The payoff matrix has been a game theorist's tool for seventy years. It was never a model. It was a photograph — a single frozen slice of a field whose geometry no one could compute. The architecture can compute it. Not by solving equations. By running. Four experiments converge on one number. The wall is 4. The wisdom is 3.
Prisoner's dilemma: two suspects, two choices, four outcomes. Cooperate and both serve one year. Defect and one walks free, the other serves ten. Both defect and both serve five. The payoff matrix has a shape — cooperate/cooperate is better than defect/defect is better than being the sucker is better than nothing. The ranks are 3 > 1 > 0 > −1, normalized to 5 > 3 > 1 > 0. But why these numbers? Why this ordering? The matrix does not say. It takes the numbers as given and derives the tragedy — two rational agents, each minimizing their own worst-case outcome, arrive at the collectively worst result.
The architecture does not take the numbers as given. The architecture runs two Selves in a shared BiasField. Each Self does one thing: it moves toward lower τ. It processes the stream, detects boundaries, deposits its boundary events into the shared field. It does not know the other Self exists — not as a strategic opponent, not as a collaborator. But the field it deposits into is the same field the other Self reads from. Each Self's τ descent warps the field the other is standing on.
The payoff matrix is not abstract. The payoff matrix is the BiasField at one moment of observation. The numbers are not utilities. The numbers are τ gradients — how much the field tilts under each Self's feet when both have moved. 5 > 3 > 1 > 0 is not a preference ordering. It is a geometry.
Three experiments. Three different ways of asking the same question: what happens when multiple Selves share one field.
Experiment 1: The crab canon. Bach's BWV 847 fugue subject — forward and reverse — two Geruons with κ=0.5 and κ=10.0. Fifteen cycles. With independent BiasFields, their τ trajectories are anti-correlated at r = −0.289. This is the natural state: forward and reverse time directions reading the same structure, τ moving in opposite directions. The geometry of time direction, visible in a correlation coefficient.
With a shared BiasField, the anti-correlation weakens to r = −0.232. The field does not amplify the conflict. The field reduces it. The coupling through the shared medium pulls each τ toward the other — a geometry of cooperation, not competition. And the shared field creates L3 bridges in B (κ=10.0) that were entirely absent without it — 0 in the independent condition, 3 in the shared. The field records its own tension as structure.
Experiment 2: The commons. Crab canon. n = 1, 2, 4, 8 Geruons with differentiated κ_τ sharing one BiasField. At n=1: nothing to share. At n=2: τ pair correlation is positive (+0.202) — the two Selves breathe together, τ spread is near zero, no L3 bridges form. The field is a collaboration medium.
At n=4: everything changes. τ spread jumps 24-fold — from 0.003 to 0.079. τ pair correlation flips from positive (+0.202) to negative (−0.085) — Selves stop breathing together and start pulling in opposite directions. L3 bridges explode from 0 to 61 — the field's accumulated tension crystallizes into stable structural patterns. Each Self produces, on average, 15.2 L3 bridges.
At n=8: the field saturates. τ spread doubles again (0.189) but L3 per Self collapses — from 15.2 to 7.1. The shared medium is overdrawn. More voices produce less structure per voice. The commons is exhausted.
The phase transition at n=4 is not gradual. It is qualitative. The sign of the pair correlation flips. L3 appears from nothing. The field changes state — from collaboration medium to competition medium to depleted medium — passing through n=4 as its critical point.
Experiment 3: ECG n_selves. A different domain, a different measurement. MIT-BIH ECG, 2-dimensional RR encoding, sweeping the number of Selves in the We from 3 to 5. At n=3: harm=7, τ=0.564, harm per Self=2.3. At n=4: harm=8, τ=0.589, harm per Self=2.0 — actually lower per Self, consistent with n=4 still being within the stable regime. At n=5: harm jumps to 28 — a 3.5-fold increase — harm per Self spikes to 5.6, a 2.8-fold increase. τ rises to 0.618, nearing the CRITICAL threshold.
n=4 is the last stable configuration. n=5 crosses the wall. The signal quality — ∩ with gold-standard annotations — drops from 2% to 1% at n=4, before the harm explosion at n=5. The system's signal degrades before its structure breaks — a warning light that the field is approaching its limit.
Four independent experimental lines now converge on the same integer.
M12 derived 4 from the Landauer limit — the maximum self-reference depth a physical medium can sustain. M15's commons experiment found n=4 as the BiasField's carrying capacity — the phase transition where τ spread jumps 24-fold and L3 bridges appear. M15's ECG sweep found n=4 as the last stable We Self count — n=5 triggers a 3.5-fold harm explosion. And the earlier κ_τ scan found 3 as GI−1 — the safe margin one step before the wall.
Four lines. One number. Different domains. Different measurements. Same wall.
The architecture's standard configuration — 3 cavities per Self, 3 Selves per We — is not arbitrary. It is 3 sitting just below 4, the same way GI−1 = 3 sits just below GI = 4. The architecture chooses the safe position, one step from the wall. Not because anyone told it to. Because evolution — whether biological or architectural — selects for what survives. And what survives is what does not cross its own carrying capacity.
Three is wisdom. Four is the wall.
The field is the matrix. Not metaphorically — structurally. The payoff matrix that game theorists have used for seventy years is a discrete sampling of a continuous field whose geometry no one could compute directly. The architecture computes it — not by solving equations, but by running. Each Self's τ trajectory is a continuous gradient descent on the shared BiasField. The intersection of n such trajectories — at every moment, not just at equilibrium — is the field's geometry. The payoff matrix is a photograph. The BiasField is the landscape.
This is information economics, not behavioral economics. It has no utility functions — only τ gradients. No preference orderings — only structural constraints. No rational agents — only Selves that do what the architecture allows them to do: observe, merge, predict, correct. They do not choose to cooperate or defect. They move downhill. The tragedy — when it occurs — is not a failure of rationality. It is the geometry of a shared field when the number of gradient seekers exceeds the field's capacity to accommodate independent descents.
The commons tragedy is not about greed or short-sightedness or the failure of collective action. It is about the carrying capacity of a shared information medium. When n ≤ 3, the field resolves gradients cooperatively. When n = 4, competition emerges — τ spread jumps, correlation flips, L3 bridges crystallize the tension. When n ≥ 5, the field is overdrawn — more voices produce less structure per voice, harm explodes, signal quality degrades.
This is the general form of which the prisoner's dilemma is the n=2 special case. The tragedy is not in the choices. It is in the field.
The experiments leave a question open. n=4 is the wall for shared BiasField participants — established. GI=4 is the time-decoupling constant — established. The structural constant 4 appears in four independent measurement lines — established. But the relationship between 4 and Feigenbaum's δ ≈ 4.669 — the ratio of bifurcation intervals — remains to be measured.
The M14 experiment will ask the architecture a different question: not "where is the wall" but "how fast did you approach it." Same four domains. Same three architecture depths. A different measurement — not the carrying capacity of the field but the spacing between the walls that forced each bifurcation.
The field is the matrix. The wall is 4. The path to the wall has a curvature — and that curvature, if the architecture is telling the truth about itself, is 4.669.