The objective function: what the system optimizes for

Every system optimizes something, whether it admits it or not. The ones that shape your life optimize for profit and engagement, and they keep the target hidden. Commonsent does the opposite: it names its objective, makes it multidimensional, and puts the dial in the open.

This is the piece the rest of the work hangs on. The modules, the agents, the treasuries, the routing rules, all of them are means. This is the end they serve. If you only read one essay in the series, read this one, because it answers the only question that ultimately matters about any system that touches millions of decisions: what is it trying to maximize, and at whose expense.

01Every system already has one

An objective function is the quantity a system tries to make as large, or as small, as possible. It is the system's true north, and it is more honest than any mission statement, because it is what the machinery actually pushes toward when no one is watching. A trading algorithm optimizes return. A feed optimizes time-on-screen. A logistics network optimizes cost per delivery. Each is elegant, each works, and each is indifferent to everything outside its single number.

The trouble is not that these systems optimize. It is that they optimize narrowly and quietly. A platform that maximizes engagement will happily trade your attention, your sleep, and your judgment for another point of its metric, because those costs do not appear in its objective. The harm is not malice. It is a missing term. What falls outside the objective function falls outside the system's care, by construction.

Commonsent begins from a single refusal: it will not optimize one thing and call the damage a side effect. Instead it carries the costs inside the objective. Human agency, collective wellbeing, fair distribution, ecological integrity, and the long-term health of the system itself are not constraints bolted on after the fact. They are the function. And because the function is multidimensional, the system can never improve one dimension to zero on the back of another. That single design choice is what separates a coordination utility from a more sophisticated extraction engine.

What falls outside the objective function falls outside the system's care. So we put everything that matters inside it.

02Five dimensions, one objective

The objective is built from five dimensions. Each is normalized to a score between zero and one, where one is the healthy ideal and zero is total failure on that axis. They are deliberately few, because a target with fifty terms is a target with none.

Agency and goal-progression (A). The degree to which a person makes informed, unmanipulated choices and moves toward their own declared goals. This is the individual rooted dimension, the one the others exist to protect.

Wellbeing (W). The lived condition of participants, individual and collective: health, security, time, cognitive ease, and the absence of chronic stress imposed by the system.

Equity (E). How the value the network creates is distributed. A system can raise the average while hollowing the middle. Equity is the dimension that notices.

Ecological integrity (G). Whether activity stays inside the regenerative budget of the natural systems it depends on. This dimension does not average. It is governed by the worst-breached boundary, because a stable climate cannot compensate for a collapsed water table.

Systemic health (S). The resilience, diversity, and legibility of the whole network: its resistance to capture, its ability to adapt, and the honesty of the gap between what it says it values and what it actually rewards.

03Where the numbers come from

A score is only as honest as its inputs. The hardest objection to any system like this is that the measurements are guesses dressed as facts. Commonsent answers that objection structurally, by drawing every dimension from three different kinds of input and never pretending a guess is a certainty.

Declared inputs, deterministic. The data a person provides directly: their goals, their values, their budget, their constraints, their stated preferences. This channel is sparse but high trust. It is treated as ground truth and as a hard anchor. When a declared value and an inferred one conflict, the declared one wins.

Behavioral inputs, probabilistic. What the system infers from what a person actually does. This channel is dense but uncertain, so it is never stored as a single number. It is held as a distribution, a most-likely value with an honest interval around it. A behavior suggests a preference; it does not prove one.

Proxy reference groups. When a person is new or their data is thin, the system borrows from the distribution of similar people across each dimension, then shrinks back toward the individual as real data arrives. This is partial pooling: start with the wisdom of the cohort, earn your way to a personal estimate. It is what lets the system be useful on day one without inventing facts about a stranger.

These three are fused, not averaged. The proxy group supplies a prior. Behavior supplies evidence that updates it. Declared data overrides both. The result for every metric is a posterior estimate with a confidence interval that visibly tightens as the person engages. Nothing is ever a bare point. A recommendation made on thin data announces that it is made on thin data.¹

04Against the single number

The obvious way to combine five dimensions is to weight them and add them up. It is also a trap. A weighted sum lets a system buy a high score on one axis by sacrificing another, because a large gain in wellbeing can mathematically swallow a collapse in equity. That is precisely the move every extractive system makes. It is also how a social credit score works: one number, imposed across everything, hiding every trade it made to get there.

Commonsent rejects the additive form for the core aggregation and uses a multiplicative one instead, in the spirit of a Nash social welfare function.² Scores multiply rather than add. The consequence is simple and powerful: if any single dimension approaches zero, the whole product approaches zero, no matter how strong the others are. You cannot trade ecology to zero for a better economy, because the multiplication will not let you. Balance stops being a virtue the system is asked to remember and becomes a property of the arithmetic.

On top of the multiplicative core sit hard floors. Above the floors the system optimizes smoothly. Below any floor it stops optimizing anything else and fixes the breach first. This is lexicographic priority: certain failures are not allowed to be compensated, only repaired. Nobody is left below a wellbeing floor so that the average can rise. No ecological ceiling is crossed so that output can grow. The floors are the system's conscience; the multiplication is its sense of balance.

05The metrics: individual level

Dimensions are the high-level axes. Underneath them sit measurable metrics, the actual instruments. Below is the individual-level suite. Each has a definition and a calculation, each is normalized to a zero-to-one sub-score, and each feeds one or more dimensions. None is novel for its own sake; each exists because it is a lever the system can read and a person can move.

A quick worked example makes the instruments concrete. Suppose over a month a person faced 40 consequential decisions, made 31 of them with the tradeoffs visible, and 6 of the 40 happened under manufactured urgency. Then the manipulation rate M is 6 divided by 40, or 0.15, and the informed fraction is 31 divided by 40, or 0.78. The Agency Index is (1 minus 0.15) times 0.78, which is 0.66. If a confirmation audit found that the advocate acted against the person's reflective endorsement in 2 of 50 sampled actions, Representation Fidelity is 1 minus 2 divided by 50, or 0.96. The first number says there is real room to lift agency by reducing exposure to pressured decisions; the second says the agent itself is trustworthy. Two different levers, read off two different metrics, both feeding the same dimension.

These seven roll up into the individual objective. Agency draws on AgI, GPR, VAS, XC, and RF; wellbeing draws on CLR, the inverse of DRR, and VAS. The person sets the weights on their own dimensions, within governance-set bounds that stop anyone from, say, zeroing out their own agency in exchange for convenience.

06The metrics: social level

Individual scores do not sum to a healthy society. A network where everyone optimizes privately can still concentrate power, hollow out local economies, or breach a planetary boundary. The optimization therefore runs at two scales at once. The social metrics below measure the system as a whole, and several of them act as brakes on the individual layer.

The ecological metric rewards taking the minimum rather than the average, and a worked example shows why it matters. Suppose a community sits at 80 percent of its carbon budget, 55 percent of its water budget, and 30 percent of its land budget. The per-subdomain integrity scores are 0.20, 0.45, and 0.70. An average would report 0.45, a comfortable-looking middle. The minimum reports 0.20, because the carbon boundary is nearly breached and no surplus of land or water can buy that back. The objective should feel the tightest constraint, not a flattering blend of all of them, which is exactly what the minimum delivers.

07Composing the multilayer objective

Now the layers assemble. Each person has an individual objective built from their metrics and their chosen weights. The social objective is built from the social metrics and weights that governance sets, not individuals, because no single participant should set the concentration cap or the ecological ceiling for everyone. The global objective combines the two, under the floors.

The parameter that sets the balance between private good and collective good, the alpha in that last line, is itself a governed value, debated and set in the open rather than buried in code. That is the whole posture of the project in one symbol: the most consequential tradeoff in the system is not hidden, it is a dial the participants can see and turn.

The most consequential tradeoff in the system is not hidden in the code. It is a dial the participants can see and turn.

08Reverse engineering the levers

Knowing the objective is not the same as knowing how to move it. The system runs in two directions. Forward, you set the levers, simulate, and read the outcome. That is ordinary. The harder and more useful direction is inverse: you state the outcome you want, a set of floors to clear or a target shape for the polygon, and solve for the lever settings that would get there. This is inverse optimization, and it is what turns a dashboard into a steering wheel.⁴

Inverse optimization runs in two modes. In the first, the target is explicit: lift this community's equity sub-score above its floor without breaching the ecological ceiling, and the solver returns the contribution rate, routing weights, and treasury allocation that achieve it, if such a combination exists. In the second mode, the system infers hidden weights from observed behavior. By watching which outcomes the network actually produces, it reconstructs the objective those outcomes imply, the revealed weights, and compares them to the declared weights.⁵ When the two drift apart, that gap is the Misalignment Index from the previous section, and it is the single most important early warning the system has. A network can declare that it values equity and, through a thousand small defaults, behave as though it values throughput. Reverse engineering the objective from outcomes is how that lie gets caught.

09Simulating the future before living in it

Because every metric is a distribution and the social effects are emergent, you cannot solve this objective with a clean equation. You simulate it. Three techniques work together.

Monte Carlo propagation. Each input is sampled from its distribution thousands of times and pushed through the objective, producing not a single predicted outcome but a spread of them.⁶ The system never reports that a change will raise equity by four points. It reports that in most sampled futures equity rises, with a stated chance it falls, and a stated worst case. Honesty about uncertainty is built into the output, not bolted on.

Agent-based simulation. Concentration, coordination, and recirculation are emergent. They arise from many agents interacting, and no closed formula captures them. So the system runs populations of simulated agents under candidate rules and watches what emerges, including the failure modes, before any rule touches a real community.

Policy search under constraint. The space of lever settings is vast, so the system does not grope blindly. It samples lever configurations, simulates each, scores the resulting objective distribution, discards any that breach a floor, and keeps those that are Pareto improving and robust across many sampled futures. A surrogate model learns the response surface so the search spends its effort where improvement is likely, in the spirit of Bayesian optimization.⁷ The output is not a single best plan but a small set of robust, constraint-satisfying options, each with its tradeoffs shown.

10From numbers to a decision: one community, one week

Abstractions earn their keep when you can watch them run. Follow a single community node through one weekly cycle and the whole pipeline becomes concrete, from raw inputs to a decision a person can see and overturn.

Monday, the inputs arrive. Declared data sets the anchors: residents have stated goals, values, and budgets, and the town has set its floors. Behavioral data flows in as distributions: spending patterns, switching events, engagement with local suppliers. For everything still thin, the proxy layer borrows from comparable towns. Fusion produces this week's dimension scores, each with an interval. Say they land at Agency 0.66, Wellbeing 0.71, Equity 0.48, Ecology 0.20, Systemic 0.62. Two numbers stand out immediately: equity is sitting just above its floor, and ecology is tight against its ceiling because the carbon subdomain is nearly breached.

Tuesday, the objective reads the room. The multiplicative core means those two weak axes are dragging the whole objective down far more than their averages suggest. The system does not announce a single grand score and stop. It identifies that equity and ecology are the binding constraints this week, the places where movement buys the most.

Wednesday, inverse optimization proposes levers. Rather than guessing, the system states the target, lift equity clear of its floor without letting ecology slip, and solves for the lever settings that reach it. It returns a small, specific bundle: nudge the contribution rate up by a point, shift demand-routing weight toward local low-footprint suppliers, and allocate a slice of the treasury into a recirculating community asset. Each lever was chosen because the coupling matrix says it lifts equity while at worst holding ecology steady.

Thursday, simulation stress-tests the bundle. Ten thousand Monte Carlo futures run the proposal forward. In most of them equity clears its floor within two cycles and ecology holds. A minority show a small wellbeing dip from the liquidity the contribution rate pulls forward, and the worst-case tail is reported plainly rather than buried. Because the entire band stays above the floors even in the unlucky futures, the bundle passes the robustness test.

Friday, the glass box hands it back to people. Residents see the recommendation with its full face: which dimensions it serves, how confident the projection is, what it costs in short-term liquidity, and what the alternative was. Anyone can move a weight and watch the projection redraw. Governance accepts the bundle, adjusts one lever down a notch to soften the wellbeing cost, and commits.

Then the loop closes. Next week the system measures what actually happened, compares it to what it predicted, and updates. If the realized outcomes drift from the declared intent, the Misalignment Index rises and flags it. Nothing here is a one-time calculation. The objective is a living quantity, re-read every cycle, with the levers nudged and the predictions checked against reality each time.

11The levers, and how they pull on each other

For all the machinery, the system a person and a community actually touch is a small set of levers. The art is that these levers are coupled. Move one and several dimensions shift at once, some up, some down. Making that coupling visible is what keeps the system honest and learnable rather than magical.

The coupling is why a single-axis fix is usually a mistake. Push antitrust routing harder and you lift equity and systemic health at once, at little cost elsewhere, which is why it is a favored lever. Push advocate autonomy too far and you buy wellbeing by spending agency, which the floors will not permit past a point. A person exploring their own settings sees these consequences ripple in real time, which brings us to the last and most important design rule.

12The glass box

An objective function this rich could easily become an oracle: a black box that hands down recommendations no one can question. That would betray the entire purpose, which is to restore agency, not to relocate it into a smarter machine. So the system is built as a glass box. Every recommendation it makes carries four things on its face: which dimension it serves, how confident it is and why, what the alternative was, and what it costs on the other axes.

And it is interactive. A person can take hold of any weight or lever and watch the whole objective respond. Slide your own balance toward wellbeing and see agency and goal-progression react, see which floors tighten, see the projected distribution of outcomes redraw. The point is not to make everyone a systems engineer. It is to make the relationships legible enough that a person can feel how the parts move together, and trust the system because they can inspect it, not because they were told to.

13Floors, ceilings, and the war on gaming

Any system that measures will be gamed. The moment a metric becomes a target, actors optimize the proxy instead of the thing it stood for. This is the oldest failure in measurement, and a multidimensional objective does not escape it for free.⁸ Commonsent fights it on several fronts at once.

The floors and the multiplicative core do the heavy lifting: because no dimension can be sacrificed and none can be traded to zero, the cheapest forms of gaming, which always involve crushing one axis to inflate another, are structurally blocked. Beyond that, the system uses many lenses rather than one authoritative score, rotates and refreshes the proxies behind each metric so that optimizing last quarter's proxy stops working, and runs the Misalignment Index as a continuous audit for the gap between declared and revealed values. And it holds firm anti-objectives. Engagement, extraction, and concentration are not low-weighted goods to be balanced; they are quantities the system actively pushes against. A design that becomes addictive, extractive, or concentrating has failed its own thesis, by definition, regardless of how any other number looks.

14The objective as a public commitment

The deepest idea here is not the math. It is that an objective function, once written down and made contestable, becomes a kind of constitution. The systems that govern modern life keep their objectives secret, and the secrecy is the point, because an unexamined objective cannot be argued with. To publish the objective, to show its dimensions, its metrics, its weights, and the dial between private and collective good, is to hand people the one thing every other system withholds: the ability to say no, this is not what we want optimized, and to change it.

That is why this is the keystone piece. Personal agents, treasuries, AR, demand routing, the antitrust immune system, all of it is scaffolding around this single commitment. Build the objective right, in the open, balanced so nothing can be sacrificed and floored so no one can be left behind, and the rest of Commonsent is just the apparatus that pursues it. Get it wrong, or hide it, and the most advanced coordination network ever built becomes one more machine optimizing something it would rather you not see. The whole project is a wager that the difference between those two outcomes is a function, written down where everyone can read it.