↩ SocioRhetoric · A New Catalogue of Speech Acts
sociorhetoric · a speech act
“Thoughts and prayers” once meant something; now it is a reflex that can signal not caring at all. How does a phrase that genuinely indexed a substance drain into an empty token — and why is the draining almost inevitable?
Say the words to a grieving family in 1970 and you performed a small, real thing: you were, in fact, thinking of them; you may, in fact, have prayed. The phrase was a token — a marker — and behind it stood a substance, the felt concern it pointed to. For a while the marker was a reliable index: hearing it, you could bet that the substance was there. That reliability is exactly what made the phrase worth saying, and worth rewarding with warmth, trust, belonging.
But a marker that earns reward is a marker worth counterfeiting. The politician who feels nothing still gains by saying it; the account with no prayer in it still collects the likes. Each such utterance is surrogation — letting the token stand in for the substance it once indexed — and as the counterfeits accumulate the correlation between saying the words and meaning them falls. The phrase does not get louder or softer. It stays exactly the same string of syllables while the thing behind it evaporates, until “thoughts and prayers” becomes a byword for not caring — the very opposite of what it once certified. This is not a moral failing of any one speaker. It is a drift with an engine, and the engine is reward. The instrument below lets you run that engine and watch a live token hollow out.
the honest instrument
A population of speakers, each with a hidden substance level (genuine grief, skill, belief) and a live decision to emit the token (the phrase). Press play. Because emitting earns reward, low-substance speakers adopt the phrase too; the correlation between token and substance and its information value both drain toward zero. The displayed values are computed from the model. Drag a lever, re-anchor, reseed, and watch what holds and what hollows.
Surrogation drift simulator
400 speakers · substance ~ Uniform[0,1] · token = emit the phrase
← rounds →
low substance ← speakers → high substance
Recognition / belonging earned by saying the phrase. Higher reward recruits mimics faster — the drift accelerates.
How hard the token is to emit without the substance. A costly, hard-to-fake signal resists surrogation — raise it and the phrase holds its meaning.
How quickly low-substance speakers who decide to adopt the phrase actually take it up each round.
Number of speakers. Larger populations give smoother, steadier curves (reseeds the population).
the marker outlives the substance
Every reliable marker starts life keyed to reality: the phrase tracks the grief, the metric tracks the performance, the ritual tracks the devotion. That is what an index is — a sign bound to its object by a real dependence, the way smoke is bound to fire. But a marker is also visible, and what is visible can be rewarded, and what is rewarded gets produced for its own sake. Over time the utterance stops answering to the object and starts answering to the audience. It slides, in the catalogue's terms, from being keyed to reality toward being keyed to belonging — from a report about the world to a password for a room. That slide is surrogation, and it is the general engine behind a hundred specific hollowings: the applause line, the compliance checkbox, the credential, the “thoughts and prayers.”
The drift is directional and hard to reverse because the two things the marker could serve — tracking the substance and earning the reward — come apart the moment faking is cheaper than having. Once that gap opens, the population of speakers is re-sorted. The genuine were always going to emit; now the reward pulls in everyone else, and since there are always more people who want the reward than who have the substance, the emitters come to look like the whole population rather than a select, informative slice of it. The marker outlives the substance it once pointed to — and, worse, keeps drawing the reward that was only ever meant for the substance.
what to try
Load the drift and press play. Now drag reward on the token up toward 1 and reset. The correlation falls off a cliff — the phrase hollows out in a handful of rounds. The more the marker is worth, the faster it is counterfeited. Success is what recruits the mimics.
Keep the reward high, then raise cost to fake. Watch the correlation refuse to collapse — it settles high and stays there. A token that is expensive to emit without the substance behind it holds its meaning. This is costly signalling: the cost is the anti-surrogation.
Let a token fully drain to pure ceremony, then hit ⚓ re-anchor — an authenticity check that re-links token to substance. The correlation leaps back up. Then keep playing: with the reward still on, it decays all over again. Re-anchoring buys time, not a cure.
success recruits its own mimics
The very success of a signal is what hollows it out. A marker earns reward precisely because, for a while, it reliably indicated the substance — and that reward is a standing invitation to everyone who wants the indication without the substance. High reward is not a sign of health; past a point it is a countdown. The stronger the incentive to be seen thinking and praying, the larger the crowd that will say the words with nothing behind them, and the faster the words stop meaning anything. A signal carries, in its own payoff, the seed of its own emptying.
This is Goodhart's law read from the speech-act side. Goodhart's version — when a measure becomes a target, it ceases to be a good measure — is the same drift watched from the measurement booth: optimise the proxy and it decouples from the thing it proxied. Surrogation is what that decoupling feels like from inside language, where the “measure” is a phrase and the “target” is belonging. And the model names the one force that resists it: cost to fake. A signal that is hard to counterfeit — expensive, dangerous, or possible only for those who actually have the substance — cannot be flooded by mimics, because the mimics cannot afford it. Costly signalling depends on the price of the token. Lower the price and you invite the drift; raise it and the correlation holds. The instrument's cost-to-fake lever is that price, made draggable.
model
Each speaker i is given a hidden substance sᵢ drawn from a uniform distribution on [0,1]. Their initial tendency to emit the token is xᵢ = σ(6·(sᵢ − 0.5)) — a soft threshold, so at the start the phrase is emitted mostly by high-substance speakers and the token is an honest index. Each round, every speaker weighs an incentive to adopt the token, reward − cost·(1 − sᵢ): the reward pulls everyone in, but the cost-to-fake term bites hardest on the low-substance speakers, who have the most to fake. That incentive passes through a logistic to a drive, and emission climbs xᵢ ← xᵢ + rate·drive·(1 − xᵢ). Low-substance speakers flood in exactly when reward beats cost.
The two headline numbers are then read straight off the population. Correlation is the honest point-biserial correlation between emitting the token (a Bernoulli event with probability xᵢ) and substance sᵢ — it includes the emission noise, so when everyone emits regardless of substance the correlation is exactly zero. Information value is the mutual information (in bits) between emitting and being high- versus low-substance: how much learning that someone said the phrase tells you about whether they meant it. Both start positive and drain to zero as the mimics arrive. Re-anchor resets every xᵢ to its substance-keyed starting value — an authenticity check that re-links marker to substance — after which, with the reward still on, the decay simply restarts. Press play and the arithmetic runs in your browser, one round at a time.
the act ↔ the model
| The speech act | The model |
|---|---|
| the substance | the real thing the marker once indexed — genuine grief, skill, or belief — the hidden level sᵢ each speaker carries. |
| the token | the phrase or marker that indexed it: the decision to emit, whose probability is xᵢ. |
| reward on the token | the recognition or belonging earned by saying it — why mimics find it worth adopting at all. |
| correlation(token, substance) | how much the marker still means: how reliably an emission still indicates the substance behind it. |
| cost to fake | what slows the drift — the price of emitting without the substance; costly signalling's anti-counterfeit. |
| the drained equilibrium | pure ceremony: everyone emits regardless of substance, correlation at zero — surrogation failure. |
how this act misfires
misfire · surrogation failure
The catalogue's named misfire for this act, and the equilibrium the instrument drives toward. Once the correlation has drained, the phrase is emitted by everyone regardless of what they feel, so rewarding it now rewards nothing — or worse, rewards the mimicry directly, since the counterfeit is indistinguishable from the genuine and cheaper to produce. The listener who still hears “thoughts and prayers” as an index of care is reading a marker that has already outlived its object. Surrogation failure is not that the words are false; it is that they have stopped being about anything but the saying of them.
caveat · substance is not a scalar
The model treats “substance” as a measurable number on [0,1], which real meaning stubbornly resists — grief and belief are not dials, and no authenticity check reads them cleanly. The decaying correlation is therefore a picture of a mechanism — how reward on a visible marker re-sorts who emits it — not a literal sincerity meter for any actual phrase. Trust the direction and the levers (reward speeds the drift, cost-to-fake resists it, re-anchoring only delays it); do not trust the third decimal place as a verdict on anyone's heart.
Reward the marker and the mimics come; the phrase that meant everything ends up meaning only that you said it.
can you use it?
RECOGNITION — Which is surrogation? A: a friend texts “thinking of you,” then drives over with dinner. B: a brand posts “thinking of you all” after a disaster because the metrics reward it. C: a politician claims a relief donation he never made.
B. The token is emitted for the reward, with no substance behind it — one more counterfeit draining the correlation. A is a healthy index; C is false about the world, a lie.
THE NEAREST NEIGHBOR — What single criterion separates a drained token from a lie?
What fails. A lie is false; a drained token has stopped being about anything but the saying of it — emitted by everyone regardless of substance, its correlation with meaning it sits at zero.
PRODUCTION — “Great work!” in your team chat has drained to ceremony. Using the model's levers, design a repair, then check it.
Raise the cost to fake: praise must name the specific thing done and why it was hard — producible only by someone who engaged. Yours works if it prices out mimics; a one-time re-anchor decays again while the reward stays on.