A world made of little agreements

Imagine three adjacent rooms, labeled $A$, $B$, and $C$, arranged in a triangle around a central courtyard.

Each room contains a simple two-position light switch, which can point either up ($+1$) or down ($-1$).

On each of the three connecting doorways, a landlord posts a straightforward rule:

• Between rooms $A$ and $B$: the two switches must point in opposite directions.
• Between rooms $B$ and $C$: the two switches must point in opposite directions.
• Between rooms $C$ and $A$: the two switches must point in opposite directions.

Each individual rule is entirely reasonable and easy to satisfy.

Yet the moment you try to set all three switches simultaneously, a baffling frustration occurs:

1. Start with switch $A$ pointing up ($+1$).
2. To satisfy the first doorway, switch $B$ must point down ($-1$).
3. To satisfy the second doorway, switch $C$ must point up ($+1$).
4. But now inspect the third doorway between $C$ and $A$: both switches point up! The rule is violated.

Flip switch $A$ down, and the conflict merely leaps to a different doorway. No matter how you arrange the three switches, at least one doorway is left unsatisfied.

Every local instruction made perfect sense in isolation. Yet the collection of all three instructions cannot coexist in any single universe.

When I encountered this simple architectural paradox, it opened up a profound realization: a collection of valid local agreements does not automatically assemble into a global reality. This single phenomenon unifies the physics of frustrated magnets, the geometry of quantum contextuality, the consistency of distributed databases, and the dynamics of social consensus.

Let us trace how these local tensions give rise to global geometry.

The impossible triangle: When local harmony breeds global paradox

Let us translate our three doorway rules into elementary algebra.

Represent each switch by an algebraic sign variable: $A,B,C\in \{-1,+1\}$.

The product of two signs measures their agreement:

• If two switches point the same way, their product is $(+1)(+1)=(-1)(-1)=+1$.
• If two switches point in opposite directions, their product is $(+1)(-1)=(-1)(+1)=-1$.

Our landlord’s three doorway rules demand:

$$AB=-1,BC=-1,CA=-1.$$

Now multiply all three equations together:

$$(AB)(BC)(CA)=(-1)(-1)(-1).$$

Group the variables on the left side:

$$A^2{B}^2{C}^2=-1.$$

Because the square of any real sign is strictly positive ($(+1{)}^2=1$ and $(-1{)}^2=1$), the left side evaluates to:

$$1=-1.$$

Our three simple rules demand a logical impossibility!

Now introduce probability. Suppose an observer does not inspect the rooms simultaneously, but measures them two at a time.

Across many independent visits, the observer finds that each individual switch is up half the time: $\mathbb{E}[A]=\mathbb{E}[B]=\mathbb{E}[C]=0$.

Furthermore, whenever the observer checks a doorway, the two switches are always opposite: $P(A\ne B)=1$, $P(B\ne C)=1$, and $P(C\ne A)=1$.

Each of the three pairwise reports defines a completely valid bivariate probability table: an even split between $(+1,-1)$ and $(-1,+1)$. Each report matches the individual $50/50$ statistics of every room.

Yet there exists no joint probability distribution $P(A,B,C)$ over all three rooms that can reproduce these pairwise observations! Local observations can be completely self-consistent while their global synthesis is mathematically impossible.

How much disagreement can a universe hold? The 2/3 bound

What happens when we relax our absolute demands?

Let $d_{AB}=P(A\ne B)$ denote the probability that switches $A$ and $B$ disagree. Define $d_{BC}=P(B\ne C)$ and $d_{CA}=P(C\ne A)$ similarly.

Consider any deterministic configuration of the three switches $(A,B,C)\in \{-1,+1{\}}^3$. There are $2^3=8$ possible configurations.

Let us count the number of disagreeing edges in each configuration:

• In the two uniform configurations $(+1,+1,+1)$ and $(-1,-1,-1)$, all three pairs agree: the number of disagreeing edges is 0.
• In the other six configurations (such as $(+1,-1,+1)$), exactly two pairs disagree and one pair agrees: the number of disagreeing edges is 2.

Notice an extraordinary combinatorial fact: a triangle can have 0 disagreeing edges or 2 disagreeing edges, but it can never have 1 or 3!

Because every individual state has at most 2 disagreeing edges, the expected sum of disagreements across any mixture of states must satisfy:

$$\boxed{d_{AB}+d_{BC}+d_{CA}\le 2.}$$

In addition, disagreement probabilities satisfy metric triangle inequalities. If $A$ disagrees with $B$, then for any intermediate switch $C$, either $A$ must disagree with $C$, or $C$ must disagree with $B$ (or both):

$$d_{AB}\le d_{BC}+d_{CA}.$$

Rotating the labels gives two companion inequalities:

$$d_{BC}\le d_{AB}+d_{CA},d_{CA}\le d_{AB}+d_{BC}.$$

Now ask a fundamental question: what is the maximum symmetric disagreement rate that all three pairs can share equally ($d_{AB}=d_{BC}=d_{CA}=d$)?

Substitute $d$ into our total sum inequality:

$$d+d+d\le 2⟹3d\le 2⟹d\le \frac{2}{3}.$$

The maximum classical disagreement probability that three variables can simultaneously maintain is exactly $2/3$!

No physical system described by classical joint probabilities can ever exceed a mutual disagreement rate of $66.7\mathrm{\%}$.

Three rooms, one possible world

Can all three wishes fit together?

Reset

Each room points up half the time. Choose how often each pair should disagree, then see whether any joint distribution can fulfill all three requests.

A and B disagree

85.0%

B and C disagree

85.0%

C and A disagree

85.0%

These local wishes cannot all hold together. A negative weight is required. Total negative amount: 27.5%.

A slice through the possible worlds

The required mixture

Each row groups two opposite sign assignments. The signs are ordered A, B, C.

All equal-27.5%

+++ / ---

A differs42.5%

-++ / +--

B differs42.5%

+-+ / -+-

C differs42.5%

++- / --+

Split each nonnegative row weight equally between its two assignments to build a joint distribution. A negative row identifies an impossible ordinary mixture.

Every pair oppositeEqual compromiseIndependent fair roomsAll together

$$d_{AB}+d_{BC}+d_{CA}=2.550$$

The sum must be at most two. Each individual rate must also be at most the sum of the other two. The four pattern weights check all four conditions.

In the interactive laboratory above, test this geometric boundary. Adjust the three disagreement rates, and watch how the shaded region of feasible classical worlds changes shape. Notice that attempting to push all three rates toward 1 exits the allowed polytope.

The tetrahedron of possible worlds: Decomposing into four patterns

To see why these inequalities are both necessary and sufficient, let us construct the exact joint distribution.

Although there are 8 distinct sign assignments, flipping all three signs simultaneously preserves every pairwise agreement. Therefore, the 8 assignments group naturally into 4 symmetric patterns:

1. Pattern 0 (All Equal): $(+1,+1,+1)$ and $(-1,-1,-1)$.
2. Pattern A (Only A Differs): $(-1,+1,+1)$ and $(+1,-1,-1)$.
3. Pattern B (Only B Differs): $(+1,-1,+1)$ and $(-1,+1,-1)$.
4. Pattern C (Only C Differs): $(+1,+1,-1)$ and $(-1,-1,+1)$.

Let $w_0,w_A,w_B,w_C$ be the probabilities assigned to these four patterns.

We can express the three pairwise disagreement rates in terms of these weights:

• Pair $AB$ disagrees in patterns $A$ and $B$: $d_{AB}=w_A+w_B$.
• Pair $BC$ disagrees in patterns $B$ and $C$: $d_{BC}=w_B+w_C$.
• Pair $CA$ disagrees in patterns $C$ and $A$: $d_{CA}=w_C+w_A$.

Because the probabilities must sum to one ($w_0+w_A+w_B+w_C=1$), we can invert this linear system to solve for the pattern weights:

Look at what these formulas reveal:

• The condition $w_0\ge 0$ is precisely the total sum bound $d_{AB}+d_{BC}+d_{CA}\le 2$.
• The conditions $w_A\ge 0,w_B\ge 0,w_C\ge 0$ are precisely the three triangle inequalities!

In the three-dimensional space of disagreement coordinates $(d_{AB},d_{BC},d_{CA})$, the space of all possible classical worlds forms a regular tetrahedron.

The four vertices of this tetrahedron correspond to the four pure patterns:

• $w_0=1$: $(0,0,0)$ (complete agreement).
• $w_A=1$: $(1,0,1)$ (only $A$ opposes).
• $w_B=1$: $(1,1,0)$ (only $B$ opposes).
• $w_C=1$: $(0,1,1)$ (only $C$ opposes).

Any point inside this tetrahedron represents a genuine, physically realizable classical distribution. Any point outside it forces at least one pattern weight to become strictly negative (), signaling an impossible demand.

Geometric frustration: The antiferromagnet in physics

We do not have to look far to find this exact geometry in the physical world. It lives inside solid-state magnetic materials.

Consider an antiferromagnetic triangular lattice, where magnetic atoms (Ising spins $s_i\in \{-1,+1\}$) occupy the vertices of a triangle.

In an antiferromagnet, neighboring spins repel each other’s magnetic alignment. The Hamiltonian (energy function) is:

where $J$ is the positive exchange coupling constant.

Each pair of spins wants to point in opposite directions to minimize its energy contribution to $-J$.

If all three bonds could be simultaneously satisfied, the ground state energy would be:

$$E_{\mathrm{ideal}}=(-J)+(-J)+(-J)=-3J.$$

However, as we proved algebraically, no spin configuration can satisfy all three bonds.

For any of the six configurations where two spins point one way and the third points the opposite way, two bonds are satisfied (contributing $-2J$) and one bond is frustrated (contributing $+J$):

$$E_{\mathrm{actual}}=(-J)+(-J)+(+J)=-J.$$

The true ground state energy is $-J$, which is $2J$ higher than the unconstrained ideal!

Physicists call this phenomenon geometric frustration.

Because all six partially satisfied configurations have the exact same ground state energy $-J$, this triangle retains sixfold degeneracy even at absolute zero ($T=0\text{ K}$). Equal weighting gives a residual entropy of $k_B\ln 6$, where $k_B$ is Boltzmann’s constant. What happens when we tile many such triangles? Some frustrated lattices retain a number of ground states that grows exponentially with system size. Add quantum dynamics, and suitable models can host an exotic quantum phase called a spin liquid. Our little classical triangle is the doorway into that larger question!

The missing coupling: Fréchet bounds on joint beliefs

Even before three variables form a triangle, a fundamental ambiguity hides inside any pair of probabilities.

Suppose an analyst reports that hypothesis $X$ is true with probability $p_X=0.90$, and hypothesis $Y$ is true with probability $p_Y=0.90$.

How often are $X$ and $Y$ true together?

Most people instinctively multiply them: $q=(0.90)(0.90)=0.81$.

However, that multiplication assumes independence: an enormous and often unjustified assumption!

The true joint probability $q=P(X\cap Y)$ is constrained only by the Fréchet inequalities:

$$max(0,p_X+p_Y-1)\le q\le min(p_X,p_Y).$$

For our example:

$$max(0,0.90+0.90-1)\le q\le min(0.90,0.90)⟹0.80\le q\le 0.90.$$

The joint probability can be anything from $80\mathrm{\%}$ (maximal negative correlation) to $90\mathrm{\%}$ (maximal positive correlation).

A probabilistic system that records only marginal probabilities for individual propositions is structurally incomplete. To understand how beliefs interact, a notebook must preserve the lineage of shared trials: which observations occurred together on the same experimental run.

Contextuality and sheaves: When local sections refuse to glue

In modern mathematics, the tool designed to handle local compatibility and global obstruction is sheaf theory.

In a sheaf-theoretic description of an experiment:

• A context is a set of compatible measurements that can be performed simultaneously (such as measuring the pair $A,B$).
• A local section is an assignment of outcomes to that context.
• Restriction maps project local sections onto sub-contexts (for example, looking only at $A$ from the pair $A,B$).

A global section is a single, unified assignment across all possible measurements.

When every local observation can be explained as a marginal of a single global probability distribution, the system is non-contextual.

When valid local distributions cannot be glued into a global probability measure, the system is contextual.

In quantum foundations, Abramsky and Brandenburger’s framework proved that quantum non-locality and contextuality are precisely such topological obstructions. Entangled quantum particles violate classical bounds because physical reality does not possess predetermined values independent of measurement context.

Carry that viewpoint into a scientific research network. One laboratory studies a material under extreme pressure, another measures its conductivity at high temperatures, and a third maps its magnetic response. The pairwise overlaps between experiments form the restriction maps of a sheaf. When the overlapping data refuse to glue into a single classical phase diagram, the disagreement points directly toward the missing joint condition that needs to be measured. Related mathematical work develops a general theory of local agreement and global disagreement across diverse informational domains.

Gauge connections on networks: Curvature and loop residuals

How can an interconnected network detect these inconsistencies dynamically?

Let us introduce a gauge theory on an evidence graph.

Suppose each agent $i$ measures a continuous property (such as phase, coordinate frame, or price baseline) relative to a local reference frame ${\varphi }_i$.

When agent $i$ communicates with neighbor $j$, they do not transmit their raw coordinates; they measure the relative translation across their shared edge:

$$g_{ij}={\varphi }_j-{\varphi }_i.$$

Now calculate the total directed sum of translations around our closed triangle $A\to B\to C\to A$:

$$h=g_{AB}+g_{BC}+g_{CA}=({\varphi }_B-{\varphi }_A)+({\varphi }_C-{\varphi }_B)+({\varphi }_A-{\varphi }_C).$$

Notice that the intermediate references telescope and cancel:

$$h=0.$$

The quantity $h$ is the loop holonomy, or curvature residual. When local translations derive from genuine node potentials, the loop residual vanishes identically.

Now suppose that due to noisy measurements, local biases, or contextual discrepancies, the measured edge translations around the loop sum to a non-zero value:

$$h=g_{AB}+g_{BC}+g_{CA}\ne 0.$$

A non-zero loop residual $h\ne 0$ indicates that the network contains topological curvature!

Notice the fundamental property of gauge invariance: if any individual node shifts its local coordinate system by an arbitrary offset (${\varphi }_i\to {\varphi }_i+{\delta }_i$), the edge connections transform as:

$$g_{ij}^{\prime }=g_{ij}+({\delta }_j-{\delta }_i).$$

Yet the loop residual remains completely unchanged:

$$h^{\prime }=g_{AB}^{\prime }+g_{BC}^{\prime }+g_{CA}^{\prime }=h+({\delta }_B-{\delta }_A)+({\delta }_C-{\delta }_B)+({\delta }_A-{\delta }_C)=h.$$

The loop residual $h$ is an invariant physical observable: it separates arbitrary local conventions from objective global strain.

Let these translations evolve as agents learn. Penalize loop residuals $h^2$, transport trial histories through the network, and allow new observations to reshape the connections. An evidence geometry can now have something like physical strain, circulation, and an energy landscape. Its dynamics describe how a distributed world negotiates its own consistency.

A gauge change you can put your finger on

Move the local zeros. Keep the loop.

Reset the network

Repair doorway CA so the round trip returns to zero, then walk all three edges to test it. Room-zero handles relabel the map. Can they repair the mismatch?

Round-trip mismatch +1.0. Adjust the highlighted repair doorway.

Next repairFree doorway editing

One network, two coordinate choices

Each edge label reads “original → shifted.” The ring marks the room reached by your walk.

Compare both coordinate walks

Original zerosShifted zeros

Walk one edge →Start the walk again

Try moving the local zeros

Move a handle, or focus it and use arrow keys. A local zero is just a choice of coordinate origin.

Restore the original zeros

Change a measured doorway

These buttons change the measured relation itself. Predict which changes can alter h.

AB−1

+2.0

+1

BC−1

+1.0

+1

CA−1

-2.0

+1

Show a repair hint

Change the repair edge by -1.0. Local zero shifts cannot change the loop sum.

Why the loop sum survives coordinate changes

Original loop: +1.0. Shifted loop: +1.0. The round trip does not close. Relabeling local zeros preserves the mismatch; changing a measured doorway can remove it.

$$g_{ij}^{\prime }=g_{ij}+{\delta }_j-{\delta }_i,h^{\prime }=g_{AB}^{\prime }+g_{BC}^{\prime }+g_{CA}^{\prime }=h$$

δ is a local zero shift; g is a directed translation. This is an additive network model. A nonzero residual diagnoses a loop inconsistency here; its physical cause needs its own model.

Quasiprobabilities: Negative weights as topological shadows

What if we insist on writing a global joint table for the fully frustrated triangle where $d_{AB}=d_{BC}=d_{CA}=1$?

If we evaluate our pattern weight formulas for $d=1$:

$$w_0=1-\frac{1+1+1}{2}=1-\frac{3}{2}=-\frac{1}{2}.$$ $$w_A=w_B=w_C=\frac{1+1-1}{2}=\frac{1}{2}.$$

The weight on the all-equal pattern is negative: $w_0=-1/2$!

We can write down an explicit signed probability distribution (quasiprobability) over all 8 states $(a,b,c)\in \{-1,+1{\}}^3$:

$$q(a,b,c)=\frac{1}{8}(1-ab-bc-ca).$$

Let us check the properties of this signed distribution:

1. Sum of weights: Summing over all 8 configurations yields $\sum _{a,b,c}q(a,b,c)=1$.
2. Marginal recovery: If you sum away switch $c$, the marginal distribution over $a$ and $b$ is: $$P(a,b)=q(a,b,+1)+q(a,b,-1)=\frac{1}{4}(1-ab).$$ This yields $P(+1,-1)=P(-1,+1)=1/2$, and $P(+1,+1)=P(-1,-1)=0$. Every pairwise marginal is positive and perfectly reproduces the requested 100% disagreement!
3. Negative states: For the two all-equal states $(+1,+1,+1)$ and $(-1,-1,-1)$, the weight is: $$q(+1,+1,+1)=\frac{1}{8}(1-1-1-1)=-\frac{2}{8}=-\frac{1}{4}.$$

In quantum mechanics, this is precisely how the Wigner quasiprobability distribution operates: it uses negative probabilities to preserve marginal consistency across non-commuting observables!

Negative probabilities are the mathematical shadow cast by geometric frustration when forced into a global description.

Three reports, eight possible worlds

Add the hidden worlds. Recover the pair.

Restore perfect disagreement

Every room is up half the time. The three pair tables below each look ordinary. Can one joint table explain all of them? Change a disagreement rate, then tap a pair-table cell to see the two hidden worlds that add up to it.

Every pair opposite The 2/3 compromise Independent fair rooms All together

Rooms A and B

− 100.0%disagreement +

Row A, column B

	+	−
+	0.0%	50.0%
−	50.0%	0.0%

Rooms B and C

− 100.0%disagreement +

Row B, column C

	+	−
+	0.0%	50.0%
−	50.0%	0.0%

Rooms C and A

− 100.0%disagreement +

Row C, column A

	+	−
+	0.0%	50.0%
−	50.0%	0.0%

No ordinary joint distribution fits these three reports. The displayed reconstruction needs 50.0% of negative weight.

Eight worlds behind the three reports

Signs are ordered A, B, C.

Room C is up (+)

A ↓ / B →	+	−
+	+++-25.0%	+−+25.0%
−	−++25.0%	−−+25.0%

Room C is down (−)

A ↓ / B →	+	−
+	++−25.0%	+−−25.0%
−	−+−25.0%	−−−-25.0%

Blue outlines trace the selected pair outcome through both values of the unobserved room.

World +++-25.0%

+

World ++−25.0%

=

Pair AB: ++0.0%

Each pair table sums to one and keeps both rooms fair. The joint reconstruction splits each pattern equally between opposite sign assignments. Negative paired-pattern weights make every ordinary joint reconstruction impossible; a positive reconstruction may have other forms.

$$q(a,b,c)=\frac{1+(1-2d_{AB})ab+(1-2d_{BC})bc+(1-2d_{CA})ca}{8}$$

Here a, b, c are signs, and d is a pair's disagreement rate. Signed weights can cancel when a hidden room is summed away.

The architecture of collective understanding

We began with three switches in three rooms, and discovered that:

• Local pairwise agreements cannot always be satisfied globally.
• Classical joint distributions are constrained by the tetrahedron bound $d_{AB}+d_{BC}+d_{CA}\le 2$, limiting mutual disagreement to at most $2/3$.
• Physical spin lattices embody this constraint through geometric frustration and residual zero-point entropy.
• Sheaf theory and quantum contextuality formalize why local observational sections refuse to glue together.
• Gauge theory on graphs separates arbitrary local conventions from invariant topological curvature $h$.
• Quasiprobabilities with negative weights emerge as the exact mathematical representations of frustrated systems.

That is an exciting invitation for Epistemic Physics: give each local observer a valid evidence world, then let the links between worlds carry what a single global probability table cannot accommodate. Signed structure, contextual variables, and evolving gauge translations become different materials for building with the tension.

Perhaps a world is made from countless little agreements whose relationships keep negotiating a larger whole. Our three switches show the first challenge clearly: to understand what can exist together, we must keep more than what each part believes alone.

In our next excursion, we will investigate how individual trials ripple across a network in One trial, many echoes, uncovering how redundant signals reshape the geometry of shared evidence.