Alternative Scenarios at the Federal Reserve from 1968 to 2020: Data, Interpretation, and Evaluation

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Alternative Scenarios at the Federal Reserve from 1968 to 2020: Data, Interpretation, and Evaluation Edward Herbst, Scott Konzem, and Cristina Scofield 2026-033 Please cite this paper as: Herbst, Edward, Scott Konzem, and Cristina Scofield (2026). “Alternative Scenarios at the Federal Reserve from 1968 to 2020: Data, Interpretation, and Evaluation,” Finance and Economics Discussion Series 2026-033. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2026.033. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Alternative Scenarios at the Federal Reserve from 1968 to 2020: Data, Interpretation, and Evaluation ∗ Edward Herbst, Scott R. Konzem, and Cristina Scofield † Federal Reserve Board April 2026 Abstract We comprehensively document 1,265 Federal Reserve staff alternative scenarios presented to the Federal Open Market Committee in publicly released materials from 1968 to 2020. Scenarios grew in frequency and sophistication, typically spanning a range of outcomes around the baseline. We construct a taxonomy with six categories: aggregate demand, aggregate supply, external risks, financial conditions, fiscal policy, and expectation shifts. Staff qualitative risk assessments complemented the scenario composition. Comparing scenario forecasts to realized outcomes, the most accurate scenarios often anticipated major macroeconomic developments even when magnitudes were missed, revealing the value and limits of scenario analysis for central bank risk management. 1 Introduction Central banks rely heavily on forecasts in their internal deliberations, often emphasizing a single “baseline” projection for key macroeconomic variables. Yet monetary policy is ultimately an exercise in risk management, in which the challenge is not only to forecast the ∗Theanalysisandconclusionssetfortharethoseoftheauthorsanddonotindicateconcurrencebyother members of the research staff or the Board of Governors. We thank Isabel Cair´o, Ben Johannsen, Edward Nelson, and Fabian Winkler for helpful comments and Karen Page for research assistance regarding the Summary of Economic Projections. All errors are our own. †Email: edward.p.herbst@frb.gov, scott.r.konzem@frb.gov, cristina.m.scofield@frb.gov. 1

most likely path of the economy but also to take into account the possibility of less likely outcomes. Robust policymaking requires careful consideration of uncertainty and the range of riskssurroundingthebaselineprojection. Againstthisbackdrop,“alternativescenarios”have become an increasingly important tool in both policy deliberations and communications.1 Alternative scenarios are structured point forecasts constructed under different assumptions about the evolution of the economy—for example, a sharp slowdown in productivity growth or a sudden rise in energy prices. They provide policymakers with concrete illustrations of how key risks might unfold—unlike purely statistical tools for describing uncertainty such as fan charts—thereby complementing the baseline by highlighting vulnerabilities and possible trade-offs. Indeed, recent research, including Bernanke (2024), Wessel (2024), and Bordo, Levin, and Levy (2020), has advocated a more prominent role for alternative scenarios in monetary policy deliberations and communications. Recent Federal Reserve communications illustrate the current practical relevance of scenario analysis. In December 2024, when asked about tariff policy at the press conference following the Federal Open Market Committee (FOMC) meeting, Chair Powell referred to staff scenarios prepared years earlier; a month later, he described the alternative scenarios as “one of the best things” the staff does because they help policymakers think through uncertainty.2 These remarks motivate the historical questions addressed in this paper: how has the Federal Reserve used alternative scenarios, what risks have those scenarios emphasized, and how informative have they been in light of subsequent outcomes? Alternative scenarios have a long history at the Federal Reserve and other central banks.3 The first known mention of “alternative scenarios” in the publicly available FOMC historical materials came at the April 1968 meeting, immediately following the collapse of the Gold Pool and arrangements to manage the private-sector world gold price—a pivotal event in the unraveling of the Bretton Woods System.4 Robert Solomon, director of the Division of International Finance, laid out the monetary policy implications for the FOMC:5 What I shall try to do this morning is to place these meetings in perspective and sketch out for the Committee two alternative scenarios for the period ahead. One 1Bauer and others (2025) lay out a taxonomy of uncertainty and risks relevant for monetary policy analysis and discuss tools, including alternative scenarios, that can be used to assess and quantify uncertainty and risks. Garga and others (2025) discuss advantages and disadvantages of alternative scenarios in monetary policy deliberation and communications. Cair´o and others (2025) propose a framework to incorporate alternative scenarios into the assessment of monetary policy strategies. Adrian and others (2025) describe a statistical framework for inferring the likelihoods of alternative scenarios. 2See Powell (2024) and Powell (2025). 3SeeKochandCrawford(2019)foranearlieranalysisofalternativescenariosattheFederalReserveand Ciccarelli and others (2025) for a discussion of scenario analysis at the European Central Bank. 4For background on the Gold Pool see Solomon (1977) and Bordo, Monnet, and Naef (2019). 5Memorandum of Discussion of the Federal Open Market Committee Meeting on April 2, 1968, p. 13. 2

scenario involves a return to international monetary stability based on returning confidence that the existing price of gold will be maintained; the other scenario involves a recurrence of instability based on a failure to correct the U.S. balance of payments problem. Fromsuchcrisis-motivatedbeginnings, alternativescenariosevolvedintoaregularfeature of monetary policy analysis, expanding in both frequency and analytical sophistication. In this paper, we analyze the period from 1968 through 2020, capturing all years for which relevant FOMC historical materials are publicly available.6 We examine the complete set of publicFOMChistoricalmaterialstocomprehensivelyidentifyalternativescenariospresented by the staff in advance written briefing materials (the Greenbook, Bluebook, and Tealbook) as well as during the FOMC meetings (as documented in the transcripts, memoranda of discussion [the predecessor of the transcripts], and presentation materials). We compile a dataset covering 1,265 alternative scenarios presented to the FOMC from 1968 to 2020, including each scenario’s name and model used, if available.7 For the 1,132 scenarios appearing in a more structured form in the Greenbook and Tealbook A from 1989 to 2020, we record narrative descriptions of the scenarios along with tabulated numerical forecasts for a selected set of macroeconomic variables. In addition, we record the narrative balance-of-risks statements presented just prior to the presentation of scenarios: these risk statements began to appear in the January 2008 Greenbook and were included in every Tealbook from March 2013 through the end of the sample. These texts provided qualitative assessments of upside and downside risks around the baseline forecast, complementing the quantitative alternative scenarios. Our analysis reveals a systematic expansion over the sample period in both the types of risks that were considered in alternative scenarios and the analytical sophistication with which they were modeled. Our paper makes three contributions. First, we provide a descriptive overview of the available historical record, documenting the number of alternative scenarios presented to the FOMCovertime,themodelsusedtoconstructthem,andtheBoardstaff’soverallassessment offorecastuncertaintyandthebalanceofrisks. Second,wecategorizethescenariosaccording to the nature of their shocks, providing a taxonomy of risks emphasized in the scenarios. Using the quantitative information, we examine the joint distribution of output and inflation expressed as deviations from the baseline forecast. This allows us to characterize the risks 6See Federal Open Market Committee (2026). FOMC historical materials are available at https://www. federalreserve.gov/monetarypolicy/fomc_historical.htm. 7We exclude scenarios that differed from the baseline only in the setting of the federal funds rate or balance sheet policy. Additionally, when a scenario was presented both with and without a monetary policy response,wekeeponlytheversionthatincludedamonetarypolicyresponse. InthecaseofsomeGreenbook scenarios, information on the numerical values was incomplete. 3

capturedbythescenariosandtheirco-movements, withoutreferencetosubsequenteconomic outcomes. Third, weinformallyevaluatetheperformanceofthealternativescenariosrelative to realized macroeconomic outcomes, providing illustrative examples of scenarios that shed light on key turning points. 2 Data A typical Greenbook or Tealbook A in the first two decades of the 21st-century contained around six or seven alternative scenarios, also known as alternative simulations. Staff briefings given at FOMC meetings and Bluebooks (later Tealbook B) either referred to these scenarios or introduced additional scenarios, up to a maximum of 10 total scenarios during an FOMC round. Each alternative scenario represented a potential alternative trajectory for the economy, relative to the main “baseline” forecast, which was consistently described over the years as the modal or most likely projection.8 A review of available FOMC materials reveals that the baseline forecast represented the collective judgment of a large team, informed byavarietyofmodelsandsupportedbysubstantialresources. Itwasaccompaniedbyacomprehensive write-up, comprising a large portion of the Greenbook or Tealbook, that provided detailed analysis and context. By contrast, each alternative scenario was typically built using a single model, supplemented by judgment, and developed with more limited resources. Their narratives were shorter, less detailed, and intended to illustrate risks associated with the baseline rather than to provide a full-fledged parallel forecast. As shown in Figure 1, most alternative scenarios from the late 1990s through 2020 used FRB/US, a large-scale econometric model for the United States introduced in Brayton and Tinsley (1996),9 but a variety of models were used over time. The debut of alternative scenario analysis in 1968 coincided with the development of the Federal Reserve’s first largescale econometric model, MIT-Penn-SSRC (MPS),10 which enabled periodic briefings to the 8See, for example, the January 1985 comment by James Kichline, director of the Board’s Division of Research and Statistics (R&S), “We have alleged for years that we have a modal forecast.” (Transcript of the Federal Open Market Committee Meeting on February 12–13, 1985, p. 13.) A decade later, R&S Director Michael Prell remarked, “I would characterize our forecasts over the years as an effort to present a meaningful, modal forecast of the most likely outcome. When we felt that there was some skewness to the probability distribution, we tried to identify it.” (Transcript of the Federal Open Market Committee MeetingonJuly2–3,1996,p.7.) InresponsetoaquestioninDecember2016abouthowthestaffdecidedto incorporate new fiscal assumptions in the forecast, Stephanie R. Aaronson, R&S Assistant Director, replied, “I think there are two issues. One is, how do we make the decision to put something in? And then what do we decide to put in? Because once we decide to do something, it’s a modal forecast, so we need to write downsomethingspecific.” (TranscriptoftheFederalOpenMarketCommitteeMeetingonDecember13–14, 2016, p. 32.) 9See also Brayton, Laubach, and Reifschneider (2014). 10See Brayton and Mauskopf (1985). 4

FOMC on alternative fiscal scenarios throughout the 1970s. The addition of the FRB Multicountry Model11 in the late 1970s allowed for a broader range of scenarios, notably exchange rate scenarios, and the two models were frequently used together as the “Board model” in the 1980s and early 1990s.12 The limitations of the MPS model prompted frequent complaints on the part of the staff and FOMC participants alike, but it was used out of necessity, with Governor Alan Blinder remarking, “Nobody has to believe these numbers, but I don’t know where else to get numbers other than trying to put this policy through an econometric model.”13 It was the July 1996 launch of FRB/US, featuring an improved treatment of expectations, that enabled a rapid expansion in the number of scenarios presented to the FOMC. During the 2010s, alternative scenarios were increasingly constructed using a range of New Keynesian dynamic stochastic general equilibrium (DSGE) models. The most commonly used DSGE model in our sample was SIGMA,14 with others including EDO,15 DGS,16 GST,17 GEMUS,18 Gertler-Karadi,19 and Smets-Wouters.20 In addition, scenarios were developed using event-driven extensions of baseline models or using a composition of multiple models. For example, a December 2020 alternative simulation was constructed with “a small-scale New Keynesian model with a SIR (susceptible, infected, recovered) model of viral propagation to capture the endogenous response of consumption and production to the progression of the pandemic.”21 We record the alternative trajectories of the key macroeconomic variables reported in tables accompanying the Greenbook and Tealbook A alternative scenarios. Both the forecast horizons and the set of variables varied across time, with earlier briefing materials typically presenting fewer scenarios, shorter projection horizons, and a more limited set of variables. Figure 2 illustrates this evolution: the March 1995 Greenbook showed only a couple of exchange-rate scenarios affecting a handful of variables, while later vintages (such as 1999, 2006, and 2018) featured a richer set of scenarios, longer horizons, and more detailed projections. For most of the sample period from 1989 to 2020, real GDP growth, the 11See Edison, Marquez, and Tryon (1987). 12See Brayton and others (1997) regarding the evolution of macro models at the Federal Reserve. 13Transcript of the Federal Open Market Committee Meeting on November 19, 1994, p. 32. 14First used in June 2010 to consider international risk scenarios. See Erceg, Guerrieri, and Gust (2005). 15A domestic DSGE model first introduced in April 2011. See Edge, Kiley, and Laforte (2008). 16See Del Negro, Giannoni, and Schorfheide (2015). 17See Gertler, Sala, and Trigari (2008). 18BeginningAugust2018,GEMUSisa“two-countryopeneconomyDSGEmodelwithrelativelystandard New Keynesian features.” See Bodenstein and others (2021). 19See Gertler and Karadi (2011). 20See Smets and Wouters (2007). 21TealbookAfromDecember4,2020,p.109. FordetailsontheSIRmodel,seeLepetitandFuentes-Albero (2022). 5

unemployment rate, and a measure of core inflation (CPI before 2000 and PCE after) were provided. Coverage of additional variables expanded over time, with the federal funds rate added in 2006, followed later by total PCE inflation in 2011. The evolution of the alternative simulations tables accompanied a gradual expansion and formalization of the use of alternative scenarios. A key development in this formalization was the launch of the Risks & Uncertainty chapter in the inaugural Tealbook in June 2010. This new chapter consolidated alternative simulations that had been presented separately in the domestic and international forecast sections of the Greenbook. The introduction to the Risks & Uncertainty chapter highlighted a standardized and recurring assessment of forecast uncertainty and balance of risks, which had appeared quarterly alongside the Greenbook’s domestic forecast since 2008. (Similar assessments were introduced in the FOMC Summary of Economic Projections (SEP) associated with the October 2007 meeting.) Theassessmentofforecastuncertaintyandbalanceofrisksservedasacomplementtothe scenario analysis from 2008–2020, providing a consistent summary of how the staff viewed the degree and asymmetry of uncertainty surrounding the baseline projection. Specifically, the staff evaluated overall forecast uncertainty relative to a historical benchmark—typically defined as the average magnitude of forecast errors over the previous 20 years—and classified it as being the same or higher than this benchmark (they never classified it as lower in the available sample). These assessments were stated explicitly, but they appeared in a variety of sentence structures and could be accompanied by additional context or nuance.22 We used AI methods to extract (or, as needed, infer) the risk and uncertainty assessments as documented in the appendix. In most cases, the horizon for these judgments was not explicitly stated, though a recurring footnote to a table of historical Tealbook forecast errors indicatedthat“TheTealbookforecasthorizonhastypicallyextendedabout2years.”23 Since these judgments were expressed relative to Tealbook forecast errors, they plausibly refer to a similar horizon, unless otherwise noted. In some Tealbooks later in our sample, risks are characterized for the period “over the next year or so.”24 22For example, a section labeled “Assessment of Forecast Uncertainty” in Tealbook A from October 27, 2010 stated, “We continue to see the risks around our projection for economic activity as elevated relative to the average experience of the past 20 years (the benchmark used by the Committee). [...] All told, we continue to judge the risks to our projection of real activity as skewed to the downside. We also continue to see the risks around our inflation projection as elevated relative to the experience of the past 20 years. [...] Weighingtheseriskstoboththeupsideanddownside,wecontinuetoseetherisksaroundourinflation projection as roughly balanced” (p. 79). In perhaps the most nuanced example, the December 2012 and March2013Tealbooksstated,“Withregardtoinflation,weseesignificantuncertaintyaroundourprojection but do not view these risks as unusually high” (Tealbook A from December 5, 2012, p. 83; Tealbook A from March13,2013,p.77). Wecodedbothoftheseas“typical,”overridingtheAI-assistedcodingforDecember, in order to reflect the comparison with the historical benchmark. 23See, for example, Tealbook A from October 22, 2014, p. 72. 24See,forexample,TealbookAfromOctober26,2018,p.81: “Wecontinuetoviewtheuncertaintyaround 6

Figure 3 displays these assessments in two panels, with each panel showing bars for both real GDP growth and inflation. The figure includes staff judgments given in Tealbook A as well as FOMC participant assessments in the Summary of Economic Projections. The top panel shows assessments of forecast uncertainty (elevated, typical, or diminished). Both the staff and participants judged uncertainty to be elevated for both variables from spring 2008 through 2011. Thereafter, the staff returned its assessment of uncertainty about inflation to “typical” in January 2012, while the SEP did so in September 2012. Uncertainty about real GDP returned to typical levels in both the staff’s and SEP participants’ assessments in 2013. From 2013 through early 2020, uncertainty was predominantly assessed as typical, before spiking to elevated again with the onset of the COVID-19 pandemic in March 2020. The bottom panel shows the balance of risks (downside, balanced, or upside)—whether the distribution of possible outcomes is viewed as asymmetric or skewed. The staff and participant assessments often diverged during the recovery period. For real GDP, staff judgments were predominantly skewed to the downside from 2008 through mid-2017, but the SEP’s risk assessment was balanced from 2013 onward. With respect to inflation, staff assessments were balanced throughout most of the post–Global Financial Crisis period, with a downside tilt from late 2014 until late 2016. Participant assessments of inflation risks evolved differently, underscoring how the staff dataset provides complementary information to the SEP assessments. Taken together, these two sets of judgments—about uncertainty and skewness—formed the backbone of the Tealbook’s formal treatment of forecast risk in the later part of the sample. They offered a compact, qualitative summary that complemented the richer narrative content of the alternative scenarios, providing policymakers with both structured (what-if) and probabilistic (how likely) perspectives on the risks surrounding the baseline outlook. These qualitative assessments complemented the alternative scenarios but did not determine their composition. For example, a staff judgment that real GDP risks are skewed to the downside was not associated in the sample with a larger number of downside scenarios. And although, in the case of real GDP, elevated uncertainty was associated with a larger number of scenarios (in addition to a larger range of outcomes), this relationship did not hold for inflation. Overall, this weak association underscores the interpretation of alternative scenarios as illustrative narratives of specific risks rather than probabilistic statements about the likelihood of different outcomes, a point emphasized by both Bauer and others (2025) and Garga and others (2025). thestaffforecastofeconomicactivityoverthenextyearorsoasbeinginlinewiththeaverageoverthepast 20 years, the benchmark used by the FOMC.” 7

3 Interpretation What types of scenarios did the Federal Reserve staff present to policymakers? We categorizedeachscenariousingthenarrativedescriptionsprovidedintheGreenbookandTealbook. We developed a detailed classification schema with six main categories and used a large language model to provide initial classifications, which we then systematically reviewed and adjusted. The final categories in our dataset reflect the judgments that we reached after this review process. In doing so, we contribute to a growing literature on the use of LLM-based classifiers in macroeconomic research—see Dunn and others (2024) and Kwon and others (2025) for recent examples—though we emphasize the role of human judgment in our exercise. The appendix provides details on the classification exercise. We identify six main classes of scenarios: • Aggregate Demand: Scenarios primarily related to changes in overall spending, consumption, investment, or economic activity levels. Example scenario titles: Faster Recovery, Housing Slump, Stronger Investment. • Aggregate Supply: Scenarios primarily related to the economy’s productive capacity, structural factors, or long-term economic potential. Example scenario titles: Faster Productivity Growth, Greater Supply-Side Damage, Lower Natural Rate of Unemployment. • External/Global Risks: Scenarios primarily related to international economic conditions, global shocks, or external factors affecting the domestic economy. Example scenario titles: European Crisis with Severe Spillovers, Higher Oil Prices, Stronger Dollar. • Financial Markets and Stability: Scenarios primarily related to conditions in financial markets (including equity prices), systemic financial risks, or the stability of the financial system. Example scenario titles: Higher Bond Premiums, Stock Market Correction, Recession with Financial Amplification. • Fiscal and Regulatory: Scenarios primarily related to uncertainty about future nonmonetary government policies or their effects. Example scenario titles: Fiscal Cliff, Larger Fiscal Stimulus, Political Gridlock. • Public Expectations: Scenarios primarily related to how economic agents (consumers, businesses, investors) form expectations about future economic conditions. Example scenario titles: Lower Inflation Expectations, Unanchored Inflation Expectations, Brighter Expectations. 8

Figure 4 shows the volume of scenarios of each category over time. From 2000 through 2020, Aggregate Demand, Aggregate Supply, External/Global Risks, and Public Expectations scenarios appear regularly. In contrast, Fiscal and Regulatory scenarios appear somewhat infrequently, concentrated around large fiscal debates in Congress. What were the forecasted effects of these scenarios? Figure 5 shows the outcomes for GDP and inflation in deviations from the baseline projection for each category at a horizon of one year after the Tealbook. As expected, real GDP and inflation generally moved in the same direction in Aggregate Demand scenarios and in opposite directions in Aggregate Supply scenarios. External/Global Risks scenarios featured a wide spread of inflation and output outcomes. Public Expectations scenarios, which most frequently reflect changes to inflation expectations, primarily affected inflation. Both Financial Markets and Stability scenariosandFiscalandRegulatoryscenariosprimarilyaffectedrealGDPbutwithasmaller dispersion than Aggregate Demand scenarios. Alternative scenarios presented a range of distinct risks to the baseline outlook. We categorize each scenario into four quadrants reflecting upside or downside risks to inflation and output relative to the baseline.25 Since 2000, 42 percent of the 168 Greenbooks and Tealbooks featured scenarios in all four quadrants, and 88 percent featured scenarios in three or more quadrants. How did these scenarios evolve? Alternative scenarios became more analytically sophisticated through 2020. An August 1995 scenario mechanically shifted the natural rate of unemployment up or down 0.5 percentage points, assuming policymakers knew the true value and producing symmetric results. By January 2016, natural rate scenarios incorporated learning and misperception: policymakers observed noisy signals and discovered the true natural rate only gradually, leading to persistent policy errors. Financial scenarios showed similar evolution. The November 1999 “20 Percent Stock Price Decline” scenario presented a simple wealth effect. The June 2018 “Financial Correction with Return to Effective Lower Bound” scenario modeled multiple transmission mechanisms—higher risk premiums, credit curtailment, sentiment shocks—and specified the policy response, including a return to the effective lower bound. Scenarios increasingly modeled uncertainty about economic structure itself. A June 2002 scenario simply posited a different value for the long-run equilibrium real federal funds rate (r∗), but scenarios starting in March 2015 modeled gradual recognition of a change in r∗. 25We first characterize the scenarios in terms of outcomes for GDP and core PCE inflation at a one-year horizonrelativetothebaseline. Ifthisisnotdeterminative,webreaktiesbythenumberofforecasthorizons in which the variable is higher or lower than the baseline, by related variables (total PCE inflation or the unemploymentrate),andbytheoveralltrendofthevariable(orarelatedvariable)overtheforecasthorizon. 9

4 Evaluation Evaluating scenario performance is inherently challenging. Scenarios are not predictions but rather illustrations of plausible alternatives. Nonetheless, retrospective analysis sheds valuable light. We consider the period between 2000, when alternative simulations became more numerous, and the onset of the pandemic. Figure 6 shows a scatterplot of alternative scenarios at a one-year horizon in deviations from the baseline, overlaid with realized values. Realized values incorporate all subsequent data revisions in order to compare forecasts with the best understanding of what happened.26 The spreads of scenario and realized values were comparable, suggesting that the alternative scenarios present a range of empirically plausible outcomes. For each meeting, we identify which scenario (including the baseline) came closest to actual outcomes over the subsequent year, using a distance metric based on equally weighted squared deviations in GDP growth and core inflation using current-vintage data. Figure 7 shows a time series of the most accurate scenarios and their associated forecast errors. In light of the ad hoc nature of our distance metric and the horizon chosen, this measure should be interpreted as a heuristic measure of ex post plausibility rather than the result of a statistical horse race among competing forecasts. The figure shows that even the most accurate scenario was associated with large prediction errors during periods of economic disruption. The Global Financial Crisis and COVID-19 pandemic produced forecast errors exceeding 2.5, underscoring the inherent difficulty of forecasting during major turning points or structural breaks, consistent with findings in Bauer and others (2025). However, during the relatively stable period between 2012 and 2017, forecast errors compressed dramatically, even as the number of alternative scenarios fell slightly from earlier in the sample. That said, the alternative scenarios demonstrated meaningful value in identifying the direction of economic risks, even when they failed to capture their magnitude. The baseline forecast proved closest to actual outcomes only 16 percent of the time—barely above the 14 percent success rate expected from random selection among the typical six or seven scenarios per meeting.27 This result indicates that the alternative scenarios represented genuinely plausible paths for the macroeconomy. The categorical patterns are also informative. During the financial crisis period (2006– 2009), scenarios related to financial markets and stability consistently outperformed the 26Realized values are measured as of the fourth subsequent quarter for meetings in the first half of the quarter, and as of the fifth subsequent quarter for meetings in the second half of the quarter. This ensures comparable data maturity by recognizing the lag in economic data availability. 27This is meant only as a very rough reference; a comprehensive statistical analysis depends on the range of outcomes in the scenario set. 10

baseline. In contrast, scenarios emphasizing external or global risks proved to be the most accurate in the early COVID period (2020–2021). By contrast, during the relatively stable 2012–2017 period, the baseline forecast frequently proved the most accurate. Table 1 further distills the information provided in Figure 7 by displaying the scenario (or baseline) that proved to have the lowest forecast error in any given year. In retrospect, the scenarios that came closest to the actual outcomes often featured narratives that align with conventional readings of economic developments in the first two decades of the 21st century, even if these scenarios still failed to capture the magnitudes of changes. For example, the “11,000 Wilshire” scenario of February 2000 anticipated the equity market crash one year ahead, while in June 2006 “Housing Slump” identified the peak of the housing cycle. During the financial crisis period, “Business Pessimism” (March 2007) and “Labor Market Damage” (September 2009) captured the downturn and its persistent effects. More recently, “Lower Long-Term Inflation Expectations” (October 2014) correctly identified the low inflation of the mid-2010s, while “Early Moderation” (April 2020) proved more accurate than baseline pessimism about the pandemic recovery trajectory. These examples demonstrate that the scenario framework successfully captured directional risks even when the baseline forecast proved too optimistic or pessimistic. The varied results highlight both the usefulness and inherent limitations of alternative scenarios: they excel at identifying plausible risks and directional changes but struggle—like all forecasting approaches—to anticipate the magnitude of major disruptions. 5 Conclusion Thispaperconstructsandanalyzesanewhistoricaldatasetofalternativescenariospresented to the FOMC from 1968 through 2020. From sporadic beginnings, scenarios became more frequent and standardized, and were constructed using an increasingly diverse set of models. Therisksconsideredbroadenedfromfiscal, exchange-rate, andfinancialmarketuncertainties to a wider set of demand and supply shocks and structural changes in the economy, including changes in expectations formation. Three main findings emerge from our analysis. First, in a typical Tealbook in the later part of our sample, alternative scenarios collectively represented a broad range of risks and typically spanned multiple quadrants in the output-inflation risk space. Second, the staff’s qualitative assessments of uncertainty and the balance of risks complemented, rather than mechanically reflected, the information in the alternative scenarios. Third, the ex post performance of alternative scenarios was informative: the baseline forecast was typically not the closest path to the realized outcomes, and the narratives of the closest alternative 11

Table 1: Scenario with the Lowest Forecast Error by Year Year Meeting Scenario Root Squared Error 2000 February 11,000 Wilshire 0.70 2001 November Stronger Sentiment, Policy Tightening 0.10 2002 September Baseline 0.12 2003 January Faster Productivity Growth 0.51 2004 August Less-Favorable Supply Conditions 0.05 2005 November Stagflation with Monetary Policy Response 0.10 2006 June Housing Slump 0.08 2007 March Business Pessimism 0.04 2008 December Bigger Fiscal Package 0.32 2009 September Labor Market Damage 0.26 2010 March Persistent Caution 0.25 2011 December Greater Supply-Side Damage 0.25 2012 October Baseline 0.07 2013 July Baseline 0.04 2014 October Lower Long-Term Inflation Expectations 0.14 2015 July Weaker Labor Productivity, Slower Output Growth 0.12 2016 April Stronger Foreign Growth and Weaker Dollar 0.09 2017 September Baseline 0.09 2018 March Recession 0.17 2019 October Recession 0.15 2020 April Early Moderation 0.94 scenarios often anticipated conventional accounts of macroeconomic developments. That said, the exercise also highlights the limits of scenario analysis. Even the most accurate scenarios often failed to capture the magnitude of changes. The dataset also opens several avenues for future research: whether scenarios had a measurable impact on FOMC deliberations, how scenario design changed after major macroeconomic events, and how these scenarios might be combined with probabilistic assessments. 12

60 50 40 B 30 20 B O B U 10 O B U U 0 U 1968- 80 1981- 85 1986- 90 1991- 95 soiranecS latoT F F F F F F F F F F F F F F F F G F F F F F S S S S F S F U S S E S F G B O U F U U G U U G U G G G S G G S G S S S S S O E S O E O E O U E O O O U E 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 10 8 6 4 2 0 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 Year soiranecS gniteeM rep F FRB/US S SIGMA G FRB/Global B Board Model E EDO O Other U Unspecified Figure 1: Macro Model Usage in Alternative Scenarios Over Time Note: Alternative scenarios exclude baseline projections. Models with fewer than 15 total uses across all years are grouped into “Other.” “Unspecified” represents scenarios in which model attributionwasnotdocumentedinthesourcematerials. “BoardModel”includesMPS(MIT-Penn- SSRC model) and the Multi-Country Model, the Federal Reserve’s primary staff macroeconomic models before FRB/US. The top panel shows annual counts of scenarios by model. The bottom panel shows the number of scenarios presented per FOMC meeting. Letters within bar segments identify models. Bars are stacked in the same order as the legend, though not all models appear in every year. Source: Authors’calculationsusingpublicFederalReserveGreenbooks,Bluebooks,Tealbooks,and other publicly available FOMC historical materials. 13

I-21 ALTERNATIVE EXCHANGE RATE SIMULATIONS Alternative Federal Funds Rate (Percent change, Q4 to Q4, unless otherwise noted) and Stock Market Assumptions 1994 1995 1996 (Percent change, Q4 to Q4, except as noted) Real GDP Measure 1999 2000 2001 Baseline 4.0 2.1 2.3 Stronger dollar ... 1.8 1.6 Real GDP Weaker dollar ... 2.4 3.1 Baseline 3.8 3.5 3.5 Civilian unemployment rate1 Faster productivity growth 3.8 4.2 5.6 B S a t s r e o l n i g n e e r dollar 5 . . . 6 . 5 5 . . 7 8 5 6 . .2 8 Slower productivity growth 3.8 2.8 1.6 Weaker dollar ... 5.6 5.4 Flat funds rate 3.8 3.6 4.1 CPI Tighter policy 3.8 3.2 2.7 Baseline 2.6 3.2 3.2 15,000 Wilshire 3.8 3.7 4.1 Stronger dollar ... 2.7 2.6 Weaker dollar ... 3.7 3.9 20 percent stock price decline 3.8 2.4 2.4 1. Average for the fourth quarter. Civilian unemployment rate1 (a) March 1995 (b) Nov ember 19 99 Alternative Scenarios Alternative Scenarios (Percent change, annual rate, from end of preceding period except as noted) (Perce nt change , annu al rate , from en d of preceding period ex cept as no ted) 2006 2007 Measure and scenario H1 H2 H1 H2 2008 Measu re an d sce nario 2 H 01 2 8 20 19 202 0 2021 2022 2 0 2 2 4 3 - Real GDP Greenbook baseline 4.1 1.7 2.0 2.4 2.5 Real GDP Faster labor force growth 4.1 1.7 2.2 2.8 2.9 Tealbook baseline a nd exte nsio n 2.8 2.5 1 .9 1.5 1.2 1.1 L G o re w a e te r r N w A a I g R e U a cceleration 4 4 . . 1 1 1 1 . . 7 7 1 1 . . 8 9 2 2 . . 1 4 2 2 . . 1 8 I R n e f c la e t s i s o i n o n fe ars 2 2 . . 8 8 1 2 . . 5 5 1 1 . . 3 9 1. . 2 1 1 - . . 1 8 1 2 . . 1 0 Stronger growth 4.1 2.1 2.9 3.1 2.9 Faster wage growth, su pply const raints 2.9 2 .5 1. 7 1.3 1.1 1.1 Extended housing decline 4.1 1.7 1.6 1.8 2.3 Faster wage growt h, higher produ ctivity 3.5 4.0 1. 8 .8 .5 .9 T M i a g r h k t e e t r f b i a n s a e n d c i f a e l d c e o ra n l d f i u ti n o d n s s rate 4 4 . . 1 1 1 1 . . 7 7 2 1 . . 0 8 2 1 . . 6 7 1 3 . . 9 1 H EM igh E e t r u t r r b a u d l e e n b c a e rr a ie n r d s stronger do llar 1 2 . . 8 8 2 -. . 3 0 1 1 . . 1 4 1 1 . . 3 4 1 1 . . 1 3 1 1 . . 0 3 Higher trade barriers--see through 2.2 .4 .9 1.0 .8 .9 Unemployment rate1 Unemplo yment rat e1 (c) December 2006 (d ) S ep temb er 20 1 8 Fi gure 2: Ev olut ion of A lter native Sce na rio Table Presenta tion s Note: The figure shows excerpts from alternative scena rio ta bles at four diff ere nt t ime pe riod s, illustr atin gc han gesinpresen tati onfo rma t,nu mberofsce na rios,andprojection hor izon sov ern ear ly two an d a half decades. Pa nel (a) M ar ch 1 995 shows the co mp lete t able. P ane l s ( b ) N ov e mb er 1999, (c ) D ecem ber 2006, a nd (d) Sept emb er 2018 sh ow cli pped excerpts d ispl ayin g o nly Re al GDP proj ecti ons a s given in lar ger tabl es. P anel (a) March 1995: The original Greenbook table incorrectlylabeledonevariableas“CPI”whenitshouldhaveread“CPIexcludingfoodandenergy,” which was corrected in an attached erratum to that Greenbook. Source: Pu blicly available F eder al R eser ve G reenbooks and T ealb ook s. 14

Assessment of Forecast Uncertainty GDP: Staff Elevated GDP: SEP Typical Inflation: Staff Diminished Inflation: SEP 2009 2011 2013 2015 2017 2019 2021 Date Assessment of the Balance of Risks GDP: Staff Upside GDP: SEP Balanced Inflation: Staff Downside Inflation: SEP 2009 2011 2013 2015 2017 2019 2021 Date Figure 3: Assessment of Forecast Uncertainty and the Balance of Risks Note: The figure contains two separate panels showing qualitative assessments over time, including both staff assessments and FOMC participant assessments, the latter given in the Summary of Economic Projections (SEP). Within each panel, bars are shown for both GDP and inflation, with the staff and participant (SEP) assessments displayed separately. The top panel shows the forecast uncertainty assessment: elevated uncertainty (red with crosshatch pattern), typical uncertainty (blue with diagonal stripes), and diminished uncertainty (green solid). The bottom panel shows the balance-of-risks assessment: downside risks (red with crosshatch pattern), balanced risks (blue with diagonal stripes), and upside risks (green solid). Each vertical bar represents one FOMC meeting, with SEP participant assessments submitted quarterly and carried forward to subsequent meeting dates. Staff assessments began in 2008; FOMC participant assessments began with the start of the SEP in the fourth quarter of 2007 and are submitted four times per year. SEP values are carried forward to subsequent meeting dates until the next SEP submission. The color and pattern combinations differ between the two panels. Source: Authors’ calculations using public Federal Reserve Greenbooks, Tealbooks, and the Summary of Economic Projections. 15

Aggregate Demand Aggregate Supply External/Global Risks Financial Markets and Stability Fiscal and Regulatory Public Expectations 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 Date Shading intensity: 1 scenario 2 scenarios 3+ scenarios Figure 4: Alternative Scenarios Presented to the FOMC by Risk Category, 1968-2020 Note: Each horizontal band represents a scenario category. Shading intensity indicates the number of scenarios in that category at each FOMC meeting: solid shading represents 3 or more scenarios, medium shading represents 2 scenarios, and faint shading represents 1 scenario. No shading indicates zero scenarios in that category at the meeting. Source: Authors’calculationsusingpublicFederalReserveGreenbooks,Bluebooks,Tealbooks,and other publicly available FOMC historical materials. 16

Aggregate Demand Aggregate Supply External/Global Risks 6 3 0 -3 -6 Financial Markets and Stability Fiscal and Regulatory Public Expectations 6 3 0 -3 -6 -1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0 Inflation PDG Figure 5: Alternative Scenario Deviations from the Baseline by Category Note: Values show deviations from the baseline in percentage points. Each circle represents one alternative scenario. The six panels correspond to different scenario categories. Black reference lines mark zero deviation on both axes. Source: Authors’ calculations using public Federal Reserve Greenbooks and Tealbooks. 17

6 3 0 -3 -6 -1 0 1 Inflation PDG Alternative scenarios Realized values Figure 6: Alternative Scenarios and Realized Values Note: Values show deviations from the baseline in percentage points. Small gray dots represent alternative scenarios from FOMC meetings between January 2000 and March 2020; positions are slightly jittered for visibility. Hollow orange diamonds represent realized deviations from baseline forecasts. Black reference lines mark zero deviation on both axes. Scenarios developed following the onset of COVID feature a wider range of outcomes than would fit on this chart. Source: Authors’ calculations are based on public Federal Reserve Greenbooks and Tealbooks; realized values are obtained from Bureau of Economic Analysis and Bureau of Labor Statistics via FRED. 18

3 2 1 0 2000 2005 2010 2015 2020 Date rorrE derauqS tooR Baseline Aggregate Supply Financial Markets and Stability Public Expectations Aggregate Demand External/Global Risks Fiscal and Regulatory 1-Year Moving Average Figure 7: Realized Forecast Error of Most Accurate Scenario by Meeting Note: Each point represents the most accurate alternative scenario at a given FOMC meeting from 1997 onward, showing the root squared forecast error relative to realized outcomes for GDP and inflation. Colors and shapes indicate the scenario category. The gray line shows the one-year moving average of forecast errors. Only meetings with three or more alternative scenarios are included. Source: Authors’ calculations are based on public Federal Reserve Greenbooks and Tealbooks; realized values are obtained from Bureau of Economic Analysis and Bureau of Labor Statistics via FRED. 19

References Adrian, Tobias, Domenico Giannone, Matteo Luciani, and Mike West (2025). “Scenario Synthesis and Macroeconomic Risk,” Finance and Economics Discussion Series 2025- 036. Washington: Board of Governors of the Federal Reserve System, May, https: //dx.doi.org/10.17016/FEDS.2025.036. Anthropic (2024). Claude 3.5 Sonnet. Large language model, accessed via AWS Bedrock, model ID: anthropic.claude-3-5-sonnet-20240620-v1:0. Bauer, Michael, Travis Berge, Giuseppe Fiori, Francesca Loria, and Molin Zhong (2025). “Accounting for Uncertainty and Risks in Monetary Policy,” Finance and Economics Discussion Series 2025-073. Washington: Board of Governors of the Federal Reserve System, August, https://dx.doi.org/10.17016/FEDS.2025.073. Bernanke, Ben S. (2024). Forecasting for Monetary Policy Making and Communication at the Bank of England: A Review. Bank of England Independent Evaluation Office, April, https://www.bankofengland.co.uk/independent-evaluation-office/forecasting -for-monetary-policy-making-and-communication-at-the-bank-of-england-areview/forecasting-for-monetary-policy-making-and-communication-at-thebank-of-england-a-review. Bodenstein, Martin, Pablo Cuba-Borda, Jay Faris, and Nils Goernemann (2021). “Forecasting During the COVID-19 Pandemic: A Structural Analysis of Downside Risk,” FEDS Notes. Washington: Board of Governors of the Federal Reserve System, February 1, https://dx.doi.org/10.17016/2380-7172.2806. Bordo, Michael, Eric Monnet, and Alain Naef (2019). “The Gold Pool (1961–1968) and the Fall of the Bretton Woods System: Lessons for Central Bank Cooperation,” The Journal of Economic History, vol. 79 (4), pp. 1027–1059. https://dx.doi.org/10.1017/S0022 050719000548. Bordo, Michael D., Andrew T. Levin, and Mickey D. Levy (2020). “Incorporating Scenario Analysis into the Federal Reserve’s Policy Strategy and Communications,” NBER Working Paper 27369. Cambridge, Mass.: National Bureau of Economic Research, June, https://dx.doi.org/10.3386/w27369. Brayton, Flint, Thomas Laubach, and David Reifschneider (2014). “The FRB/US Model: A ToolforMacroeconomicPolicyAnalysis,”FEDSNotes.Washington: BoardofGovernors of the Federal Reserve System, April 3, https://dx.doi.org/10.17016/2380-7172.00 12. Brayton, Flint, Andrew Levin, Ralph Tryon, and John C. Williams (1997). “The Evolution of Macro Models at the Federal Reserve Board,” in Carnegie-Rochester Conference Series 20

on Public Policy, vol. 47, pp. 43–81. https://dx.doi.org/10.1016/S0167-2231(98)0 0004-9. Brayton, Flint, and Eileen Mauskopf (1985). “The Federal Reserve Board MPS Quarterly Econometric Model of the US Economy,” Economic Modelling, vol. 2 (3), pp. 170–292. https://dx.doi.org/10.1016/0264-9993(85)90022-7. Brayton,Flint,andPeterTinsley,eds.(1996).A Guide to FRB/US: A Macroeconomic Model of the United States, Finance and Economics Discussion Series 1996-42. Washington: Board of Governors of the Federal Reserve System, July, https://dx.doi.org/10.170 16/FEDS.1996.42. Cairo´, Isabel, Christopher Gust, James Hebden, Edward Herbst, Scott Konzem, and Giovanni Nicol`o (2025). “Risk-Adjusted Optimal Policy for Scenario Analysis,” Unpublished paper, Board of Governors of the Federal Reserve System, Division of Monetary Affairs. https://edherbst.net/research/pdfs/risk-adjusted-oc.pdf. Ciccarelli, Matteo, Matthieu Darracq Pari`es, Bettina Landau, and Joa˜o Sousa (2025). “Exploring an Uncertain Future with the Help of Scenarios,” January 15. https://www.ecb. europa.eu/press/blog/date/2025/html/ecb.blog.20250115~f016f263dd.en.html. Del Negro, Marco, Marc P. Giannoni, and Frank Schorfheide (2015). “Inflation in the Great Recession and New Keynesian Models,” American Economic Journal: Macroeconomics, vol. 7 (1), pp. 168–196. https://dx.doi.org/10.1257/mac.20140097. Dunn, Wendy, Ellen E. Meade, Nitish Ranjan Sinha, and Raakin Kabir (2024). “Using GenerativeAIModelstoUnderstandFOMCMonetaryPolicyDiscussions,”FEDSNotes. Washington: Board of Governors of the Federal Reserve System, December 6, https: //dx.doi.org/10.17016/2380-7172.3678. Edge, Rochelle M., Michael T. Kiley, and Jean-Philippe Laforte (2008). “Natural Rate Measures in an Estimated DSGE Model of the U.S. Economy,” Journal of Economic Dynamics and Control, vol. 32 (8), pp. 2512–2535. https://dx.doi.org/10.1016/j.jedc.20 07.09.011. Edison, Hali J., Jaime R. Marquez, and Ralph W. Tryon (1987). “The Structure and Properties of the Federal Reserve Board Multicountry Model,” Economic Modelling, vol. 4 (2), pp. 115–315. https://dx.doi.org/10.1016/0264-9993(87)90015-0. Erceg, Christopher J., Luca Guerrieri, and Christopher Gust (2005). “SIGMA: A New Open Economy Model for Policy Analysis,” International Finance Discussion Paper 835. Washington: Board of Governors of the Federal Reserve System, July (revised January 2006), https://dx.doi.org/10.17016/IFDP.2005.835. FederalOpenMarketCommittee(2026).“Transcriptsandotherhistoricalmaterials,”Collection of historical FOMC documents including Greenbooks, Tealbooks, Bluebooks, tran- 21

scripts, and other materials. https://www.federalreserve.gov/monetarypolicy/ fomc_historical.htm. Garga, Vaishali, Edward Herbst, Alisdair McKay, Giovanni Nicol`o, and Matthias Paustian (2025). “Monetary Policy, Uncertainty, and Communications,” Finance and Economics Discussion Series 2025-074. Washington: Board of Governors of the Federal Reserve System, August, https://dx.doi.org/10.17016/FEDS.2025.074. Gertler, Mark, and Peter Karadi (2011). “A Model of Unconventional Monetary Policy,” Journal of Monetary Economics, vol. 58 (1), pp. 17–34. https://dx.doi.org/10.1016 /j.jmoneco.2010.10.004. Gertler, Mark, Luca Sala, and Antonella Trigari (2008). “An Estimated Monetary DSGE ModelwithUnemploymentandStaggeredNominalWageBargaining,”Journal of Money, Credit and Banking, vol. 40 (8), pp. 1713–1764. https://dx.doi.org/10.1111/j.1538 -4616.2008.00180.x. Koch, Christoffer, and Jackson Crawford (2019). “In Uncertain Times, Fed Sometimes Turns to’Insurance’,”FederalReserveBankofDallas,DallasFedEconomics,August13.https: //www.dallasfed.org/research/economics/2019/0813. Kwon, Byeungchun, Taejin Park, Phurichai Rungcharoenkitkul, and Frank Smets (2025). “Parsing the Pulse: Decomposing Macroeconomic Sentiment with LLMs,” BIS Working Paper 1294. Basel, Switzerland: Bank for International Settlements, October 3, https: //www.bis.org/publ/work1294.htm. Lepetit, Antoine, and Cristina Fuentes-Albero (2022). “The Limited Power of Monetary Policy in a Pandemic,” European Economic Review, vol. 147, pp. 104168. https://dx. doi.org/10.1016/j.euroecorev.2022.104168. Powell, Jerome H. (2024). “Transcript of Chair Powell’s Press Conference, December 18, 2024,” Washington: Board of Governors of the Federal Reserve System, December 18. https://www.federalreserve.gov/mediacenter/files/FOMCpresconf20241218 .pdf. ——————(2025). “Transcript of Chair Powell’s Press Conference, January 29, 2025,” Washington: Board of Governors of the Federal Reserve System, January 29. https: //www.federalreserve.gov/mediacenter/files/FOMCpresconf20250129.pdf. Smets, Frank, and Rafael Wouters (2007). “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach,” American Economic Review, vol. 97 (3), pp. 586–606. https://dx.doi.org/10.1257/aer.97.3.586. Solomon,Robert(1977).The International Monetary System, 1945–1976: An Insider’s View. New York: Harper & Row. 22

Wessel, David (2024). “Could the Fed Replace the Dot Plot with Scenarios?,” Brookings Institution Hutchins Center Blog, August 1. https://www.brookings.edu/articles/ could-the-fed-replace-the-dot-plot-with-scenarios/. 23

Appendix: Classification Methodology For the staff assessments of forecast uncertainty and balance of risks shown in Figure 3, we used Claude Sonnet 3.5 (Anthropic, 2024) to extract these qualitative judgments from narrative text in the Tealbook across 84 meetings from 2008–2020. To assess reliability, we conducted 9 independent AI coding iterations. Intercoder reliability was very high, with perfect agreement (100 percent) across all iterations for risk assessments to GDP and inflation, and near-perfect agreement (99 percent and 88 percent) for uncertainty assessments. After taking a majority vote across iterations, one manual change was made to the AI majority classifications after human review (discussed in footnote 22 supra). For the classification of scenario types shown in Figures 4, 5, and 7, we developed a classification schema iteratively. We provided Claude Sonnet 3.5 with the full list of scenario titles and asked it to suggest natural groupings based on the economic risks they appeared to address, along with a codebook to assist in classification. We then refined these initial suggestions based on our knowledge of Federal Reserve forecasting practices and macroeconomic theory. We then used Claude Sonnet 3.5 with a detailed prompt (including category definitions and examples) to classify the scenarios based on their narrative descriptions. This initial pass provided a starting point but revealed ambiguities in our category definitions, so we refined the prompt and ran the classification a second time. We then systematically reviewed the classifications, making corrections when the LLM output did not align with our intended category definitions. Based on this review, we refinedourcategorydefinitionsandappliedhumanjudgmenttoproducethefinalclassifications includedinthedataset. Approximately2percentofscenarioswerereclassifiedduringthisreview. Inaddition, categoriesweremanuallysuppliedforanadditional16percentofscenarios for which AI did not supply a classification, including cases in which multiple scenarios were described in one narrative as well as scenarios laid out in sources other than the Greenbook or Tealbook A that were identified later in the research process. The categories in our dataset represent our final judgment about the primary focus of eachscenario. WhiletheLLMprovidedausefulfirstpassforalargecorpus, theclassification ultimately reflects judgment informed by careful reading of the scenario narratives. 24