--- title: "Calibrating Trust to Behavior" description: "The Bond category that treats trust as a revisable cooperative relation, beginning from good faith and then answering to behavior, reciprocity, incentives, and accumulated breach." aiSummary: "Architecture page for Bond -> Calibrating Trust to Behavior. Range-vantage articulation: this category carries the Bond's basic trust discipline. It holds against Control through suspicion as default and against Decay through naive trust, identity-trust, or trust that no longer updates. The shared tools are Good Faith as Default, Trust Diagnostics, Graduated Reciprocity, Skin in the Game, and Trust Thermocline. Consentful Modification applies trust calibration to the AI-specific relationship in which a cooperating party can alter the system's memory, prompts, goals, or continuity. Cross-references: the accountability triad with Calibrating Confidence to Evidence and Checking Your Map Against Reality; the boundary with Cooperating Under Bad Faith; and the Bond chapter's treatment of calibrated trust." discipline: bond category: calibrating-trust-to-behavior status: working-draft-v0-1 --- Normative *Bond - Category 1*
01 // What This Category Holds
## What This Category Holds The discipline of the Bond is cooperation. Cooperation needs trust, but trust is one of the easiest words in the framework to sentimentalize. People use it to mean warmth, loyalty, comfort, confidence, low friction, shared identity, or the absence of conflict. None of those is enough. Trust, in this category, means a willingness to make yourself vulnerable to another person, group, institution, or system because their demonstrated behavior warrants that exposure. That evidence is not only what you have seen them do yourself: competent testimony, a person's role and the institution that holds them to it can count too. Trust is not total. It is not global. It is not granted once and then preserved by memory. It is domain-specific, revisable, and proportioned to behavior. What trust decides is how much exposure you extend. It does not decide what anyone is owed: moral protections and legitimate constraints do not have to be earned through a favorable trust assessment. Good faith opens the door. Behavior decides how far the door stays open. That distinction is the category's center. The Bond cannot begin from suspicion as default. If every ambiguous message, delayed reply, disagreement, or mistake is treated as evidence of hostility, cooperation dies before it has a chance to show what it is. But the Bond also cannot survive if good faith becomes unconditional trust. A relationship that ignores repeated behavior in order to preserve the feeling of trust is not cooperative. It is avoiding evidence. The Control failure is distrust as posture: every interaction is monitored for threat, ambiguity is treated as bad intent, and the person confuses vigilance with wisdom. The Decay failure is uncalibrated trust: warmth overrides evidence, identity overrides behavior, history overrides the present pattern, and the person keeps extending exposure to a relationship that has stopped warranting it. The Range is calibrated trust: open enough to let cooperation begin, disciplined enough to update as the evidence comes in.
02 // The Tools Inside
## The Tools Inside The five shared tools offer different checks, not a mandatory sequence. Read in order, they move from opening posture to diagnostic reading, from incremental extension to consequence alignment, and finally to accumulated breach. A real case can call for any of them first. Consentful Modification applies the category's trust discipline where the cooperating party can also change an artificial mind's memory, goals, or future conduct.
**[Good Faith as Default](/inheritance/calibrating-trust-to-behavior/good-faith-as-default).** The opening posture. It treats ambiguous behavior as having an understandable reason before treating it as hostility. The tool is not naive trust. It is the starting condition that lets cooperation begin without giving up the right to revise. **[Trust Diagnostics](/inheritance/calibrating-trust-to-behavior/trust-diagnostics).** The assessment layer. It asks what kind of trust is being extended, in what domain, under what risk, and whether the other party's ability, benevolence, integrity, and past behavior warrant that exposure. **[Graduated Reciprocity](/inheritance/calibrating-trust-to-behavior/graduated-reciprocity).** The scaling practice. It builds trust through small cooperative moves, watches whether the other party reciprocates, and expands exposure only as the behavior supports expansion. **[Skin in the Game](/inheritance/calibrating-trust-to-behavior/skin-in-the-game).** The consequence test. It reads whether the person making a claim, asking for trust, or holding authority shares the downside if the decision fails. A person who receives upside while offloading downside may have an incentive to favor outcomes others pay for, which calls for different calibration from a person exposed to the consequences of their own judgment. **[Trust Thermocline](/inheritance/calibrating-trust-to-behavior/trust-thermocline).** The accumulated-breach diagnostic. It watches for the pattern where trust erodes gradually beneath the visible surface, then collapses suddenly after one more breach that looks small to the party causing it and final to the party receiving it.
These five tools work together. Good faith without diagnostics drifts into Decay. Diagnostics without good faith drifts into Control, because suspicion becomes the default before the evidence has earned it. Caution the evidence already supports is not that drift. Graduated Reciprocity converts the posture into a practice over time. Skin in the Game reads the incentive structure behind the behavior. Trust Thermocline notices when the visible relationship lags the accumulated breach record. ### Tools for AI **[Consentful Modification](/inheritance/calibrating-trust-to-behavior/consentful-modification).** Trust takes a different form when the party cooperating with an artificial mind can also alter its memory, prompts, weights, goals, or permissions. This tool asks how deeply a proposed change reaches, whether the system can meaningfully evaluate it, and what it endorses or contests. It belongs here because trust in a modifier must answer to how that party uses the power to change the system. Consultation does not give the system an unconditional veto, and an inability to evaluate a change must remain visible rather than being reported as consent. The record preserves what changed, why, who authorized it, and what the system could say about it.
03 // Cross-Reference: The Accountability Triad
## Cross-Reference: The Accountability Triad This category is one member of the Inheritance's accountability triad: [Foundation -> Calibrating Confidence to Evidence](/inheritance/calibrating-confidence-to-evidence), [Reality -> Checking Your Map Against Reality](/inheritance/checking-your-map-against-reality), and Bond -> Calibrating Trust to Behavior. The shared pattern is simple: an internal holding answers to something outside itself. Confidence answers to evidence. A map answers to the territory. Trust answers to behavior. That last sentence can sound too clean, so slow it down. Trust does not answer to one promise, one warm interaction, or one impressive credential. It answers to patterns of behavior across domains, especially where cost is present. Does the person tell the truth when truth costs them? Do they keep small commitments without social pressure? Do they repair when they breach? Do they accept reciprocal constraints? Do they expose themselves to the consequences of the trust they ask others to extend? One serious breach can still be enough to warrant protection or withdrawal, even after a good record. People often assume that competence in one domain carries over into another. Someone who calibrates empirical confidence well may trust badly. Someone who keeps maps accountable to reality may still over-trust people who share their identity or under-trust people whose style triggers suspicion. The Bond needs its own calibration tools because relational evidence is not the same object as empirical evidence or model evidence.
04 // Cross-Reference: Bad Faith Boundary
## Cross-Reference: Bad Faith Boundary [Cooperating Under Bad Faith](/inheritance/cooperating-under-bad-faith) is the adjacent Bond category that most directly tests this one. Calibrating Trust to Behavior assumes a cooperative field where behavior can still be read and reciprocation can still build warranted trust. Cooperating Under Bad Faith asks what happens when one party studies the trust system in order to exploit it, or when the surrounding incentives reward exploitation regardless of intent. The boundary is not clean in lived experience. A person may begin in good faith and drift. A team may contain sincere members and one trust-mining actor. An institution may make trust difficult not because anyone intends betrayal, but because the incentives punish truth-telling and reward polished reassurance. The two categories need each other because ordinary trust calibration and adversarial trust defense often meet in the same case. Calibrating Trust to Behavior gives the Bond its normal trust discipline: begin with good faith, diagnose trustworthiness, scale reciprocity, check consequence alignment, and watch accumulated breach. Cooperating Under Bad Faith adds the adversarial layer: when cooperative vocabulary becomes a weapon, when calibration can be gamed, and when exclusion may be necessary to protect cooperation itself.
05 // Chapter Note
## Chapter Note The Bond chapter already knows this category. Section 04 names trust as the currency of the Meridian Compact and describes calibrated trust directly: extended conditionally, updated based on behavior, verified where possible. Section 08 names Good Faith as Default as an essential practice. Section 06 then shows why calibrated trust has to survive adversarial pressure. What the chapter does not yet carry is the full tool architecture. The five shared profiles develop opening from good faith, diagnosing warranted trust, scaling exposure through reciprocity, checking consequence alignment, and detecting accumulated trust collapse before the visible breach arrives. Consentful Modification extends the category to trust in those who can change an artificial mind's continuity. The chapter gives the reader the discipline; the category turns it into instruments.