Helen Park has been making her own investment decisions since she retired from teaching mathematics fifteen years ago. She is 73, lives alone in a ranch house outside Tulsa, and manages a portfolio she built methodically over three decades. She is sharp. She is also, by her own account, terrified of losing money. She held cash through the 2020 rally because the possibility of loss outweighed the probability of gain. She pays more for insurance than actuarial tables justify because the security of coverage outweighs the expected savings of self-insuring. She chose a fixed annuity over an indexed one because the guaranteed floor mattered more to her than the potential ceiling.
Helen is not confused. She is not cognitively impaired. She is loss-averse, and her loss aversion is a consistent feature of how she reasons about risk, not a defect in her reasoning.
A system that treats Helen’s loss aversion as an error to correct has misunderstood both Helen and the problem. A system that ignores her loss aversion when an external agent exploits it to sell her an overpriced protection product has also failed her. The space between those two failures is where Irrationality Vector Quantization operates.
Cognitive biases as features#
The term “irrationality” in IVQ is a technical label for deviations from expected-utility-maximizing decision-making. It does not mean the person is irrational. It means the person’s decision-making has consistent patterns that diverge from a theoretical rational agent, and those patterns are features of the person’s cognition that the system should understand.
Loss aversion is one such pattern. Helen’s preference for guaranteed outcomes over expected-value-higher risky outcomes is well-documented in behavioral economics. It is not a mistake she is making. It is how she processes risk. An AI system that presents financial options to Helen as if she were an expected-utility maximizer is communicating in a language she does not speak. The recommendation may be technically optimal. It will be ignored, because the framing does not connect with how Helen actually evaluates tradeoffs.
Anchoring bias is another pattern. A person who sees a $500 item first and a $200 item second evaluates the $200 item differently than a person who sees the $200 item first. The anchor shapes the evaluation. This is not a reasoning failure. It is how humans process comparative information. A system that understands anchoring can present options in an order that helps the person evaluate them fairly. A system that does not understand anchoring can inadvertently create anchors that skew the person’s decisions.
Status quo bias, the preference for current arrangements over changes that might be beneficial, affects healthcare decisions with particular force in aging adults. Margaret has been taking the same blood pressure medication for eight years. Her physician recommends switching to a newer medication with fewer side effects. Margaret’s status quo bias makes the switch feel risky even though the evidence favors it. The system that understands this bias does not override Margaret’s preference. It presents the recommendation with framing that acknowledges the comfort of the current arrangement while making the evidence for the change accessible: “Your current medication is working. Dr. Patel suggests a newer one that may work equally well with fewer side effects. You do not have to switch. Here is what the research shows, and we can discuss it with Dr. Patel at your next visit.”
Sunk cost reasoning, hyperbolic discounting, authority bias, availability heuristic: each is a documented pattern that shapes how specific individuals make specific decisions. IVQ does not catalog these as pathologies. It models them as dimensions of the person’s cognitive profile that inform two distinct system functions: internal communication adjustment and external exploitation protection.
Two directional uses#
The internal use of IVQ is framing translation. The system communicates with the person’s cognitive style, not against it.
For Helen, this means financial recommendations are framed in terms of downside protection rather than upside potential. “This option protects your principal with a guaranteed 3% floor” connects with Helen’s loss aversion better than “this option has an expected return of 7%.” Both statements may describe the same product. The framing difference determines whether Helen engages with the recommendation or dismisses it.
For a person with strong authority bias, the system surfaces the credentials of the recommending source more prominently. “Your cardiologist, Dr. Rivera, reviewed your recent echocardiogram and recommends…” lands differently than “based on your recent results, you might consider…” The content is identical. The framing matches how this specific person evaluates information.
The framing translation is not manipulation. Manipulation changes what the person decides. Framing translation changes how the information is presented so that the person can engage with it through her natural cognitive patterns. The distinction is testable: if the same information, presented through framing translation, produces a different decision than the same information presented neutrally, the system has communicated more effectively. If framing translation produces a different decision than the person would make with unlimited time and full information, it has crossed into manipulation.
The external use of IVQ is exploitation protection. The system protects the person from external agents that exploit her cognitive patterns.
Helen’s loss aversion makes her a target for fear-based marketing. “Your home warranty expires in 48 hours. Without coverage, a furnace replacement could cost $8,000.” The urgency is manufactured. The loss framing exploits Helen’s specific vulnerability. The manipulation detector (BMT-03.06) catches the urgency claim. IVQ adds a layer: the system knows that Helen is particularly susceptible to loss-framed urgency because her IVQ profile shows high loss aversion. The system intervenes more aggressively for Helen than it would for a person with lower loss aversion, because the exploitation risk is higher.
The asymmetry is critical and non-negotiable: external agents never see IVQ profiles. The pharmacy does not know that Helen is loss-averse. The insurance agent does not know that Margaret has high status quo bias. The financial advisor does not know that Dorothy has strong authority bias. IVQ profiles are internal-only, used by the system to serve the person and to protect the person, never disclosed to the agents that might exploit them. The privacy architecture (BMT-04.07) enforces this at the maximum protection tier. IVQ data is among the most sensitive information the system holds, because it is a map of the person’s cognitive vulnerabilities.
The protection mechanism works through calibrated intervention thresholds. When an external agent’s communication pattern matches the person’s IVQ vulnerability profile, the intervention threshold drops. The manipulation detector uses the same detection logic for all subscribers, but the sensitivity is adjusted: a loss-framed urgency claim from a vendor triggers a standard detection score for a TIER_ANALYST subscriber and a higher detection score for a TIER_GUARDIAN subscriber. The content of the message is identical. The exploitation risk differs because the person’s cognitive pattern differs. The system calibrates its defense to match the person’s specific exposure.
Consider a concrete scenario. A supplemental insurance agent contacts Helen through the membrane with an offer. The offer is legitimate: a dental coverage plan with reasonable terms. But the framing is designed to convert: “Without dental coverage, a single root canal could cost $2,500 out of pocket. That is money you cannot get back.” The language is factually accurate. It is also a textbook loss-framing strategy that targets exactly the cognitive pattern Helen’s IVQ profile describes.
The manipulation detector flags the loss framing. The IVQ layer recognizes that Helen’s TIER_GUARDIAN profile makes her more susceptible to this framing than the subscriber population average. The system does not block the offer. It does not decide for Helen. It reframes: “Hartford Dental is offering a supplemental dental plan at $28 per month. The plan covers cleanings, X-rays, and major procedures including root canals. Your current dental expenses over the past year averaged $340. Here is a comparison of the plan cost versus your current spending.” The loss framing is replaced with a factual comparison. Helen can still buy the coverage. She is evaluating it against her actual spending pattern, not against a hypothetical catastrophe designed to trigger her loss aversion.
Cognitive tiers, not scores#
IVQ quantizes cognitive bias profiles into four tiers that characterize reasoning style, not reasoning quality.
TIER_ANALYST describes a person whose decisions are primarily data-driven, with lower sensitivity to framing effects and higher tolerance for probabilistic reasoning. The system communicates with this person using data tables, probability distributions, and explicit tradeoff comparisons. External exploitation risk for this tier is lower for emotional manipulation but higher for data-based deception (fabricated statistics, misleading comparisons).
TIER_GUARDIAN describes a person whose decisions are primarily protection-oriented, with high loss aversion and strong status quo bias. Helen fits this profile. The system communicates with this person by leading with safety, stability, and downside protection before presenting opportunity. External exploitation risk is highest for fear-based marketing and manufactured urgency.
TIER_CONNECTOR describes a person whose decisions are primarily relationship-driven, with high authority bias and strong influence from social proof. The system communicates with this person by surfacing trusted source attribution and community patterns. External exploitation risk is highest for social engineering and false authority claims.
TIER_EXPLORER describes a person whose decisions are primarily novelty-seeking, with low status quo bias and high tolerance for uncertainty. The system communicates with this person by leading with what is new, different, or unfamiliar. External exploitation risk is highest for unfounded opportunity claims and “too good to be true” offers.
The tiers are not fixed. A person may be TIER_GUARDIAN for financial decisions and TIER_EXPLORER for social activities. IVQ maintains domain-specific tier assignments, consistent with the contextual approach in TVQ (BMT-11.02). Helen’s financial IVQ tier is TIER_GUARDIAN. Her cooking IVQ tier, where she experiments freely with new recipes and cuisines, might be TIER_EXPLORER. The system adapts its communication and protection by domain, not by a single whole-person classification.
What IVQ does not do#
IVQ does not diagnose cognitive impairment. The Cognitive State Estimator (BMT-04.05) handles cognitive capacity assessment. IVQ operates on a different axis: it models decision-making style, not decision-making capacity. A person can have high cognitive capacity and high loss aversion simultaneously. A person can have declining cognitive capacity and stable decision-making style. The two assessments are independent, and the system treats them as independent.
IVQ does not correct the person’s biases. The system that decides Helen’s loss aversion is “wrong” and overrides it to present expected-value-maximizing options has violated her autonomy. Helen’s loss aversion is hers. It reflects her values, her experiences, and her risk tolerance. The system respects it by communicating through it and protecting against its exploitation. It does not attempt to rationalize her into a different decision-making style.
IVQ does not produce a vulnerability score that external parties can use. The IVQ profile, the tier assignments, the domain-specific patterns: all of this is internal. The person herself can request to see her IVQ profile in plain language. She can contest it. She can request that specific bias patterns be removed from her profile if she believes they are inaccurate. The system that models the person’s cognition without giving her visibility and control over the model has built a tool for surveillance, not service.
IVQ operates regardless of the subscriber’s deployment path. A subscriber on Path F (Zone 3 only, no Local Pane device) receives the same cognitive pattern modeling and the same exploitation protection as a subscriber on Path A (full stack with dedicated Local Pane). The difference is where the IVQ computation runs: for Path A subscribers, the cognitive state signals that feed IVQ originate in Zone 1 and the IVQ profile is stored locally. For Path F subscribers, the signals are processed in Zone 3 and the profile is stored under the same privacy controls that govern all MoC context in the cloud tier. The protection level is identical. The privacy posture for the underlying data differs by path, consistent with the deployment architecture (BMT-09.01). The equity monitoring framework (BMT-11.04) tracks whether IVQ protection outcomes differ by deployment path and flags any disparity.
Helen’s financial advisor recommended a variable annuity with a 7% expected return. Helen declined. Under IVQ, the financial concierge would present the same recommendation differently: “This option has a floor of 3.5% that protects your principal in any market condition, with the potential for higher returns in strong years.” The information is identical. The framing connects with how Helen evaluates risk. Whether she accepts or declines is still her decision. The system served her cognition. It did not correct it.
Cross-References#
Attack Resistance (BMT-03.06). The manipulation detection architecture that IVQ extends with cognitive vulnerability awareness for more precise exploitation detection.
Cognitive Capacity and Consent (BMT-04.05). The capacity assessment architecture that operates on a different axis from IVQ: capacity measures how well the person can reason, while IVQ models how the person reasons.
The Cognitive Concierge (BMT-01.07). The agent architecture that generates the cognitive state estimates IVQ uses for domain-specific tier assignment.
Trust Vector Quantization (BMT-11.02). The complementary framework: TVQ models trust in external agents, while IVQ models the person’s cognitive patterns that external agents might exploit.
Technical Appendix BMT-11.03-A is available to partners and investors at partners.bluemirror.tech.