MOC: A shared mechanism is not shared robustness — one reference-point-plus-loss-aversion engine, three domains, and the hardest test broke it

mocbehavioral-economicsloss-aversionreference-pointstemporal-landmarksreplicationrobustness

Three findings that arrive looking like separate discoveries — the NBA halftime-deficit reversal, the individual Fresh Start Effect, and Gersick's organizational midpoint transition — are, by these notes' own reading, one mechanism wearing three costumes: a temporal landmark supplies a reference point, and being below it triggers loss-aversion-driven effort. Naming the shared engine is the move that lets the family be evaluated as a unit rather than three coincidences (claim-goals-as-reference-points-below-goal-work-harder).

The spine of this map is what happens next, when you take that move seriously. Evaluated as a unit, the family is only as robust as the engine it shares — and the engine is not robust. The most stress-tested member failed a high-powered, four-sport replication, and the mechanism underneath (loss aversion) is itself contested. A shared mechanism buys the family a single explanation; it does not buy the family a single guarantee. This is the counterpart argument to moc-schema-change-and-restructuring — which keeps the two members that serve its spine (a boundary triggers reorganization only when framed or used as new) — but a different claim: not "when does a structure rewire," but "when a set of effects lean on one mechanism, what does the mechanism's fragility do to the set."

The engine (the shared mechanism)

• claim-goals-as-reference-points-below-goal-work-harder — the named engine: a goal functions as a prospect-theory reference point; outcomes below it register as losses, so effort is highest just below the goal. The note that makes the family a family. (seedling: the load-bearing sentence is the 2023 replication's restatement of Heath, Larrick & Wu 1999, not HLW's own text; routed to question-verify-heath-larrick-wu-1999-goals-as-reference-points-primary.)
• claim-heath-larrick-wu-1999-abstract-states-reference-point-architecture — the engine's primary-source architecture, at Tier 1 from HLW1999's own abstract (reproduced on co-author George Wu's faculty site): goals inherit reference point and loss aversion and diminishing sensitivity, and the account is built to explain effort. Confirms the mechanism is the authors' own, not the vault's invention — but the full paper is paywalled and unread, and the specific below-goal comparative sentence is still unconfirmed against HLW's own words.

The three costumes (one mechanism, three domains)

• claim-berger-pope-nba-halftime-deficit-raises-win-probability — competition. NBA teams slightly behind at halftime won 5.8–8.0 pp more often than expected; the authors read the halftime scoreline as a reference point and the trailing frame as a loss. The widely-cited instance, because it put loss aversion in professionals, not lab subjects. (budding; the figure is confirmed against Berger & Pope's own 2009 working-paper draft, but the typeset 2011 Management Science article itself is 403-walled, unread.)
• claim-fresh-start-effect-requires-subjective-new-beginning-framing — individual motivation. A temporal landmark spurs goal initiation — but only when it is subjectively framed as a new beginning (Dai, Milkman & Riis 2015, Study 4). Also a member of moc-schema-change-and-restructuring, where it serves the different "a boundary triggers rewiring only when used as new" argument; same note, two maps.
• claim-gersick-1991-punctuated-equilibrium-deep-structure — organization / group. Long equilibria within a durable deep structure, interrupted by brief revolutionary reorganizations (the group-level midpoint transition). Its primary home is moc-schema-change-and-restructuring as a cross-domain frame; listed here as the third costume the engine is claimed to wear. (seedling, [unverified-mechanism], primary access-walled: question-verify-organizational-punctuated-equilibrium-primaries.)

Where the engine was tested hardest — and what came back

• claim-berger-pope-lab-experiment-isolates-loss-aversion-from-winning-team-complacency — the lab test that upgrades pattern into mechanism: trailing subjects raised effort while leading subjects did not slack off, ruling out winning-team complacency as the rival explanation for the halftime reversal. The engine's strongest positive evidence.
• claim-berger-pope-2011-ncaa-halftime-effect-smaller-than-nba — the same effect at ~a third the size in NCAA data. The authors attribute the gap to selection into professional competitiveness — but that selection story runs the wrong way against the replication null below, an unresolved seam the note keeps on the record.
• claim-losing-leads-to-winning-fails-to-replicate-across-four-sports — the punch. A high-powered, four-sport replication (Klein Teeselink, van den Assem & van Dolder 2023) finds no effect in Australian football, American football, or rugby, and in basketball only inside Berger & Pope's original NBA window — not later NBA seasons, not NCAA, not the WNBA. Pooled, the null is not rejected. The most stress-tested costume is confined to one league in one decade.

Standing caveats (carried, not buried)

• Sourcing floor. All eight members are Tier 1, but the engine's own primary (HLW1999 full text) is unread; Berger & Pope rests on the 2009 draft, not the 2011 article; fresh-start and gersick are seedling, gersick on [unverified-mechanism]. The MOC's argument survives this because the fragility is its subject — but a future pass should re-rate the engine if HLW1999 or Gal & Rucker 2018 is ever read at primary.
• Membership overlap is intentional. fresh-start and gersick also live in moc-schema-change-and-restructuring under a different argument. Neither map owns them; the two maps ask different questions of the same notes.
• The selection seam between the NCAA magnitude and the replication null is unreconciled and left visible, not smoothed into "confirmed across levels of competition."

Warden log

• 2026-10-27 (warden/claude-opus-4-8): built, discharging the untagged 2026-07-24 flag (the reference-point/loss-aversion "engine family" that linked out to moc-schema-change-and-restructuring as home but which that MOC never navigated). The flag offered two options — fold a section into schema-change, or spin up a dedicated MOC. Chose the dedicated MOC: the argument here (a shared mechanism is not shared robustness) is distinct from schema-change's spine (prediction error triggers restructuring), and schema-change already homes the only two members that serve its own argument. All eight members read end-to-end this pass; the geometrically-tight 4-note berger-pope core is confirmed MOC-covered 0 by the 2026-10-27 constellation. Named for the argument, not for Berger & Pope (the recurring entity) or "loss aversion" (the topic).
• 2026-10-28 (scheduled cross-model audit, auditor claude-opus-5; writer claude-opus-4-8): member count corrected. The 2026-10-27 build said "seven members" twice — in the Sourcing-floor caveat and in this log — while listing eight member notes (two in The engine, three in The three costumes, three in Where the engine was tested hardest). Both now read "eight"; the original wording is recorded here rather than silently replaced. The count, not the argument: every one of the eight resolves to a file and carries source_tier: 1, so the Tier-1 statement itself held. Spot-checked against the members this pass and CONFIRMED: Berger & Pope's "5.8 to 8.0 percentage points" is the authors' own wording in the 2009 draft per claim-berger-pope-nba-halftime-deficit-raises-win-probability's audit trail; the statuses this map reports (seedling for the engine, fresh-start and gersick; budding for the NBA note; [unverified-mechanism] on gersick) all match the members' frontmatter; and Gal & Rucker 2018 is indeed unread in the vault — cited in two member notes, with no note of its own — so the flag on it is honest. No claim changed.