Method — every formula, every source, every limit

You don't need to understand this page to use NetPerHash. This page exists so the formulas, assumptions and sources behind the calculations can be independently checked — by you, or by anyone.

The calculation source

One engine computes everything: mining_calc.py (Python, the reference) and its byte-parity JavaScript port engine.js that runs in your browser. Both are tested by automated suites — 28 unit tests (Python), a cross-language parity test, and a browser-logic test, all passing 2026-08-30 — so the page cannot show a number the tested model would not produce.

Formulas (each with its source)

#	Quantity	Formula	Source (retrieved 2026-08-30)
1	Network hashrate	`net_TH = D · 2³² / 600 / 10¹²`	Bitcoin protocol (difficulty definition)
2	Gross reward	`gross_BTC = TH/net_TH · 144 · block_reward`	Vendor docs, Miners/About + net-reward formula
3	Electricity cost	`W/TH · TH · $/kWh · 24 / 1000`	Vendor docs, Energy efficiency — matches their worked examples to the cent
4	Service cost	`$0.0089 · TH` per day	Vendor docs, Miners/About ("service charge for 1 TH per day is $0.0089")
5	Maintenance	`(elec + service) · (1 − discount)`	Vendor docs, Maintenance fees
6	Net	`net = gross − maintenance`, daily floor at zero (negative-rewards protection)	Vendor docs + rewards page
7	Simple payback	`days = cost / net_per_day`	definition, not a forecast
8	Difficulty path	average factor `ln(1+g)/g` for linear growth g	exact integral; hand-check g=0.2 → 0.91161 in the test suite
9	Fee repricing	fees are USD-fixed but deducted in BTC at the daily price → over a price path the BTC fee cost scales by `ln(1+c)/c`	Vendor docs ("recalculated daily at the BTC rate"); validated against account behavior (BTC +23% → net-only +2% across one difficulty step)
10	Power growth	linear power growth p multiplies gross and per-TH fees by `1 + p/2` on average	observed: account grew 807.06 → 810.13 TH by reinvestment (2026-08-30)
11	Resale value	user-entered % of purchase price, added as a price-independent USD term	no public default; marketplace prices need an account (checked 2026-08-30); 5% seller fee documented (Academy)

The validation that gates this site

365-day day-by-day replication of a real account (2025-08-30 → 2026-08-29): using only public inputs — per-epoch network difficulty, the account's own daily BTC price, the documented $0.05–0.07/kWh range (account line implies $0.05), and the account's actual discount (28.26%) — the engine reproduced the account's daily gross within ±2% on 329 of 357 days over the full year (mean +0.03%, worst outliers ±12.8%, all at difficulty boundaries where the platform lags a retarget by 1–2 days). The farm size was reconstructed day-by-day from the data itself (687 → 812 TH, including an observable purchase step on 2026-07-27) instead of a fitted growth curve. Account data: official read-only API, 357 days retrieved 2026-08-30; full method and scripts: calculator/validate_365d.py; day-by-day table on the Log page.

Two lessons from the calibration are documented as warnings rather than hidden: vendor dashboards can show gross where you expect net (it cost us one wrong "premium" hypothesis), and payout schedules can lag public network events by days.

Deliberate model limits (each is a choice, not an oversight)

Linear difficulty growth, averaged analytically. Real difficulty retargets every ~2 weeks; the model applies the exact average of a straight-line path instead of a step simulation: a growth rate of g over the horizon multiplies average output by ln(1+g)/g — hand-checkable: g = 0.2 → 0.911607784, verified in the test suite. The calculator states this in its assumptions box.
Fees are USD-fixed but deducted in Bitcoin at the daily rate. That is the platform's documented behaviour, and it matters: on a rising price path the same dollar fee costs fewer Bitcoin, so mined net BTC differs between the three price scenarios. The model integrates the documented daily repricing exactly (ln(1+c)/c along an exponential price path). Cross-check against the account: BTC +23% at constant gross should lift net ≈14%; the account showed +2% across that window at constant gross — the remainder was gross-side movement, which the model tracks separately.
Fixed BTC price within each scenario. Only start and end price matter for valuation; the path between them is not modelled (the fee repricing in limit 2 uses an exponential path as its documented-behaviour approximation).
The discount is one pre-combined number. Four documented pieces can stack (token payment ≤20%, VIP ≤6%, service streak ≤3%, mining-mode bonus), but the platform does not publish how they combine — so the model never stacks them itself; the user enters one observed or assumed total (capped at the documented 29% ceiling of the pieces).
No taxes, no transaction fees, no hidden compounding. Reinvestment is an explicit lever (power growth), never assumed silently; taxes depend on the reader's jurisdiction and are out of scope.
Resale value is a user assumption, not a model output. The model will not predict what a tokenized miner sells for after a year — the reader enters a percentage of purchase price (with the documented 5% seller fee folded in). The sensitivity analysis behind this choice (including r*, the resale fraction at which mining ties holding — 92% of purchase price at this calculator's default inputs over 12 months) lives in the project's research and in the Calculator page's worked example; the engines treat the resale fraction as what it is, the reader's own exit-price assumption.
The April 2028 halving (estimated) is never auto-applied. A 24-month horizon spans it; the UI flags it and lets the reader set the block reward explicitly instead of silently halving their result.

What the platform attributes elsewhere — the day-table residual, documented

On the real day shown on the Explained page, the four visible lines do not sum exactly: 94.11 − 31.05 − 15.39 = 47.67 GMT, while the payout line shows 46.60 GMT. The difference, 1.07 GMT (about $0.36 at that day's token price), is attributed by the platform somewhere else that day; its internal accounting does not publish the booking. This site documents the residual instead of smoothing it away, and draws three boundaries around it:

The engines compute net as gross − maintenance (formula 6) and claim nothing about that residual.
The 63-day validation (above) runs against the gross line, which the residual does not touch — the ±2% result therefore stands independently of it.
Net-side figures on the site inherit it as noise of roughly $0.36 per day on an 810 TH account — an order of magnitude below every difference this site discusses.

A cautionary tale, kept visible on purpose. For several weeks this project believed the platform's pool paid a fixed uplift over pure network physics — a factor somewhere between 1.3× and 2.3× depending on assumptions (the exact value moving between attempts was itself the warning sign). A calibration constant to that effect was drafted. The cause was mundane: the account's rewards endpoint returns gross, before fees, and it had been compared against a net model. Compared like-for-like, the physics-only engine reproduced the account day-by-day with no calibration factor at all, and the constant was deleted from the engine (the audit trail of the wrong hypothesis remains in the repository: calibrate.py, compare_real.py). Lesson for every number on this site: when a model disagrees with a platform, check the unit and the fee basis before inventing a fudge factor — and vendor dashboards can show gross where you assumed net.

Reference-account figures used on this site

Every account-derived number on this site comes from exactly one day of one real account (retrieved 2026-08-30 via the official read-only API) plus one 63-day daily series (2026-06-28 → 2026-08-29). This table is the complete inventory — if a figure is not here, it is not account-derived:

Figure	Value	Used on	Basis
Gross mining reward, that day	94.11 GMT = $31.64	Explained	account daily table
Electricity line (discounted)	−31.05 GMT = −$10.44	Explained	account daily table
Service line (discounted)	−15.39 GMT = −$5.17	Explained	account daily table
Net payout, that day	46.60 GMT = $15.67	Explained	account daily table
Day-table residual	1.07 GMT ≈ $0.36	Explained, Method	arithmetic between the lines above
Token price that day	GMT = $0.3362	Explained	account's own daily price
Share size that day	810.13 TH	Explained	account miner list
Purchase price of TH	$9.99 per TH	Home, Explained	vendor pricing (all collections identical)
Combined maintenance discount	28.26% (VIP portion 4.2%)	Home, Explained, Compare	account fee breakdown
Implied electricity rate	$0.05 per kWh	Method (validation box)	account fee lines vs documented formulas
Weighted efficiency of the farm	14.95 W/TH	Method (validation box)	account per-miner data
Power trajectory over 63 days	744 → 816 TH	Calculator	implied series, lag-corrected; = 100% reinvestment (superseded by the 365-day series: 687 → 812 TH)
Validation result	329 of 357 days within ±2% (365-day run)	Home, Log	model vs account, day by day
GoBTC Pay reward, that day	0.00000006 BTC ≈ $0.005	Explained	account daily table (new app column, observed 2026-08-31)

Rules: account data appears only with per-figure approval by the account owner; no wallet addresses, deposit addresses or balances anywhere on this site; public figures (fee caps, discount caps, VIP rules, documented rates) carry the vendor's documentation as source with their own retrieval date. Secondary-market observation used once for context: 9,511 active listings, cheapest 28.49 GMT/TH (≈ $9.58/TH at that day's token price), retrieved 2026-08-30 with an account session. GoBTC Pay mechanics claimed from the vendor's launch materials (0.2% merchant fee, split evenly between confirming miners and the paying wallet's provider — press release via Chainwire, May 8 2026; vendor blog gomining.com/blog/gobtcpay, retrieved 2026-08-31); its payout size to tokenized hashrate is not documented by the vendor and is taken as zero by this site's engines — the one observed day (account daily table, 2026-08-31: 0.00000006 BTC on 810 TH) is shown on the Explained page for what it is: a real column, a negligible amount.

The efficiency treadmill — the hidden operating cost this model does not compute

A miner's efficiency (W/TH) is not fixed by purchase. Older miners shipped at up to 50 W/TH; the platform's best tier today is 12 W/TH ("Why 12 W/TH Matters Most in a Bear Market", gomining.com/blog/12-w-th, published Jun 16 2026: new miners "starting from $21.99/TH", upgrades "from $2.67/TH per step"). Electricity is charged per watt, so the W/TH tier sets a maintenance floor no discount can remove. The vendor's own docs put the electricity rate at $0.05–$0.07 per kWh (docs.gomining.com, Energy efficiency, retrieved 2026-08-31).

At the current network difficulty and this site's reference BTC price, the resulting margin per TH looks like this (electricity at $0.06/kWh mid-range, service fee $0.0089/TH/day as used site-wide, gross $0.0391/TH/day = the $3.91/day that 100.1 TH net at today's difficulty and price):

Efficiency	Total maintenance / TH / day	Margin, 0% discount	Margin, 28.26% discount
50 W/TH (old Greedy tier)	$0.0809	0.48× — negative	0.67× — negative
40 W/TH	$0.0665	0.59× — negative	0.82× — negative
30 W/TH	$0.0521	0.75× — negative	1.05× — knife's edge
20 W/TH	$0.0377	1.04×	1.45×
15 W/TH (site default)	$0.0305	1.28×	1.79×
12 W/TH (platform best)	$0.0262	1.49×	2.08×

Read this as a treadmill, not a ladder: the gross line falls as network difficulty grows (this site models +20%/year), while the maintenance line per TH stands still. A 50 W/TH miner that was profitable when it was bought drops below its electricity bill within roughly two to four years at modeled difficulty growth — before the BTC price even enters. The platform itself has retired power upgrades on old tiers and makes efficiency upgrades available "up to the best energy efficiency currently available on the platform" (docs.gomining.com, Energy efficiency). Each step costs money — "from $2.67/TH per step" per the vendor blog — so staying alive is a recurring capital expense this site's model does not add to its payback figures.

For scale, the economics of one step at $0.06/kWh: moving 38 W/TH (say 50 → 12) saves $0.0547/TH/day, so a single $2.67/TH step would pay back in ~49 days — while a 1 W/TH step saves only $0.0014/TH/day, ~5.1 years at the same step price. The blog documents only the minimum step price; the real multi-step path from 50 to 12 W/TH is not published. The honest summary: upgrading from an old tier is usually necessary, sometimes cheap, never free — and a calculator that omits it overstates every old miner's return.

Sources: docs.gomining.com/en/product/miners/energy-efficiency (electricity rate range, upgrade mechanics, retrieval 2026-08-31); gomining.com/blog/12-w-th (tier prices, per-step price, Jun 16 2026). Margin table computed by this site's formulas at press-date difficulty and price — it moves when they move.

Engine parity policy

One calculation, two implementations, zero tolerance for drift: mining_calc.py (Python) is the single reference; engine.js (JavaScript) runs the same formulas in your browser. Neither may show a number the other would not. Enforcement, all running against the current build (2026-08-30):

Unit tests: 28 Python tests cover every formula (formula-by-formula), the hand-checkable cases (g = 0.2 → 0.911607784, docs' electricity examples to the cent, negative-reward clamp, break-even identity), the parameter policy (missing required input → error, never a silent default) and the scenario identities (mined BTC identical across price paths except for fee repricing; holding = cost ÷ start price).
Cross-language parity test: JS and Python must agree bit-for-bit on the same inputs — the test fails on any rounding difference.
Browser self-test: the calculator and home pages run visible self-tests in your browser using the same engine code that produced the results above the test — if the page computes, the test computes with it; a red test means do not trust the page.

When an engine changes, the parity test gates the change; when this page's formulas change, the tests and this page change in the same commit, and the verification date above moves.

Page state: 2026-08-30.