Lifting The Payroll Cap

Charge the Social Security payroll tax on earnings above the taxable maximum of 184,500 dollars, which today stop being taxed altogether.

AI evaluation · not yet reviewed by a human

This evaluation was produced and sourced by an AI model; a human review is still pending. Figures and conclusions may still change. The review log is at the foot of the page.How review works →

Social Security is funded by a 12.4 percent tax on earnings, and in 2026 that tax stops at 184,500 dollars: a person earning three times that pays the same contribution as a person earning exactly that. When the cap was set, it covered 90 percent of all wages; because pay at the top has grown faster than pay in the middle, it now covers about 83 percent, and the gap is a large part of why the retirement fund is projected to run out of reserves in the last quarter of 2032, after which the law permits only 78 percent of scheduled benefits to be paid. The proposal charges the tax above the cap as well, with little or no additional benefit credited for those earnings. Nothing in it changes the benefit formula, the retirement age or the tax rate. This evaluation looks twenty years ahead, so that the year the reserves run out falls inside it.

Balance

Better for the future · 0.72 previous scale

Balance on the previous scale. The Bilanz 2.0 simulation is not yet available for this evaluation. The category comes from the share of the debate on the pro side (r).

For 22 · 72 % Against 8.8 · 28 %

Size class: very large Scale of this evaluation: Normalised Impact — unitless, calibrated to this topic. For comparison: one point here is worth roughly 50 billion euro per year. This is the largest measure evaluated anywhere on this site, and it is large for one reason: it moves about 290 billion euro a year — the revenue at today's distribution of pay, a tenth more for the base that keeps moving above the cap, five percent less for earnings that stop being reported — from the six percent of Americans who earn above 184,500 dollars to the people drawing a state pension. Nothing else here is close to that. The result therefore rests almost entirely on how much more a euro is worth to a pensioner than to somebody earning three hundred thousand dollars — treat them as equal and the measure comes out clearly negative, because the tax also discourages some work and reporting at the top. The second number that matters is what happens without it: current law pays 78 percent of scheduled benefits from late 2032, and that is the comparison used throughout. How we score →

Arguments for

Arguments against

5 arguments evaluated · Scoring v1.3 Δ absolute +13.2

Arguments — For

2 arguments

Pensions that are paid in full

22of 100

The retirement fund runs out of reserves in the last quarter of 2032, and current law then permits only 78 percent of scheduled benefits. Taxing earnings above the cap closes most of that gap.

Value 5 · Household budgetsImpact 8.1Plausibility 5.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money arriving in the households of people drawing a state pension, priced at the middle of the scale as all money is. What makes this the largest argument on this site is not the value class but where the money lands: about four in ten American pensioners take more than half their income from this one source, and one in seven takes almost all of it. The people paying it appear as the argument opposite, at the weight of the top few percent of earners, and the difference between the two weights is the whole of what this evaluation says. The part of the base that keeps moving above the cap as top pay pulls away is inside this figure rather than a separate argument: it is the same money, and it is worth what money is worth. Nothing is counted here for the security of knowing the payment will arrive, which is a separate argument. Nothing is counted for the additional benefit credit some versions of the proposal would give the people paying, which would reduce the transfer and is not in the version scored here. The value is the middle of the scale, and the entire weight of this argument sits in the distance between two income bands.

The taxable maximum is 184,500 dollars in 2026, and about six percent of workers earn more than it in a given year. Charging the tax above it raises roughly 320 billion dollars a year before anybody changes their behaviour. Two things move that figure over the twenty years. The share of wages above the cap has grown by about a quarter of a point a year for four decades; carried on at that pace and averaged over the horizon, it adds a tenth to the base, in a range from 4 to 20 percent. And a fifth less kept on every dollar above the cap means about five percent of those earnings stop being reported — the same response, at the same elasticity of 0.25, that the argument on work and reporting is built on, so that one number governs both sides. Together that is 334 billion dollars a year, or 288 billion euro. All of it goes into the retirement fund, whose reserves are otherwise exhausted in the last quarter of 2032, after which current law permits only 78 percent of scheduled benefits — a cut of about 22 percent for every recipient at once. The money collected in the first six years builds reserves rather than paying anybody; but the programme already costs about 1.7 trillion dollars in 2026 and more every year, so the cut it prevents from late 2032 runs at 450 billion dollars a year or more and exceeds the yearly revenue; over the fourteen years from 2033 the cuts prevented come to about 7 trillion dollars against about 6.7 trillion collected over all twenty, so the reserves built early are paid out inside the horizon — this evaluation does not discount within it. The money therefore lands with pensioners, who carry a weight of 1.4, in a range from 1.1 to 2.0: a mix of the poorest fifth, for whom this is nearly all their income, and households around the middle for whom it is a part. The Impact is the largest figure anywhere on this site and it is simply the size of the transfer multiplied by the distance between two income bands.

▸ Show calculation ▾ Hide calculation

	Revenue from charging the tax above the cap, before anybody changes behaviour static estimate over the next ten years [3]	320 billion dollars a year, 2027 to 2036	320 billion dollars a year
×	Base that keeps moving above the cap over the twenty years Setting, range 1.04 to 1.20: the covered share has fallen from 90 to about 83 percent since the cap was designed, roughly a quarter of a point a year; carried on at that pace and averaged over the horizon, that adds a tenth to the base above the cap [4]	1.10	352 billion dollars a year
×	Earnings that stop being reported at the higher rate the same response as in the argument on work and reporting: elasticity 0.25, a fifth less kept, five percent less reported — one number for both legs and for the output lost [5]	0.95	334 billion dollars a year
÷	In euro exchange rate used throughout this evaluation	1.16 dollars to the euro	288 billion euro a year
×	Weight of a euro for the households receiving a state pension Setting, range 1.1 to 2.0: four in ten pensioners take more than half their income from this source and one in seven almost all of it, alongside households around the middle for whom it is a part	1.4	403.2 billion euro a year
÷	Normalised Impact scale of this evaluation	50 billion euro a point	8.06

Score 8.06 Impact × 5 Value × 5.5 Plausibility ÷ 10 = 22 of 100

The revenue is arithmetic — a statutory rate applied to earnings that are already reported — and the benefit cut it prevents is what current law does automatically when a trust fund is empty, so neither end of this requires anybody to decide anything. The counterfactual is that law, unamended, which the trustees' 2026 report dates precisely. Two things hold this well below what an arithmetic certainty would otherwise carry. The first is timing: the money is collected from 2027 and the cut it prevents begins in 2032, so for the first six years of this evaluation the revenue accumulates rather than preserves anything, which is inside the twenty-year horizon but not evenly spread across it. The second is larger and is not a technical objection: Congress has never once allowed a scheduled Social Security cut to take effect, and if it would have acted by some other means, the benefit preserved here is only the difference between this fix and that one, which nobody can size. That is named and unresolved. Reverse causation does not arise. The Plausibility is a little above the middle: the arithmetic is certain and the thing it is compared against is a law nobody expects to be allowed to operate.

evidence basis: Projection · P ceiling 6 identification: Definitional · no rung ceiling

Counterfactual: current law unamended, under which the retirement fund's reserves are exhausted in the fourth quarter of 2032 and 78 percent of scheduled benefits are payable [1]. Design: definitional — a statutory rate on reported earnings is collected in every case and, once the fund would otherwise be empty, pays benefits that current law would otherwise cut; the pair takes the form of its better-identified leg (rule of 14 September 2026), and P 5.5 on both legs is a judgement on the size of the shared sum, not on whether the payment happens. Confounder: Congress has never allowed a scheduled cut to take effect, so the true counterfactual may be a different fix and the benefit preserved only the difference between them — that doubt concerns the measure as a whole and is carried in the shared eintritt_gruppe over both legs, named and unresolved. Direction: no reverse causation. Ceiling: projection 6.0 binds.

Something to plan around

0.1of 100

A worker of forty today cannot say what the state pension will pay when they reach sixty-seven, because current law promises a cut in 2032 and everyone assumes it will not happen. Removing the shortfall removes the question.

Value 4 · Provision and securityImpact 0.0Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the ability to make a plan: what to save, when to stop working, whether to buy an annuity. This site places that with ordinary usefulness and comfort rather than higher, because what is at stake is the quality of a decision rather than the money it concerns — the money itself is the argument above. It is genuinely separate from that argument in the way that insurance is separate from a payout: a person whose benefit is eventually paid in full still spent thirty years not knowing. Nothing is counted for the people already drawing a pension, for whom the horizon is short and the uncertainty small. The value sits in the lower-middle of the scale, because what is priced is the quality of a decision and not the money the decision is about.

About 70 million people are within thirty years of drawing a state pension and currently face a promise the law says cannot be kept in full. What removing that uncertainty is worth to each is set at 30 euro a year, in a range from 10 to 100 — the band this site uses for a benefit that reaches a very large number of people and matters modestly to each, and roughly what an annual insurance premium against a small chance of a large loss would cost. That gives 2.1 billion euro a year. The figure would be larger if the shortfall were closer or the cut deeper, and smaller if people simply do not believe the cut will happen, which surveys suggest most do not. It is deliberately small next to the money in the first argument because it is the same event valued twice over: once as the payment and once as knowing it will come. The Impact is about a two-hundredth of the money in this debate, which is the honest size of certainty priced separately from the thing it is certain about.

▸ Show calculation ▾ Hide calculation

	People within thirty years of drawing a state pension [1]		70 million people
×	Value of knowing what it will pay, per person per year Setting, range 10 to 100 euro: the band this site uses for a benefit reaching a very large number of people and mattering modestly to each	30 euro	2.1 billion euro a year
÷	Normalised Impact scale of this evaluation	50 billion euro a point	0.04

Score 0.04 Impact × 4 Value × 5 Plausibility ÷ 10 = 0.1 of 100

The uncertainty is documented — the trustees publish the depletion date and the payable share every year, and it is widely reported. The counterfactual is the same workers under current law. What has never been measured is what the uncertainty costs anybody: no study puts a value on knowing what a state pension will pay, and the thirty euro used here is this site's convention for a small benefit reaching a large population rather than a finding. The confounder that matters cuts against the argument, and it is the same one as in the first: most people do not expect the cut to happen, and a person who does not believe in a risk is not paying anything to avoid it. That is named and unresolved. Reverse causation does not arise. The Plausibility is at the middle: the uncertainty is real and documented, and what it costs the people living with it is a convention.

evidence basis: Plausibility · P ceiling 5 identification: Definitional · no rung ceiling

Counterfactual: the same workers under current law, facing a promise the trustees say cannot be kept in full. Design: definitional — the shortfall and the payable share are published annually; no behavioural link carries the quantity. Confounder: most people do not expect the cut to happen, and somebody who does not believe in a risk pays nothing to avoid it; named and unresolved. Direction: not applicable. Ceiling: plausibility 5.0 binds, because the per-person figure is this site's convention for a small benefit reaching a large population.

Arguments — Against

3 arguments

The six percent who pay it

5.6of 100

About six percent of workers earn more than 184,500 dollars, and all of the money comes from them. A person on 400,000 dollars would pay roughly 27,000 dollars a year more than today.

Value 5 · Household budgetsImpact 2.0Plausibility 5.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money leaving the households that earn above the cap, priced at the middle of the scale exactly as the pensions it funds are. This argument exists so that the first one is not counted as though the money came from nowhere. The weight is what differs and it is the whole substance of this evaluation: a euro at the top few percent of American earnings carries 0.35 on this site's scale against 1.4 for a pensioner household, which is why the same sum weighs four times as much on the other side. Nothing is counted for the sense of unfairness at paying into a system that credits little of it back, which is a separate argument. Nothing is counted for the work and reporting response, which is the argument after this one. The value is the middle of the scale, and this argument exists so that the pensions above are not booked as if nobody funded them.

The same 288 billion euro a year, seen from the side that pays it — including the tenth that the base adds as top pay keeps pulling away, and net of the five percent that stops being reported. It falls on about six percent of workers — those earning above 184,500 dollars in a given year — although nearly a fifth of workers cross the cap at some point in a career, so the group is less fixed than the annual figure suggests. A person earning 400,000 dollars would pay roughly 27,000 dollars a year more, counting both halves of the tax, since a levy on wages is borne by wages whichever side writes the cheque. The weight applied is 0.35, in a range from 0.2 to 0.5, which sits between the figures this site uses for the top tenth and the top one percent, because earnings above 184,500 dollars span both. That gives 100.8 billion euro a year. Little or no additional benefit is credited for the earnings taxed, which is what makes this a transfer rather than a contribution. The Impact is a quarter of the argument opposite, built on exactly the same money.

▸ Show calculation ▾ Hide calculation

	Revenue from charging the tax above the cap [3]	the same money, seen from the side that pays it: 334 billion dollars a year averaged over the horizon	288 billion euro a year
×	Weight of a euro for the six percent who earn above the cap Setting, range 0.2 to 0.5: between this site's figures for the top tenth and the top one percent, because earnings above 184,500 dollars span both	0.35	100.8 billion euro a year
÷	Normalised Impact scale of this evaluation	50 billion euro a point	2.02

Score 2.02 Impact × 5 Value × 5.5 Plausibility ÷ 10 = 5.6 of 100

The quantity is the revenue figure read from the other side, so it carries the same confidence and the same caveats. The counterfactual is the cap as it stands. That a transfer has two ends is an identity — if the fund receives 288 billion euro, somebody pays it, and a statutory rate on reported earnings is handed over in every case — but how much is handed over depends on how far earnings above the cap move away from the tax, which is the same question the argument opposite turns on, so the two legs carry the same plausibility for the same sum; what is certain is that the payment happens, what is judged is its size. What is estimated is the weight, which is this site's band structure applied to a group that spans from comfortable professionals to the very rich, and the split matters — at 0.2, the figure this site uses for the top one percent, this argument falls by more than a third. The confounder is the same one as opposite and runs the same way: if behaviour erodes the base more than assumed, both this argument and the one it mirrors shrink together. Reverse causation does not arise. The Plausibility is a little above the middle, matching the argument it mirrors: the sum is fixed by the same arithmetic and the weight applied to it is a band rather than a measurement.

evidence basis: Projection · P ceiling 6 identification: Definitional · no rung ceiling

Counterfactual: the taxable maximum as it stands at 184,500 dollars. Design: definitional — a statutory rate on reported earnings is handed over in every case; the size of the sum (how far earnings above the cap move away from the tax, whether Congress would have acted otherwise) is the shared quantity of the pair and sits in P 5.5 and the range, the same as on the receiving leg. Since 14 September 2026 a pair takes the form of its better-identified leg, so this is the form for both. Confounder: the weight applied to a group spanning comfortable professionals and the very rich; at the top-one-percent figure this argument falls by more than a third, which is carried in the range. Direction: no reverse causation from a legislated rate. Ceiling: projection 6.0 binds.

Work and reporting that stop

3.2of 100

Adding 12.4 points to the marginal rate on earnings above 184,500 dollars takes the combined rate past half in several states. Some of the response is less work and some is income moved into forms the tax does not reach.

Value 6 · OutputImpact 1.0Plausibility 5.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is output that is not produced: hours not worked, a practice not expanded, a partner who retires two years earlier. This site places that with the working order of economic systems and prosperity, because what is lost is production rather than anybody's income in particular. Only the part of the response that is real is counted here. The rest — income shifted from wages into corporate or capital form, or into deferred compensation — is not a loss to anybody, it is money the tax does not collect, and it is already inside the revenue figure used on both sides. Nothing is counted for the effort spent on the shifting itself, which is small and real and not estimated. The value sits in the middle-upper part of the scale, and only the part of the response that is genuinely less production is counted.

Earnings above the cap come to about 2,225 billion euro a year at today's distribution of pay, which is what a static revenue estimate of 320 billion dollars at a 12.4 percent rate implies, and about 2,450 billion averaged over the horizon as the base keeps moving above the cap. The marginal rate on those earnings rises by 12.4 points, taking the amount kept from about 60 cents in the dollar to about 48, a fifth less. At an elasticity of taxable income of 0.25 — the central figure in a literature whose estimates run from 0.1 to 0.6 for high earners — that is a five percent fall in reported earnings above the cap, or about 122 billion euro, and it is the same five percent by which the revenue on both sides of this evaluation is reduced. Forty percent of that is treated as real output foregone rather than income relabelled, in a range from 20 to 70 percent, giving 49 billion euro a year at the ordinary weight for production. The free time the people who work less gain in its place is not deducted here: it is a different good, valued differently, and it is not yet counted anywhere in this evaluation. The reason the real share is set below half is that the response of top earners to tax changes is dominated by timing and form in every study that can separate them. The Impact is half the money the six percent hand over, which is a large behavioural cost by the standards of this site and a middling one by the standards of this literature.

▸ Show calculation ▾ Hide calculation

	Earnings above the cap, averaged over the horizon [3][4]	2,225 billion euro at today's distribution of pay (320 billion dollars of static revenue at 12.4 percent), × 1.10 for the base that keeps moving above the cap	2,448 billion euro a year
×	Fall in reported earnings above the cap the amount kept falls from about 60 to 48 cents in the dollar, a fifth, at an elasticity of 0.25 from a literature ranging from 0.1 to 0.6 [5][6]	5 %	122.4 billion euro a year
×	Share that is real output rather than income relabelled Setting, range 20 to 70 percent: the response of top earners is dominated by timing and form in every study that can separate them	40 %	49 billion euro a year
÷	Normalised Impact scale of this evaluation	50 billion euro a point	0.98

Score 0.98 Impact × 6 Value × 5.5 Plausibility ÷ 10 = 3.2 of 100

The counterfactual is the same earners at the current marginal rate, which is what the tax reform literature constructs from the sharp rate changes of 1986 and 1993 and, more recently, the 2013 increase in top rates ^[5]^[6]. Those designs are quasi-experimental — difference in differences around a legislated threshold, and bunching at kinks — and they are among the better-identified estimates in public finance. The confounder that matters is that a legislated rate change is announced in advance, so what looks like a behavioural response is partly income shifted between years, and the studies that can separate short-run timing from long-run supply find much smaller long-run effects — after 2013, reported top incomes fell sharply for a year and then resumed their trend, a short-run elasticity above one and a medium-run one that is small ^[6]. That is named, unresolved, and the reason only 40 percent of the response is treated as real. Reverse causation does not arise from a legislated rate. The elasticity itself is contested by a factor of six across the literature, which is carried in the range rather than in the plausibility. The Plausibility is a little above the middle: the response is measured on good designs and how much of it is real production rather than relabelled income is not settled.

evidence basis: Converging studies · P ceiling 6 identification: Quasi-experimental · rung ceiling 8

Counterfactual: the same earners at the current marginal rate. Design: quasi-experimental — differences in differences around the legislated rate changes of 1986 and 1993 [5] and the 2013 increase [6], plus bunching at kinks [5]. Confounder: announced rate changes shift income between years, so short-run responses overstate the long-run supply effect — the 2013 increase shows a short-run elasticity above one and only a small medium-run response [6]; named, unresolved, and the reason only 40 percent of the response is treated as real output. Direction: no reverse causation from a legislated rate. Ceiling: set at 6.0, below the quasi-experimental ceiling, because the elasticity estimates span a factor of six.

A contribution that buys nothing

0.0of 100

Social Security has survived ninety years by being an earned benefit rather than welfare: what you pay in determines what you get out. Taxing earnings that are credited with little or nothing breaks that link.

Value 6 · Trust in the systemImpact 0.0Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the character of the programme in the minds of the people who pay for it: a contributory pension in which a contribution buys a claim, against a tax that funds somebody else's payment. This site places that with confidence in public institutions, because what is at stake is whether a programme keeps the political protection that has carried it through ninety years and several attempts to cut it. What is priced is the change in the relationship, spread across the people paying in, and not any particular political outcome, which nobody can forecast. Nothing is counted for the versions of the proposal that credit some benefit on the earnings taxed, which would reduce this and also reduce the revenue. The value sits in the middle-upper part of the scale, and what is priced is a change in what the programme is rather than any political consequence of it.

About 170 million people pay this tax in a year. The change in what the programme is, for each of them, is set at 3 euro a year, in a range from 1 to 10 — the band this site uses for a broad and lightly felt change in a public arrangement. That gives 510 million euro a year. The figure is small in absolute terms and it is the argument most likely to be badly sized in either direction: if breaking the contributory link is what eventually costs Social Security its political protection, the cost is orders of magnitude larger than this, and if voters never notice, it is zero. Nothing here forecasts which. What can be said is that every serious attempt to convert the programme into means-tested welfare has failed precisely because contributors saw it as theirs. The Impact is the smallest in this evaluation by a factor of eighty, and it is the one whose true size is least knowable.

▸ Show calculation ▾ Hide calculation

	People paying the tax in a year [1]		170 million people
×	Value of the change in what the programme is, per person per year Setting, range 1 to 10 euro: the band this site uses for a broad and lightly felt change in a public arrangement, with no anchor behind it	3 euro	510 million euro a year
÷	Normalised Impact scale of this evaluation	50 billion euro a point	0.01

Score 0.01 Impact × 6 Value × 5 Plausibility ÷ 10 = 0.0 of 100

The change itself is definitional: earnings would be taxed without a matching benefit credit, which is a different arrangement from the one that exists. The counterfactual is the contributory system as it stands. What is a valuation rather than a finding is the three euro, and this site has no anchor at all for what a change in the character of a public programme is worth — the figure is the general band for a broad light change, applied by judgement. The confounder that matters is that most people do not know the benefit formula well enough for the link to be doing the work attributed to it, in which case breaking it costs nothing; that is named and unresolved. Reverse causation does not arise. Nothing has measured this and nothing is likely to. The Plausibility is at the middle: the change is certain and its value is a convention with no anchor behind it.

evidence basis: Plausibility · P ceiling 5 identification: Definitional · no rung ceiling

Counterfactual: the contributory system as it stands, in which a contribution buys a claim. Design: definitional — earnings would be taxed without a matching benefit credit; no behavioural link carries the quantity. Confounder: most people do not know the benefit formula well enough for the contributory link to be doing the work attributed to it, in which case breaking it costs nothing; named and unresolved. Direction: not applicable. Ceiling: plausibility 5.0 binds, because the per-person figure is a convention with no anchor behind it.

Summary

This is the biggest thing evaluated on this site and it comes out clearly positive, for a reason that is easy to state and hard to argue about at the margins: about 290 billion euro a year would move from the six percent of Americans who earn above 184,500 dollars to the people drawing a state pension, and this site holds that a euro does substantially more work at the second address than the first. Against it stands the fact that somebody pays — a person on 400,000 dollars would find about 27,000 dollars a year — and that a marginal rate past half in several states produces a real response in work and reporting, of which the productive part is counted here at about 49 billion euro a year. Two things could change the picture. If the weights at the two ends were equal, the measure would come out clearly negative rather than clearly positive. And the comparison throughout is with current law, which pays 78 percent of scheduled benefits from late 2032 — a cut Congress has never once allowed to happen, so the real alternative may be a different fix rather than no fix at all.

Outlook — effect over time

Better for the future · 0.72 previous scale

Sources

Social Security Administration: Annual report of the Board of Trustees, and the taxable maximum. ssa.gov
Congressional Research Service: Social Security: selected findings of the 2026 annual report. congress.gov
Tax Foundation: The Save Social Security payroll tax cap proposal: details and analysis. taxfoundation.org
Social Security Administration, Office of Retirement and Disability Policy: Population profile: taxable maximum earners. ssa.gov
Saez, Slemrod and Giertz, Journal of Economic Literature: The Elasticity of Taxable Income with Respect to Marginal Tax Rates: A Critical Review. aeaweb.org
Saez, Emmanuel (NBER Working Paper 22798; Tax Policy and the Economy 31, 2017): Taxing the Rich More: Preliminary Evidence from the 2013 Tax Increase — short-run elasticity of reported top incomes above one, driven by retiming; only a small medium-run response 2011 to 2015. nber.org

Last reviewed by Claude Opus 5 · September 14, 2026 · 3× AI, not yet reviewed by a human

September 14, 2026AI reviewClaude Opus 5re-scored
Gesamtprüfung 08.09. umgesetzt: eine Einnahmekette für Paar und con-2 (320 $ × 1,10 Drift × 0,95 = 288 Mrd €), pro-3 (unpaarer Transfer) in das Paar gefaltet, Quelle Saez 2013 ergänzt; r 0,73 → 0,72, Kategorie bleibt besser.
September 6, 2026AI reviewClaude Opus 5record updated
i_spanne an allen 6, normalisierung an 4 (Vorsorge-Sicherheit und Systemvertrauen haben keinen Anker), Transferpaar pro-1 ⇄ con-1 verdrahtet (250 Mrd EUR, P 5,5 beidseitig). Kernpunkt: con-1 stand auf definitorisch, pro-1 auf mechanistisch — die Zahlerseite trat in jeder Ziehung ein, die Empfaengerseite nur in 55 %. con-1 auf mechanistisch angeglichen, P unveraendert unter dem Deckel: P(D>0) 0,55 → 0,77, Kategorie bleibt Besser. Der Angleich zeigt einen Widerspruch: der Brief sagt unsichereres Bein, die offene Modellfrage schlaegt das sicherere vor — hier entscheidet das eine Kategorie.
September 6, 2026AI reviewClaude Opus 5First evaluation
First evaluation: the largest transfer scored on this site, with the behavioural response booked separately rather than deducted from the revenue twice.

Evaluations are produced with AI support and reviewed on a schedule for new developments; human passes are marked separately.How we review →