Chapter 21: Geometric Clinical Ethics — Triage, Consent, and the Clinical Geodesic
RUNNING EXAMPLE — Priya’s Model
Dr. Osei points out that TrialMatch effectively assigns QALYs: Mrs. Voss’s ‘expected quality-adjusted life-years from trial participation’ drives her score. By the QALY Irrecoverability Theorem, this number cannot capture that Mrs. Voss values knowing she fought independently of outcome. Her dignity-interest Iₘ at μ = 5 contracts with a different obligation component than her survival-interest at μ = 1. The scalar cannot hold both.
This chapter applies the Geometric Ethics framework to clinical medical ethics. The central claim is that the field's foundational tools — principlism, QALYs, and narrative deliberation — share a common deficiency: they lack the mathematical structure needed to make clinical-ethical reasoning precise, consistent, and auditable. The framework developed in the preceding chapters provides that structure. Clinical decisions are pathfinding on a nine-dimensional clinical decision complex, informed consent is a gauge-invariance condition, moral injury is cumulative manifold damage, and the QALY is a scalar projection that destroys eight dimensions of clinically relevant information.
21.1 The Failure of Scalar Clinical Ethics
Clinical ethics currently operates with four principal tools, each of which is inadequate for distinct reasons.
Principlism. Beauchamp and Childress's four principles — autonomy, beneficence, non-maleficence, and justice — provide a vocabulary for ethical deliberation but no algorithm for resolving conflicts among them. When autonomy and beneficence collide (a competent patient refuses life-saving treatment), principlism offers no formal method for determining which principle prevails. The resolution depends entirely on the clinician's judgment, which is neither transparent nor reproducible.
QALYs. The Quality-Adjusted Life Year compresses the full clinical-moral state into a single scalar: years of life weighted by health-related quality. This is precisely the scalar projection that Chapter 15 proved destroys information. The QALY maps a nine-dimensional clinical state to a one-dimensional score, with a kernel of dimension eight. All information in that kernel — autonomy, trust, dignity, justice, epistemic status — is mathematically irrecoverable.
Casuistry. Case-based reasoning identifies morally relevant similarities between current and paradigm cases but provides no formal measure of "similarity." Two clinicians may judge different features as morally relevant and reach opposite conclusions, with no principled method for adjudication.
Narrative ethics. The narrative approach enriches understanding of patients' lived experience but is inherently non-quantitative. It cannot support the kind of systematic policy analysis that institutional decision-making requires.
The geometric framework addresses all four deficiencies: it provides an algorithm (A* search), a multi-dimensional cost function (Mahalanobis distance on the clinical decision complex), a formal similarity metric (graph distance), and quantitative policy analysis (the Clinical Bond Index).
21.2 The Clinical Decision Complex
Definition 21.1 (Clinical Decision Complex). The clinical decision complex C is a weighted simplicial complex with the following structure:
Vertices (0-simplices): Each vertex v_i represents a clinical state — a configuration of diagnosis, prognosis, treatment status, patient preferences, and institutional context. The vertex carries an attribute vector a(v_i) in R^9 scoring the nine clinical-moral dimensions.
Edges (1-simplices): An edge (v_i, v_j) represents a clinical action — a treatment, test, conversation, referral, or decision to wait. The edge carries a weight w(v_i, v_j) >= 0 representing the total clinical-moral cost of the transition.
Higher simplices: A k-simplex [v_0, ..., v_k] represents a care bundle — a coordinated set of interventions (e.g., intubation + sedation + family notification).
The nine clinical-moral dimensions are the moral manifold dimensions of Chapter 5, instantiated for clinical contexts:
d_1: Clinical outcomes — survival, morbidity, functional status, symptom burden. This is the dimension QALYs target.
d_2: Rights and obligations — the patient's right to treatment, duty of care, non-maleficence, the right to refuse.
d_3: Justice and fairness — equitable access, distributive justice, non-discrimination, proportionality of resource allocation.
d_4: Autonomy — self-determination, voluntariness, freedom from coercion, capacity for informed decision-making.
d_5: Trust — the therapeutic relationship, fiduciary duty, confidentiality, institutional trustworthiness.
d_6: Social and relational impact — effects on family and caregivers, community health implications, social determinants, social role preservation.
d_7: Dignity and identity — the patient's sense of self, bodily integrity, alignment with personal values, quality of dying, and (for clinicians) professional identity.
d_8: Institutional legitimacy — standard of care, clinical guidelines, regulatory compliance, legal defensibility.
d_9: Epistemic status — diagnostic certainty, prognostic accuracy, patient understanding, health literacy.
Definition 21.2 (Clinical Edge Weights). The weight of an edge (v_i, v_j) in C is:
w(v_i, v_j) = ΔaT Σ−1 Da + Σk βk * 𝟙[boundary k crossed]
where Da = a(v_j) - a(v_i) is the attribute-vector difference, Sigma is the 9x9 clinical-moral covariance matrix estimated from clinical behavioral data, and βk is the penalty for crossing the k-th clinical-moral boundary.
The covariance matrix captures critical clinical-moral interdependencies. Sigma_{1,5} (outcomes x trust): uncertain outcomes cost more when trust is fragile. Sigma_{4,9} (autonomy x epistemic): respecting autonomy requires adequate patient understanding. Sigma_{3,7} (justice x dignity): allocation decisions that are fair but dignity-stripping have high cross-dimensional cost.
Definition 21.3 (Clinical-Moral Boundaries). Five boundary types carry distinctive penalties:
Harm boundary (beta_harm): "First, do no harm." Very high but finite — surgery crosses it acceptably because the clinical geodesic through the harm boundary reaches a state with sufficiently lower total cost.
Futility boundary (beta_futile): Treating patients who cannot benefit. Crosses d_3 (justice — resources consumed without commensurate benefit), d_7 (dignity — subjecting patients to interventions that cannot help), and d_5 (trust — performing procedures the clinician knows are futile).
Consent boundary (beta_consent): Approximately infinite for competent, non-emergency patients. Finite for genuine emergencies (implied consent) and for patients lacking capacity (surrogate consent at reduced boundary cost).
Abandonment boundary (beta_abandon): Withdrawing care from dependent patients without appropriate transition. One of the highest-cost crossings in clinical practice.
Sacred-value boundary (beta_sacred = infinity): Non-consensual experimentation, execution, deliberate non-palliative hastening of death. No clinical benefit can offset an infinite boundary penalty; these boundaries define the absolute limits of clinical action.
21.3 Clinical Decision as A* Search
A clinical decision is pathfinding on C. The clinician stands at vertex v_0 (the patient's current clinical-moral state) and seeks a path to a goal region G (the set of acceptable clinical-moral outcomes). The evaluation function is f(n) = g(n) + h(n), where g(n) is the accumulated clinical cost from v_0 to n (the evidence-based medicine component — what has been done so far) and h(n) is the moral-heuristic estimate of remaining cost from n to G (the clinical wisdom component — judgment about what remains).
Definition 21.4 (Clinical Geodesic). The clinical geodesic gamma* from clinical state v_0 to goal region G is the minimum-cost path:
gamma* = arg min_gamma Σ w(v_i, v_{i+1})
subject to gamma(0) = v_0 and gamma(end) in G. This is a geodesic in the sense of Chapter 10, computed via A* search as in Chapter 11.
Theorem 21.1 (Clinical Heuristic Admissibility). If boundary penalties βk are calibrated by clinical training as lower bounds on true moral cost (βk <= βk* for all k), then the clinical heuristic h_c(n) is admissible and A* search finds the optimal clinical path.
Proof. This is the Heuristic Truncation Theorem of Chapter 11, applied to the clinical domain. If h_c(n) <= h_c*(n) for all n, A* with h_c is admissible and returns the minimum-cost path. Clinical training calibrates boundary penalties as lower bounds: "do not harm" is taught as a strict underestimate of actual harm costs, ensuring admissibility. []
This yields a hierarchy of clinical heuristic quality:
Strictly admissible: Prohibitions whose true cost always exceeds the heuristic estimate (non-consensual treatment, non-therapeutic experimentation). The heuristic underestimates, so A* explores more paths but never misses the optimal one.
Epsilon-admissible: Normal well-trained clinicians. Heuristics are slightly miscalibrated but within tolerance, producing near-optimal paths (within epsilon of optimal cost).
Inadmissible: Heuristics inflated by moral injury or trauma. A clinician who has been forced to perform futile interventions may develop an inflated beta_futile that causes avoidance of beneficial interventions in borderline cases. The heuristic overestimates, and A* may miss the optimal path.
Gauge-variant: Heuristics sensitive to framing rather than clinical reality. Omission bias (acting feels worse than not acting, even when not acting causes more harm) is a gauge-variant heuristic — it violates the Bond Invariance Principle because the evaluation depends on description (act vs. omit) rather than outcome.
21.4 The QALY Irrecoverability Theorem
Theorem 21.2 (QALY Irrecoverability). Let Q: R^9 -> R be any scalar QALY function mapping the nine-dimensional clinical-moral attribute vector to a scalar score. Then:
Q is not injective: multiple clinically and morally distinct states map to the same QALY.
The information destroyed by Q is irrecoverable: no function psi: R -> R^9 satisfies psi(Q(a)) = a for all a in R^9.
The clinical geodesic on C differs from the QALY-maximizing path, and the difference can be arbitrarily large.
Proof. (1)-(2): Q maps R^9 to R. By the rank-nullity theorem, the kernel of dQ has dimension at least 8 at every regular point. Information in the kernel is destroyed, and by the data processing inequality, it is irrecoverable. (3): Construct two clinical states a and b with Q(a) = Q(b) but ||a - b|| arbitrarily large on dimensions d_2 through d_9. The QALY-maximizing path is indifferent between a and b; the clinical geodesic is not. The difference in manifold cost can be made arbitrarily large by choosing a and b with large non-d_1 differences. []
Corollary 21.1 (QALY Discrimination Is Mathematical). The well-documented bias of QALY-based allocation against elderly, disabled, and minority populations is not an implementation failure. It is mathematical. Populations whose needs are concentrated on non-outcome dimensions — d_4 (autonomy) for disabled persons, d_5 (trust) for racial minorities with historical betrayal, d_7 (dignity) for the elderly — are systematically disadvantaged by a scalar projection that privileges d_1 (clinical outcomes). The discrimination is in the dimensionality, not in the calibration.
Proposition 21.1 (When QALYs Are Adequate). QALY analysis is adequate if and only if three conditions hold simultaneously: (a) the decision activates primarily d_1 with negligible activation of d_2 through d_9; (b) the patient population is homogeneous on d_2 through d_9; and (c) no clinical-moral boundaries are crossed. These conditions hold for routine cost-effectiveness comparison of pharmacologically similar drugs in homogeneous populations. They fail for end-of-life care, mental health, disability, rationing under scarcity, vulnerable populations, and cross-cultural care.
21.5 The Mathematical Theory of Moral Injury
Moral injury is not burnout. Burnout is resource depletion — the exhaustion of g(n) computation capacity. Moral injury is heuristic damage — the corruption of h(n) by forced boundary crossings. A well-rested ICU physician forced to deny ventilators to patients she judges salvageable has high moral injury and low burnout. An overworked rural physician in an ethically supportive environment has low moral injury and high burnout. The geometric framework makes this distinction precise.
Definition 21.5 (Moral Injury). A clinician experiences moral injury when the institutionally mandated clinical geodesic gamma*_inst requires crossing boundary k with penalty βkclin exceeding the clinician's threshold τ.
Definition 21.6 (Moral Injury Increment). The moral injury increment at time t is:
ΔMI(t) = Σk max(0, βkclin - τ) * 𝟙[boundary k crossed at time t]
The cumulative moral injury is MI(T) = Sum_{t=1}^{T} ΔMI(t).
Theorem 21.3 (Moral Injury Accumulation). Moral injury satisfies three structural properties:
Cumulative: MI(T) is monotonically non-decreasing. No event subtracts from cumulative moral injury.
Irreversible above threshold: Once MI(T) > MI_crit, the clinician's heuristic h(n) is permanently altered. Boundary penalties shift (some inflate, others degrade), trust in institutional legitimacy (d_8) erodes, and professional identity (d_7) is damaged. This is the clinical phenomenon of "moral residue."
Structurally distinct from burnout: Burnout is g(n)-depletion (computational exhaustion from accumulated clinical work). Moral injury is h(n)-damage (the heuristic itself is corrupted by forced boundary crossing). The two are empirically dissociable: high MI with low burnout in ethically violating but adequately staffed settings; low MI with high burnout in ethically supportive but understaffed settings.
Proof. (1) follows from the definition: ΔMI(t) >= 0, so MI(T) is non-decreasing. (2) is an empirical claim formalized as follows: h(n) is a function of the clinician's internal boundary-penalty estimates; forced crossing of boundary k with cost exceeding τ permanently alters the estimate (either inflating it via avoidance learning or deflating it via desensitization), changing subsequent path computations. (3) follows from the definitions: burnout affects g(n) capacity, moral injury affects h(n) calibration; these are independent components of f(n) = g(n) + h(n). []
Definition 21.7 (Moral Injury Index). The moral injury index of institutional policy P is:
MI(P) = E[Σt Σk max(0, βkclin - τ) * 𝟙[policy P requires crossing boundary k at time t]]
Proposition 21.2 (MI Is Predictable). MI(P) can be estimated before policy implementation by (a) surveying clinicians for βkclin values, (b) analyzing policy P for boundary-crossing scenarios, and (c) computing expected frequency from historical case-mix data. This makes moral injury a prospectively manageable institutional risk, not merely a retrospectively observed pathology.
21.6 Geometric Informed Consent
Definition 21.8 (Consent as Manifold Calibration). Informed consent is valid when the patient's clinical decision complex C_patient is calibrated: the patient's edge weights approximate the true edge weights within tolerance epsilon:
||w_patient(v_i, v_j) - w_true(v_i, v_j)|| < epsilon
for all relevant edges. The patient need not compute exact weights — only weights close enough that the patient's clinical geodesic matches (or closely approximates) the geodesic on the true manifold.
Theorem 21.4 (Consent as Gauge Invariance). Valid informed consent is equivalent to a gauge-invariance condition: the patient's decision is invariant under meaning-preserving transformations of the clinical information.
Formally: if consent is valid, then for any meaning-preserving reframing τ (e.g., "90% survival rate" vs. "10% mortality rate"):
gamma*_patient(τ(x)) = gamma*_patient(x)
If the patient's decision changes under reframing, then either d_9 (understanding) or d_4 (voluntariness) is inadequately calibrated, and consent is not genuinely informed.
Proof. The Bond Invariance Principle (Chapter 12) requires that ethical evaluations be invariant under meaning-preserving transformations. If the patient's decision varies under reframing, h(n) is gauge-variant — it depends on framing rather than clinical reality. This implies d_9 is not calibrated to the true attribute vectors. []
Corollary 21.2 (The Consent Paradox in Low Health Literacy). Patients with low health literacy can sign consent forms (legally valid) while having d_9 so low that C_patient is grossly miscalibrated. The consent is legally valid but ethically vacuous: the patient's clinical geodesic on their miscalibrated manifold may radically differ from what they would compute with adequate understanding. The gauge-invariance test detects this: if the decision changes under clinically equivalent reframing, consent is not genuinely informed regardless of the signature on the form.
Shared Decision-Making as Manifold Alignment. Valid shared decision-making (SDM) requires three alignments: (a) Epistemic alignment (d_9): the patient understands clinical facts with sufficient accuracy. (b) Value alignment (d_2 through d_8): the clinician understands the patient's boundary penalties βk^{patient} so that the recommended geodesic respects the patient's manifold, not just the clinician's. (c) Goal alignment: patient and clinician agree on the goal region G. Each alignment can fail independently, requiring different interventions: education for epistemic misalignment, values exploration for value misalignment, and negotiation for goal misalignment.
21.7 Geometric Triage
Definition 21.9 (Triage as Constrained Multi-Agent Pathfinding). A triage problem is a tuple (N, C, R, G) where N = {p_1, ..., p_n} is a set of patients, C = {C_1, ..., C_n} their clinical decision complexes, R a resource constraint, and G = {G_1, ..., G_n} their goal regions. The triage decision is a feasible allocation minimizing total clinical-moral friction:
(gamma_1*, ..., gamma_n*) = arg min Sum_{i=1}^{n} CF(gamma_i) subject to Sum_{i=1}^{n} r(gamma_i) <= R
where CF(gamma_i) is the clinical friction (total path cost for patient i) and r(gamma_i) is the resource cost of patient i's path.
Theorem 21.5 (Divergence of Utilitarian and Manifold-Optimal Triage). Let gamma_U* denote the utilitarian triage (maximize lives or life-years saved) and gamma_C* the manifold-optimal triage (minimize total clinical-moral friction). Then gamma_U* = gamma_C* when decisions activate only d_1 and no boundaries are crossed. However, gamma_U* differs from gamma_C* whenever boundaries or non-outcome dimensions are activated:
Ventilator withdrawal: Removing a ventilator from a dying patient to give it to a patient with better prognosis is utilitarian-optimal (saves more life-years) but manifold-suboptimal when the abandonment boundary beta_abandon and trust boundary costs exceed the d_1 gain.
Age-based deprioritization: Deprioritizing elderly patients maximizes expected life-years (utilitarian-optimal) but violates d_3 (justice: equal moral worth regardless of age) — manifold-suboptimal.
Lottery allocation: When clinical prognoses are comparable, lottery allocation is utilitarian-equivalent (no outcome difference) but manifold-superior: it preserves d_3 (fairness), d_8 (institutional legitimacy), and avoids d_7 (dignity) violations inherent in discriminatory selection.
Proof. When only d_1 is active and no boundaries are crossed, CF(γ) reduces to the d_1 component of the Mahalanobis distance, and utilitarian optimization coincides with manifold optimization. When boundaries are active, boundary penalties add positive cost to utilitarian-optimal paths that cross them, causing the manifold-optimal path to diverge. Each of (1)-(3) is constructed by identifying a specific boundary whose penalty exceeds the d_1 advantage of the utilitarian path. []
Why Clinicians Resist Utilitarian Triage. The widespread resistance of clinicians to utilitarian crisis standards of care is not irrationality. It is the clinician's heuristic h(n) correctly identifying that the utilitarian path crosses moral boundaries whose cost exceeds the utilitarian benefit. Clinicians are right on the full manifold; utilitarian algorithms are right on the scalar projection. The conflict is dimensional mismatch, not ethical disagreement.
21.8 The Clinical Bond Index
Definition 21.10 (Clinical Bond Index). The clinical Bond Index BI(P, S) for policy P applied to population S is the expected deviation between the manifold-optimal and policy-mandated clinical paths:
BI(P, S) = E_{s in S}[CF(gamma_P*(s)) - CF(gamma_C*(s))]
where gamma_P*(s) is the path mandated by policy P for patient s, and gamma_C*(s) is the clinical geodesic on the full manifold.
Proposition 21.3 (Bond Index Detects Structural Injustice). If BI(P, S_1) >> BI(P, S_2), then policy P imposes systematically higher clinical-moral cost on population S_1 than on S_2. This makes "policy P is structurally unjust toward population S_1" an empirical claim with a numerical value.
Three applications illustrate the diagnostic power:
Racial disparities: If Black patients have systematically lower trust (d_5) due to historical institutional betrayal, and policy P assumes uniform trust, the clinical geodesic for Black patients is more costly than the policy recognizes. BI(P, S_Black) > BI(P, S_White), and the difference quantifies the structural injustice.
Disability discrimination: If QALY-based allocation underweights autonomy (d_4) and dignity (d_7) relative to function (d_1), disabled patients bear systematically higher manifold cost than the QALY score reflects.
Age discrimination: If triage protocols underweight equal moral worth (d_3) and dignity in aging (d_7), elderly patients bear higher cost, with the Bond Index quantifying the magnitude.
BI(P, S) is computable from measurable quantities: edge weights from clinical decision data, boundary penalties from clinician surveys, and population covariance matrices from behavioral data. Health equity policy becomes evidence-based in a precise mathematical sense.
21.9 Worked Examples
Example 21.1 (Jehovah's Witness with Gastrointestinal Bleed). A 45-year-old competent Jehovah's Witness presents with acute GI bleed. Without blood transfusion, mortality risk is approximately 40%. The patient refuses transfusion on religious grounds.
Principlism: Autonomy versus beneficence — no algorithm for resolution.
QALY analysis: Transfusion gains approximately 20 QALYs. QALY-optimal: transfuse.
Clinical geodesic analysis: Path A (transfuse against refusal): Da_1 = +large (survival), but Da_4 = -infinity (autonomy violation), Da_2 = -infinity (rights violation), Da_5 = -large (trust destroyed), Da_7 = -large (identity violated). Edge weight: w_A = infinity (consent boundary beta_consent = infinity for competent patient). Path B (respect refusal, optimize alternatives): Da_1 = -moderate (mortality risk), but Da_4 = 0, Da_2 = 0, Da_5 = +moderate (trust reinforced), Da_7 = 0. Edge weight: w_B = (mortality risk)^2 / sigma_1^2.
The clinical geodesic selects Path B. This is not a "concession to autonomy" but manifold-optimal: Path A has infinite cost because beta_consent = infinity. The geodesic further computes: aggressive exploration of transfusion alternatives (erythropoietin, cell salvage, volume expanders, interventional radiology) to reduce Da_1 while maintaining zero cost on d_2 through d_9.
Example 21.2 (Withdrawing Life Support). An 82-year-old patient has been on mechanical ventilation for three weeks with no neurological recovery, no advance directive, and a divided family. Utilitarian analysis: withdraw — negligible survival benefit. Clinical geodesic analysis: Da_1 is approximately 0 (negligible survival benefit), but d_6 (family impact), d_5 (trust), and d_7 (dignity) are heavily activated.
The clinical geodesic is not an immediate binary decision but a path through time: family meeting -> values conversation -> time-limited trial -> transition to comfort care. This path has lower total manifold cost than the immediate withdrawal path, even though the clinical endpoint is the same. The geodesic optimizes process, not just outcome — a distinction invisible to scalar analysis.
21.10 Falsifiable Predictions
The framework generates six predictions that distinguish it from existing clinical ethics approaches:
Prediction 1 (Dimensional Moral Distress): Clinician moral distress (measured by MDS-R) should correlate with the number of manifold dimensions activated in the clinical scenario, not with severity on any single dimension. Test: clinical vignettes matched for clinical severity (d_1) but varying in dimension count. Prediction: distress increases monotonically with dimension count. Falsified if: distress depends only on d_1 severity.
Prediction 2 (Moral Injury vs. Burnout Dissociation): Moral injury (MISS-HP) and burnout (MBI) should be statistically dissociable. Test: identify clinicians with frequent boundary-crossing but adequate staffing (predicted high MI, low burnout) and clinicians with high volume but no boundary-crossing (predicted low MI, high burnout). Falsified if: MI and burnout are empirically indistinguishable.
Prediction 3 (Consent Gauge-Invariance Test): Genuinely informed patients should show framing invariance — treatment preferences should not change under clinically equivalent reframing ("90% survival" vs. "10% mortality"). Prediction: framing-variant patients have lower health literacy scores. Falsified if: framing variance is independent of health literacy.
Prediction 4 (Bond Index Predicts Disparities): BI(P, S) computed from policy analysis should predict patient-reported experience disparities (HCAHPS scores) across race, age, and disability status. Falsified if: BI is uncorrelated with experience disparities.
Prediction 5 (Triage Resistance Correlates): Clinician resistance to utilitarian triage protocols should correlate with pre-crisis boundary-penalty profiles (βkclin values). Falsified if: resistance is uncorrelated with boundary penalties.
Prediction 6 (Process-Path Superiority): End-of-life satisfaction (family and clinician reported) should be higher when the clinical geodesic includes process steps (family meetings, values conversations, time-limited trials) than when it proceeds directly to the same endpoint. Falsified if: satisfaction is identical regardless of path when endpoint is the same.
21.11 Connection to the Framework
The Geometric Clinical Ethics program extends the parent framework in four directions:
1. Chapter 11 established A* pathfinding as the formal model of moral reasoning. This chapter shows that clinical decision-making is a domain-specific instance: f(n) = g(n) + h(n), where g(n) is evidence-based medicine and h(n) is clinical moral wisdom.
2. Chapter 12 established the Bond Invariance Principle. This chapter shows that informed consent is a gauge-invariance condition: valid consent requires that the patient's decision be invariant under meaning-preserving reframing.
3. Chapter 15 established the Scalar Irrecoverability Theorem. This chapter applies it to QALYs: the QALY is a rank-0 contraction of the clinical-moral tensor, and the eight dimensions it destroys include precisely the dimensions on which vulnerable populations' needs concentrate.
4. Chapter 12 established the Conservation of Harm. This chapter extends it to moral injury: cumulative boundary-crossing damage is a conserved quantity in the sense that it can be created but never destroyed, accumulating monotonically with forced ethical violations.
21.12 Summary
This chapter has shown that the geometric ethics framework, when applied to clinical medicine, yields:
1. A formal construction of the clinical decision complex C as a domain-specific instantiation of the moral manifold.
2. Clinical heuristic admissibility: clinical training as heuristic calibration, with a hierarchy from strictly admissible to gauge-variant.
3. The QALY Irrecoverability Theorem: scalar clinical measures destroy eight dimensions of information, with the destruction systematically disadvantaging vulnerable populations.
4. A mathematical theory of moral injury: cumulative, irreversible, and formally distinct from burnout.
5. Informed consent as gauge invariance: a testable condition that goes beyond the legal checklist.
6. Geometric triage: manifold-optimal allocation diverges from utilitarian allocation whenever boundaries or non-outcome dimensions are activated.
7. The Clinical Bond Index: a quantitative measure of structural injustice that makes health equity policy evidence-based.