Pre-commitment / Ulysses contract

Status: Folded · Evidence: P · Family: Risk, failure, and resilience · Verdict: fold (2026-06-09)

Use instead: WOOP / MCII

What it is

A pre-commitment (or Ulysses contract, after Odysseus lashing himself to the mast so he could hear the Sirens without steering onto the rocks) is a self-imposed constraint that a present, clear-headed self places on a future self that it does not trust. The defining move is to act now to remove or penalize a future option you predict you will be tempted to take - to bind tomorrow’s behavior in advance against a known weakness. The diagnosis behind it is akrasia: time-inconsistent preferences, where what you will want in the moment (skip the workout, raid the savings, miss the deadline, keep escalating a losing bet) predictably diverges from what you want for yourself on reflection. The remedy is to make the unwanted future choice harder, costlier, or impossible before the moment of temptation arrives.

Concretely, the classic instantiations are external and enforced: lock money in an account you cannot withdraw from until a goal date (a commitment savings product); stake a cash deposit you forfeit if you fail (a deposit contract); hand a friend or a service the authority to impose a penalty; or set a bright-line rule (“never the first drink”, “no email after 6pm”) that converts a thousand small willpower battles into one standing decision. In a clinical setting a Ulysses contract is a psychiatric advance directive: a patient, while well, pre-authorizes specified treatment during a future episode of incapacity.

The honest description has to separate two things that travel under one banner, because they land on different artifacts and different owners:

1. The personal follow-through reading - bind your own future self against an internal obstacle so a decided goal survives the moment of friction. The product is a pre-bound response to a foreseen cue: an if-then plan.
2. The decision / risk reading - pre-decide, before sunk cost and momentum distort judgment, the condition at which you will stop, divest, or kill a course of action. The product is a tripwire and a kill criterion on a risk register.

Both are the same abstract stance (“decide now to constrain a future self I do not trust”) pointed at different problems. As the verdict section argues, each concrete artifact the stance produces is already owned by a shipped skill - and the genuinely distinct kernel of a Ulysses contract (a costly, externally enforced penalty that physically removes the option) is precisely the part a single reasoning agent cannot produce.

When it helps / when it misleads

As a stance, pre-commitment helps when three conditions hold together: the present self can foresee the future temptation, the present self would on reflection rather not yield to it, and there is some lever available now that makes yielding harder or costlier later. It is the right frame for the intention-action gap (the goal is chosen, the follow-through keeps failing) and for guarding a decision against the predictable pull of escalation and sunk cost.

It misleads or wastes effort when:

• There is no enforcement, so the “contract” is just a wish. The mechanism the evidence credits is the cost or impossibility of defecting - the locked account, the forfeited deposit, the third party who will actually pull the trigger. A pre-commitment with no teeth (“I’ll just decide to be disciplined”) is an intention, not a Ulysses contract, and does not buy the effect.
• It is applied where there is no time-inconsistent future self to bind. The frame is built for a present self constraining a later, weaker-willed self across time. A one-shot reasoning task with no temptation, no episodic incapacity, and no across-time preference reversal has nothing to lash to the mast - the move degenerates into a plain if-then plan or a plain stop rule.
• The binding is over-tight or the wrong option is removed. A hard commitment is double-edged: it helps a present self with a self-control problem but can trap a future self who has learned something new and legitimately should change course. Bryan, Karlan and Nelson make this the central caution - hard commitments can hurt naive users and time-consistent users who never needed them. A bright line set in the wrong place (“never raise prices”) can be worse than the judgment it replaces.
• People set the constraint sub-optimally. Even when self-binding works, the binder tends to mis-calibrate it - Ariely and Wertenbroch found people who self-imposed deadlines beat having no intermediate deadlines, but set them worse than an optimal external schedule would.

What the evidence says

The honest read splits, and the split is the whole story for this entry. For the real-world institutional intervention - lock the money, stake the deposit, impose the deadline, sign the advance directive - the evidence is genuinely respectable, M-leaning, well above what most catalog candidates carry. For the thing this library could actually ship - a single agent, in one reasoning session, designing a self-binding constraint - the supporting evidence does not transfer, and the governing grade is capped at P.

What the record supports (M, for the human intervention). Pre-commitment / commitment devices have a real experimental and field-experimental base. Ariely and Wertenbroch (2002) is a true experiment: students allowed to self-impose costly deadlines used them and outperformed a no-intermediate-deadline condition. Ashraf, Karlan and Yin (2006), the study that gave the Ulysses framing its empirical anchor (“Tying Odysseus to the Mast”), is a field RCT in which a locked commitment savings account raised savings balances ~81% over a control group after one year. Bryan, Karlan and Nelson (2010) is the canonical review establishing that demand for commitment is real and that hard and soft commitments behave differently. The theory is old and serious (Strotz 1956 and Schelling 1956 on binding a future self; Ainslie 1975 and Thaler and Shefrin 1981 on the self-control problem it solves).

What the record does NOT support, and the cap. Two facts hold the grade down. First, even the celebrated human results are qualified, not clean: in Ariely and Wertenbroch self-imposed deadlines were worse than externally imposed evenly-spaced ones and people set them sub-optimally; take-up of the SEED commitment account was only ~28%; Bryan, Karlan and Nelson and Rogers, Milkman and Volpp (2014) both document that commitment devices can backfire (binding the wrong self, harming naive users). Second and decisively, every one of these findings measures a human with a time-inconsistent future self, binding that self across weeks or months through a real external enforcement mechanism - a locked account, money at stake, a third-party enforcer, a clinical directive. None of it measures the move performed by or with an AI agent, and the operand it depends on (an akratic future self, temptation across time, an enforceable penalty the move itself creates) has no clean analog in a stateless single-session reasoning task. Promoting this entry to M on the strength of the savings and deadline studies would be exactly the laundering this library exists to prevent: borrowing the robustness of a human-behavior-over-time intervention to grade an agent reasoning move it never tested. Honest read M for the intervention, capped at P for the move, because the strong evidence is on a different operand and even there it is mixed.

Transfer caveat (required). All of the supporting evidence is from human subjects in lab, field, and clinical settings, binding their own future behavior over time through real enforcement. None studies pre-commitment performed by or with an AI agent in a reasoning session. The evidence is transferred from human contexts and not validated for AI-augmented use; the part that is strongest (external enforcement) is the part that transfers worst.

Excluded figures. The 81% savings effect (Ashraf, Karlan and Yin 2006) and the deadline-condition ordering (Ariely and Wertenbroch 2002) are reported here with their named primary sources and are not in dispute - but they are presented as evidence for the human intervention, explicitly NOT counted toward an agent-move grade. No unsourced statistic moved this grade; the popular “commitment devices boost success by N%” framings that circulate without a primary citation are excluded under the evidence rule and did not influence the tier.

Why it is / is not a skill here

Verdict: Fold into woop. This overturns the catalog’s prior cand / build / P tag (“Candidate; complements WOOP and premortem tripwires”). The concrete reason is the one that tag already half-conceded: the move complements WOOP and premortem because everything it concretely produces is already WOOP’s Plan or premortem’s tripwire, and what it would add on top - a costly, externally enforced penalty - is not a reasoning artifact a single agent can emit.

The Build burden is to name one distinct, durable cognitive move that no shipped skill produces, and show no skill (or short chain) already produces it above the ~20% ceiling. Pre-commitment fails that burden because it is a stance, not a procedure with its own separable artifact, and splitting it the way the “What it is” section splits it shows each artifact is owned:

• The personal follow-through reading IS WOOP’s Plan. “Bind your own future self against a foreseen internal obstacle so a decided goal survives the moment of friction” is, made concrete, an implementation intention: “if [obstacle or its cue], then I will [pre-committed action].” That is not merely adjacent to WOOP - it is WOOP’s signature deliverable, and WOOP’s own evidence base (Gollwitzer’s implementation intentions, Oettingen’s mental contrasting) is the precommitment literature for the single actor. WOOP also already names the “known weakness” as a mandatory step (the internal Obstacle), which is the exact diagnosis pre-commitment runs on. The shared mechanism - foresee the weakness, pre-bind the response to its cue, emit a standing if-then - is far above the ~20% overlap ceiling. woop is status: shipped, so the fold target resolves.

• The decision / risk reading is already folded into premortem. “Pre-decide, before sunk cost distorts you, the condition at which you stop, divest, or kill” is a tripwire plus a kill criterion - and kill-criteria-tripwires (“pre-decided stop signals and conditions”) is already a fold into premortem, living in its register as the per-cause conversion step. So the risk flavor of pre-commitment cannot rescue a standalone skill either; the catalog has already decided where pre-decided stop signals live.

• The genuinely distinct kernel of a Ulysses contract is not a reasoning artifact at all. What separates a true commitment device from a plain if-then plan or a plain stop rule is the costly, externally enforced penalty that removes the option - the locked account, the forfeited deposit, the third-party enforcer. That is exactly (a) the part with the strongest human evidence, (b) the part that transfers worst to an agent, and (c) a real-world arrangement a model in a reasoning session cannot create or enforce. Strip it out and what an agent can actually produce collapses back into WOOP’s Plan or premortem’s tripwire. There is no third, separable artifact that is uniquely “pre-commitment.”

So this is a fold, not a build. Why fold rather than recipe or reject: it is not a fixed chain (it is one stance that maps onto one existing move depending on whether the weakness is a personal obstacle or a decision threshold - the same shape that made inversion a fold, not a recipe). And reject would be less informative than fold: the move is real, famous, and worth locating, so the honest service is to point the reader to where it already lives. Fold into woop as the single-actor home for binding a foreseen weakness with a pre-committed response, and let the dossier record that the decision-stop-signal flavor lives in premortem’s register (via kill-criteria-tripwires). The learning value of the NO: a genuinely well-evidenced behavioral intervention is not automatically a thinking skill - when the part that the evidence credits (external enforcement across time) is the part an agent cannot perform, the residue that an agent can perform turns out to be a move the library already ships.

Lineage and who to read

The idea is ancient in image (Homer’s Odysseus and the Sirens) and modern in theory. The economic formulation of binding a future self is owed independently to R. H. Strotz (“Myopia and Inconsistency in Dynamic Utility Maximization”, 1955-56) and Thomas Schelling (on commitment and self-command, from 1956 onward; see his later “Egonomics, or the Art of Self-Management” and “Enforcing Rules on Oneself”). The self-control problem it solves was deepened by George Ainslie (“Specious Reward”, 1975, on hyperbolic discounting and impulse control) and modeled by Richard Thaler and H. M. Shefrin (“An Economic Theory of Self-Control”, 1981). For the experimental and field evidence, read Dan Ariely and Klaus Wertenbroch on self-imposed deadlines and Nava Ashraf, Dean Karlan and Wesley Yin on the Philippine commitment savings product (the paper that supplied the “tying Odysseus to the mast” framing); Gharad Bryan, Dean Karlan and Scott Nelson (2010) is the review to start with, and Todd Rogers, Katherine Milkman and Kevin Volpp (2014) is the health-behavior synthesis with the backfire cautions. For the clinical Ulysses-contract literature, see the bioethics reviews of self-binding advance directives. For where the single-actor move already lives in this library, read the WOOP / implementation-intentions work (Gollwitzer; Oettingen), and for the decision-stop-signal flavor, Gary Klein on the premortem register. “Pre-commitment” and “Ulysses contract” are generic descriptive terms in common scholarly use - no trademark, no attribution required beyond crediting the originating economists - so this entry is documented descriptively and is not flagged as branded.

Named sources

• Thomas C. Schelling, work on commitment and self-command (from 1956; e.g. “Egonomics, or the Art of Self-Management”, American Economic Review 1978; “Enforcing Rules on Oneself”, JLEO 1985). Foundational theory of binding a future self; not an effectiveness study of the move as a reasoning skill. Practitioner / foundational. (P)
• R. H. Strotz, “Myopia and Inconsistency in Dynamic Utility Maximization”, Review of Economic Studies 23(3) (1955-56): 165-180. The original economic account of pre-commitment as a response to time-inconsistent preferences. Foundational theory, not an agent-move study. (P, foundational)
• George Ainslie, “Specious Reward: A Behavioral Theory of Impulsiveness and Impulse Control”, Psychological Bulletin 82(4) (1975): 463-496. Establishes the hyperbolic-discounting self-control problem pre-commitment addresses; explanatory, not an effectiveness trial of the skill. (P, foundational)
• Dan Ariely and Klaus Wertenbroch, “Procrastination, Deadlines, and Performance: Self-Control by Precommitment”, Psychological Science 13(3) (2002): 219-224. True experiment: people self-impose costly deadlines and beat a no-intermediate-deadline condition, but set them sub-optimally and below an optimal external schedule. The nearest direct experimental evidence - on humans, over time, and mixed. (M, human intervention - not the agent move)
• Nava Ashraf, Dean Karlan and Wesley Yin, “Tying Odysseus to the Mast: Evidence from a Commitment Savings Product in the Philippines”, Quarterly Journal of Economics 121(2) (2006): 635-672. Field RCT; a locked commitment savings account raised balances ~81% versus control after a year, with ~28% take-up. Strong field evidence for the institutional device, on humans with real enforcement. (M, human intervention - not the agent move)
• Gharad Bryan, Dean Karlan and Scott Nelson, “Commitment Devices”, Annual Review of Economics 2 (2010): 671-698. Canonical review: real demand for commitment, hard vs soft commitments differ, and hard commitments can harm naive and time-consistent users (the backfire caution). Review of human-subject evidence. (M, review)
• Todd Rogers, Katherine L. Milkman and Kevin G. Volpp, “Commitment Devices: Using Initiatives to Change Behavior”, JAMA 311(20) (2014): 2065-2066. Health-behavior synthesis; commitment devices can change behavior but have unintended effects. Human health-behavior context, not agent-validated. (P/M, viewpoint synthesis)

Excluded under the evidence rule: the ~81% savings effect (Ashraf, Karlan and Yin 2006) and the deadline-condition ordering (Ariely and Wertenbroch 2002) are real and sourced, but are counted as evidence for the human intervention only and explicitly NOT toward an agent-move grade; the popular “commitment devices improve success by N%” framings that circulate without a primary source are excluded and did not move the tier.