Canonical: https://thonly.org/research/program · Licence: CC0 1.0
This document states a research program and the conditions under which it should be abandoned. The program's hard core is a single claim about direction: that value can be organized to move only forward — from giver to receiver to the next receiver — and that a system built on that constraint circulates better than one that permits return to the source. Four chapters name the ways the core can break: whether receiving creates the capacity to give, whether the constraint survives a change of currency, whether it survives past the family, and whether it survives the giver. Each chapter is attached to pre-registered predictions, published here as a register of sixty-six items with their falsifiers, their instruments, and their status.
The program is published at a deliberate moment: almost nothing in it has been run. Two desk censuses have returned results, both null or partial-null. There have been no field tests. The first is gated on a product launch in August 2027. A register published after the data arrives cannot be distinguished from a register assembled to fit it; this one is published while the outcome is unknown, which is the only condition under which it constitutes evidence of anything.
A gift and a trade are not the same operation. A trade closes: goods move one way, value moves the other, and the parties are square. A gift does not close. Something moves, nothing comes back along the same line, and what the receiver owes — if the word applies at all — is owed forward, to someone else.
Most economic infrastructure is built for the first operation. This program asks what can be built on the second.
The question is not whether gifts are good. It is narrower and harder: whether the forward-only constraint is a mechanism or an ornament. A system can be designed so that value never returns to its source — thanks route into a fund the recipient may spend forward but never withdraw, never apply against what they owe. That constraint is expensive. It forecloses rebates, refunds, reciprocal settlement, loyalty credit, and every familiar device for making generosity pay. The program's core claim is that the constraint buys more than it costs: that a system forbidding return circulates more than one permitting it.
That claim can be wrong. If it is wrong, the constraint is an aesthetic preference dressed as engineering, and the right response is to abandon it rather than to explain it. This document exists to make that abandonment possible — to state in advance what would compel it, so that the compelling evidence, if it arrives, cannot be absorbed.
A program named for its answer becomes a lie when the answer is wrong. The Gift Economy Program would have to be renamed or defended. Which Way Value Moves stays true either way: if value turns out to move back, the question was still the right one, and the answer is a finding rather than an embarrassment.
This is the same discipline the register applies to individual predictions. A prediction is only worth registering if its author can survive the null. A program is only worth naming if its name can.
The program carries no trademark, deliberately. The corpus is dedicated to the public domain, and a research question held as property is not a research question.
Value moves forward, never back.
Stated as an operation: the atom is give-as-gift, not exchange. Its circulation form is receive → give forward. The claim attached is that this operation is invariant — that it holds across substrate (money, time, object), across scale (person to institution), and across time (beyond the participation of the original giver).
The core is what the program protects and what the program tests. Both halves matter. A hard core that is protected but never tested is what makes a research program degenerate: the core survives by absorbing every anomaly after the fact, and the absorption is mistaken for confirmation. The safeguard adopted here is that the core carries its own falsifiers, listed in §4, each marked publish either way.
A core that excludes nothing has selected nothing. This one excludes, and the exclusions are checkable against any design:
Each of these has already excluded a real design in this corpus, not a hypothetical one. That is the test of whether a purpose is doing work: not does this serve the core — everything can be made to — but what has the core forbidden.
A chapter is admitted on two conditions: it must name a way the core breaks, and it must name an instrument that could show it. A proposed chapter with no instrument is a theme, and themes do not enter.
The four are ordered by the length of their feedback loops, which is also the order in which they can be answered.
The break: if receiving does not reliably produce onward giving, forward-only motion is not circulation. It is depletion with better manners. The second arrow in receive → give forward fails, and the system runs down.
Status: the only chapter with data. In the founding household, re-tips (434) overtook self-thanks (189) over the observation window — receiving converting into giving, at n=1, confounded, and reported as such. The pilot's own summary of the finding was giving is a gift too.
Honest limit: one family, no control, an observer who wanted the result.
The break: if the atom is money-specific, the invariance claim is false and the dual-currency architecture rests on an analogy rather than a mechanism.
Status: running, unread. The instrument is a single nullable field (moments.timelogId) in the Treasury product, pre-registered 2026-08-03 before any data existed. The predictions (P-C0a, P-C0b, P-C0c) are read at a pre-committed point; reading early, or prompting participants toward the predicted behavior, voids the test.
Disclosed defect: P-C0b cannot be evaluated as originally written. Its metric — completion rate — is biased against the hypothesis, because time-sacrifice recordings are plausibly longer and longer recordings score lower by construction. P-C0b would read confirmed (thesis wrong) for a mechanical reason. It is restricted to views-per-week and re-tip rate until the instrument is repaired. This defect is published before the data is read, which is the only time such a disclosure is worth anything.
The break: if forward-only circulation is a small-group phenomenon that collapses into exchange once participants stop being kin, then everything here is a description of families and not of economies.
Status: unrun. Gated on a product launch in August 2027. The predictions (P-S1 through P-S14) are specified and stamped; none has an observation attached.
Honest limit: this is the chapter carrying the most weight and the least evidence. One household is the entire empirical base beneath a set of scale-invariance claims.
The break: if a gift requires a living receiver to complete, then transmission across generations is a metaphor, and an institution designed to hand itself to a successor is built on one.
Boundary already ruled: the question is not metaphysical. Merit is not transferred to the dead; what the canon describes is relief arising within a destination, never conveyance between them. The operational form is the tractable one: you cannot give to the dead, so you give to the living in their name. The chapter asks whether a gift keeps circulating once the person who began it can no longer receive anything back.
Status: unrun, and unrunnable on the founder's timescale. This chapter is inherited rather than completed — a fact stated here plainly rather than treated as poignant.
These are distinguished from the sixty-six in the register by a single property: if one of them fails, the program ends rather than adjusts. They are the reason this is a program and not a worldview.
P-GO1 — the forward-only constraint pays for itself. In matched populations, a system that forbids backward return — value routed to a forward-spendable fund, never withdrawable, never applied against what the recipient owes — produces more onward giving per unit received than an otherwise-identical system permitting withdrawal or offset. Falsifier: equal or greater onward giving under the permissive condition. If it fails, the forward-only constraint is an aesthetic preference and the program's organizing question has been answered no. Publish either way. (Registered 2026-08-27. First public timestamp: this document.)
P-K1 — kindness is inelastic to subsidy. As external subsidy for a kind act approaches zero, the rate of the act does not approach zero. Falsifier: kindness rate tracks subsidy proportionally, i.e. what looked like generosity was priced behavior throughout. If it fails, the second-singularity thesis and the Right Livelihood telos both fall, and the institution is a payment scheme. Publish either way. (Registered 2026-08-27. First public timestamp: this document.)
P-L2 — the founding bet. Witnessed presence generates more downstream giving than unwitnessed presence of equal duration. Falsifier: if unwitnessed presence produces equal downstream gratitude, the chronicle is decorative and the institution's core claim is wrong. Publish either way. (Registered 2026-08-07 in co-presence-gated-redemption §10.)
P-R2 — the raft must become unnecessary. Product-dependence begins to fall after the inflection. Falsifier: if dependence never falls, the raft is a boat — and the two-phase spiral was a way of never being wrong. (Registered in the user-arc material; first public timestamp: this document.)
P-B3 — the offering does not erode. The proportion of creators electing to dedicate their share does not decline over time. Falsifier: dedication collapses once money is visible, meaning the default was carrying the work and the offering frame is eroding. (Registered in dedicatory-generation.)
A register maintained by an author's discipline is worth what the author's discipline is worth on the worst day. This one is built so that the guarantee does not depend on anyone's restraint, because an institution intended to outlive its founder cannot rest on the founder's care.
The property, not the rule. Every prediction is fixed by an external timestamp at the moment of registration. Forty-five of the sixty-six are already carried by the published papers that state them, each of which bears an OpenTimestamps proof anchored in Bitcoin and RFC 3161 tokens from three independent authorities, one of them eIDAS-qualified. The remaining twenty-one receive their first public timestamp from this document and its register.
The consequence is checkable by a stranger: a reader can verify, without trusting the author, that a prediction said what it says now and said it before the data existed. Editing a prediction after the fact does not produce a quiet correction; it produces a divergence between the file and its stored proof, which any verifier reports. Revisions are additive and separately stamped. The register is append-only not because a policy forbids overwriting but because overwriting is visible.
No document in this corpus records its own hash. A hash of a file cannot live inside that file, and a claim to the contrary is a claim that cannot be true. The proofs are external artifacts, and the verification instructions are published with them.
What this does not protect against. It does not stop a prediction from being vague enough to survive any outcome. It does not stop an instrument from being chosen badly, as P-C0b was. It does not stop a prediction from going unrun forever. Timestamping guarantees that a claim was made in advance; it guarantees nothing about the claim's quality.
The corpus contains one hundred documents. The admission rule is mechanical, so that membership is a property of the file rather than a judgment by whoever is filing:
A paper is instrumented if it names at least one registered prediction.
Applied today, that returns thirteen instrumented papers of one hundred. The remaining eighty-seven divide in two:
the-gift-operation — the paper that states the hard core, which until today carried nothing falsifiable at all.The thirteen-of-one-hundred result is reported rather than repaired. The rule was not softened to produce a flattering number, and the most uncomfortable consequence — that the paper stating the core was, by the program's own test, the least tested document in it — is the finding that most justifies keeping the rule.
A paper moves from open to instrumented when someone registers a prediction against it. That is an event with a timestamp, not an opinion.
The evidence base is approximately zero. Sixty-six predictions; two desk censuses run, both returning null or partial-null; no field tests; the first gated on August 2027. Every mechanism in this corpus is currently supported by argument and by one household.
The coherence is not evidence. The design grammar underlying this program is self-similar and tension-dissolving, which is exactly what a deeply correct system and a deeply seductive one both look like from inside. The elegance earns the experiment; it does not replace it. Each structural property is individually falsifiable; the pattern they make is not, and the pattern is never cited as support for a member of it.
Most instances are retro-fitted. The signatures organizing this work were recognized in existing material rather than predicted before it. That is a legitimate way to find a pattern and an illegitimate way to confirm one.
One prediction has already been contradicted by our own data. The kids-as-triggers hypothesis — that children would be the primary triggers of self-thanking — was contradicted in the founding household, where parents self-thanked more than children. It is confounded and has not been hardened into a finding. It is listed in the register as contradicted rather than quietly dropped, and the essay stating it has not been withdrawn.
The register is incomplete by construction. Twelve predictions were found during its assembly living only in published papers, invisible to the internal index; the true count moved from fifty-five to sixty-seven during a single afternoon's audit. Neither system held the whole set. That failure is the reason this register exists, and it is not evidence that the register is now complete.
Every 7 January — the date on which the institution's ledger resets — the full prediction set is walked and each item marked unrun · running · resolved. This commitment predates the register; what is new is that the walk's output is published rather than filed.
A pre-registration that is never revisited rots into decoration. The walk is the mechanism that prevents it, and publishing the walk is the mechanism that prevents the walk from being skipped quietly.
Living. Predictions are added by registration and never removed; resolved items keep their original text beside their outcome. Revisions are separately timestamped, and no revision may alter a prediction's registered wording — a correction is a new entry that cites the one it corrects.
The register accompanying this document is at prediction-register.md.