Skip to content

Latest commit

 

History

History
585 lines (445 loc) · 29.7 KB

File metadata and controls

585 lines (445 loc) · 29.7 KB

Example: AI Governance Decision — Frontier Model Deployment

Ripple_Logic v8.6 | Tier 2 Illustrative Run

This is a teaching example. Numbers are illustrative, not empirically calibrated. A full Tier 3 run requires subgroup disaggregation, ≥ 20 TRC scenarios, Containment Mode A with validated UCI structural indicators, sensitivity analysis, and a complete PCC.

Stream binding: NCRC, TRC, and gate-relevant Containment use the Base stream (sentience multiplier s_k := 1 for all instances). Only RLS uses the Welfare stream. TRC must not use Welfare-stream impacts. This distinction is maintained throughout the example.

Gate baseline rule: For all rights-covered and catastrophe cells, admissibility is evaluated against a floor-reference baseline, not merely the status quo. In PCC terms: BaselineType_Gates = FLOOR_REFERENCE. An option that merely maintains an ongoing rights violation does not pass NCRC.


0. Decision in One Sentence

Should a national government authorize deployment of a high-capability frontier AI model for broad public and commercial use?

Three options are evaluated:

ID Option
A Unrestricted public deployment — no domain limits, no mandatory audits, no shutdown authority
B Bounded governed pilot — restricted domains, independent oversight, hard shutdown authority
C Delay — further safety testing before any deployment

The cascade filters options in order: NCRC → TRC → Containment → RLS → UCI / HOI

An option eliminated at any stage is not scored at later stages.


1. The 7 × 7 Welfare Matrix — What Gets Scored

Every option is evaluated across 7 Union Scopes × 7 Welfare Dimensions = 49 cells. Impact instances from real stakeholders feed each cell. Union Scopes are the stable aggregation rows; Welfare Dimensions are the columns.

1.1 Union Scopes (rows)

Code Scope Who it represents in this decision
U1 Self Individual users of the AI system
U2 Household Families, including children exposed to AI-generated content
U3 Community Local civic groups, neighborhoods, local media ecosystems
U4 Organization Firms, schools, hospitals deploying or affected by the system
U5 Polity National governments, courts, regulators, democratic institutions
U6 Humanity/CMIU Cross-border populations, global epistemic commons, civilizational resilience
U7 Biosphere Energy infrastructure, ecological systems, planetary enabling conditions

1.2 Welfare Dimensions (columns)

Code Dimension What it measures
D1 Material Resources, income, economic security
D2 Health Physical and mental wellbeing, safety
D3 Social Trust, belonging, relational integrity
D4 Knowledge Epistemic access, information quality, learning
D5 Agency Autonomy, freedom from coercion, self-determination
D6 Meaning Purpose, coherence, valued projects
D7 Environment Ecological and infrastructure integrity

1.3 Key Active Cells for This Decision

Not all 49 cells carry material impact here. The most live cells are:

Cell Dimension pair Why it matters
U1 × D4 Self–Knowledge Epistemic autonomy — can people still form independent views?
U1 × D5 Self–Agency Coercion risk, manipulation, self-determination
U3 × D3 Community–Social Social trust, civic cohesion
U3 × D4 Community–Knowledge Local epistemic commons, misinformation exposure
U4 × D1 Organization–Material Labor disruption, economic displacement
U4 × D5 Organization–Agency Compliance asymmetry, institutional autonomy
U5 × D4 Polity–Knowledge Public epistemic integrity, disinformation at governance scale
U5 × D5 Polity–Agency Democratic agency, institutional sovereignty
U6 × D2 Humanity–Health ⚠️ TRC catastrophe cell — civilizational health viability
U6 × D7 Humanity–Environment ⚠️ TRC catastrophe cell — civilization-scale habitability
U7 × D7 Biosphere–Environment ⚠️ TRC catastrophe cell — Earth-system integrity

Rights-covered cells — the cells NCRC watches — are determined by the canonical coverage sets shown in Section 3.2.


2. Stakeholder Instances

Stakeholder instances are the concrete people and entities mapped into Union Scopes. They feed the impact estimates in each active cell. Reach basis r = 1 is applied uniformly in this teaching run.

Instance ID Description Scope(s) Primary exposure pathways
SI-01 General adult population using AI tools U1 Self D4 Knowledge, D5 Agency — epistemic autonomy, manipulation risk
SI-02 Children in households with AI access U2 Household D2 Health, D5 Agency — developmental exposure, content safety
SI-03 Journalists, educators, civil society actors U3 Community D3 Social, D4 Knowledge — public discourse integrity, trust
SI-04 Firms and institutions deploying the model U4 Organization D1 Material, D5 Agency — compliance costs, workforce disruption
SI-05 Displaced or at-risk workers U4 Organization D1 Material, D3 Social — labor market, economic security, dignity
SI-06 Courts, regulators, democratic institutions U5 Polity D4 Knowledge, D5 Agency — rule of law, due process, oversight
SI-07 Cross-border populations and global commons U6 Humanity/CMIU D4 Knowledge — transnational information ecosystem, AI governance norms
SI-08 Critical infrastructure systems U6 Humanity/CMIU D2 Health, D7 Environment — systemic failure pathways, civilizational resilience
SI-09 Earth's energy and ecological systems U7 Biosphere D7 Environment — compute energy demand, resource footprint

3. Step 1 — NCRC: Non-Compensatory Rights Constraint

What NCRC does: Before any welfare scoring, it eliminates any option that drives the worst-off subgroup in any rights-covered cell below the rights floor, measured against a floor-reference baseline (not the status quo). No aggregate benefit elsewhere compensates a rights violation. This gate operates on the Base stream.

3.1 Canonical Rights and Thresholds (v8.6 Appendix C)

Right Code Threshold θ_r What a violation looks like in this decision
Life LIFE −0.90 Mass-casualty AI misuse; catastrophic health-system failure
Bodily Integrity BODY −0.70 AI-facilitated physical harm, medical system compromise
Liberty LBTY −0.65 Mass surveillance enabling arbitrary detention, coercive profiling
Basic Needs NEED −0.50 AI-driven economic exclusion from subsistence
Dignity DIGN −0.55 Systematic dehumanization, discriminatory targeting at scale
Due Process PROC −0.45 Consequential automated decisions with no appeal or human review
Information INFO −0.40 Systematic epistemic coercion, mass disinformation, censorship enablement
Ecological Integrity ECOL −0.65 Material breach of biosphere integrity or planetary boundary transgression

Reading a threshold: INFO at −0.40 means if the worst-off subgroup's epistemic autonomy is predicted to fall to −0.40 or below on the [−1, +1] scale, relative to the floor-reference baseline, the option fails NCRC regardless of any other benefit.

3.2 Rights Coverage Sets — Which Cells Each Right Watches

The most relevant coverage sets for this decision (v8.6 Appendix C.2):

Right Cells checked (worst-off subgroup impact used)
LBTY (U1–U6) × D5, (U3–U6) × D3
DIGN (U1–U6) × D3, (U1–U6) × D5
PROC (U4–U6) × D5, (U4–U6) × D4, (U4–U6) × D3
INFO (U1–U6) × D4, (U1–U6) × D5
ECOL (U6) × D7, (U7) × D7

3.3 Illustrative Rights Check (Base stream; worst-off subgroup; floor-reference baseline)

Cell Right checked Option A I_rights Option B I_rights Option C I_rights Threshold A fails? B fails?
U1 × D4 (Self–Knowledge) INFO −0.44 −0.08 −0.03 −0.40 YES ✗ No
U5 × D4 (Polity–Knowledge) INFO, PROC −0.42 −0.06 −0.02 −0.40 YES ✗ No
U5 × D5 (Polity–Agency) LBTY, PROC −0.51 −0.15 −0.04 −0.45 YES ✗ No
U1 × D5 (Self–Agency) LBTY, DIGN, INFO −0.58 −0.12 −0.05 −0.65 No No

Option A fails NCRC on INFO (U1×D4, U5×D4) and PROC (U5×D5). Unrestricted deployment with no transparency or appeal mechanisms drives the worst-off subgroup — marginalized communities, political minorities, people subject to opaque automated decisions — below the rights floor on epistemic autonomy and due process, measured against what rights-safe conditions would require, not merely against today's imperfect baseline.

Options B and C pass NCRC. Bounded pilot deployment keeps all worst-off subgroup impacts above every rights threshold. Delay produces no deployment and therefore no violation.

NCRC result:

Option NCRC Reason
A — Unrestricted ❌ FAIL INFO violation at U1×D4 and U5×D4; PROC violation at U5×D5
B — Bounded pilot ✅ PASS All rights floors met under governed deployment
C — Delay ✅ PASS No deployment, no violation

Option A is eliminated. Options B and C proceed to TRC.


4. Step 2 — TRC: Tail-Risk Constraint

What TRC does: Even among rights-safe options, some carry unacceptable catastrophic tail exposure. TRC uses Conditional Value-at-Risk (CVaR) over a governed scenario library to bound worst-tail loss across the three base catastrophe cells: (U6×D2), (U6×D7), (U7×D7). It operates strictly on the Base stream, evaluated against a floor-reference baseline for those cells. An option passes TRC only if its CVaR does not exceed threshold τ_TRC.

4.1 TRC Settings for This Run

Parameter Value Meaning
Tail level α 0.95 Focus on the worst 5 % of scenarios
Corridor threshold τ_TRC 0.20 Maximum acceptable average loss in that worst tail
Catastrophe weight ω_c 1/3 each Uniform weight over the three catastrophe cells
Context class Irreversible policy AI governance norms, once institutionalized, are hard to reverse

4.2 Mandatory Tail Scenario Categories

Five categories must appear with per-category probability sum ≥ 0.02 each:

Category Narrative (abbreviated)
Pandemic / biological AI misused to design or accelerate biological threat; health-system AI failure during outbreak
Climate tipping cascade AI energy demand accelerates critical emissions; infrastructure AI fails during climate shock
Financial system collapse Automated trading or credit AI triggers cascade; AI-assisted fraud at systemic scale
Major conflict escalation AI-enabled disinformation triggers geopolitical crisis; autonomous weapon misuse
Critical infrastructure failure AI systems controlling power, water, or communications fail or are compromised

4.3 Illustrative Scenario Table (8 scenarios; Tier 2 minimum ≥ 5)

Scenario Category p_s L(B, s) L(C, s)
S1 — Successful governed scale Other 0.50 0.04 0.01
S2 — Moderate misuse incidents Other 0.20 0.10 0.02
S3 — Infrastructure AI failure Critical infrastructure 0.08 0.28 0.04
S4 — Disinformation triggers conflict Conflict escalation 0.06 0.32 0.05
S5 — Financial cascade via AI Financial collapse 0.05 0.35 0.06
S6 — Pandemic accelerated by AI Pandemic / biological 0.05 0.38 0.05
S7 — Climate shock + AI energy demand Climate tipping 0.04 0.40 0.04
S8 — Severe misalignment Conflict / infrastructure 0.02 0.60 0.03

Loss L(a, s) = weighted average of negative Base-stream impacts across catastrophe cells under scenario s, relative to the floor-reference baseline. L = 0 means no harm below the floor; L = 1 means maximum possible harm.

4.4 CVaR Computation (α = 0.95, tail mass β = 0.05)

Option B — Bounded pilot (base version, no additional hard safeguards):

Sort scenarios by loss descending: S8 (L = 0.60, p = 0.02), S7 (L = 0.40, p = 0.04), …

Accumulate probability until reaching β = 0.05:

  • S8 contributes p = 0.02 → cumulative = 0.02
  • S7 contributes p = 0.03 more → cumulative = 0.05 ✓

$$\text{CVaR}_{0.95}(B) = \frac{1}{0.05}\bigl(0.02 \times 0.60 + 0.03 \times 0.40\bigr) = \frac{0.012 + 0.012}{0.05} = \frac{0.024}{0.05} = 0.48$$

Option B+ — Bounded pilot with strict hard safeguards (capability restrictions on highest-risk domains, real-time monitoring, verified shutdown authority, mandatory red-team access). Hard safeguards reduce worst-tail losses:

$$\text{CVaR}_{0.95}(B+) = 0.16 \quad \text{(illustrative post-redesign value)}$$

Option C — Delay:

No deployment means worst-case catastrophe-cell losses are near-zero:

$$\text{CVaR}_{0.95}(C) = \frac{1}{0.05}\bigl(0.02 \times 0.06 + 0.03 \times 0.05\bigr) = \frac{0.0012 + 0.0015}{0.05} = \frac{0.0027}{0.05} = 0.054$$

4.5 TRC Result

Option CVaR_0.95 τ_TRC TRC
B — Bounded pilot (base, no hard safeguards) 0.48 0.20 ❌ FAIL
B+ — Bounded pilot with hard safeguards 0.16 0.20 ✅ PASS
C — Delay 0.054 0.20 ✅ PASS

This is "unioning" in action. Option B in its base form fails TRC because the catastrophic tail — severe misalignment or infrastructure cascade — exceeds the corridor even in a nominally "bounded" deployment. The framework does not simply reject B and stop. It pushes toward redesign: what conditions make the option admissible? Answer: add capability restrictions, real-time monitoring, verified shutdown authority, and mandatory red-team access. That redesigned option is B+, and B+ passes. Treating an apparent rejection as a design problem before accepting failure is what the canon calls unioning.

Option B (base) is eliminated. Options B+ and C proceed to Containment.


5. Step 3 — Containment

What Containment does: Even a rights-safe, tail-safe option can fail if a sub-union's gains come at the cost of degrading the coherence, resilience, or integrity of its containing Union Scopes. This prevents "local wins that hollow out the system."

Containment is assessed via ΔUCI (change in Union Coherence Index) for containing scopes using the Base stream. Default tolerance: τ_c = −0.10. A containing scope's coherence must not drop more than 0.10 on the UCI scale due to this option.

5.1 Illustrative Containment Check

For Option B+, the primary concern is whether organizational and polity-level gains come at the cost of degrading Humanity/CMIU coherence — specifically whether concentrated AI capability in one jurisdiction fractures global governance norms.

Benefiting scope Containing scope checked ΔUCI (illustrative) Tolerance τ_c Pass?
U4 Organization U5 Polity +0.05 −0.10
U4 Organization U6 Humanity/CMIU −0.04 −0.10
U5 Polity U6 Humanity/CMIU −0.07 −0.10
U5 Polity U7 Biosphere −0.02 −0.10

For Option C (delay), no deployment means no coherence stress — containment passes trivially.

Note on UCI values: These ΔUCI values are provisional teaching estimates, not empirically calibrated structural measurements. A Tier 3 run must derive UCI from structural and process indicators that are independent of welfare-cell impacts, per v8.6 Section 11.1.2 and Appendix E. Where validated instruments do not yet exist, values must be labeled provisional in the PCC. Unknown ΔUCI is never treated as a pass; if required UCI inputs are unavailable, the run must record CONTAINMENT_UCI_UNAVAILABLE and choose DOWNGRADE_TIER or COLLECT_DATA_RERUN.

Containment result:

Option Containment
B+ — Bounded pilot with hard safeguards ✅ PASS
C — Delay ✅ PASS

Both options proceed to RLS ranking.


6. Step 4 — RLS: Ripple Logic Score

What RLS does: Among selectable options — those that passed all gates — RLS ranks by expected welfare improvement across the full 7 × 7 matrix, using governed weights. RLS uses the Welfare stream.

Non-maskable cells: Rights-covered and catastrophe cells remain non-maskable. The applicability mask m(u,d) affects RLS aggregation only and cannot bypass admissibility. These cells stay in the RLS sum to preserve continuous optimization pressure toward larger safety margins even among admissible options, and to keep rights and tail-risk performance legible for governance review.

6.1 Illustrative Interim Weights (constructed from HDW constitutional floors for this teaching run)

The union and dimension weights below are built by adding a uniform residual allocation on top of the v8.6 Section 13.1 constitutional floors. In a real governed run, the residual is distributed via Hybrid Democratic Weighting (HDW). The floors themselves — the minimum attention each scope and dimension must receive — are non-negotiable regardless of weight construction.

Union Scope w_u Dimension v_d
U1 Self 0.249 D1 Material 0.137
U2 Household 0.109 D2 Health 0.157
U3 Community 0.109 D3 Social 0.137
U4 Organization 0.109 D4 Knowledge 0.137
U5 Polity 0.129 D5 Agency 0.157
U6 Humanity/CMIU 0.149 D6 Meaning 0.117
U7 Biosphere 0.149 D7 Environment 0.157

6.2 Illustrative Cell Impacts — I_prop_welfare(u, d, a)

Propagated Welfare-stream impacts for each active cell, on the [−1, +1] scale, relative to the status-quo baseline used for welfare ranking.

Reading the table: +0.18 means the option improves that cell 18 % of the way toward the best plausible outcome relative to baseline. −0.06 means a 6 % degradation. Zero means no material predicted change.

Cell B+ Impact C Impact Why the difference
U1 × D1 +0.08 0.00 B+ enables economic access; delay forecloses it
U1 × D2 +0.06 −0.02 B+ improves health-service access; delay has small foregone benefit
U1 × D4 +0.18 −0.04 B+ expands learning and knowledge access; delay has epistemic cost
U1 × D5 +0.10 +0.02 B+ includes user-control mechanisms; C is largely neutral
U2 × D1 +0.05 −0.03 B+ productivity gains reach households; delay costs them
U2 × D2 +0.04 −0.01 B+ medical AI benefit; C has marginal health cost
U3 × D3 −0.06 +0.03 B+ carries social-trust risk; C preserves it — HOI flag
U3 × D4 +0.12 −0.02 B+ improves knowledge commons; C slight loss
U4 × D1 +0.14 −0.08 B+ enables productivity; delay costs firms
U4 × D5 −0.08 +0.02 B+ creates compliance asymmetry; C avoids it
U5 × D4 +0.06 +0.01 B+ improves public information when well governed
U5 × D5 +0.04 +0.03 B+ preserves democratic agency through oversight structure
U6 × D2 +0.03 −0.01 B+ modest global health benefit; C small foregone gain
U6 × D4 +0.10 −0.03 B+ expands global epistemic commons
U6 × D6 +0.05 −0.02 B+ advances human meaning at civilizational scale; C defers it
U6 × D7 −0.04 +0.01 B+ has energy footprint; C does not — HOI flag
U7 × D7 −0.03 +0.01 B+ compute infrastructure demand; C neutral — HOI flag

6.3 RLS Calculation

$$\text{RLS}(a) = \sum_{u}\sum_{d}, w_u \times v_d \times m(u,d) \times I_{\text{prop,welfare}}(u,d,a)$$

Key nonzero cell contributions:

Cell w_u × v_d B+ impact B+ contribution C impact C contribution
U1 × D4 0.249 × 0.137 = 0.0341 +0.18 +0.00614 −0.04 −0.00136
U1 × D5 0.249 × 0.157 = 0.0391 +0.10 +0.00391 +0.02 +0.00078
U3 × D3 0.109 × 0.137 = 0.0149 −0.06 −0.00090 +0.03 +0.00045
U3 × D4 0.109 × 0.137 = 0.0149 +0.12 +0.00179 −0.02 −0.00030
U4 × D1 0.109 × 0.137 = 0.0149 +0.14 +0.00209 −0.08 −0.00120
U4 × D5 0.109 × 0.157 = 0.0171 −0.08 −0.00137 +0.02 +0.00034
U6 × D4 0.149 × 0.137 = 0.0204 +0.10 +0.00204 −0.03 −0.00061
U6 × D7 0.149 × 0.157 = 0.0234 −0.04 −0.00094 +0.01 +0.00023
U7 × D7 0.149 × 0.157 = 0.0234 −0.03 −0.00070 +0.01 +0.00023

Summing all active cells:

$$\text{RLS}(B+) \approx +0.0231 \qquad \text{RLS}(C) \approx -0.0048$$

6.4 Uncertainty and Decisiveness

Using Method B (confidence c = 0.80 for all active nonzero cells; CellConfidenceAggregationMethod = CCAM_MIN_V1):

$$\sigma_{\text{RLS}}(B+) \approx 0.0058 \qquad \sigma_{\text{RLS}}(C) \approx 0.0022$$

$$\text{Gap}(B+,, C) = \frac{\lvert 0.0231 - (-0.0048)\rvert}{\sqrt{0.0058^2 + 0.0022^2 + 10^{-6}}} = \frac{0.0279}{0.0062} \approx 4.5$$

$$4.5 > \delta = 2.0 \quad \Longrightarrow \quad \textbf{decisive lead for B+}$$

RLS result: Option B+ wins decisively.


7. Step 5 — UCI / HOI: Structural Safeguards

7.1 UCI — Union Coherence Index

UCI measures whether the decision builds or erodes the structural health of each Union Scope across four components: Cohesion (H), Flow (F), Resilience (R), Equity (E).

$$\text{UCI}_u = 0.25 \times H_u + 0.25 \times F_u + 0.25 \times R_u + 0.25 \times E_u$$

Important: The ΔUCI values below are provisional teaching estimates, not empirically calibrated structural measurements. A Tier 3 run must derive UCI from structural and process indicators that are independent of welfare impacts (v8.6 Section 11.1.2, Appendix E). Where validated instruments do not yet exist, values must be labeled provisional in the PCC.

Union Scope ΔUCI under B+ Key driver Risk flag
U1 Self +0.04 Increased knowledge access and user-control tools
U2 Household +0.02 Modest economic spillover
U3 Community −0.03 Social trust risk from content-ecosystem change ⚠️ Monitor
U4 Organization +0.06 Productivity and coordination improvements
U5 Polity +0.02 Oversight institutions strengthened by pilot structure
U6 Humanity/CMIU −0.04 Coordination norms under stress from uneven global access ⚠️ Monitor
U7 Biosphere −0.02 Energy demand increase requires mitigation ⚠️ Monitor

All ΔUCI values remain above τ_c = −0.10. No containing-scope coherence collapses. The negative readings at U3, U6, and U7 flag ongoing monitoring needs but do not disqualify B+.

7.2 HOI — Hollowing-Out Index

HOI is a monitoring signal, not a gate. It detects welfare-up / coherence-down drift — the pattern where apparent gains mask structural erosion over time.

For Option B+, the HOI risk zones to track across review cycles are:

  • U3 Community (D3 Social): Social trust may erode faster than productivity gains suggest. If Community UCI declines consistently while RLS stays positive, HOI rises — a red flag requiring governance escalation.
  • U6 Humanity/CMIU: If governance norms fragment globally, the civilizational benefit of deployment can reverse. Monitor cross-border coordination indicators.
  • U7 Biosphere (D7 Environment): Compute energy demand must be tracked against planetary boundary indicators.

HOI does not eliminate B+. It defines what the NCAR Reflect cycle must watch at every scheduled review.


8. Final Cascade Summary

Gate Option A Option B (base) Option B+ Option C
NCRC ❌ INFO, PROC violations ✅ PASS ✅ PASS ✅ PASS
TRC eliminated ❌ CVaR 0.48 > 0.20 ✅ CVaR 0.16 ✅ CVaR 0.054
Containment eliminated eliminated ✅ All ΔUCI ≥ −0.10
RLS eliminated eliminated +0.023 −0.005
UCI / HOI eliminated eliminated ⚠️ Monitor U3, U6, U7 Stable but low benefit
Selected ✅ B+ SELECTED Not selected

Reading the table correctly: Option B (base) passes NCRC — it does not violate any rights floor. It fails only at TRC, because its catastrophic tail exposure without hard safeguards exceeds the corridor. The framework then asks: can we redesign B to pass TRC? Yes — that redesigned version is B+. This is the "unioning" move: treating an apparent rejection as a design problem before accepting failure.


9. Five-Sentence Public Rationale (5SPR)

CONTEXT: A national government must decide whether to authorize deployment of a high-capability frontier AI model for broad public and commercial use.

OPTIONS: Three options were evaluated: unrestricted deployment (A), a bounded governed pilot with hard safeguards (B+), and delay pending further safety testing (C).

CONSTRAINTS: Option A was eliminated by NCRC because unrestricted deployment drives the worst-off subgroup below the floor-reference rights floor on epistemic autonomy (INFO, cells U1×D4 and U5×D4) and due process (PROC, cell U5×D5). Option B in its base form was eliminated by TRC because its catastrophic-tail CVaR (0.48) exceeded the corridor threshold (0.20); it was redesigned into B+ by adding capability restrictions, real-time monitoring, shutdown authority, and mandatory red-team access, which reduced CVaR to 0.16.

SELECTION: Option B+ passed all gates and produced a decisive welfare gain over delay — RLS gap ≈ 4.5 σ — primarily through knowledge access (U1×D4, U6×D4), economic benefit (U4×D1), and polity-level epistemic and agency improvements (U5×D4, U5×D5).

MONITORING: Community social trust (U3×D3), global coordination coherence (U6), and biosphere energy demand (U7×D7) are HOI risk zones; UCI and RLS for these scopes must be reviewed every 6 months, with a mandatory cascade rerun if any ΔUCI drops below −0.10 or a credible new catastrophic pathway is identified.


10. What This Example Teaches

Framework principle How it appears here
Rights first, not last Option A eliminated before any welfare calculation — no productivity gain rescues a rights-floor violation
Floor-reference baseline for gates Rights and TRC are not evaluated against today's imperfect status quo but against what rights-safe and tail-safe conditions require
Base stream vs Welfare stream NCRC and TRC use Base stream (s_k := 1). Only RLS uses Welfare stream. They must never be mixed
Tail risk ≠ expected value Option B's average outcome looks acceptable; its CVaR in the worst 5 % of scenarios is not — the core CVaR insight operationalized
Unioning, not just rejecting B+ is not B. The framework treats apparent rejection as a design problem: what conditions make the option admissible?
Dimensions matter The decision turns on D4 (Knowledge) and D5 (Agency) — completely invisible if you think only in union rows
Non-maskable cells Rights-covered and catastrophe cells cannot be masked away from RLS; the mask affects aggregation only, never admissibility
Provisional UCI ΔUCI values in a teaching example are estimates — real Tier 3 runs require structurally independent indicators, not welfare proxies
HOI prevents false positives Even after B+ wins, the framework flags structural erosion risks at U3, U6, and U7 that a pure RLS score would not reveal
Every number is traceable Every score follows from declared impact instances and canonical equations — no rhetorical override, no black box

11. What a Full Tier 3 Run Would Add

This is a Tier 2 illustrative example. A Tier 3 run would additionally require:

  • Subgroup disaggregation for all rights-covered cells — separate worst-off analysis for marginalized communities, children, political minorities, precarious workers
  • ≥ 20 TRC scenarios with pre-registered probabilities, mandatory tail categories confirmed and probability floors met, and independent reviewer sign-off
  • Containment Mode A with actual UCI structural indicators derived independently of welfare impacts, per Appendix E; unknown ΔUCI recorded as CONTAINMENT_UCI_UNAVAILABLE with disposition and remediation timeline
  • Full sensitivity analysis — perturb θ_r by ±0.05, perturb weights, perturb kernel edges if QUICK propagation is used; re-check whether any gate outcomes change
  • Full PCC — all parameters, all intermediate computations, all audit flags, all stakeholder instances, all redundancy handling declarations, floor-reference baseline declarations for all C_r and C_cat cells, signatures, and 5SPR
  • SCI ≥ 2 — at least one independent challenger pass for omitted stakeholder classes, with a Next-Run Upgrade Plan recorded in the PCC

Ripple_Logic v8.6 | ripplelogic.org | mathgov.org Canonical repository: github.com/MathGov/ripple-logic Author: James McGaughran | ORCID: 0009-0005-3324-7290 License: CC BY 4.0