The Oversight Dilemma: Who Watches the Watchmen?

The Oversight Dilemma: Who Watches the Watchmen?

Framework Interpretation

This case study analyzes oversight mechanisms through the lens of our delegation risk framework. The examples are based on real governmental structures, but the mathematical framing is our own interpretation.

The U.S. President commands nuclear weapons, controls the world’s largest military, directs $6 trillion in annual spending, and can issue executive orders affecting 330 million people. This concentration of power creates enormous Delegation Risk.

The obvious solution: add oversight. But here’s the problem—any overseer powerful enough to check the President becomes a massive trust liability themselves.

This document works through the mathematics of oversight design, exploring increasingly sophisticated solutions to this fundamental dilemma.

The Core Insight

Oversight doesn’t eliminate trust exposure—it redistributes it. The art is in finding architectures where the redistribution reduces total system risk.

---

Part 1: The Baseline — Presidential Power Unmonitored

Estimating Presidential Delegation Risk

Let’s quantify the trust exposure of an unchecked President:

Authority Domain	P(catastrophic misuse/term)	Damage Potential	4-Year Delegation Risk
Nuclear launch authority	0.0001	$50 trillion (civilization)	$5B
Military deployment	0.01	$500B (major war)	$5B
Executive orders (domestic)	0.05	$100B (economic/rights)	$5B
Appointments (judiciary, agencies)	0.10	$200B (institutional damage)	$20B
Foreign policy/treaties	0.03	$300B (alliance collapse)	$9B
Pardon power	0.05	$10B (justice system)	$500M
Emergency powers	0.02	$1T (constitutional crisis)	$20B
Total Presidential Delegation Risk			~$65B/term

Calibration Note

These numbers are illustrative. The key point is that presidential Delegation Risk is enormous—likely among the highest for any individual on Earth. Even small probability catastrophic outcomes dominate the calculation.

The Trust Topology Without Oversight

flowchart TB
    V[Voters<br/>Principal] -->|"trust: 0.52<br/>(election margin)"| P[President<br/>Agent]
    P -->|"unchecked authority"| A1[Nuclear]
    P -->|"unchecked authority"| A2[Military]
    P -->|"unchecked authority"| A3[Executive Orders]
    P -->|"unchecked authority"| A4[Appointments]

Current state:

• Trust granted: 0.52 (bare electoral majority)
• Delegation Risk: $65B/term
• Verification: Elections every 4 years (extremely slow feedback)
• Revocation mechanism: Impeachment (requires 67 senators—very difficult)

Problem: The trust architecture grants enormous authority with minimal real-time verification and near-impossible revocation.

---

Part 2: The Naive Solution — Add a Powerful Overseer

Option A: Single Presidential Ombudsman

Create a powerful overseer with authority to veto or expose presidential actions.

flowchart TB
    V[Voters] -->|"0.52"| P[President]
    V -->|"0.60"| O[Ombudsman<br/>Independently elected]
    P -->|"actions"| GOV[Government Actions]
    O -->|"can veto/expose"| GOV
    O -.->|"monitors"| P

Ombudsman Powers:

• Access to all presidential communications and decisions
• Authority to delay or publicize any action
• Cannot be fired by President
• 6-year term, independently elected

Calculating the New Delegation Risk

Presidential Delegation Risk with Ombudsman:

The Ombudsman reduces the probability of presidential misuse by catching violations:

P(misuse succeeds) = P(President misuses) × P(Ombudsman fails to catch)

If Ombudsman catches 80% of violations:

Domain	Original P(misuse)	With Ombudsman	New Delegation Risk
Nuclear launch	0.0001	0.0001 × 0.20 = 0.00002	$1B
Military deployment	0.01	0.01 × 0.20 = 0.002	$1B
Executive orders	0.05	0.05 × 0.20 = 0.01	$1B
Appointments	0.10	0.10 × 0.20 = 0.02	$4B
Foreign policy	0.03	0.03 × 0.20 = 0.006	$1.8B
Pardon power	0.05	0.05 × 0.20 = 0.01	$100M
Emergency powers	0.02	0.02 × 0.20 = 0.004	$4B
Presidential subtotal			$12.9B

Reduction: 80% — looks great!

But Now Calculate Ombudsman Delegation Risk

The Ombudsman now has:

• Access to all classified information
• Effective veto over presidential action
• Power to trigger constitutional crises
• Near-zero accountability (6-year term, hard to remove)

Ombudsman Failure Mode	P(occurrence/term)	Damage	Delegation Risk
Corruption (bribed/compromised)	0.05	$100B (sells secrets, enables enemies)	$5B
Partisan abuse (blocks legitimate actions)	0.15	$50B (governance paralysis)	$7.5B
Incompetence (fails to catch real problems)	0.20	$30B (false confidence)	$6B
Power grab (uses information for personal gain)	0.03	$80B (blackmail, manipulation)	$2.4B
Leaks classified information	0.10	$20B (intelligence damage)	$2B
Ombudsman subtotal			$22.9B

Total System Delegation Risk With Single Ombudsman

Total Delegation Risk = Presidential Delegation Risk + Ombudsman Delegation Risk
         = $12.9B + $22.9B
         = $35.8B

Compared to baseline ($65B): 45% reduction

The Oversight Paradox

Adding an Ombudsman reduced Presidential Delegation Risk by 80%, but the Ombudsman added $22.9B in new exposure. Net improvement: only 45%. The overseer became a major liability.

Why This Happens

The Ombudsman requires:

1. Access to everything the President sees (to monitor effectively)
2. Power to stop presidential actions (to matter)
3. Independence from the President (to not be captured)

But these same properties make the Ombudsman dangerous:

• Access → Can leak or sell information
• Power → Can abuse veto for personal/partisan gain
• Independence → Hard to hold accountable

This is the fundamental oversight dilemma.

---

Part 3: Distributing Oversight — The Committee Approach

Option B: Oversight Committee (5 Members)

Instead of one powerful overseer, distribute the role:

flowchart TB
    V[Voters] -->|"0.52"| P[President]

    subgraph Committee["Oversight Committee"]
        O1[Member 1<br/>Appointed by House]
        O2[Member 2<br/>Appointed by Senate]
        O3[Member 3<br/>Appointed by Chief Justice]
        O4[Member 4<br/>Appointed by State Governors]
        O5[Member 5<br/>Random citizen lottery]
    end

    V -.->|"indirect"| Committee
    P --> GOV[Government Actions]
    Committee -->|"majority veto"| GOV

Rules:

• 3 of 5 members must agree to veto any action
• Staggered 5-year terms
• Different appointing authorities (diverse principals)
• Each member sees only partial information (need-to-know compartmentalization)

Calculating Committee Delegation Risk

Key insight: Corruption now requires corrupting 3+ members, not just 1.

Probability of committee corruption:

If each member has P(corrupt) = 0.10, the probability that 3+ are corrupt:

P(≥3 corrupt) = C(5,3)×0.1³×0.9² + C(5,4)×0.1⁴×0.9¹ + C(5,5)×0.1⁵
             = 10×0.001×0.81 + 5×0.0001×0.9 + 1×0.00001
             = 0.0081 + 0.00045 + 0.00001
             = 0.00856
             ≈ 0.86%

Compare to single Ombudsman: P(corrupt) = 10%

Committee corruption is 11.6× less likely than single-overseer corruption.

Updated Delegation Risk Calculation

Committee failure modes:

Failure Mode	Single Ombudsman P	Committee P	Damage	Committee Delegation Risk
Corruption	0.05	0.0086	$100B	$860M
Partisan abuse	0.15	0.03	$50B	$1.5B
Incompetence	0.20	0.25*	$30B	$7.5B
Power grab	0.03	0.005	$80B	$400M
Leaks	0.10	0.18**	$20B	$3.6B
Subtotal	$22.9B			$13.86B

*Committees can be more prone to groupthink and diffused responsibility **More people = more leak surface

Presidential Delegation Risk with committee oversight:

Committee catches violations at 70% rate (lower than single expert, due to coordination costs):

Domain	Original Delegation Risk	With Committee (30% passthrough)	New Delegation Risk
Nuclear launch	$5B	× 0.30	$1.5B
Military deployment	$5B	× 0.30	$1.5B
Executive orders	$5B	× 0.30	$1.5B
Appointments	$20B	× 0.30	$6B
Foreign policy	$9B	× 0.30	$2.7B
Pardon power	$500M	× 0.30	$150M
Emergency powers	$20B	× 0.30	$6B
Presidential subtotal	$65B		$19.35B

Total System Delegation Risk With Committee

Total Delegation Risk = Presidential Delegation Risk + Committee Delegation Risk
         = $19.35B + $13.86B
         = $33.2B

Architecture	Total Delegation Risk	vs. Baseline
No oversight	$65B	—
Single Ombudsman	$35.8B	-45%
5-Person Committee	$33.2B	-49%

Modest improvement. The committee reduces corruption risk but increases leak risk and reduces oversight effectiveness.

---

Part 4: Adversarial Oversight — Competing Watchmen

Option C: Two Competing Overseers

What if we create adversarial overseers who watch each other?

flowchart TB
    V[Voters] -->|"0.52"| P[President]

    subgraph Adversarial["Adversarial Oversight"]
        O1[Overseer A<br/>Appointed by Majority Party]
        O2[Overseer B<br/>Appointed by Minority Party]
        O1 <-->|"mutual monitoring"| O2
    end

    P --> GOV[Government Actions]
    O1 -->|"can expose"| GOV
    O2 -->|"can expose"| GOV

Key insight: Overseers from opposing parties have incentive to expose each other’s failures and the President’s misdeeds (if President is from opposing party) or their own party’s overseer’s failures.

The Game Theory

If Overseer A covers for President (same party):

• Overseer B (opposing party) has incentive to expose both
• A’s cover-up becomes evidence of A’s corruption
• A faces consequences

If both overseers are corrupt:

• Each knows the other knows
• Neither can safely cover for the other
• Mutual blackmail creates unstable equilibrium
• Defection (exposing) is dominant strategy in iterated game

Calculating Adversarial Delegation Risk

Corruption probability:

For corruption to succeed, both adversarial overseers must be corrupt AND choose to collude:

P(successful corruption) = P(A corrupt) × P(B corrupt) × P(they collude | both corrupt)
                        = 0.10 × 0.10 × 0.30
                        = 0.003

(Collusion is unlikely because they’re from opposing parties and each has incentive to defect)

Partisan abuse probability:

Each overseer might try partisan abuse, but the other will likely expose it:

P(successful partisan abuse) = P(A abuses) × P(B fails to expose)
                             = 0.20 × 0.15
                             = 0.03

Adversarial system Delegation Risk:

Failure Mode	P	Damage	Delegation Risk
Corruption (both collude)	0.003	$100B	$300M
Partisan abuse (successful)	0.03	$50B	$1.5B
Incompetence (both miss)	0.10 × 0.10 = 0.01	$30B	$300M
Leaks (either)	0.15 + 0.15 = 0.28*	$20B	$5.6B
Gridlock (both veto everything)	0.08	$40B	$3.2B
Subtotal			$10.9B

*Leak risk increases—two overseers means two potential leak sources

Presidential Delegation Risk with adversarial oversight:

Adversarial overseers catch ~85% of violations (high motivation to find problems):

Presidential Delegation Risk = $65B × 0.15 = $9.75B

Total System Delegation Risk With Adversarial Oversight

Total Delegation Risk = Presidential Delegation Risk + Adversarial Oversight Delegation Risk
         = $9.75B + $10.9B
         = $20.65B

Architecture	Total Delegation Risk	vs. Baseline
No oversight	$65B	—
Single Ombudsman	$35.8B	-45%
5-Person Committee	$33.2B	-49%
Adversarial (2 parties)	$20.65B	-68%

Significant improvement! Adversarial incentives align oversight with exposure rather than protection.

The Adversarial Insight

When overseers have opposing interests, their mutual monitoring is self-enforcing. Neither can safely be corrupt because the other profits from exposing them.

---

Part 5: Randomized Oversight — Unpredictable Watchmen

Option D: Rotating Random Oversight

What if we make oversight unpredictable?

flowchart TB
    V[Voters] -->|"0.52"| P[President]

    subgraph Lottery["Random Selection Pool"]
        direction LR
        C1[Citizen 1]
        C2[Citizen 2]
        C3[...]
        C4[Citizen N]
    end

    Lottery -->|"random selection<br/>monthly rotation"| O[Current Overseer<br/>Identity unknown in advance]
    P --> GOV[Government Actions]
    O -->|"reviews"| GOV

Mechanism:

• Large pool of pre-cleared citizens (1,000+)
• Each month, 5 are randomly selected
• Selection revealed after they arrive for duty
• President cannot know who will oversee which actions

Why Randomization Helps

Corruption becomes expensive:

• Can’t pre-corrupt the specific overseer (unknown until activation)
• Would need to corrupt large fraction of pool (expensive, detectable)
• Each corrupt member only serves briefly (limited damage window)

P(random overseer corrupt):

If 5% of pool is pre-corrupted (50 of 1,000):

P(selected overseer corrupt) = 50/1000 = 0.05

But corruption only helps during their 1-month term:

P(corrupt overseer during specific presidential action) = 0.05 × (1/12) = 0.004

Effective corruption probability over 4-year term:

Each action faces independent random oversight:

P(corrupt oversight for any given action) ≈ 0.004

This is 12.5× better than a single permanent overseer with 5% corruption probability.

The Competence Problem

Random citizens may lack expertise to recognize violations:

Overseer Type	P(catches violation)	Why
Expert Ombudsman	0.80	Deep knowledge
Committee	0.70	Collective expertise
Adversarial	0.85	Motivated to find problems
Random citizen	0.40	Untrained, unfamiliar

Hybrid: Random + Expert

Solution: Pair random citizens with expert staff:

flowchart TB
    subgraph "Random Oversight Unit"
        R[Random Citizens<br/>5 monthly] -->|"decision authority"| DEC[Decisions]
        E[Expert Staff<br/>Permanent, non-voting] -->|"analysis, briefing"| DEC
    end

Random citizens decide, but experts help them understand what they’re seeing.

• Expert staff can be monitored for corruption (they don’t make final calls)
• Random citizens can’t be pre-corrupted (unknown selection)
• Combined: harder to corrupt than either alone

Calculating Randomized + Expert Delegation Risk

Oversight effectiveness:

• Expert staff catch 75% of violations in analysis
• Random citizens approve/publicize staff findings 90% of the time
• Combined catch rate: 0.75 × 0.90 = 0.675

Corruption probability:

• Expert staff: 0.08 (can be corrupt, but can’t unilaterally suppress)
• Random citizens: 0.004 (per-action, due to rotation)
• Both simultaneously corrupt: 0.08 × 0.004 = 0.00032

Randomized oversight Delegation Risk:

Failure Mode	P	Damage	Delegation Risk
Corruption (both levels)	0.00032	$100B	$32M
Partisan abuse (random selection is neutral)	0.01	$50B	$500M
Incompetence	0.20	$30B	$6B
Leaks	0.15	$20B	$3B
Gaming (President delays until favorable oversight)	0.05	$20B	$1B
Subtotal			$10.53B

Presidential Delegation Risk with randomized oversight:

Presidential Delegation Risk = $65B × (1 - 0.675) = $65B × 0.325 = $21.1B

Total System Delegation Risk With Randomized Oversight

Total Delegation Risk = Presidential Delegation Risk + Randomized Oversight Delegation Risk
         = $21.1B + $10.53B
         = $31.6B

Architecture	Total Delegation Risk	vs. Baseline
No oversight	$65B	—
Single Ombudsman	$35.8B	-45%
5-Person Committee	$33.2B	-49%
Randomized + Expert	$31.6B	-51%
Adversarial (2 parties)	$20.65B	-68%

Randomized is better than committee but worse than adversarial. However, randomized might be combined with adversarial…

---

Part 6: Combining Approaches — The Optimal Architecture

Option E: Adversarial + Randomized + Compartmentalized

What if we combine the best features?

flowchart TB
    V[Voters] -->|"0.52"| P[President]

    subgraph Oversight["Layered Oversight System"]
        subgraph Layer1["Layer 1: Real-Time (Adversarial)"]
            A1[Party A Monitor] <-->|"mutual watch"| A2[Party B Monitor]
        end

        subgraph Layer2["Layer 2: Random Audit"]
            R[Random Citizen Panel<br/>Quarterly rotation]
            E[Expert Staff]
        end

        subgraph Layer3["Layer 3: Specialized"]
            N[Nuclear: Separate chain]
            M[Military: Separate chain]
            F[Financial: Separate chain]
        end
    end

    P --> GOV[Government Actions]
    Layer1 -->|"immediate monitoring"| GOV
    Layer2 -->|"random deep audits"| GOV
    Layer3 -->|"domain-specific veto"| GOV

Architecture features:

1. Adversarial real-time monitors: Opposing parties watch everything in real-time
2. Random audit panels: Quarterly deep dives into random decisions
3. Specialized oversight: Nuclear, military, and financial actions have separate oversight chains
4. Compartmentalization: No single overseer sees everything

Calculating Combined System

Layer 1 (Adversarial) effectiveness: 85% catch rate, corruption P = 0.003

Layer 2 (Random Audit) effectiveness: 60% catch rate on audited items (25% of actions audited)

• Overall catch rate contribution: 0.60 × 0.25 = 0.15 (for things Layer 1 missed)

Layer 3 (Specialized): For highest-damage categories only

• Nuclear: Additional 90% catch rate with dual-key requirement
• Military: Additional 70% catch rate through military chain
• Financial: Additional 75% catch rate through GAO/IG

Combined catch rate:

For a presidential violation to succeed, it must pass ALL applicable layers:

P(violation succeeds) = P(passes Layer 1) × P(passes Layer 2) × P(passes Layer 3 if applicable)

Domain	L1 Pass	L2 Pass	L3 Pass	Combined Pass
Nuclear	0.15	0.85	0.10	0.013
Military	0.15	0.85	0.30	0.038
Executive orders	0.15	0.85	1.0 (no L3)	0.128
Appointments	0.15	0.85	1.0	0.128
Foreign policy	0.15	0.85	0.50	0.064
Pardon power	0.15	0.85	1.0	0.128
Emergency powers	0.15	0.85	0.40	0.051

Presidential Delegation Risk with combined oversight:

Domain	Original Delegation Risk	× Pass Rate	New Delegation Risk
Nuclear	$5B	0.013	$65M
Military	$5B	0.038	$190M
Executive orders	$5B	0.128	$640M
Appointments	$20B	0.128	$2.56B
Foreign policy	$9B	0.064	$576M
Pardon power	$500M	0.128	$64M
Emergency powers	$20B	0.051	$1.02B
Presidential subtotal	$65B		$5.12B

Oversight system Delegation Risk:

Now we must sum Delegation Risk across all oversight components:

Oversight Component	Delegation Risk
Adversarial monitors (Layer 1)	$3B
Random audit (Layer 2)	$2B
Nuclear chain (Layer 3a)	$500M
Military chain (Layer 3b)	$1B
Financial chain (Layer 3c)	$800M
Coordination failures between layers	$1.5B
Oversight subtotal	$8.8B

Total System Delegation Risk With Combined Oversight

Total Delegation Risk = Presidential Delegation Risk + Oversight Delegation Risk
         = $5.12B + $8.8B
         = $13.9B

Architecture	Total Delegation Risk	vs. Baseline	Reduction
No oversight	$65B	—	0%
Single Ombudsman	$35.8B	-45%	45%
5-Person Committee	$33.2B	-49%	49%
Randomized + Expert	$31.6B	-51%	51%
Adversarial (2 parties)	$20.65B	-68%	68%
Combined Layered	$13.9B	-79%	79%

The Power of Layered Defense

The combined system achieves 79% Delegation Risk reduction—not because any single layer is perfect, but because violations must evade all layers simultaneously. Defense in depth works.

---

Part 7: The Limits of Oversight

Irreducible Risks

Even the best oversight architecture has limits:

1. Speed-of-Decision Problems

Some presidential actions must happen faster than oversight can review:

Action	Decision Window	Oversight Feasible?
Nuclear response	15 minutes	No
Hostage situation	Hours	Partial
Market intervention	Hours-days	Yes
Troop deployment	Days	Yes
Judicial appointments	Months	Yes

For fast decisions, oversight must be pre-positioned (people already in the room) rather than reactive.

2. Information Asymmetry

The President has access to intelligence no overseer can fully verify:

President: "I ordered the strike because classified source X said..."
Overseer: [Cannot independently verify source X]

This creates an irreducible trust floor—at some point, someone must be believed.

3. Collusion Across Branches

If President + Congressional leadership + Supreme Court majority all collude, no oversight within the system can stop them. This is a constitutional trust assumption, not addressable by oversight architecture alone.

4. Public Attention Limits

Oversight mechanisms that rely on public exposure (adversarial monitors, random audits) only work if the public:

• Pays attention
• Understands the violation
• Has memory longer than news cycles
• Has mechanisms to act (elections, pressure)

The Oversight Cost Curve

                     │
                     │              ╭─── Diminishing returns
    Total System     │            ╭─╯
    Delegation Risk              │          ╭─╯
                     │        ╭─╯
                     │      ╭─╯
                     │    ╭─╯
                     │  ╭─╯
                     │╭─╯
                     ├─────────────────────────
                     │ Oversight Complexity →

                     Zone 1: High ROI (simple oversight)
                     Zone 2: Moderate ROI (layered systems)
                     Zone 3: Low ROI (diminishing returns)
                     Zone 4: Negative ROI (oversight costs exceed benefits)

At some point, adding more oversight creates more liability than it prevents.

Our combined system ($13.9B Delegation Risk) is likely near the optimal—additional layers would add complexity and new failure modes without proportional benefit.

---

Part 8: Design Principles for Oversight Systems

Based on this analysis:

Principle 1: Adversarial > Aligned

Overseers with opposing interests are more reliable than aligned overseers:

Overseer Relationship	Corruption Probability	Why
Overseer loyal to President	0.15	Incentive to cover up
Overseer neutral	0.08	No strong incentive either way
Overseer adversarial to President	0.03	Incentive to expose

Implication: Staff oversight with opposing political parties, not “nonpartisan” appointees.

Principle 2: Rotate > Permanent

Rotating oversight reduces:

• Time for corruption relationships to develop
• Value of corrupting any individual
• Institutional capture

Optimal rotation: Frequent enough to prevent capture, slow enough for competence

Rotation Speed	Corruption Resistance	Competence	Sweet Spot
Daily	Excellent	Terrible	No
Monthly	Very good	Poor	Partial
Quarterly	Good	Moderate	Yes
Annually	Moderate	Good	Maybe
Never	Poor	Excellent	No

Principle 3: Layer > Single Point

Multiple independent oversight layers compound effectiveness:

P(violation succeeds) = P(evades L1) × P(evades L2) × P(evades L3)

Three 70%-effective layers: 0.3 × 0.3 × 0.3 = 2.7% success rate

One 90%-effective layer: 10% success rate

Layering beats perfection.

Principle 4: Specialize High-Damage Domains

The highest-damage actions deserve bespoke oversight:

Damage Potential	Oversight Investment
< $1B	Standard layered system
$1B -$100B	Domain-specific chain
> $100B	Dedicated system, multiple vetos, hardware interlocks

Nuclear launch ($50T damage potential) warrants extreme measures that would be absurd for routine executive orders.

Principle 5: Accept Irreducible Trust

At some point, you must trust someone:

Minimal Trust Floor = P(all overseers collude) × P(public fails to respond)
                    ≈ 0.001 × 0.10
                    ≈ 0.0001

Trying to eliminate this last 0.01% often costs more than the risk is worth. Accept a minimal trust floor and focus resources on higher-ROI improvements.

---

Summary: The Oversight Menu

Option	Mechanism	Delegation Risk Reduction	Key Benefit	Key Weakness
Single Ombudsman	One powerful overseer	45%	Simple	Overseer becomes liability
Committee	Multiple appointed members	49%	Harder to corrupt	Groupthink, leaks
Adversarial	Opposing party monitors	68%	Self-enforcing vigilance	Gridlock risk
Randomized	Lottery selection, rotation	51%	Unpredictable, uncorruptible	Less competent
Combined Layered	Multiple independent systems	79%	Defense in depth	Complex, expensive

Key Takeaways

1. Oversight redistributes, not eliminates, trust exposure
2. Powerful overseers become powerful liabilities
3. Adversarial incentives > good intentions
4. Unpredictability defeats corruption planning
5. Layered defense compounds effectiveness
6. Specialize oversight for highest-damage domains
7. Accept irreducible trust floor and optimize above it

---

See Also

• Trust Across Civilizations — Historical oversight architectures
• Organizational Trust — Business applications
• Risk Inheritance — How trust flows through networks
• Mechanism Design — Incentive-compatible structures
• Anti-patterns — What not to do