid backup; adding storage costs would erode the reported margins.","The reported profits assume network hash rate and block reward stay at historical levels; a large rise in network hash rate or the April 2024 halving to 3.125 BTC per block would shrink the daily BTC harvest, so the 2023-based estimates likely overstate future revenue.","The same fleet-sizing framework could be stress-tested against hourly solar generation data or applied to wind and hydroelectric surplus; if profitable under those conditions, the national-strategy case would strengthen considerably."],"forward_implications":["If the estimate holds, KEPCO could earn roughly $24.5 million per month in the most conservative case (Random Forest price predictions with a fixed 30,565-miner fleet).","The revenue stream would convert currently wasted solar surplus into an asset that can be applied to KEPCO's debt and unpaid settlement obligations.","The strategy requires amending KEPCO Act Article 13, which currently restricts KEPCO to electricity sales, so legislative change is a precondition for implementation.","The approach is portable to other renewables and to any region with net-metering surplus, not just South Korea.","Both price-prediction models and both fleet sizes produce positive profit, so the conclusion is not sensitive to the choice of forecasting model or fleet configuration."],"supporting_citations":[{"why":"Provides the KEPCO surplus electricity data (2021-2023) that gives the kWh amounts used to size the miner fleet.","marker":"[16]"},{"why":"Supplies the Antminer S21 XP Hyd specifications, fixing each miner's hash rate and power draw.","marker":"[21]"},{"why":"Gives the 10-year transmission-loss average (3.59%) used to discount available power.","marker":"[14]"},{"why":"Supports the 10-minute block time that sets the daily block count at 144.","marker":"[20]"},{"why":"Provides the ASIC miner lifespan estimate (5-7 years, up to 10+) used for the 7.5-year depreciation schedule.","marker":"[13]"},{"why":"Defines the random forest regression method used as one Bitcoin price prediction model.","marker":"[17]"},{"why":"Defines the LSTM architecture used as the alternative price prediction model.","marker":"[18]"},{"why":"Supplies the technical indicator feature set (SMA, WMA, momentum, K%, D%, RSI) used for price prediction.","marker":"[11]"}],"fun_headline_variants":["Surplus solar in Korea can mine $294M in Bitcoin yearly","KEPCO's surplus solar power could mine $294M in Bitcoin annually","Turning surplus solar into Bitcoin: $294M yearly for South Korea","Surplus solar Bitcoin mining could earn South Korea $294M yearly"],"cache_read_input_tokens":14208,"weakest_assumption_plain":"The paper assumes surplus electricity is available to miners 24 hours a day, but household solar surplus is produced mainly in daylight and varies by season, so without storage the miner fleet cannot run around the clock.","fun_headline_variants_meta":{"raw":{"variants":["Surplus solar in Korea can mine $294M in Bitcoin yearly","KEPCO's surplus solar power could mine $294M in Bitcoin annually","Turning surplus solar into Bitcoin: $294M yearly for South Korea","Surplus solar Bitcoin mining could earn South Korea $294M yearly"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000675,"raw_usage":{"total_tokens":3089,"prompt_tokens":980,"completion_tokens":2109,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":596,"completion_tokens_details":{"reasoning_tokens":2032}},"tokens_in":596,"tokens_out":2109,"duration_ms":16935,"temperature":1.0,"reasoning_tokens":2032,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T04:46:24.164575+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Use hourly or sub-daily solar surplus output from KEPCO's net-metered households (or the Jeju project) to compute how many hours per day the fleet can actually run; if the result is materially below 24 hours, the 30,565-miner base case and its $294 million profit figure collapse unless storage costs are added.","supporting_citations":[{"cited_title":"Surplus electricity trade by administrative district (korean),","cited_arxiv_id":null,"evidence_quote":"Provides the KEPCO surplus electricity data (2021-2023) that gives the kWh amounts used to size the miner fleet."},{"cited_title":"Bitcoin miner s21 xp hyd","cited_arxiv_id":null,"evidence_quote":"Supplies the Antminer S21 XP Hyd specifications, fixing each miner's hash rate and power draw."},{"cited_title":"Annual power transmission and distribution loss of 1.7 trillion won... equivalent to a year of operation of 21 nuclear power plants (korean),","cited_arxiv_id":null,"evidence_quote":"Gives the 10-year transmission-loss average (3.59%) used to discount available power."},{"cited_title":"How to mine 1 Bitcoin daily: Understanding the concept of 10 minutes in bitcoin mining,","cited_arxiv_id":null,"evidence_quote":"Supports the 10-minute block time that sets the daily block count at 144."},{"cited_title":"Tips to expand asic miner lifes- pan,","cited_arxiv_id":null,"evidence_quote":"Provides the ASIC miner lifespan estimate (5-7 years, up to 10+) used for the 7.5-year depreciation schedule."}],"review_version":1}