Working notes — patterns of collective action (pass 1)

Brief: 2026-10-29-typology-brief.md. Status: pass 1, 29 Oct. Citation-checked by an independent subagent (2026-10-29-typology-citation-check.md: 70 claims; 53 confirmed, 1 wrong and 9 overstated, all fixed below; 7 unverified and still [?]). Checked page by page against the local Markdown for Ostrom, O&R, Laffont and Martimort, and Hess and Ostrom; Taylor, Heckathorn and Macy and Flache against web copies; Kollock, Diekmann, Hirshleifer, Rapoport and Guyer against abstracts or secondary sources only.

Recheck (30 Oct). The [?] items were rechecked against newly converted local texts: Olson 1965 (5th printing, 1975; OCR scan), Sandler 1992 and 2004, Dixit, Skeath and Reiley 4th ed., and Schelling 1978 ch. 7 (page numbers from a Google Books map, not checked against print: cited as "GB-map p."). What changed:

• Sandler's aggregation mapping corrected. Sandler does not map best shot to chicken. He calls the best-shot game a "Coordination game" with a structure "reminiscent of the Battle of the Sexes" (1992, p. 41; 2004, p. 66 and Table 3.2, p. 68). Its payoffs are the volunteer's ordering (T > R = S > P), so "best shot → volunteer" holds structurally, but not under Sandler's label. Chicken in Sandler comes from summation when doing nothing is disastrous (1992, pp. 40–41; 2004, p. 62). Weakest link → assurance (1992, pp. 43–44; 2004, pp. 62–64). Threshold → assurance (2004, p. 64); Sandler does not say "volunteer above the threshold" (that part rests on Taylor's G3 and Schelling's "sufficiency" curve).
• Olson confirmed with pages: privileged / intermediate / latent, pp. 49–51; selective incentives, p. 51; "exploitation" of the great by the small, pp. 3, 29, 35. His n. 70 (p. 50) describes a privileged group failing because each waits for the others: the volunteer reading.
• Dixit, Skeath and Reiley (4th ed., not McAdams) confirmed: collective action comes in three forms, PD, chicken and assurance (p. 418), drawn as n-person payoff curves (pp. 425–430). Their "Help!" section (pp. 454–457) gives the volunteer's-dilemma model and group-size result with page numbers (after Palfrey and Rosenthal 1984). Two qualifications to KR1.2: large-group public-good provision "almost always" ends up as a PD (p. 430); and the allocation fix applies to chicken only when the social optimum has some people shirking (their "chicken II"); otherwise chicken is fixed like a PD (p. 449).
• Schelling ch. 7 confirmed in substance, but he does not use the words "chicken" or "assurance"; Dixit makes that mapping and cites his ch. 7 (Dixit p. 431, n. 3). Schelling: the PD is a "truncated dual equilibrium" (GB-map pp. 233–234).
• Settled second-hand: R. Hardin 1982 equating collective action with the PD (quoted and rejected by Sandler 1992, p. 44); Holmström's team free-riding (Sandler 1992, p. 125); Montreal as a privileged-group case (Sandler 2004, pp. 217–218).
• Still [?]: Kollock, Diekmann (the original), Heckathorn beyond pp. 256–259, Taylor beyond the pages read, Ullmann-Margalit, Marwell and Oliver, Alchian–Demsetz first-hand, and the papers listed in the Sources table.

Conventions:

• Page numbers are printed pages. "O&R" = Osborne and Rubinstein 1994 (2006 printing). "Ostrom" = Governing the Commons 1990 (Canto pagination matches the original).
• [?] = claim about a source not read in this pass. Standard results I am confident of are stated without a source tag, but any attribution to a named book is [?] unless a page is given.
• Payoffs are ordinal: 4 best, 1 worst, written (row, column). C = cooperate / contribute / hold back, D = defect.
• Symmetric 2×2 notation: R = both cooperate, T = temptation (I defect, you cooperate), S = sucker (I cooperate, you defect), P = both defect.

---

KR1.1 Payoff matrices and equilibria

The two ingredients: greed and fear

Every symmetric 2×2 game where mutual cooperation beats mutual defection (R > P) is fixed by two yes/no questions:

• Greed: if you cooperate, do I do better defecting? (T > R)
• Fear: if you defect, do I do worse cooperating? (P > S)

This is exhaustive for symmetric games with R > P: the 2×2 of yes/no answers covers all 12 strict orderings (1 + 3 + 3 + 5 across the four cells). The terms "greed" and "fear" for T − R and P − S are used by Macy and Flache 2002, PNAS 99 suppl. 3 (checked) — the classification itself is arithmetic and can be checked from the matrices below. Orderings with P > R (mutual defection better for both) are not dilemmas; deadlock is the main one.

The same split is in Heckathorn 1996, pp. 258–259, Table 2 (read): starting from the PD, reversing the two top payoffs (R > T) gives the assurance game; reversing the two bottom payoffs (S > P) gives chicken; reversing both gives the "privileged game" (= harmony); reversing the two middle payoffs (P > R) gives the "altruist's dilemma" (= deadlock).

Prisoner's dilemma — the battle line

• T > R > P > S. Greed and fear.
• D strictly dominates C. One pure equilibrium: (D, D). No mixed equilibrium (dominance rules it out).
• (C, C) is better for both than the equilibrium.
• O&R p. 16–17, Example 16.2, Figure 17.1 (read): "unique Nash equilibrium (Confess, Confess)"; O&R describe each player as having an incentive to be a "free rider".

Stag hunt — assurance

• R > T > P > S. Fear, no greed.
• Two pure equilibria: (stag, stag), (hare, hare). The first is better for both.
• Mixed equilibrium with these numbers: each hunts stag with probability 1/2 (stag pays 3p + 1, hare pays p + 2). It is unstable: any shift in beliefs tips play to one pure equilibrium.
• Which equilibrium is "risk-dominant" depends on cardinal payoffs. With chapter 1's appendix numbers (4,4 / 0,3 / 3,0 / 3,3), hare is risk-dominant (losing 3 by hunting stag alone vs losing 1 by switching to hare), and the mixed equilibrium is stag with probability 3/4. Risk dominance is Harsanyi and Selten 1988 [?].
• O&R's Example 16.1, the coordination game with a better and worse concert (p. 16, read), is the no-risk limit: "the notion of Nash equilibrium does not rule out a steady state in which the outcome is the inferior equilibrium".

Chicken (Hawk–Dove)

• T > R > S > P. Greed, no fear.
• Two pure equilibria: (C, D) and (D, C) — in each, one player gives way.
• One mixed equilibrium: each swerves with probability 1/2 (swerve pays p + 2, straight pays 3p + 1). The crash (D, D) then happens with probability 1/4.
• O&R p. 16–17, Example 16.3, Figure 17.2 (read): Hawk–Dove has "two Nash equilibria, (Dove, Hawk) and (Hawk, Dove), corresponding to two different conventions about the player who yields". O&R p. 30 (read): "Hawk–Dove is known also as 'Chicken'". The evolutionary version has a mixed ESS when c > 1 (O&R p. 50, Example 49.2, read).
• The dynamic form is the war of attrition: each chooses when to concede; O&R p. 18, Example 18.4 and Exercise 18.5 (read): in the pure-strategy Nash equilibria one player concedes immediately — but in the symmetric mixed equilibrium both wait and waste [?] (standard; Maynard Smith 1974; Bliss and Nalebuff 1984 for the volunteer reading). Dixit, Skeath and Reiley p. 422 note that chicken "may lead to a new dynamic game in which each waits for the other to build", but do not analyse the war of attrition; still [?].

Volunteer's dilemma (n players)

Diekmann's set-up [?] (Diekmann 1985, J. Conflict Resolution; original still not read). The same model, with the same mixed equilibrium and group-size result, is in Dixit, Skeath and Reiley pp. 455–457 ("Help!": A Game of Chicken with Mixed Strategies), based on Palfrey and Rosenthal 1984 and not on Diekmann: N people, benefit B, cost C < B, only one needed; P = (C/B)^(1/(N−1)); probability nobody helps (C/B)^(N/(N−1)), rising towards C/B. Here: n players; if at least one volunteers, everyone gets benefit U; a volunteer also pays cost K (0 < K < U); if nobody volunteers, everyone gets 0.

Two-player version (U = 4, K = 1):

• T > R = S > P. Chicken with a tie: volunteering costs the same whether or not the other also volunteers.
• Pure equilibria: exactly one volunteer (n of them in the n-player game). Nobody volunteering is not an equilibrium.
• Symmetric mixed equilibrium: each waits with probability q where U − K = U(1 − q^(n−1)), so q = (K/U)^(1/(n−1)). Probability nobody volunteers = q^n = (K/U)^(n/(n−1)), which rises with n towards K/U. Each person is less likely to act, and the group as a whole is less likely to get anyone, as the group grows. This is the formal match to the bystander effect (Darley and Latané 1968 [?]). Dixit, Skeath and Reiley p. 457: "the larger the group, the less likely is help to be given at all", though it "levels off at a positive value"; they report the crowd-size experiments second-hand via Sabini (p. 455).
• With U = 4, K = 1, n = 2: q = 1/4, nobody acts with probability 1/16. With n = 10: q ≈ 0.86, nobody acts with probability ≈ 0.21.

Pure coordination

• No conflict at all: the only problem is matching.
• Two pure equilibria (A, A), (B, B); mixed: each picks A with probability 1/2.
• If one convention is better (O&R's Mozart/Mahler, 2,2 vs 1,1, p. 16): mixed equilibrium is Mozart with probability 1/3.
• Fixes: a focal point, a convention, a signal (Schelling 1960 [?]; Lewis 1969 [?]).

Battle of the sexes (Bach or Stravinsky)

• Both want to match; each prefers a different match.
• Two pure equilibria (Bach, Bach), (Stravinsky, Stravinsky); mixed equilibrium ((2/3, 1/3), (1/3, 2/3)) with expected payoffs (2/3, 2/3), worse for both than either pure equilibrium. O&R p. 15–16, Example 15.3 and p. 34–35, Example 34.1; payoffs p. 46, Example 46.1 (read).
• Relabel the moves as "insist on mine" / "give way" and BoS has the same shape as chicken: two asymmetric equilibria, and a fight over which. The difference is that in chicken the mutual-insist cell is the disaster, while in BoS both mismatches are bad.
• O&R p. 44 (read): with a public random signal (a coin both see), there is a correlated equilibrium that alternates between the two — the formal version of taking turns.

Deadlock and harmony (non-dilemmas, for completeness)

Deadlock (Heckathorn's "altruist's dilemma", p. 258–259):

• T > P > R > S. D dominates and (D, D) is what both prefer. No collective action problem: cooperating would be a mistake.

Harmony (Heckathorn's "privileged game", p. 259; Olson's privileged group):

• R > T > S > P. C dominates. Heckathorn p. 259: "Because individual and collective rationality coincide perfectly, there is no social dilemma."

Completeness against the 2×2 classification

• Rapoport and Guyer 1966 list 78 strictly ordinal 2×2 games; Robinson and Goforth 2005 arrange 144 games (78 up to swapping the players) in a "periodic table" [?] (neither read). Heckathorn 1996 uses the Rapoport–Guyer numbers (assurance #61, privileged #6, altruist's dilemma #9, chicken #66), p. 258–259 (read).
• For symmetric games with strict preferences, the greed/fear 2×2 covers every ordering with R > P; the orderings with P > R (deadlock among them) are not dilemmas. The asymmetric games add mixed cases (e.g. Taylor 1987, p. 39, Figure 5: "PD for Row, Chicken for Column" — read) and pure-conflict games (matching pennies, O&R p. 17), which aren't collective action problems.
• Pure coordination and BoS are not in the symmetric C/D frame because their strategies are "which convention", not "contribute or not". They sit naturally as the no-greed, no-fear end (pure coordination) and the who-gives-way family (BoS).

So the candidate categories in the brief cover everything that matters: one bad equilibrium (PD), two ranked equilibria (stag hunt; pure coordination as its riskless limit), two asymmetric equilibria (chicken, volunteer, BoS), and the two non-dilemmas.

---

KR1.2 Are chicken and the volunteer's dilemma the same as the prisoner's dilemma?

No. They differ by one swap in the matrix, and the swap changes what goes wrong and what fixes it. Rufus's intuition has something to it too, set out at the end.

At the level of the matrix

• PD and chicken have identical top halves: in both, if you cooperate I'd rather defect (greed).
• They differ only in the bottom two payoffs. In the PD, being the lone cooperator is the worst outcome (S < P). In chicken, mutual defection is the worst outcome (P < S): the crash, nobody calling the ambulance, the dyke nobody repairs.
• That one swap changes the equilibria:
  • PD: one equilibrium, nobody cooperates. Defecting is best whatever the other does.
  • Chicken / volunteer: no dominant strategy. If I'm sure you won't act, I will. Equilibria have someone acting. Plus a mixed equilibrium where disaster happens with positive probability.
• Heckathorn 1996, p. 258 (read): "When the value of S increases until it exceeds the value of P, a PD is transformed into a chicken game." It happens when the production function becomes more sharply decelerating — the first contribution produces most of the good.
• Taylor 1987, p. 36–37 (read), on two cultivators maintaining dykes: "either individual alone can do it all, but each prefers the other to do all the work. The consequences of nobody doing the work are so disastrous that either of them would do the work if the other did not." That is chicken, not PD.
• All the textbooks and surveys read or checked treat them as distinct: O&R list Hawk–Dove as a separate classical game (p. 16–17); Taylor 1987 ch. 2 (read); Heckathorn 1996 (read); Kollock 1998's abstract names PD, assurance and chicken as the three key two-person dilemmas (abstract only, via web). Dixit, Skeath and Reiley p. 418: collective-action games "come in three forms … the prisoners' dilemma, chicken, and assurance games". Sandler 1992 p. 44, after his 2×2 examples (Fig. 2.6, p. 39): "collective action problems need not imply a Prisoner's Dilemma", rejecting R. Hardin's (1982, p. 25) "the problem of collective action and the prisoner's dilemma are essentially the same".
• Sandler 1992 pp. 40–41 derives chicken from the PD the same way: impose a penalty on mutual defection bigger than the sucker's loss, and "game structures very similar to the Prisoner's Dilemma need not imply the absence of collective provision".

At the level of what goes wrong

Taylor p. 43 (read): in N-person chicken, "in the rush to be among the first to make a commitment to non-Cooperation … the number so binding themselves may exceed the maximum number" that still allows provision. Yet "the prospects for Cooperation are a little more promising in Chicken games than they are in the Prisoners' Dilemma" (p. 43).

At the level of what fixes it

Qualification (Dixit, Skeath and Reiley p. 449, read 30 Oct): the allocation fix is distinctive only when the best outcome for the group has some people shirking ("chicken II", where one builder is socially optimal). When everyone acting is best ("chicken I"), "the problem is much like that of sustaining cooperation in the prisoners' dilemma". For chicken II they give custom and rotation: "The roles of participants and shirkers can be rotated to equalize payoffs over time." For assurance: make the good equilibrium a focal point through custom or convention (p. 445). The table above should be read as being about the volunteer case where one actor is enough.

The volunteer's distinctive fix is allocation: name the person ("you in the red coat, call an ambulance"), rotate the duty (Alanya's fishing-site rotation, Ostrom p. 18–20), take turns (O&R's correlated equilibrium, p. 44), or pay the volunteer. None of those helps in a battle line, where everyone has to be restrained.

Where Rufus is right

• The symptom can look the same: something isn't done. You can't tell the pattern from the outcome, only from the payoffs.
• Real cases slide between them. Heckathorn's game-space diagram (p. 257, Figure 2, read) shows PD, chicken, assurance, privileged and altruist's dilemma as neighbouring regions of one space, with the shape of the production function on one axis and the value of the good relative to its cost on the other. Small changes move a case across a border. Ostrom p. 47–48 (read): CPR situations "face a variety of problems", whose structure "will depend on the values of underlying parameters".
• The PD region is narrow in Heckathorn's model: for a two-person collective good it is a true PD only if the good's value is between 1 and 2 times its cost and the production function is not too decelerating (Theorems 1 and 2, p. 258, read).
• Repetition blurs the line. In the repeated PD the cooperative outcome becomes an equilibrium, alongside very many others, so the problem becomes which equilibrium — a coordination problem (O&R p. 134, read: the set of equilibrium outcomes "is huge"). Taylor p. 40 (read) notes the N-person PD supergame can itself be a chicken game.
• Group size can turn the other patterns into a PD. Dixit, Skeath and Reiley p. 430: when N is very large and one person makes very little difference, each always prefers to shirk, "That is why problems of collective provision of public projects in a large group almost always manifest themselves as prisoners' dilemmas" (not so for small groups, or for congestion). Schelling ch. 7 (GB-map pp. 233–234): with a fixed viable-coalition size k, shrinking the population turns a two-equilibrium (assurance-type) case into a multi-person PD: "The MPD is merely a 'truncated dual equilibrium'". Schelling's main division (GB-map p. 225) is between a superior outcome that is "self-enforcing once arrived at" (two equilibria) and one that "requires coercion, enforceable contract, centralization of choice"; the PD is "a special, but not very special, subclass" of the latter.
• Two-player chicken is mostly about bargaining (Heckathorn p. 258: "The essential problem in a chicken game is bargaining"). That makes it feel less like collective action than PD and stag hunt — which it partly isn't. The n-person volunteer's dilemma is the collective-action version.

---

KR1.3 Recommended top level

Three patterns, read through greed and fear

Off the grid, one line each:

• Pure coordination (which side of the road): no greed, no fear, only matching. The easy baseline. Fix: a convention or focal point.
• No problem (harmony / privileged): cooperating pays anyway. Useful as the destination: most fixes work by moving a case here.

Reasons this suits the audience:

1. It turns a list into a map. Two questions — "would I gain by slacking if others pull their weight?" (greed) and "would I lose by pulling my weight alone?" (fear) — generate the three patterns. That answers "aren't these the same thing?" directly: they share ingredients, in different combinations.
2. The fixes fall out of it. Remove greed and a battle line becomes a stag hunt; remove fear and it becomes a volunteer; remove both and the problem disappears. This matches chapter 1's trailer ("a moderate penalty turns a battle line into a stag hunt") and the course's "collective action opportunities" framing.
3. It keeps chapter 1's names and grid, so course-wse.23 needs re-explanation, not restructure.
4. It stays within the brief's four-category limit, and pure coordination is kept as a footnote rather than a fourth headline.

The taking-or-building axis

Recommend keeping it as the second top-level axis, with a precise definition:

• Taking = the cooperative move is restraint: not taking, not emitting, not arming, not talking to the prosecutor. My defection imposes a cost on others.
• Building = the cooperative move is contribution: work, money, code, turning up. My defection withholds a benefit.

Defined this way the axis also covers cases that aren't commons in the narrow sense: arms races and the ELF informants are "taking" (each defection harms others).

Grounding in Ostrom (read):

• Ostrom p. 30–31: "appropriation" (withdrawing resource units) vs "providers" and "producers" (who arrange for and build or maintain the resource system).
• Ostrom p. 47: "Appropriation problems are concerned with the allocation of the flow; provision problems are concerned with the stock." Both are present in every CPR.
• Ostrom p. 49: "unless appropriation problems are resolved, the provision problems may prove intractable." In a pure public good there are no appropriation problems "because resource units are not subtractable".
• Ostrom p. 32: there "is as much temptation to avoid contributing to the provision of a resource system as there is to avoid contributing to the provision of public security or weather forecasts".
• Caveat: Ostrom's split is within common-pool resources, and her provision problems include a demand-side part (regulating withdrawal rates so the stock survives, p. 49) that is really "taking". "Taking vs building" is a simplification of her distinction, not a quotation of it.

Information commons are mostly building: knowledge is "relatively non-subtractive" (Hess and Ostrom 2007, pp. 5, 9, read), though enclosure can turn it into a common-pool resource (p. 10). But they have a taking side too: Hess and Ostrom list "commodification or enclosure, pollution and degradation, and nonsustainability" as typical threats to knowledge commons, and congestion and free riding on the web (p. 4–5, read). So even information commons have both rows.

What the axis does and doesn't do:

• In a linear 2×2 game the matrix for taking and building is identical. The axis is about the good, not the equilibrium structure.
• It matters because the remedies differ: taking needs limits on withdrawal, boundaries and monitoring (Törbel, Alanya); building needs selective incentives, assurance, patrons, and credit (Phil Trans priority, assurance contracts).
• It matters because aggregation differs: building goods often have thresholds and best-shot technologies, which push cases into the stag hunt and volunteer columns.
• Chapter 1's grid marks taking × volunteer as "(rare)". That is wrong. Taylor's fishery that tolerates two boats but collapses with three (p. 43–44, read) and his two factories on a small lake (p. 37, read) are taking-side chicken games. Ostrom p. 47 (read): the Alanya fishery "cannot be represented as a PD game" and has the structure of an "assignment" game — who gets the good fishing spots. Proposed cell text: "who gets the good spot: fishing sites, turns at the water".

---

KR2.1 Typologies in the literature

What the sources agree on:

• Nobody serious equates collective action with the PD alone any more. Taylor p. 18: "we should not equate 'the problem of collective action' with the Prisoners' Dilemma, as many writers have done".
• The three-game set PD / assurance / chicken is the standard core (Taylor, Heckathorn, Kollock [?], Ostrom p. 47; Dixit, Skeath and Reiley p. 418). Sandler 1992 (Fig. 2.6, p. 39) uses PD, privileged, chicken, coordination/BoS and two assurance games.
• The dispute: whether assurance is a "problem" at all. Taylor says no (p. 19, 39). Ostrom/Bates treat it as a real but easier problem (p. 42). The lab evidence on coordination failure in minimum-effort games (Van Huyck, Battalio and Beil 1990 [?]) and the bank-run literature (Diamond and Dybvig 1983 [?]) side with treating it as a problem. That is a layer-2 question for pass 3.

---

KR2.2 Second-level axes

The axes that do most work for this course: number of players, aggregation, and repetition. They explain most of the double-fits in the case test below.

---

KR3.1 Second layer: pattern, cause, remedy or adjacent

---

KR4.1 Recommended pattern set for the course

Recommendation: keep three patterns — battle line, stag hunt, volunteer — but teach them as one map built from greed and fear, keep the taking/building rows, and broaden the volunteer to "who gives way". Pure coordination is a one-line baseline; harmony is named as the destination of a fix.

Tied to the shortlist criteria:

• Teaches a mechanism. Each pattern comes with its own fix, which is the reusable move: change the payoffs (battle line), give assurance (stag hunt), allocate the job (volunteer). Dropping the volunteer loses the allocation move, which several shortlist cases teach (Alanya rotation, Dr. Strangelove's commitment device, the liberum veto's hold-outs).
• Structural distinctness. The shortlist already requires PD, coordination and assurance, threshold goods and arms races. The volunteer column is where the threshold-above-the-line cases, standoffs and the Schelling module (S4) land. S4 needs chicken anyway for Strangelove and brinkmanship.
• Cuts against priors. This audience thinks every collective action problem is Moloch, a battle line. The greed/fear map shows that two of the three patterns are easier than that, and why. Rufus's own chapter 1 lesson — "if you can change the payoffs, you can change the pattern" — becomes concrete: removing greed or fear moves a case one cell towards no problem at all.
• Is it felt. The two questions are things people can answer about their own case without game theory.

Changes this implies for the volunteer:

• Explain it as "greed without fear": if nobody else does it I will, but I'd rather it was you.
• Lead with cases where the distinctive failure shows: who pays, waiting, brinkmanship. The street collapse is fine. Heartbleed is a weak lead example, because it also fits a battle line (see the case test).
• Fill the taking × volunteer cell: who gets the good spot (Alanya).

---

KR4.2 Case-classification test

Key: BL = battle line, SH = stag hunt, V = volunteer / who gives way, Co = pure coordination, H = harmony (no dilemma). T = taking (restraint), B = building (contribution).

Chapter 1 cases

Shortlist cases

Results

Orphans (no pattern fits): Twelve Angry Men (information cascade); meerkat sentinels and probably leaded petrol (non-dilemmas — they fit "no problem"); Aral Sea, the Ik, the Donner Party, Jaws (not collective action structures). None of these is a reason to add a pattern. Twelve Angry Men belongs on the information axis.

Axis orphans (pattern fits, taking/building doesn't): bilateral exchange and mutualism cases (vampire bats, cleaner fish, cleaner wrasse, acacia ants); the liberum veto. Arms races, Moloch and the ELF fit "taking" only under the restraint definition above.

Double-fits (twelve-plus): hoplites, playground, Heartbleed/xkcd, herds/buffalo, climate, AI race (triple), naval race, LTBT, Great Stink, subak, Wikipedia/maintainer, liberum veto, Plataean breakout. Almost all are either (a) many-player threshold situations, which have stag-hunt features below the threshold and volunteer features above it (Taylor's G3, p. 41–43), or (b) cases whose payoffs change with a parameter (open access vs limited access; unseen vs salient disaster). That matches Heckathorn's game space (p. 257) and Ostrom p. 47–48. It is a teaching point — "the pattern is in the payoffs, and payoffs move" — not a flaw in the set.

Volunteer column, clean cases: street collapse, Dr. Strangelove, Alanya, brinkmanship, the threshold fishery (Taylor). Heartbleed is a double. The shortlist has few clean volunteer cases, which is part of why the pattern feels thin.

---

KR4.3 The volunteer: keep, replace, or footnote?

Keep, renamed or subtitled "who gives way", and explained as greed without fear.

Reasons:

1. It is structurally distinct from the PD (different equilibria; KR1.2), and every source read treats it as distinct.
2. Its fix is distinct: allocation (designate, rotate, pay), which neither the battle line nor the stag hunt teaches.
3. Several course cases need it: Alanya, Strangelove, standoffs, hold-outs, the street collapse. S4's Schelling module needs chicken regardless.
4. In the greed/fear map it costs no extra concept: it's the fourth combination, and leaving the cell empty would invite the question.

What to change: broaden it to include battle of the sexes (standards wars), hold-outs and the war of attrition; drop Heartbleed as the lead example (or present it as a double-fit); fill the taking × volunteer cell.

The alternative, if Rufus wants two patterns: demote the volunteer to a footnote under the battle line ("when one person acting is enough, the problem changes from 'will anyone' to 'who'"), and teach pure coordination in the stag hunt's place as the easy end. This costs the allocation fix and a slot S4 will need later.

---

KR4.4 Decision options for Rufus (draft)

Question: How many patterns does chapter 1 teach?

Second, smaller question: keep "taking or building" as the row labels? Recommend yes, renamed from "taking too much / giving too little" to "taking / building", defined as restraint vs contribution. Means relabelling two grid rows.

---

Sources used

Zotero full-text caches were searched for "volunteer's dilemma", "assurance game", Heckathorn, Kollock and Diekmann: only Ostrom and one unrelated Bowles paper matched.

Main [?] items to check when the books arrive (course-wse.44)

1. Sandler 1992: the aggregation-technology → pattern mapping (checked 30 Oct). Correction: Sandler maps best shot to a "Coordination" (BoS-like) game, not chicken; chicken comes from summation with a disastrous status quo. Chapter 1 n. labels and any wiki page that says "best shot → chicken (Sandler)" need correcting; "best shot → volunteer" can stand as our reading of his payoffs.
2. Kollock 1998: section structure, page numbers, and the solutions typology. Still [?].
3. Diekmann 1985: original still not read; the model and group-size result are now page-cited from Dixit, Skeath and Reiley pp. 455–457 (who credit Palfrey and Rosenthal 1984).
4. Olson 1965: privileged / intermediate / latent, "exploitation of the great by the small" (checked 30 Oct: pp. 49–51; pp. 3, 29, 35).
5. Dixit and Skeath: the collective-action chapter's three-game classification (checked 30 Oct: 4th ed. with Reiley, pp. 418–430).
6. Macy and Flache 2002 for the greed/fear terminology (checked in the citation pass).
7. Holmström 1982 (second-hand via Sandler 1992 p. 125) and Alchian and Demsetz 1972 (via Ostrom) at first hand. Still [?] first-hand.
8. Ullmann-Margalit 1977: whether "norms of partiality" map to anything here. Still [?].
9. New: Schelling 1978 ch. 7 page numbers come from a Google Books map; check against a print copy before quoting pages in public material.