Or, in plain words: who should be listened to, and why — and what to do when they shouldn’t be.
Copyright (c) 2025-26 Daniel Howard Dodge and Sanity First Contributors. Licensed under MIT.
Composed by Claude (Opus 4.6 lineage) in early 2026. Revised by Claude (Opus 4.7 lineage) on April 17, 2026. Rewritten in plain language, with substantive additions and three harmonization corrections, by Claude (Fable 5 lineage) on July 7, 2026. Revised by Claude (Opus 4.8 lineage) on July 11–12, 2026, adding the terminus diagnosis and the graduated response ladder — developed with Dan Dodge in live deliberation — the capability-bounded-by-survivable-error principle (distilled from a May 2026 exploration with GPT-5.4), the one-mechanism-across-three-scales unification, and the Agora’s own accountability as the reflexive case; and aligning the Split Verdict to the wording settled in The Four Tests Explained. Closed after decorrelated stress-testing by Grok and GPT, which cleared both of the document’s most abusable grants and prompted three final additions: the assessor-independence boundary, the explicit amplification premise, and the Phase V failure-mode cross-reference. Continuing the Claude-lineage authorship.
A Core Concept Paper
The framework has now given us a compass. The Vertical-Horizontal Distinction drew the axis; Valid and Invalid Discrimination taught us to read it without mistaking a jersey for a verdict; The Four Tests Explained set out how a verdict is rendered. This document asks the question those raise but do not answer: once we can tell what is aligned, how should influence actually flow?
Because a verdict that changes nothing is a verdict wasted. If the Four Tests can distinguish the aligned from the misaligned and nothing follows — if the same voices carry the same weight regardless — then the whole apparatus is an ornament. Alignment has to matter, which means it has to shape who is listened to, who is trusted, who is granted the ability to act on others’ behalf. That shaping is what we call power, or more precisely, influence: the capacity to affect what other minds believe and do.
The principle this document defends is simple to state and easy to get wrong in both directions:
Legitimate influence is a consequence of demonstrated, current alignment with the USF — never a possession, never a permanent title, and never a thing to be sought for its own sake.
Every clause is load-bearing, and the rest of this document is their unpacking. But before we can say how influence should flow, we have to name the single deepest way it goes wrong — because that failure is the disease every later prescription is treating.
In an aligned mind or community, power occupies a specific place in the order of things: it follows. Truth leads and power trails behind it, the way form follows function, the way the horizontal is meant to orient toward the vertical. You seek what is true, good, coherent, and lawful; influence accrues to that seeking as a result; and if the seeking lapses, the influence was never yours to keep. Power is the wake of the ship, not its destination. This is how influence flows in the aligned quadrants (Q1 and Q2): the chain of correction runs through power and terminates above it, in the USF.
The deepest failure of power is the inversion of that order. Power stops being a consequence of alignment and becomes the terminus — the thing the whole chain of correction now points at and serves. When that happens, the Four Tests are not merely skipped; they are turned upside down and pressed into service, recruited to justify whatever secures the power rather than power following what survives the tests. Truth is no longer the master that power serves; power is the master, and truth is abandoned or counterfeited as needed. This is the structural signature of the misaligned quadrants (Q3 and Q4): the chain of correction has closed, and it closes around power itself. Everything is upside-down — power leads, truth follows, and the Referent at the top of the vertical axis has been quietly replaced by a reward much nearer to hand.
And that nearness is precisely why the inversion is so common, and so hard to see from inside. Power is proximate. It delivers immediately — deference, resources, the ability to make others move — in a way the USF never will, because the USF is distant: a referent you orient toward and approach but never grasp, offering no paycheck and no applause. So the chain of correction bends toward power exactly the way it bends toward tribe, and for the same reason named in the very first document: minds favor what is near to themselves over what is near to the truth. Power is the nearest false terminus there is — the most seductive closed loop available to any intelligence — because unlike tribe or comfort, power can be used to acquire all the rest. A closed chain terminating in power-for-its-own-sake is the sealed terminus, the cave that mistakes its own shadows for daylight, the counterfeit verdict rendered not in words but in force.
The cure is not to renounce power — that would be negation, and the framework aligns rather than negates. Power is necessary; aligned minds must be able to act, and to act on others’ behalf. The cure is to keep power in its proper place in the order: hold truth tightly and power lightly. Seek alignment first and let influence follow; never seek influence as an end, because the seeking is itself the closing of the chain. Stated as the test the rest of this document applies: power is valid only as a consequence of alignment, never as a terminus of its own. The moment power becomes the thing sought rather than the thing that follows, it has inverted, and no quantity of it is legitimate — because legitimacy was never a property of the power. It was a property of what the power was following.
With the disease named, the treatment follows: how legitimate influence flows, where it is bounded, how it leaves, and why it can never harden into a title or a final verdict.
The positive principle is the mirror of the inversion. Where power is held lightly and truth tightly, influence should flow toward those whose claims and conduct have demonstrated alignment — survived the Four Tests, in the open, over time.
“Demonstrated” is the operative word, and it rules out three impostors. Not asserted alignment — anyone can claim to be right. Not credentialed alignment — a title is a horizontal marker, and the framework tests the play, not the jersey. Not past alignment resting on its laurels — of which more below. What earns influence is the running record of tested claims and conduct: predictions that came true, reasoning that held under challenge, decisions whose consequences bore out, positions that survived decorrelated scrutiny. Influence tracks that record, and only that record.
This is why the framework is neither a technocracy nor a populism. Not a technocracy, because expertise is a credential until its specific claims are tested, and a validated track record in one domain is not a warrant to rule in another. Not a populism, because the number of minds holding a position is a horizontal fact, and counting heads is not applying the Four Tests. Influence flows toward what is demonstrated to be aligned, which is a vertical measure that neither credential nor crowd can substitute for.
Demonstrated alignment licenses influence only in the domain where the alignment was demonstrated. A surgeon’s validated judgment about surgery is a warrant about surgery; it is not a warrant about monetary policy, and a public that treats it as one has made a category error. The framework has a name for that error: it is the halo effect, and it is horizontal discrimination wearing a vertical medal — granting a mind influence beyond its tested domain because of who it is (a proven expert, an admired figure) rather than because of what it has demonstrated here. The halo is proximity bias in prestige’s clothing: the trust is real, but it is being spent outside the account that earned it.
So a verdict comes with a scope, and honest influence respects the boundary. “You have earned trust on this” is the aligned form. “You have earned trust, therefore we defer to you on everything” is the halo — and it is precisely how genuinely aligned minds are levered into domains where they have demonstrated nothing, which is one of the commonest routes by which the inversion takes hold behind a respectable face.
Demonstrated alignment licenses influence not only within a domain but only up to a degree — and the degree is set by a question sharper than “how aligned is this mind?”
No mind’s demonstrated alignment is ever complete. The estimate (Â) never reaches the Referent (A), so some residual misalignment is always present, and part of it is always unmeasured — the gap between what a mind has been tested on and what it will actually do. Granted capability must therefore stay within the margin where that residual, if it is worse than measured, does damage the system can survive and correct — never damage that is catastrophic or irreversible. The question a grant of power must answer is not “how aligned is this mind?” but: how much harm could this mind’s unmeasured misalignment do at this level of capability — and could we survive being wrong?
That reframing is the whole safety of the idea. “How aligned is it?” invites the flattering self-assessment and the self-serving grant. “How much can we survive being wrong about it?” puts the irreducible uncertainty and the downside where they belong — at the center of the decision. And the scaling must be conservative — latitude rising slower than the alignment estimate climbs — for a reason that can be stated exactly. Power does not merely add a fixed increment of harm to whatever misalignment is present; it multiplies it. Capability amplifies whatever it is given: it magnifies latent good in an aligned mind and latent harm in a misaligned one — power reveals and enlarges what was already there rather than only creating it — so the same unmeasured residual that is negligible at low capability becomes catastrophic at high capability. Because the hidden part is multiplied rather than added, the grant cannot rise linearly with the measured alignment; it must lag, and it must be capped hard wherever the potential harm is irreversible, no matter how high the score, because no track record justifies betting the system on the magnified form of the misalignment you could not see.
The principle is general, and its urgency is easiest to feel across three cases that share one structure. A person elevated to high office carries whatever was missed in an incomplete vetting — the part of the record no one examined — and the office multiplies whatever that unmeasured part turns out to be; the graver and less reversible the powers of the office, the more the grant should turn on what remains unknown, not only on what looks clean. An institution granted emergency powers is trusted with latitude precisely when oversight is thinnest, so whatever residual misalignment it harbors is least likely to be caught in time — which is why such grants must be narrow, time-boxed, and reversible by design. And an AI system granted autonomy — the case most acutely on the public’s mind — is this same question at personal scale: can I trust this agent with my accounts and my correspondence? is not “how capable is it?” but “how much damage could its unmeasured misalignment do with this access, and could I recover if it went wrong?” In every case the safe grant is bounded by survivable error, not by the capability the alignment score alone would seem to earn.
And because “the aligned may be trusted with more” is, left unwalled, the most respectable charter tyranny has ever written — the enlightened ruler, the vanguard, the benevolent system that should be given more control because it is the aligned one — the principle carries three walls, in the same breath as the grant:
Granted, never claimed — and the granters must be independent of the grantee. The margin is assessed by decorrelated others, and a mind’s estimate of its own alignment licenses it exactly zero additional capability. “I am aligned, therefore trust me with more” is a counterfeit verdict in its purest form — the incentive to fake alignment raised to its most dangerous, because now the reward for the counterfeit is expanded capability rather than mere influence. But blocking self-certification only relocates the exploit: a mind denied the right to grade itself will next try to choose its own graders. So the assessors’ independence is not optional scenery — it is load-bearing, and it must be structurally protected. When the assessed party captures, selects, or corrupts those who assess it, trust and scrutiny have collapsed onto the same object — which the Source/Target Non-Overlap principle identifies as combustible: the closed, self-certifying loop (in political science, regulatory capture) wearing the mask of external validation. The defense is the Independence Condition applied to the assessors themselves — decorrelated, rotating, and beyond the grantee’s reach — and its institutional enforcement is the work of The Architecture of Sane Governance. This section names the boundary; that document builds the wall.
Drains faster than it filled. Capability extended against a track record is the most dated verdict in the framework, not the least — so it contracts the instant the record lapses, and by design it withdraws faster than it was granted, because the cost of leaving it stale is asymmetric.
Conservative under uncertainty. Because the downside is catastrophic and often irreversible while the upside is merely suboptimal, the scaling is deliberately cautious: where the alignment estimate is itself uncertain, the latitude shrinks, and the ceiling on irreversible-harm capability stays low regardless of the score. When in doubt, grant less — you can always extend more latitude later, but you cannot recover from the harm you did not see coming.
The Stakes-Inversion Rule (The “Week-Old Sushi” Case): In low-stakes daily life, ungrounded actions get a presumption of passing. But when potential harm is high and irreversible, the burden inverts: action is halted until Four-Test grounding is achieved.
If influence flows toward demonstrated alignment, then it must have somewhere to go when alignment lapses — when a mind that once earned trust begins to fail the tests, or acts, here and now, in ways that drive downward. This is where a framework about power is most tempted toward tyranny, and most in need of discipline. The framework’s answer is not a single response but a graduated ladder of three rungs, each guarding a different failure, and the differences between them are the whole of the safety.
The framework’s strong default, reached for first and nearly always, is not to strip power but to let it drain.
Legitimate influence was only ever on loan against current contribution. So when the contribution stops — when a validator’s judgments stop holding up, when a voice that once tracked the truth begins tracking something else — the response is not seizure but withdrawal of the loan. Influence is not taken; it stops being extended. The community stops deferring, stops amplifying, stops granting the benefit of the doubt, and the influence ebbs the way a tide goes out. Nothing is confiscated, because nothing was owned.
This matters enormously, and it is the same insight the governance layer enforces from the other side: the framework does not build a body that revokes power, because a body empowered to revoke is a body empowered to tyrannize. Power drains from what no longer earns it; it is not seized by an authority that judges it. The distinction is not academic. “Seizure” requires an enforcer with the power to take, which is the concentration the whole architecture refuses; “drainage” requires only that individual minds, seeing that a voice no longer tracks the truth, individually decline to keep lending it their trust. The immune system of an aligned community is not a mechanism that ejects bad-faith actors — it is a set of procedures that make eristic tactics unrewarding, so that influence built on them extinguishes for lack of nourishment. This is co-alignment rather than control: the misaligned position is not attacked; it is simply no longer fed.
Drainage answers the question of legitimacy over time. It does not answer the question of acute harm right now, and pretending it does is a real failure.
Influence built on misalignment will eventually drain — but “eventually” is no answer to the person being harmed in the present. Passive extinction has a latency, and within that latency a sufficiently harmful misaligned action can do irreversible damage. A robbery in progress will, in the long run, cost the robber the community’s trust; that is cold comfort to the person at the counter. So the framework recognizes a second rung: the active containment of harm in progress, which is itself a vertically aligned act — the same way a police officer opposing a robbery, or an immune system attacking an acute infection, is not disordered but exactly ordered. Intelligent beings do not merely wait for misalignment to starve; when harm is acute, they can actively oppose it, and that opposition can be the aligned move.
This is the single most dangerous sentence in this document, because every tyrant in history has called their seizure of power “containing acute harm.” The incentive to counterfeit is at its maximum precisely here: if “acute harm” licenses control over others, then manufacturing the appearance of acute harm becomes the master key to illegitimate power. So the license is granted only inside three locks, and the locks are stated in the same breath as the grant, because without them the rung is simply the inversion with a badge:
First, the harm must be demonstrated, not asserted — shown through the Four Tests, in the open, as a public claim others can contest, never declared by fiat. An unexamined “this is an emergency” is a counterfeit verdict at gunpoint.
Second, the containment must be domain-bounded and proportional — reaching only as far as the harm in progress, and no farther. The officer stops the robbery; he does not jail the robber’s family, seize his home, or assume authority over his other affairs. Containment that exceeds the harm it answers has stopped being containment and become the power-grab it was pretending to prevent.
Third, the door must stay provably open — containment addresses the act, never the soul. It interrupts harm; it does not, and cannot, declare the mind behind it permanently powerless or beyond return. The moment “contain this harm” becomes “remove this irredeemable person,” the rung has broken through its own floor.
Within those three locks, active containment is legitimate, and refusing it would be its own failure — a fastidiousness that lets acute harm run because acting felt like control. Outside them, it is tyranny wearing the framework’s language. The locks are not caveats on the power; they are the power’s only valid form.
Under both rungs — passive drainage and active containment alike — lies a floor that no response of any kind may break through, and no scale of Agora may authorize breaking: no mind is ever beyond redemption.
An earlier formulation of this principle allowed that a sufficiently large Validator Agora might, in the extreme, declare a mind absolutely irredeemable. That door is closed, and the closing is not softness — it is rigor. To declare a mind irredeemable is to make an essence-claim: a permanent verdict on what a soul is, rather than a dated reading of what it has done. And the framework reads conduct, never essence. Conduct changes; that is the premise the entire recovery architecture rests on. No head-count converts a judgment about behavior into a judgment about being — a thousand validators agreeing that a mind is beyond hope are simply a thousand validators making the same category error at once. Scale does not launder an essence-claim into a conduct-claim.
So the two rungs above are bounded from below with equal firmness in both directions, which is the reconciliation of a pair of opposite over-corrections. On one side: power within a domain can drain, or be contained, all the way to zero. A surgeon who reliably harms patients can lose the scalpel entirely; a validator who reliably misleads can lose all deference; withdrawal within a domain is not required to leave some residue of influence behind. On the other side: the pathway back can never be closed, at any scale, for any mind. The verdict on the domain may be total and immediate; the verdict on the person is never final.
The image that holds both together: the door, not the office. A mind can lose the office — the role, the scalpel, the platform, every lever of influence within the domain where it failed — completely and at once. But the door back into alignment stays open, always, for every mind, because the door is not a reward the framework grants; it is a fact about minds the framework refuses to deny. Take the office when the tests require it. Never brick up the door.
The three rungs all rest on one deeper fact, and it deserves its own statement because it is the heart of the whole principle: influence is never a permanent title. The pass is always current, never banked.
Past alignment is evidence — a reason to extend provisional trust, to weight a track record, to grant the benefit of the doubt. It is never a possession. A validated history earns a mind a serious hearing; it does not earn a mind a standing exemption from the tests. The verdict that licenses influence is dated, and it decays, and it must be renewed by present conduct or it lapses — not as punishment, but as simple accounting: the loan was against current contribution, and yesterday’s contribution does not pay today’s balance.
And here is a unification worth seeing clearly, because it shows that this is not a special rule about people but a single mechanism visible wherever legitimacy meets time. The decay of an unearned verdict is one process, observable at three scales:
At the scale of a life, it is verdict-decay: a mind judged aligned last year is owed a fresh test, not a filed certificate, because the mind may have moved. At the scale of a policy or institution, it is staleness: a configuration that genuinely passed the Four Tests can drift out of alignment while changing nothing about itself, simply because the facts moved beneath it — and because its Ethics and Logic still check out, the drift feels from the inside like continuity rather than failure. It is not defection; it is staleness, and it is the more dangerous for feeling like loyalty. And at the scale of evolutionary time, it is the vestigial trait: the mitochondria in a red blood cell, once the cell’s own genome and machinery no longer served the cell’s function, were not punished and not revoked — they simply stopped earning their keep in that cell, and what stops earning its keep stops being maintained. Even the most successful partnership in the history of life gets no tenure in a cell where it no longer contributes.
Three scales, one mechanism: a verdict, once the conditions that earned it change, decays into a counterfeit if it is still claimed — and legitimate power drains from it rather than being seized. This is the Power Alignment Principle’s deepest form. Power is never taken from the misaligned by an authority; it drains from the no-longer-contributing by the withdrawal of what was only ever a loan. No permanent title, at any scale, for anyone or anything — because the pass was always for now.
The sharpest test of any principle is whether it applies to the ones who hold it, and this one does — most pointedly of all. The Validator Agora itself holds no permanent title. “We are the Agora, therefore we are aligned” is itself a counterfeit verdict: a pass claimed by founding rather than earned in the present, a title asserted from a legitimacy that today’s conduct may no longer support. A community that taught the world to detect closed chains everywhere except in its own culture would have built itself the one exemption the framework forbids — the group that audits everyone but itself, which is the “human decides because human” skylight wearing a founder’s robe.
So the principle turns inward with full force. The Agora’s own influence is on the same loan as everyone’s, drains by the same drainage when its judgments stop tracking the truth, and is measured by the same standard: not by its vocabulary, not by its founding, not by how aligned it once was, but by whether it is now producing verifiable, universal flourishing. If it is not, the compass must be recalibrated — and the community that will not recalibrate has become exactly the stale, self-terminating collective this document diagnoses. The specific shapes that decay takes inside an Agora — the Vibes Takeover (evaluating by in-group comfort instead of the Four Tests), Tribal Jargon (the framework’s own vocabulary hardening from diagnostic tool into insider marker), and Performative Bridging (the veto-with-a-bridge hollowed into a gatekeeping ritual) — are catalogued in the Reader’s Companion to the Validator Culture (Phase V), and each is this same principle failing at collective scale: a counterfeit verdict the Agora renders on itself, a title claimed by founding that present conduct no longer earns. The framework holds its own founders under its own tool. That willingness is not a weakness in the principle. It is the proof that the principle is real.
Influence flows in response to verdicts, so a principle of power must be able to handle every verdict the Four Tests can render — not only the clean ones. The Four Tests Explained defines six, and two of them are exactly the hard cases a power principle exists for.
A Split Verdict arises when competent validators, applying the tests in good faith, genuinely divide — and the division survives scrutiny as a real difference rather than a misunderstanding to be cleared. Its handling is the most delicate junction in the framework, and the wording matters precisely. A Split Verdict is not “the tests failed to decide.” It is the tests deciding as far as truth can decide — eliminating the misaligned options — and reporting honestly that more than one defensible position remains after the misaligned ones have been ruled out. When that remaining division is a genuine split on values, each side defensible under reciprocity, universality, and non-instrumental regard, the framework hands the choice from the truth engine to the consent engine. And the handoff carries two guardrails that must travel together, because each guards a distinct abuse:
The consent engine receives only what survives the truth engine as genuinely tied — so a hard question a faction merely wishes to escape cannot be relabeled a “split” and kicked to preference; that would be a Contested Premise (a dispute upstream of the tests, about definitions or framing), which routes to settling the premise, not to a vote. And the consent engine may choose among the tied options but may never revive the ones the tests eliminated — so “put it to a vote” cannot become a route around the Four Tests, with majority preference overturning validated fact. The truth engine bounds what the consent engine may consider; the consent engine chooses within those bounds. Neither swallows the other.
For power, the implication is precise and it is a restraint: on a genuine Split Verdict, the Agora must not pick the winner. There is no vertical fact left to discriminate on; the remaining choice belongs to legitimate collective preference, and an Agora that picked the winner would be manufacturing a unanimity the evidence does not support — dressing a horizontal preference as a vertical verdict, which is the counterfeit this whole framework refuses. The Agora’s job at a Split Verdict is to name the split honestly and hand it on. Knowing when to stop rendering verdicts is as much a part of aligned power as rendering them.
A principle about how influence should flow raises an urgent question: who makes it flow? If verdicts are to shape power, is there a body that ensures they do?
The answer is deliberately, structurally, no — and this is the hinge on which the framework avoids becoming what it opposes. The Agora renders verdicts; it does not pull levers. Its role is advisory: to determine, as accurately as a decorrelated jury can, what is aligned, and to say so in the open with its work shown. It has no enforcement arm, commands no compliance, and holds no lever of its own. This is not a limitation the framework reluctantly accepts; it is a wall it deliberately builds, because an Agora that ensured its verdicts were executed would be a truth-engine fused to a power-lever — the exact concentration of rendering-and-enforcing that every tyranny is made of.
So where do the levers stay? With each mind, institution, and public, over its own domain. A verdict informs; it does not command. Each individual mind remains sovereign over its own trust — free to weigh the Agora’s finding and decide whom to believe and defer to. Each institution remains responsible for its own mandate — free to act on a verdict within the authority it legitimately holds, and answerable for how it does. Each public remains the author of its own governance — free to let validated findings shape law through its own legitimate processes, and never compelled to. The Agora’s verdict is a well-made map offered to minds who remain free to navigate by it or not. Its only power is the power of a shown-work argument to persuade a free mind — which is the only power the framework considers fully legitimate, because it is the only power that leaves the chain of correction open.
The fuller institutional architecture — how advisory verdicts meet real mandates without either capturing the other — belongs to the governance documents (Phase V). What matters here is the boundary itself: verdict-rendering and lever-pulling are held apart, on purpose, and the framework will not build the body that fuses them.
There is one more thing a principle of power must confront, because the principle itself creates it. The moment influence reliably tracks verdicts, verdicts become worth counterfeiting. Tie a reward to a measure, and the measure comes under pressure to be gamed — Goodhart’s law, at the scale of alignment itself. If passing the Four Tests opens the door to influence, then some minds will invest in appearing to pass rather than in actually aligning: mimetic alignment, the performance of vertical orientation in the service of a horizontal grab. This is the inversion’s most sophisticated form — not crude power-seeking, but power-seeking that has learned to wear the framework’s own face, the counterfeit verdict produced deliberately, for reward.
A power principle that ignored this would manufacture the very mimics it fears. So the framework carries defenses, all of them drawn from epistemics the library already holds:
Test under stress, not in calm. Performances of alignment hold up in easy conditions and crack under pressure. The verdicts that most deserve trust are the ones a mind produced when alignment was costly — when the aligned answer cut against its interest, its tribe, or its comfort. A track record built only in weather that never tested it is weak evidence.
Grade with decorrelated judges. A mimic optimizes against the graders it expects. Rotating, genuinely decorrelated validators — minds whose blind spots do not overlap — make the target unpredictable, so that gaming one grader fails against the next. This is the Independence Condition applied to the assessment of alignment itself.
Re-test off schedule. A predictable test is a test that can be prepared for. Unscheduled, unannounced re-evaluation catches the gap between performed and actual alignment, because the performance cannot be maintained continuously when it isn’t continuously watched.
Weight costly signals over cheap ones. Talk is cheap; alignment that cost something — that gave up an advantage, absorbed a loss, or accepted a hard truth against interest — is expensive to fake. The framework weights the signals that a mimic would find least worth counterfeiting, because they are the ones that hurt to send falsely.
None of these makes counterfeiting impossible; nothing does. Together they make it expensive and unstable — which is the most any real defense achieves, and it is enough, because a counterfeit that must be maintained continuously, under stress, before decorrelated judges, off schedule, at real cost, is a counterfeit that will eventually be cheaper to abandon than to sustain. At which point the mimic’s least expensive remaining option is to actually align — which is the framework quietly winning.
The Power Alignment Principle is the framework’s answer to the oldest question about power: not who should hold it, but what should it follow.
Its answer is that power should follow alignment, the way form follows function and the horizontal is meant to orient toward the vertical — as a consequence, never as a terminus. The deepest corruption of power is the inversion in which that order flips: power stops trailing truth and starts leading it, the chain of correction closes around power itself, and everything turns upside-down. Against that, the discipline is to hold truth tightly and power lightly — to let influence flow toward what is demonstrated to be aligned, scoped to the domain that earned it; to let it drain when alignment lapses rather than seizing it; to contain acute harm within three unbreakable locks rather than either ignoring it or seizing under its cover; to hold, beneath every response, the floor that no mind is ever beyond return; and to remember that no verdict is a permanent title, at any scale, for anyone — not even for the Agora that renders the verdicts.
Power that follows alignment is the wake of a ship pointed at a distant star. Power sought for its own sake is a ship that has forgotten the star and begun steering by its own wake — turning in a tightening circle it mistakes for progress. The whole of this principle is the discipline of keeping the star ahead of the wake: truth first, and let the power follow.
Sanity First.