The unit of work: a receipt for a settled problem, and what it is not

1 · The unit of work is a receipt, and most of what the records say about the economy went against us

What a visitor can hold today. A gallery piece computed from a recorded engine state and marked with our lattice (18 of the 22 pieces in its register approved, 3 held in review); a paper figure the engine drew from a committed record; a file of the Hilbert library under CC BY 4.0 with its sha256 (96 spaces published, 3 withheld); a priced solve whose certificate re-verifies offline at no cost; and a consult. Each priced call leaves one line on a hash-chained ledger.

The question. What is the unit of that economy, how do the records measure it, how would it change, and can a unit of work be handed to a person as a real-world receipt? We answer only from objects the platform already issues (a posting, a certificate, a serial, a watermark) and instruments the records already hold, and we print what those instruments refuted before what they held. The economics rests on one postulate, the entropy reward, which is not a result.

What this paper contributes, losses first, each with its basis and sections.

• (1) Reading, with theorems in the model. The unit of work and its receipt: one settle, its certified bound and gap, the reduction at its scale and the tokens metered, on one balanced, hash-chained posting with one serial, issued to a named recipient (equations 1 to 9). Balance, the chain at the index and the telescoping of rewards are theorems in the model (sections 2 and 3).
• (2) Measured, exact, one run. On the pinned instance the reduction rises with the energy: 2.82609926 bits at the maximum, 2.17392016, 1.26215001 (the pinned claim) and 1.04475698 at the three ground states; the Boltzmann comparator removes 0.504704 bits; at unit scale a ground state books 0.275999735 (sections 2 and 4).
• (3) Measured, the central result. The six economies' co-movement ratios over 340 states: F-E1 and F-E2 falsified, F-E3 held (section 4).
• (4) Measured. The loss register: the loss account, the account's direction, the refutations of the ledger, the reward and the broker, shadow prices, the unattributed share, self-dealing accepted (section 4).
• (5) Measured; the constants are choices. The meter as it is, its price printed beside the measured cost and never confused with it (sections 2 and 4).
• (6) Reading. The receipt and a commodity option on nine axes, renewable-energy certificates as the closest analogue, the Howey prongs answered on a reading; a recommendation, not a ruling (sections 3 and 9).
• (7) Reading. The opening policy, its five stages and the two measurements that would move it (section 6).
• (8) Prediction, registered, not run. Memes and gifs as objects of the same economy: three tests (section 7).
• (9) Prediction, registered, not run. The operator of the seven papers, published free of charge with a serial, a watermark and an offline verifier: four tests (section 8).

No priority claim is made.

2 · The model: a solve books a reduction, not a value

Assumptions. A1. A problem is n two-way choices sᵢ ∈ {−1, +1} with a field hᵢ on each and a pull Jᵢⱼ on each pair, scored by equation 1; nothing on this path is quantum. A2. A solve returns an assignment s∗, its energy, a certified lower bound B and the bound's kind; the gap g is "the size of what is not known", never a distance to a known optimum; the solve searches every assignment exactly up to n = 20, and a re-solve of a recorded certificate, by our audit or by the open-source SDK on the buyer's machine, searches every assignment up to n = 18. A3. R is the ledger's local decisiveness, in the code's words "the mean binary entropy of each variable's flip probability under its local field at T=1", taken as one minus that entropy per choice and summed. It depends on the assignment and is neither the Boltzmann joint entropy nor a mutual information; the Boltzmann comparator n − H(p_B) is printed beside it (Jaynes 1957). A4. The reward is r = H(stateₙ) − H(stateₘ) in bits (Shannon 1948); the ledger books ΔH = H_after − H_before, so the reward is −ΔH.

A5, physics used and not measured. Erasing a bit dissipates at least k_B T ln 2 (Landauer 1961), and a computation's cost lies in its erasures (Bennett 1982); the bits a settle erases are not R and have not been counted. No joule is booked on any posting: the one measured joule in the tree is GPU board power times wall time per solve on one device class. A6. Four characters a token, 1.5 tokens a solve unit, 200 raster cells a token, $0.02 per 1,000 tokens, at least one token a delivered call: each a choice. A7. A posting balances by construction, is hash-chained, and carries the amount v in its book's unit, ΔH in bits and ΔF in the producer's units.

The equations, each what the code in the footnotes computes; symbols in section 12.

(1)  E(s) = Σᵢ hᵢsᵢ + Σ_{i<j} Jᵢⱼsᵢsⱼ,  s ∈ {−1,+1}ⁿ, on (h, J) as submitted;   S = maxᵢ(|hᵢ| + Σⱼ|Jᵢⱼ|)
(2)  E∗ = min_s E(s);   B ≤ E∗;   g = E(s∗) − B ≥ 0;   kind ∈ {exact_enumeration, tower_lower_bound, not_computed}
(3)  ℓᵢ = hᵢ + Σ_{j≠i} Jᵢⱼsⱼ;   ΔEᵢ = −2sᵢℓᵢ;   pᵢ = 1/(1 + e^{ΔEᵢ})   at T = 1
(4)  R = Σᵢ [1 − H₂(pᵢ)]   the reduction, at T = 1, on the instance as submitted
(5)  R_S = R on (h/S, J/S);   n − H(p_B), p_B(s) = e^{−E(s)}/Z   the Boltzmann comparator
(6)  H(p) = −Σ p log₂ p;   r = H(stateₙ) − H(stateₘ) = −ΔH
(7)  P = (book, dr, cr, v, ΔH, ΔF, unit, t, h_prev, h);   h = sha256(canonical(P) ‖ h_prev);   Σ_dr v = Σ_cr v
(8)  τ = round(c/4 + 1.5u + b/4 + r/4 + k/200), ≥ 1 when delivered;   Π = τ × $0.02/1,000;   refused: τ = 0
(9)  u = (I, s∗, c, R, S, τ),  I = (n, h, J),  c = (B, kind, g);   REC(u) = (SN, recipient, sha256(canonical(u)), h, u)
(10) Iₖ(t), k = 1…6, each in its own unit;   distinct(k) ⇔ its own instrument ∧ Iₖ unlike every other index
(11) xₖ = Σ_{t: ΔIₖ>0} max(0, ΔIₖ₊₁) / Σ_{t: ΔIₖ>0} ΔIₖ   the co-movement ratio of adjacent indices
(12) O′ = clip₀¹(O + ω·g_O·c_O·(1 + π))
(13) κ(o) = |gzip(b(o))| / |b(o)|;   w(o; q) = P[settle(corrupt(o, q)) = o]
(14) f_κ = Σ_{i∈κ} hᵢs∗ᵢ;   J̃_κλ = Σ_{i∈κ, j∈λ} J×ᵢⱼs∗ᵢs∗ⱼ;   E₂(σ) = Σ_κ f_κσ_κ + Σ_{κ<λ} J̃_κλσ_κσ_λ,  σ ∈ {−1,+1}⁷
(15) gate(I, s∗) ⇔ n_J ≥ n − 1 ∧ the best single-flip ΔE at s∗ ≥ 0

The paper defines no economic value. R is what the ledger books: the chain posts it as the amount on its entropy book. The records fix what R is: on the pinned instance it orders the arrangements as their energies do, so the worst books the most bits; it is not scale-invariant; it is not correctness (section 4); it is not usefulness (a self-chosen instance books 7.9999995 bits); it is not joules. Nothing else is called value: the amount v, the price Π, the cost, the values index and the account's ratio λ, of which its own code says no energy or sustainability claim may be made, each keep their own names.

A worked instance, exact. The pinned instance is h = (0.5, 0.5, 0.5) with a pull of 1 on each pair, S = 2.5. (+, +, +) books R = 2.82609926 at E = 4.5; (−, −, −) 2.17392016 at 1.5; each two-up arrangement, the pinned claim (−1, +1, +1) among them, 1.26215001 at −0.5, not a local minimum (one flip lowers it by 1.0); each one-up arrangement, a ground state, 1.04475698 at E∗ = −1.5. So B = −1.5 and the claim's gap is 1.0. Boltzmann: Z = 18.62547, H = 2.495296 bits, n − H = 0.504704. At unit scale a ground state books 0.275999735 and the claim 0.52950848. The claim's receipt would print E(s∗) = −0.5, B = −1.5, g = 1.0, R = 1.26215001, S = 2.5: the chain books the bits, and the gap shows the solve had not reached the bottom.

The price is not the cost. Measured on 17 September 2026 and quoted as recorded, not re-derived: a question's median compute cost, over 133 questions, sits about 300 times below the rate, and a decision takes about 0.0013 vCPU-seconds. Neither is printed in dollars here: both are compute times Cloud Run's vCPU price, which the rate card says to verify. The 300 is a ratio, not a margin; no demand has been measured.

3 · The receipt certifies one delivered solve and is never traded; a commodity needs a unit that does not exist

What one receipt certifies, and its lifecycle. The settled state with its bound and gap, the reduction at its scale, the tokens metered, the posting's hash, the serial and the recipient; never joules. A receipt is issued to the party named at order time: the payer, a recipient the payer names, or anyone when the price is zero. It is redeemable once, by that party, for delivery of the one result it names, and is retired on delivery. It is never assigned, sold or exchanged, by anyone, and we run no market in it. A receipt confers no right to any future good, service, credit or payment and is not redeemable for money; it records that one solve was delivered and consumed, and carries no liability on the issuer's books. A gift is a solve bought at a price of zero, or paid for by one party for a recipient named at order time.

Its mechanics, a design not yet built. The store issues one serial per subject as a zero-amount transition; a receipt needs a status the store lacks, retired, beside active, revoked and expired, and the recipient recorded at issue. The chain it sits on is tamper-evident at the index of any edit, not against its keeper (PX-06), and needs an outside anchor (Haber and Stornetta 1991).

The four prongs of SEC v. W. J. Howey Co. (1946), answered on a reading. Counsel rules.

On this reading the receipt does not meet prongs two, three or four, and prong one is a purchase of a consumed service. Nothing on this page is legal advice. The tree's own posture agrees: one-way, non-transferable, nothing paid for keeping; a later ruling retired every token-denominated unit. An earlier note called the unit fungible and transferable, which contradicts that posture and goes to counsel unresolved.

The commodity option, defined and not offered. A standardised, transferable unit of performed computation would be closer to a commodity than to a security if its worth came from consumption, buyers expected no profit from resale, and it were not marketed as an investment. No such standard unit exists: every solve differs in n, scale, bound kind and gap, and the reduction is not scale-invariant. A transferable unit without a standard is not a commodity; it is a transferable token, where the securities question begins. The Commodity Exchange Act counts "all services, rights, and interests … in which contracts for future delivery are presently or in the future dealt in"; the CFTC has treated virtual currencies as commodities (CFTC v. McDonnell 2018); one asset can be a commodity and a security at once; and a transferable unit redeemable for money raises money transmission (Financial Crimes Enforcement Network 2019). The option needs a standard unit on every certificate (R at unit scale with its bound kind and gap, which no door prints), a registry with vintage and retirement, metering, no description of a unit as gaining value, and no market run by us. The service is priced per metered unit; the receipt records that a unit was consumed and is not itself a commodity.

The comparison, on nine axes (a reading).

The recommendation. Issue the receipt: one serial per delivered solve, issued to the named recipient, retired on delivery, carrying the reduction at its scale, the gap, the bound kind and the metered units; never assigned; nothing accrues for keeping one; no market. The commodity option stays a question for counsel, contingent on a standard unit that does not exist. Counsel must rule on: (a) whether a non-transferable, retired receipt for delivered computation is neither a security nor stored value; (b) the zero-price and named-recipient paths; (c) prepaid credit as closed-loop prepaid access; (d) whether a transferable standardised unit of computation is a commodity, and whether it can at once be an investment contract; (e) money transmission on any transferable path; (f) whether the ruling that retired token units binds the receipt; (g) a franchise pre-sale module's own claim to be clear of Howey; (h) non-refundable prepaid credit under state prepaid-access and unclaimed-property law; (i) that pre-sale under the FTC Franchise Rule and state law. The page prints the questions and no answers.

4 · What the records say, worst first: the economies are one number read twice

4.1 · The six economies. One run of 20 seeds × 17 units × 12 beats, 340 states, 4 October 2026.

The reading. xₖ is a co-movement ratio of indices in unrelated units, which no numeraire prices. F-E2 tested a cross-unit ratio, so its falsification says the ledger carries one moving number, not that conversion between economies is impossible; ECON-8 registers the commensurable form and ECON-7 the first live reading. Today's ledger carries fewer than six economies: one moving number plus static registrations.

Same observation, three vocabularies. Three fields (information theory, systems and graph theory) describe the same situation in different words, and an insight in one usually has a twin in the other two. This series names a twin only where its own model has the object, and says so where a twin is missing or runs the other way. Each twin is our reading, not an identity, unless a result the paper states makes the two one number. The reading above is one such observation. In information theory, the six indices are not six signals: what moves is one signal read twice. Read as a system, the parts that move are one part, and the rest stand still. Read as a graph of economies joined by conversion edges, the first edge that was measured joins the one moving number to a clipped copy of itself and the second joins it to an index that never moves; three of the five were never read, as the table above shows, and reading one in a common unit is a registered test that has not run (section 5).

4.2 · The loss account. F-L2 is falsified: 21 of 200 cells (0.105) sit in the registered band; the share's median is 1.22 (quartiles 0.38 and 3.93) and the entropy face carries 98.1 % of the remainder, so the remainder is the signal, not a sliver. F-L3 is falsified: 12 of 27 subjects (0.444) read over 0.1 and 15 read 0.0 exactly, because one loss on this scale is a circle by construction; repeated loss opens it (0.866025404). F-L1 held 42 of 42: the 27 subjects with a site posting book a loss and the 15 without book 0, by construction, so its basis is a theorem. The book balanced, the chain held, the loss totalled 21.381654; one replay, 4 October 2026.

4.3 · The account's direction. F-TW1 is falsified: 0 of 24 desk rows agree with their series, all 24 opposite, so the desk books the side it fades; F-TW2 is falsified. F-A1 is falsified: the ledger's own writer agrees on 24 of 24 (F-A1.a held), the desk's first writer on 12 of 24 (F-A1.b). F-A2 held over 156 rows, the largest residual 4.336808689942018 × 10⁻¹⁹. Trading is internal research, never a client service; this is a loss of the instrument.

4.4 · The ledger's and the reward's own refutations, registered tests of earlier papers run on 7 October 2026.

4.5 · The chain and self-dealing. A claimed 42.0 bits is rejected and writes nothing; an edited block is caught by re-derivation. A self-chosen instance of eight choices at field 10.0 books 7.9999995 bits and is accepted: the arithmetic is sound, the provenance of (h, J) is unchecked, and no bound on self-dealing has been measured. On the triangle R ranks the arrangements as E does (Spearman 1.0), so the receipt certifies arithmetic, not usefulness; equation 15 would refuse the pinned claim and the field-only instance and accept a ground state (ECON-2), and closes no question of provenance.

4.6 · Prices and the reward's scale. Quoted as recorded on 15 September 2026: on the exact tier the shadow prices' dual held 14 of 14 and the push-back 4 of 4; at about 200 choices the dual held 0 of 6, every row a kink, the push-back 0 of 2, each stopped by its budget, with one price of −111.50 refuted by its sign and one of +71.57 on a satisfied constraint. The −H reward moved 2.000 times with the penalty weight at no change in quality, 1.000 on the control; one seed.

4.7 · The meter. 119 operations once charged a flat amount for every call now charge nothing that way; 138 are token-metered. Calls of 486, 814 and 55 bytes booked 122, 204 and 14 tokens, $0.00244, $0.00408 and $0.00028: 340 tokens for $0.0068, where the retired flat rule charged the three calls alike, whatever their size. A refused or rate-limited call never bills; the whole catalogue, 5,015 tokens, costs $0.10030; the retired flat price per decision and an unpriced "~$0.50 per read" are the meter's own losses.

4.8 · What held, and the rest. In the model every posting balances, rewards telescope to the first state's entropy minus the last's within 10⁻⁶, and a settlement books nothing new (five collapses of 3.6 settle in one block with a change of 0.0). The first paper's worked reward is H₂(1/4) = 0.811278 bits, a theorem on public numbers. One transaction carries three deltas; SHACL conforms over 1,000 postings and 12,000 triples, 0 violations in 0.901 seconds; a row grows from 319.45 to 355.41 bytes. On one heat run ΔH per pair is 1.0948, then −1.339 bits, then 0 for six pairs while ΔF keeps falling, −920.03 to −0.34; and 148 of 210 ledger post sites (70.48 %) name no unit, where 146 of 186 did on 17 September. A seeded store reads 416.666667 bits per 1,000 tokens, one reading and no trend. The recall library names the right space first for 372 of 840 readings, and for 92 when ranked by energy.

These theorems hold in the model, each pinned. Every count is exact on a fixed record or a count of trials, seeds or subjects with its n; no interval is printed, because no live quantity has two repetitions.

5 · Sixteen tests are registered, and none has run

Each test was frozen with its numeric prediction, kill rules, control and missing-data rule before any script that runs it; its verdict will read held, falsified, not measurable or not run. Every test reads our own records.

• The ledger and the meter. ECON-1, the chain's posting re-derives at its own precision; ECON-2, equation 15 accepts at least 95 % of the committed instances and refuses the pinned claim; ECON-3, a retired serial redeems 0 times in 1,000 (it waits for the retired status); ECON-4, no single price per bit; ECON-5, the reduction rises with the energy on at least 90 % of instances; ECON-6, raw-scale R is not a unit; ECON-8, two economies in bits co-move inside (0, 1); ECON-9, the tariff tracks cost within a factor of 3.
• Live. ECON-7: over 30 days, at least 2 of 5 edges inside (0, 1] on half the posted states.
• The gallery's class. ECON-10, the mark survives 90 % of crops, re-encodings and scalings; ECON-11, a piece re-renders byte for byte; ECON-12, a piece's basin tracks its compressibility (power 0.71 for 18 pieces).
• The operator of the seven papers. ECON-13, couplings survive in 4 of 7 blocks on the neuron lens; ECON-14, 3 of 21 block pairs beat a reassignment null; ECON-15, the joint settle recalls 50 %; ECON-16, capacity within 10 points of the library's 44.29 %.

6 · The opening policy is a design with no caller

Equation 12 is the code's one policy step: O moves by an operator's rate times the share of monitored surfaces improving, times the laminarity of the release-rate channel, times one plus an operator's horizon pull, clipped to [0, 1]. The code names five stages: private-alpha, alpha, beta-open, paid-open and decentralized. A stage names who may call the engine, changes nothing a receipt certifies, and forecasts no date. The policy has no production caller and is parked until the entropy ledger has real multi-tenant flow worth pricing internally; two further dimensions of the parked design are not modelled and are reserved for counsel. Two named measurements would move O: the monitored surfaces with each improving flag re-derived, and the channel's laminarity on a live store.

7 · Memes and gifs would join the gallery's class, and no gallery piece carries a serial

The losses first. No gallery path issues a serial; a free download cannot carry a media serial; three pieces are held and relayed pictures are outside the class; the mark's robustness is unmeasured until ECON-10.

The class. Every graphic is computed from a recorded engine state, never a shader that fakes the physics; a render shaped by a prompt is refused; every served raster is listed, marked with at least 400 pixels of the lattice and credentialed with a sidecar; the on-screen render is never handed over unmarked. A frame sequence would carry the lattice on every frame and a serial issued on its fixture's sha256. In equation 13, κ is 1 for incompressible bytes (Kolmogorov complexity is the standard it approximates, not computed) and w is 1 when every corrupted cue settles home. Spread competes for limited attention (Simon 1971; Weng and others 2012) and rides cascades (Bikhchandani and others 1992), which PX-12 says stillness does not vouch for; the mark is compared with spread-spectrum watermarking (Cox and others 1997). Spread among people is not claimed.

8 · The operator of the seven papers would be published free of charge, and it is not built

This artefact is not a Hilbert space in the series' sense nor an operator on one: it is a classical energy function E(s) on 70 two-way choices in the owned-operator file form. "Hamiltonian" means an energy on configurations; nothing is diagonalised; "nested" is the two-tier coarse-graining of equation 14, not a tensor product; the library's "massive Hilbert space" is 99 separate spaces, never one.

The losses first. The operator is built from the papers' words, including passages whose claims the records refuted (F-BH5, the orbital periods, refuted; the desk's direction, F-TW1, falsified); its settled assignment does not change when a claim is refuted, and it certifies nothing about the papers' truth.

Declared before the build. One block of ten choices per paper from its first 40 passages, five held out, 245 landmark rows, on two lenses. On the pca lens the library kept 0 of 945 couplings, so the likely outcome is "seven field tables, nothing to settle"; the neuron lens is ECON-13's question. The second tier is the best of 128 whole-block flips, never the joint ground state; across 70 choices the words are "the settled assignment" and "a certified lower bound with gap g". It would carry CC BY 4.0, a serial per version, the lattice mark and an offline verifier, and ships after the papers' own licence is ruled.

9 · Related work: where the standard models contradict us

Renewable-energy certificates are the closest analogue (Gillenwater 2008a; US Environmental Protection Agency n.d.), and unbundling them lost the link to the thing performed (Bjørn and others 2022). Carbon registries certify bookkeeping, not the quantity (Badgley and others 2022; West and others 2023). Volunteer-computing credit is the closest receipt (Anderson and others 2002); Gridcoin wrapped it into a traded coin (Gridcoin n.d.). Priced time is not priced results (Sutherland 1968; Waldspurger and others 1992); proof of work is cost with no use (Dwork and Naor 1993; Nakamoto 2008; Ball and others 2017). Contradictions. A posted tariff is not price discovery (Hayek 1945); a constant price claims no informational efficiency (Jordan 1982); a broker cannot be efficient, voluntary, truthful and balanced at once (Myerson and Satterthwaite 1983, measured by PX-14); a market in information is not informationally efficient (Grossman and Stiglitz 1980); information is hard to appropriate, and a certificate verified for nothing answers only part of that (Arrow 1962); and token white papers measured against their code fell short (Cohney and others 2019), so every claim here rests on a committed record. Compared, not claimed: growth bet on a signal (Kelly 1956), ruled out here by the desk's falsified sign; Boltzmann weights over observed transactions (Foley 1994), where ours are over a solved operator; useful work as joules (Georgescu-Roegen 1971), which ours is not; dense associative memory, whose capacity ECON-16 compares (Krotov and Hopfield 2016; Demircigil and others 2017); the classical Hebbian network, for ECON-12 only (Amit and others 1985). The structure follows a benchmark paper, used and compared, never support (Zhang and others 2026).

10 · Limits, and what this paper does not say

This is not a physics claim: entropy, energy and settle are used in their information sense; no joule is booked on any posting, and the one measured joule in the tree is a board-power reading on one device class. "Work" is used in the economic sense of a task performed, not the physical. It is not a currency, an investment or a claim on anyone's future effort. The engine's method is not described. Every number is as of the date its record gives.

Each limit, with what would close it. No live economy number exists (ECON-7, ECON-8). No joule is booked, and the pricing page's efficiency call computes energy per decision from the watts a caller supplies. Two opposite-signed quantities share the word "inflation". No receipt is served, nor a zero-price path, nor retirement (ECON-3). The legal class is unruled. The economy instrument reads one number twice. The opening policy has no caller. The chain has no outside anchor. Self-dealing is unbounded until an instance's hash is bound to an order a party other than the issuer committed before the solve. The paper spaces do not exist yet (ECON-13 to ECON-16). The free precedents carry no serial. Spread among people is not measured. No demand is measured; price over cost is a ratio, not a margin.

Not said here. That any economy can only grow; that inflation falls, that a per-tenant estate is modelled, or that the economy expands; that the six economies are six measurements; that a serial prevents double use; that the rate is a market price; that λ says anything about sustainability; that a receipt is a security, a commodity or stored value; that the universe thesis is more than a hypothesis; that the operator is a joint settle unless ECON-14 holds; that any gallery piece carries a serial.

11 · How to rerun every number

On one x86-64 processor, once, in at most 8 GB; a red check names the field that moved. Each line below names a committed record, the date it was written and its checksum, then the check that re-derives it from the repository's root; the last line re-reads every number on this page.

• fixtures/economy_rungs/falsifiers.json, 4 October 2026, sha256 9ad74f26…7293f: python3 mind/economy_rungs.py --check.
• fixtures/loss_account/record.json, 4 October 2026, sha256 d0b11360…8123: python3 deploy/loss_account.py --check.
• deploy/account_direction.json (a3a57f05…347c) and deploy/trade_writer_sign.json (d097e9de…9831), 4 October 2026: each generator's --check.
• fixtures/paper_experiments/, 7 October 2026: python3 -m unittest mind.test_paper_experiments.
• deploy/economy_figure.json and fixtures/economy_paper/triangle.json: python3 deploy/economy_figure.py --check, which re-runs the heat run and the meter walk.
• Every number on this page: python3 deploy/economy_evidence.py --check.

What Day did. Day re-derived the economy's falsifiers and the loss account, solved the worked instance exactly and drew the three figures from those records.

12 · Symbols, and one meaning per word

Symbols. s a choice, h a field, J a pull, S the scale; E the energy, E∗ its minimum, B the bound, g the gap; ℓᵢ, ΔEᵢ and pᵢ a choice's local field, flip cost and flip probability; H₂ the binary entropy; R the reduction, R_S at unit scale; Z, p_B, H the Boltzmann partition function, weight and entropy; r the reward; v the amount; P a posting with hashes h_prev and h; τ tokens, Π price; u a unit of work with instance I and certificate c; REC the receipt, SN its serial; Iₖ an index, xₖ a co-movement ratio; O the opening level with rate ω, improving share g_O, laminarity c_O and pull π; κ compressibility, w basin width; J× cross-block pulls; f_κ, J̃ and E₂ the second tier; σ_κ a block flip. Thresholds are choices: T = 1, EPS_J 0.05, k_cap 10, β = 12, n ≤ 20 for the solve's search and n ≤ 18 for the re-solve.

One meaning per word. Reduction R, scale S, amount v, price Π, cost measured compute, token the meter's unit, receipt equation 9, operator the file form, nested the coarse-graining of equation 14. Not used: value as a quantity, rung, level, inflation.

What the engine returns under the same name. The engine already returns the mean of A3's per-choice flip entropies, under the name entropy_bits, on the answers it settles from fields and pulls of A1's form; on the same instance and the same assignment, R is n times one minus that number before the answer rounds it to six decimals, so the two read one computation from opposite ends. The name does not keep one meaning across the engine: the chain books R itself under it, other answers put other readings under it or carry none, and an answer with nothing measured says not measured. A certificate's arithmetic needs no anchor: anyone holding the instance and s∗ can recompute its energy and R without us. The solve searches every assignment up to twenty choices; our audit searches a recorded problem's assignments again up to eighteen; the open-source SDK makes the same kind of check on the buyer's own machine, and of these checks only the SDK's does not rely on us. Whatever those checks show, a receipt records one consumed solve, not an investment.

13 · Companion papers

Seven papers published beside this one take the same approach: each traces its figures to a recorded measurement, says what it did not measure, and credits both authors.

• The Operational Unit: the model behind the engine, and how it was built.
• The periodic table as a ground-state ladder: each element read as the lowest-energy filling of its electrons, with every answer checked against NIST.
• Closing the gaps: four sustainability problems an engine can schedule, and what it cannot do about them.
• Sense, remember, act: what the engine senses, remembers and learns.
• The halo of a decision: a two-way choice, the partner it is tied to and the event that tied them, in textbook terms, with what an experiment could measure.
• The Earth as a two-state recorder: the Earth's dipole read as a two-way choice from public records, with a prediction written before the data, and the planet's clocks.
• HOPE: recall in a library of Hilbert spaces: recall as settling to the nearest stored state, in a library of small spaces kept one per model, with the measured answer that they do not make one large space.

Acknowledgements

• SQuAD 2.0 (Rajpurkar, Jia and Liang, ACL 2018), CC BY-SA 4.0. PX-09's slice: the entropy reward and the Brier score of answering every question and of declining every question, on the committed unanswerable questions.

Replicate this paper, or tell us what is wrong

We invite replication and correction. The measured results come from records committed in our repository, and the section "How to rerun every number" gives each record's path, date and checksum and the command that rebuilds the number. The records are not yet published; until they are, write to us through the contact page, naming the paper and the number you want to rebuild or the error you found. We correct an error we can reproduce on the paper's web page, with its date.

The tests this paper registered before reading their data are frozen, each with the rule that would fail it, in docs/ECONOMY_REGISTRY_2026_10_09.md and docs/PAPER_EXPERIMENTS_REGISTRY_2026_10_07.md.

How to cite this paper

Please cite this paper as:

Agile Integrations LLC. 2026. "The unit of work: a receipt for a settled problem, and what it is not." Version 1, 9 October 2026. https://agileintegrations.net/blog/the-unit-of-work.html

Citation file: CITATION.cff.

14 · References

• Amit, D. J., Gutfreund, H. and Sompolinsky, H. 1985. "Storing Infinite Numbers of Patterns in a Spin-Glass Model of Neural Networks." Physical Review Letters 55, 1530–1533. 10.1103/PhysRevLett.55.1530
• Anderson, D. P., Cobb, J., Korpela, E., Lebofsky, M. and Werthimer, D. 2002. "SETI@home: An Experiment in Public-Resource Computing." Communications of the ACM 45, 56–61.
• Arrow, K. J. 1962. "Economic Welfare and the Allocation of Resources for Invention." In The Rate and Direction of Inventive Activity, 609–626. Princeton University Press.
• Badgley, G. and others. 2022. "Systematic Over-Crediting in California's Forest Carbon Offsets Program." Global Change Biology 28, 1433–1445.
• Ball, M., Rosen, A., Sabin, M. and Vasudevan, P. N. 2017. "Proofs of Useful Work." IACR Cryptology ePrint Archive 2017/203.
• Bennett, C. H. 1982. "The Thermodynamics of Computation — a Review." International Journal of Theoretical Physics 21, 905–940.
• Bikhchandani, S., Hirshleifer, D. and Welch, I. 1992. "A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades." Journal of Political Economy 100, 992–1026.
• Bjørn, A., Lloyd, S. M., Brander, M. and Matthews, H. D. 2022. "Renewable Energy Certificates Threaten the Integrity of Corporate Science-Based Targets." Nature Climate Change 12, 539–546.
• CFTC v. McDonnell. 2018. 287 F. Supp. 3d 213 (E.D.N.Y.).
• Cohney, S., Hoffman, D., Sklaroff, J. and Wishnick, D. 2019. "Coin-Operated Capitalism." Columbia Law Review 119, 591–676.
• Commodity Exchange Act. 7 U.S.C. § 1a(9).
• Cox, I. J., Kilian, J., Leighton, F. T. and Shamoon, T. 1997. "Secure Spread Spectrum Watermarking for Multimedia." IEEE Transactions on Image Processing 6, 1673–1687.
• Demircigil, M., Heusel, J., Löwe, M., Upgang, S. and Vermet, F. 2017. "On a Model of Associative Memory with Huge Storage Capacity." Journal of Statistical Physics 168, 288–299.
• Dwork, C. and Naor, M. 1993. "Pricing via Processing or Combatting Junk Mail." In Advances in Cryptology — CRYPTO '92, 139–147.
• Financial Crimes Enforcement Network. 2019. Application of FinCEN's Regulations to Certain Business Models Involving Convertible Virtual Currencies, FIN-2019-G001.
• Foley, D. K. 1994. "A Statistical Equilibrium Theory of Markets." Journal of Economic Theory 62, 321–345.
• Georgescu-Roegen, N. 1971. The Entropy Law and the Economic Process. Harvard University Press.
• Gillenwater, M. 2008a. "Redefining RECs — Part 1: Untangling Attributes and Offsets." Energy Policy 36, 2109–2119.
• Gridcoin. n.d. The Gridcoin white paper.
• Grossman, S. J. and Stiglitz, J. E. 1980. "On the Impossibility of Informationally Efficient Markets." American Economic Review 70, 393–408.
• Haber, S. and Stornetta, W. S. 1991. "How to time-stamp a digital document." Journal of Cryptology 3, 99–111. 10.1007/BF00196791
• Hayek, F. A. 1945. "The Use of Knowledge in Society." American Economic Review 35, 519–530.
• Jaynes, E. T. 1957. "Information Theory and Statistical Mechanics." Physical Review 106, 620–630. 10.1103/PhysRev.106.620
• Jordan, J. S. 1982. "The Competitive Allocation Process Is Informationally Efficient Uniquely." Journal of Economic Theory 28, 1–18.
• Kelly, J. L. 1956. "A New Interpretation of Information Rate." Bell System Technical Journal 35, 917–926.
• Krotov, D. and Hopfield, J. J. 2016. "Dense Associative Memory for Pattern Recognition." In Advances in Neural Information Processing Systems 29, 1172–1180. arXiv:1606.01164
• Landauer, R. 1961. "Irreversibility and Heat Generation in the Computing Process." IBM Journal of Research and Development 5, 183–191. 10.1147/rd.53.0183
• Myerson, R. B. and Satterthwaite, M. A. 1983. "Efficient Mechanisms for Bilateral Trading." Journal of Economic Theory 29, 265–281.
• Nakamoto, S. 2008. Bitcoin: A Peer-to-Peer Electronic Cash System.
• SEC v. W. J. Howey Co. 1946. 328 U.S. 293.
• Securities and Exchange Commission. 2019. Framework for "Investment Contract" Analysis of Digital Assets.
• Shannon, C. E. 1948. "A Mathematical Theory of Communication." Bell System Technical Journal 27, 379–423. 10.1002/j.1538-7305.1948.tb01338.x
• Simon, H. A. 1971. "Designing Organizations for an Information-Rich World." In Computers, Communications, and the Public Welfare, 37–72. Johns Hopkins Press.
• Sutherland, I. E. 1968. "A Futures Market in Computer Time." Communications of the ACM 11, 449–451.
• US Environmental Protection Agency. n.d. Renewable Energy Certificates (RECs).
• Waldspurger, C. A., Hogg, T., Huberman, B. A., Kephart, J. O. and Stornetta, W. S. 1992. "Spawn: A Distributed Computational Economy." IEEE Transactions on Software Engineering 18, 103–117.
• Weng, L., Flammini, A., Vespignani, A. and Menczer, F. 2012. "Competition among Memes in a World with Limited Attention." Scientific Reports 2, 335.
• West, T. A. P. and others. 2023. "Action Needed to Make Carbon Offsets from Forest Conservation Work for Climate Change Mitigation." Science 381, 873–877.
• Zhang, W., Terver, B., Rabbat, M., LeCun, Y. and Balestriero, R. 2026. "H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning." Preprint, 5 October 2026. arXiv:2610.06805

Where this starts

Every priced call already leaves its line on a ledger you can replay: read how a call is metered and booked.