The contributions from William Weber and François Velde reveal the rich possibilities of using data to understand better the Comédie-Française’s operations. In Weber’s chapter, data help him compare nearly simultaneous crises in repertory at the Comédie-Française and the Opéra.1 Velde uses data to sketch an overview of economic planning and outcomes at the Comédie-Française, particularly by disambiguating the effects of inflation, shifting currency values, and seasonal and weekly performance patterns to elucidate real trends in attendance, incomes, and programming.2 In reply, I want to focus on a concept both writers describe in detail without fully defining: repertory.

Reading these two essays, I sensed a division between the repertory that data suggest and the repertory that the Comédie-Française’s production practices actually made in real time. Weber and Velde offer finely tuned descriptions of the effective (i.e., performed) repertory at the Comédie-Française. But how might we study repertory if we approach it not retrospectively, as something already recorded as data, but as something constantly in motion, perpetually redefined and recreated?

This essay approaches this question, rather abstractly, in two ways. First, I offer a time- and performer-dependent model of repertory, attuned to repertory as something not merely archived, but also essentially embodied. Then I discuss a framework for thinking about repertory as the result of a specific decision-making process, which I propose modeling experimentally.

I. A Diachronic, Knowledge-Dependent Model for Repertory

To begin, then, what is repertory? Neither Weber nor Velde defines the term. If we consider Velde’s discussion, repertory seems to mean “the set of works performed by the company within a given time frame,” that time frame usually being a season. Weber’s essay embraces the larger set of works not only performed by, but also available to the company at any given moment. While they differ, these definitions both suggest an archival view of repertory. Repertory here is something the company did, recorded in registers, or something the company could have done because they owned playscripts.

Repertory, however, bears a close relationship to embodiment, particularly in the thinking of performance theorists such as Diana Taylor and Tracy C. Davis. Taylor distinguishes between “repertoire” and “archive.”3 (Taylor and Davis both use the French spelling, repertoire. I will address why momentarily.) Objects fill the archive, containing, in their materiality, traces of knowledge gleaned from lived experience. The scripts in the Comédie-Française’s library form an archive, as do their collections of costumes, sets, props, prompt books, registers, etc. Taylor’s repertoire lives always in individual bodies, passed down from person to person, from master to apprentice. For Taylor, then, the repertory at the Comédie-Française lived in the performance practices (habits of gesture, of line delivery, of characterization, etc.) into which sociétaires were indoctrinated. Davis situates repertoire not only in the performer, but also between actors and audiences: “the obvious, beyond the need for comment, mundane or entirely expected aspects of performance that make theatre recognisable, predictable, and thereby comprehensible and enjoyable.”4 Repertoire brought popular theater—like that which flourished in nineteenth-century Britain—to life: a knowledge about what made theater magical that audiences anticipated and that performers provided. At the Comédie-Française, Davis’ repertoire appeared in the set of tricks/gags/turns that breathed life into the conventions of particular genres or that defined each actor’s persona, bringing the company and the crowd into unique communion.

Both definitions, with their French spelling of the word, indicate repertoire as something untranslatable, dependent on embodied knowledge. Repertoire is not something you can put into writing (though it leaves traces in the archive), but find only in particular practices that specific individuals know and experience with each other. Repertory, in the English spelling, seems then a different thing: simply a list of plays performed by a company.

But we can and should extend the idea of repertoire as embodied knowledge to our thinking about the Comédie-Française’s repertory, as a collection of plays.5 Even without embracing fully the expansive definitions from Taylor and Davis, we can relocate repertory from the archive to the bodies of the sociétaires: more precisely, to their collective knowledge of the plays they performed. I propose that we define the Comédie-Française’s repertory as a collection of performable plays, measured by their availability for performance.6 This definition marks repertory as time-specific, capturing both the work that the company can perform at will, as well as work that would require some amount of rehearsal. Repertory is thus not a stable collection of plays, definable in retrospect as the accumulated work of the season, nor is it the complete set of texts the company has performed in the past.7 Rather, plays constantly enter the repertory from the troupe’s shared attention to them, and leave the repertory through their inattention.

How might we measure a play’s availability for performance? Were I to offer a mathematical definition of repertory, I might first describe it as a function of the hours of rehearsal necessary to perform the production. Let p be a play requiring time t before the troupe can perform it for an audience. Then the repertory centrality R of p is defined as

In the minimum amount of rehearsal time, say, one minute, $R = 1$. A play requiring four hours of rehearsal has $R = \frac{1}{240}$, ten hours, $R = \frac{1}{600}$, etc. One key aspect of this equation: a play the company had already performed and since completely forgotten would be equivalent to an entirely new play, ceteris paribus. Thus this equation offers a decisively non-archival view of repertory, but rather one rooted entirely in the company’s knowledge of the play.

This formula has at least two problems, however. First, it situates knowledge in the abstract body of the company, rather than in individual performers’ bodies. Second, we cannot operationalize it, as we have no idea how long rehearsals might have taken for a given piece.

Given that rehearsal time is itself a function of the cast’s familiarity with a play, we might modify repertory centrality to be a function not of the unknowable rehearsal time, but rather of familiarity itself.8 So, let $p$ be a play with $c$ cast members. Let $t$ be the time since the cast member last performed the role. The repertory centrality $R$ of play $p$ is given as

A play performed yesterday with the same cast has $R = 1$, while a five-day-old play has $R = \frac{1}{5}$. A two-hander in which one player performed yesterday and the other ten days ago has $R = \frac{11}{20}$.

We can be even more specific, however. A role played 20 times is more familiar than one played only once previously. Furthermore, roles are not equally sized. So, let $n$ be the number of times the cast member previously performed the part. Let $l$ be the fraction of lines in the play assigned to a given cast member. Then the repertory centrality of a play is

Once cast lists have been transcribed as part of the ongoing digitization of the Comédie-Française archive, we will have data about actors’ familiarity with roles, as well as (for many plays) part length. Combining that information, we can form a diachronic, rather than synchronic, model of the Comédie-Française’s repertory. And this model will capture the Comédie-Française’s own sense of the plays they knew, rather than only the plays they, in retrospect, performed.

II. Repertory and Decision-Making

The preceding equations attempt to describe repertory mathematically as a function of a play’s availability for performance. But availability for performance, while perhaps a useful description of repertory, does not account for the actual selection of plays that make up a repertory. At the close of his essay, Velde offers that “The next step is to investigate empirically the determinants of programming choices and, if possible, model the Comédie-Française’s programming strategy.” I offer here two models that might help us take this next step by homing in on the relative importance of Velde’s “determinants” and by trying to specify the procedures that actually led to particular “programming choices.”

It is worth first recalling what we know about the Comédie-Française’s programming practices, which Velde rehearses in his essay.9 The company met weekly on Mondays to plan the next week or two of performances. It seems likely, then, that the company had a clear view of their work in the coming week, a general sense of their work for the week following, and, perhaps, some sense of when newly accepted plays might be performed even farther out. But there was not, as best we know, any real long-term planning. The company took many decisions in the weekly Monday meetings. Their repertory was thus emphatically not anything like that of a modern repertory ensemble such as a symphony orchestra or some of the larger opera companies, which plan performances years in advance. Rather, repertory arose from short-term decision making.

What factors were at play in such decisions? Velde notes that “The data we observe is the result of an interaction between audiences’ tastes and the programming choices of the actors.” We can be even more specific, despite our limited knowledge. Factors determining repertory might include (in no order):

• available company members and their roles and preferences;

• when a play was last performed (both as a matter of familiarity to the company, i.e., ease of remounting, and to the audience, for whom familiarity might have an reverse effect);

• the success or failure of other productions (e.g., one might have planned to run a new play five times, but clearly need to drop it after two performances);

• genre/type of work;

• author;

• political climate/current events;

• the balance of première and deuxième pièces;

• recent attendance figures (e.g., a bad previous week might require a riskier, more novel choice this week, to perk up attendance);10

• productions at other venues (e.g., the success of Sacchini’s Oedipe à Colone at the Opéra might encourage/discourage revivals of Voltaire’s Oedipe).11

To that list of practical, likely conscious factors, we can add a major unconscious, structural factor, namely how decisions about repertory were actually made.

I want to dwell now on this structural question, how the company actually made its decisions. We know, frankly, very, very little about this.12 And yet the process, I think, matters a great deal. I have tried to conceive of an analagous situation in the modern world: a democratically organized small group that makes collective decisions about its activities on a weekly (or even monthly) basis, with minimum constraints. Perhaps some restaurants operate in such a fashion, depending on local ingredients to set a menu weekly (or even daily). Some university curricula arise in a similar way, but only once each term.13 Variety shows such as Saturday Night Live work somewhat like the Comédie-Française, while also writing their own material. They, however, have a final arbiter, producer Lorne Michaels, and perform only once weekly for just over 20 episodes. (That would be about three weeks of programming at the Comédie-Française.) Orchestral programming is longer-term, but depends similarly on negotations among multiple parties. The New York Philharmonic’s archive documents letters in which the orchestra’s managing director explains to conductors which pieces they might viably program and which have already been claimed by other conductors for that season.14 But as a rule, organizations, particularly commercial enterprises, do not run like this anymore.

So what happened in a Monday meeting? Did someone bring in a proposed schedule for the week (or for the fortnight), which everyone then debated? Were six actors given responsibility, one for each day, and then those days harmonized? Were some slots selected monthly or quarterly or yearly, and then other works filled in around them? Did people suggest authors, and then the company suggest works? Did everyone have a clear idea of what they wanted, and then hash it out together, on the spot? What kind of compromises did people make? Who had authority in the room, when, and why?15 Would a simple majority carry the day, or did the company seek unanimity? How much attention did they actually pay to the list of factors above? Did they have any baseline (in admissions or revenue) for thinking about success and failure with respect to new or revived works? One could go on, but, clearly, we have almost no idea how the company actually chose its repertory.

Is there any way to overcome this profound ignorance? A way to reconstruct, and thus better understand, how the Comédie-Française programmed its works? I have two modest proposals to improve our knowledge, both models: (1) computer simulation and (2) role-playing. Let me explain each briefly.

The computer model would use machine learning to estimate the relative weight of different variables in choosing a week’s programming. Feed the computer model the company composition, known constraints (including new plays, illnesses, etc.) and two seasons’ worth of previous programming as priors. Then let it construct a season, week by week, balancing the weights of the given variables. The equation that minimizes deviation from the historical season as actually performed would not tell us, procedurally, how the Comédie-Française constructed its weekly repertory. But it would show us the relative importance given to the variables. That relative importance might change from season to season, thereby suggesting changes in how the company handled the information at hand.

The role-playing model attempts to answer the procedural questions: not the relative importance of different factors, but how the company weighed those factors. Begin by selecting a season to replicate and then giving a group of actors specific roles as sociétaires. The actors learn what they can about their assigned role: their age and lineage; what parts they played in recent seasons and with what success; their political and social affiliations; their financial condition; etc. They read and familiarize themselves with works in the company’s repertory and with the explicit rules and recent patterns of performance from the previous couple seasons. If company members are known to have become ill, or command performances planned, the company should be adjusted to accommodate those changes. And then the company deliberates, in a series of sessions, each a simulated Monday assembly, planning the coming week or two of performances. When new works are accepted into the repertory, they should be informed of that decision, so they can incorporate the new work in their planning. Once a session is complete, their chosen repertory should be compared to the real, historical repertory with respect to the ancientry of the plays, genres, authors, themes, etc. Then the next session should be held, but using the actual, historical repertory as a prior, including information about casting, attendance, revenues and (insofar as it is known) reception. One might even apply lessons from the computer simulation about the relative importance of different variables to the role-playing model. Ideally this experiment would be run with multiple different decision-making structures, comparing the outcomes both for their ability to reproduce the historical repertory and for their promotion of efficient and harmonious collective action. Such modeling can help us understand better the decision-making strategies the company might feasibly have used, and how they might have used information in their selection of repertory.

Both of these models can teach us only extremely limited information about the Comédie-Française’s behavior. And both would be expensive, time-consuming, and difficult to execute. But without at least some attempt to understand how season planning actually worked in practice, we can never really explain how the Comédie-Française constructed repertory from a set of complex decisions on Monday mornings throughout the season.

By way of closing, I want to reiterate the key difference I see between Weber and Velde’s excellent analyses of repertory as described by the Comédie-Française registers data, and repertory as I attempt to describe it here. We should understand repertory not only as a given, nor even as defined ex post facto, but rather always as arising from a particular set of programming decisions, each taken based on a large set of variables. To approach such a view of repertory, I offered first a mathematical measure of repertory defined diachronically, a function of the familiarity of a given work to specific company members at each moment in the Comédie-Française’s work. Even this diachronic model, however, fails to capture the process by which works come into and go out of the repertory, namely, the company’s actual decision-making. I suggested two types of simulations, one through machine learning, the other through old-fashioned role-playing, both aimed at better understanding how the Comédie-Française might actually have gone about programming each week. All three of these ideas are, in some sense, mere fantasies. But I hope they carry forward the rich data-driven sensibility in Weber and Velde’s papers in the spirit of experimentation to enliven our repertory of historiographical methods.