C_t = P_t × V_t                (αντιλαμβανόμενη σκληρότητα, κλιμακωμένη από την ορατότητα)
D_{t+1} = φ D_t + a A_t − ℓ L_t   (η βοήθεια αυξάνει την εξάρτηση· η νομιμοποίηση τη μειώνει)

ΔL_{t+1} =
   -λ A_t + ε,                 if C_t < θ      (σταθεροποίηση: καταστολή νομιμοποίησης)
   γ (C_t − θ) + η A_t,        if C_t ≥ θ      (υπέρβαση κατωφλίου: άνοδος νομιμοποίησης)

Recognition event:
R_t = 1{ L_t ≥ L* }            (εξωτερική αναγνώριση μόλις η νομιμοποίηση υπερβεί το κατώφλι)

As Adam Smith described the invisible hand, economists have been using models to define and explain the forces that shape our world. This model distills conflict’s complexity into a sequence where aid tolerated becomes dependence, dependence invites escalation, and escalation flips into legitimacy once the hand of such forces can no longer be unseen.

Prologue

I once, perhaps more than once, sat in a class taught by a great development economist. She had us run regression after regression, each model more complex than the last. By the end of the course, we were working directly with the data from her own published research — carefully randomized, bias-adjusted, and tested from every angle.

The results were rigorous and unsettling: humanitarian aid, designed to relieve suffering, can sometimes prolong conflict instead of resolving it. Yet as precise as the regressions were, I left the classroom with a broader unease. It felt as though there was something deeper at work — a layer of strategy, of anticipation and counter-anticipation, that statistics alone could not fully capture.

From Data to Strategy: A Game-Theoretic View

The regressions showed one truth: aid can change conflict dynamics in ways that are not always intuitive. But behind those numbers lies another layer — the logic of strategic interaction.

Game theory teaches us that actors rarely stop at immediate consequences. They think about what others will do, what others think they will do, and how those beliefs cascade. Economists sometimes call this second-order and higher-order thinking.

A familiar example is chess: every move is judged not only by what it accomplishes immediately but also by the counter-moves it will invite two, three, or more steps ahead.

• First-order: “Aid keeps people alive.”
• Second-order: “My opponent may exploit that aid.”
• Third-order: “If they exploit it, I can use that as justification for harsher action.”
• Nth-order: “They know that I know they may exploit it…”

What at first seemed like a statistical regularity — aid prolongs conflict — revealed itself as something more complex: a repeated game of moves and countermoves, where humanitarian flows become part of long-term strategy.

“Increases in U.S. food aid increase incidence and duration of civil conflict.” — Nunn & Qian, American Economic Review (2014)
“Insurgent violence increased in areas eligible for aid under a large program.” — Crost, Felter & Johnston, AER (2016)
“Aid shocks increase likelihood of civil conflict via bargaining shifts.” — Nielsen et al., AJPS (2011)
“Humanitarian aid … associated with higher levels of rebel violence.” — Reed Wood, AidData WP (2018)
“Under certain conditions, service provision reduces insurgent violence.” — Berman, Shapiro & Felter, Journal of Political Economy (2011)

The mechanisms identified include:

• Diversion and taxation: armed groups may capture part of the aid.
• Legitimacy: controlling relief strengthens political authority.
• Sabotage: projects may invite attacks.
• Volatility: sudden surges or cuts shift power balances.
• Service provision: in strong governance contexts, aid can indeed reduce violence.

Historical note: Somalia in the early 1990s became a tragic case study. Relief agencies documented how militias seized food aid, taxed convoys, and blocked deliveries. Human Rights Watch reported that “the looting of harvests and the theft of livestock … ensured that these communities would starve or remain at the mercy of internationally donated food” (HRW, 1993). Aid intended for famine victims was simultaneously reinforcing the control of armed groups.

Imagine a dominant actor familiar with this literature. They face a marginalized opponent: radical, dependent on humanitarian flows, unable to survive without external assistance.

Rather than always blocking aid, a dominant actor may allow or tolerate it. Historical cases suggest that this tolerance can create a fragile equilibrium:

1. Survival but weakness — the opponent lives on, but remains dependent.
2. Legitimacy through scarcity — they govern through rationing, not development.
3. Predictable aggression — aid can fuel low-level flare-ups, but these remain manageable.
4. Strategic entrapment — if the opponent escalates sharply, the dominant actor can argue: “We allowed aid to continue, and they turned it into violence. Stronger measures are now justified.”

In such a scenario, humanitarian aid is not just relief — it becomes a strategic lever in a repeated game.

This logic has echoes in Sri Lanka, where the LTTE taxed relief while Colombo tolerated flows into rebel zones; in Chechnya, where Moscow allowed convoys that insurgents siphoned.

This paradox reflects familiar economic logics:

• Moral hazard: like insurance, aid can encourage riskier behavior.
• Commitment problem: shifts in aid change bargaining power but not grievances.
• Strategic patience: a dominant actor may prefer to keep an enemy alive — weak, radical, dependent — until conditions align for decisive action.

As Nunn & Qian put it:

“Food aid may unintentionally prolong conflict by enabling belligerents to maintain fighting capacity.”

And as Crost, Felter & Johnston warn:

“Development assistance can provoke violence if insurgents perceive projects as threats to their control.”

Picture a region fractured by civil conflict. Civilians survive on aid convoys and cash infusions. A marginalized faction rules only because these flows sustain them. The dominant actor could shut them off — but chooses not to.

For years, the faction remains radical yet stagnant: feared but not legitimate, alive but not thriving. Sporadic violence persists. Then one day, escalation arrives.

The dominant actor might respond: “We tolerated their existence, we allowed aid to flow, and they turned it into violence. Now stronger measures are warranted.”

In this hypothetical, aid evolves from a lifeline to part of a strategic justification for escalation.

Yet strategy is rarely boundless. Pressure and control can work—until they don’t. Just as in finance, where stability flips suddenly into collapse, conflict has its own tipping point.

In finance, economists speak of a Minsky moment — the tipping point where leverage, which once sustained growth, suddenly flips into collapse. Something similar can occur in conflict dynamics when aid and strategy intersect.

A dominant actor may tolerate or restrict humanitarian flows to keep an opponent weak, fragmented, and dependent. Within limits, this approach can succeed: the adversary survives but remains marginal, lacking the resources to build legitimacy.

Yet strategy is bounded by perception. If pressure overshoots — if cruelty becomes too visible or disproportionate — the logic can reverse. What once sustained control can trigger a tipping point in which sympathy realigns and the opponent paradoxically gains recognition.

For radicals, survival itself can be a strategy. Enduring extreme costs signals resilience; it tests whether the international system will eventually acknowledge them. In this way, recognition becomes the delayed payoff for persistence under fire.

The paradox is stark: the same moves designed to ensure permanent weakness may, when pushed past their limit, produce legitimacy. Like in finance, where stability breeds risk until collapse, in conflict humanitarian pressure can breed control until it produces its opposite — the unintended elevation of the very actors it sought to erase.

I call this the Aid Overshoot Paradox. Aid flows, when tolerated or manipulated by a dominant actor, can initially serve as an instrument of control. Civilians survive, the weaker faction remains dependent, and stability — though fragile — persists. The very presence of aid suppresses the ability of marginalized actors to claim legitimacy: they endure but do not thrive.

📊 Aid Overshoot Paradox: Threshold Dynamics (Perceived Cruelty vs Change in Legitimacy)

Formalization. Let us define the core variables:

• A_t = humanitarian aid at time t
• P_t = pressure/control exerted by the dominant actor
• V_t = visibility of that pressure to external audiences
• C_t = perceived cruelty at time t (a function of P_t and V_t)
• D_t = dependence of the weaker faction
• L_t = legitimacy of the weaker faction
• θ = threshold beyond which cruelty becomes counterproductive
• L* = legitimacy level at which external recognition occurs

Compact representation of the dynamics:

C_t = P_t × V_t                (cruelty as perceived, scaled by visibility)
D_{t+1} = φ D_t + a A_t − ℓ L_t   (aid raises dependence; legitimacy erodes it)

ΔL_{t+1} =
   -λ A_t + ε,                 if C_t < θ      (aid stabilizes, legitimacy suppressed)
   γ (C_t − θ) + η A_t,        if C_t ≥ θ      (overshoot → legitimacy gain)

Recognition event:
R_t = 1{ L_t ≥ L* }            (acknowledgment once legitimacy crosses threshold)

The paradox shows how aid sustains dependence until perceived cruelty (C_t) crosses a threshold (θ). Below θ, legitimacy is suppressed and control holds. Once θ is breached, the mechanism inverts: excessive visibility of cruelty realigns sympathy, and legitimacy rises instead of being contained.

• x-axis: Perceived Cruelty (C_t) — how strongly external audiences register the dominant actor’s pressure
• y-axis: Change in Legitimacy (ΔL) — whether the weaker actor’s legitimacy declines (negative) or grows (positive)

As Sollenberg (2012) shows, foreign aid only raises conflict when it crosses certain volumes and in context of weak institutions.

Aid in conflict zones is never just aid.

• To civilians, it is survival.
• To marginalized groups, it can be fuel.
• To dominant actors, it may become a lever — sustaining an adversary just long enough to frame them as beyond repair.

Yet even levers have limits. Push too far, and what once reinforced control can invert into recognition. The tipping point, like a financial collapse, arrives not gradually but suddenly — the moment strategy overshoots its bounds.

This is the dilemma: relief cannot escape the logic of strategy, and strategy cannot escape the risk of overshoot. Humanitarian aid sits between the two, at once a lifeline and a liability.

At the end of the day, each of us practices strategic thinking in our own lives — anticipating others’ actions, preparing contingencies, weighing short-term trade-offs against long-term outcomes. It should not surprise us that global leaders, confronted with war and humanitarian crises, do the same. But it should remind us of something else too: even the best strategy, if pushed past its limit, may turn against its maker.

The question we too rarely ask is simple but unsettling: when we deliver humanitarian aid, do we actually help — or do we sometimes make things worse?

The paradox suggests a different way of structuring oversight. Instead of asking only “does aid reach people in need?”, we must also ask “are the patterns of its obstruction remaining hidden?”

The PAVE Framework: Pattern – Access – Visibility – Escalation

Baseline Risk Conditions (always monitored)

Certain structural factors make manipulation more plausible. They do not prove wrongdoing, but they heighten the standard of oversight:

• Aid flows into a territory partly administered by an armed political actor.
• The package includes dual-use commodities such as fuel, cement, or telecoms.
• The recipient has a history of violent confrontation or recurring escalation cycles.

When these conditions exist, even small irregularities should be treated with caution.

PAVE Triggers

• Pattern: Are disruptions, fees, or reroutes recurring in a structured way? Recurrence suggests intentional shaping rather than chance.
• Access: Who actually receives aid or salaries? If distribution lists or vendors are effectively closed circuits, aid may entrench dependency.
• Visibility: Can outsiders verify last-mile delivery? If reporting is opaque, delayed, or reliant on actor-controlled systems, the risk of hidden manipulation grows.
• Escalation: Are there contextual signals of confrontation (rhetoric, training, fortification) rising in parallel with stable aid flows? If so, the paradox moves toward overshoot.

Traffic Light Heuristic

• Green: Isolated irregularities with clear explanations.
• Amber: One or more PAVE triggers recur in combination with baseline risks.
• Red: Selective access + opacity + escalation signals under baseline conditions → warning that aid may cross the overshoot threshold (θ).

When these triggers converge, risk stops being abstract. The figure below shows the sequence as it actually unfolds — from baseline irregularities, through recurring patterns, to the point where visibility pushes the system toward overshoot.

Visibility of dominant-actor pressure (V_t) rises over time. When baseline risk conditions exist, even isolated irregularities deserve scrutiny. Recurring triggers in the PAVE framework (Pattern, Access, Visibility, Escalation) mark the path toward escalation. Once the visibility threshold (θ) is crossed, perceived cruelty flips into legitimacy for the weaker actor. Policy guardrails (bottom bar) provide layered defenses to slow or reshape this path.

• Pattern Reviews, not Incident Logs: Oversight must assess whether irregularities repeat over time; recurrence is the true signal of manipulation.
• Visibility Safeguards: Donors should invest in mechanisms that shorten reporting lags and reduce reliance on data controlled by armed actors.
• Pre-Committed Escalation: Relief operations should define in advance how Amber → Red triggers prompt independent monitoring or escalation to international review bodies.
• Legitimacy Impact Reviews: Large aid packages should be evaluated not only for humanitarian outcomes but also for how they might alter perceptions of political legitimacy if manipulated.

Why This Matters

Existing approaches either stress program ethics (Do No Harm) or focus on overt starvation and siege tactics. PAVE addresses the gray zone in between: subtle manipulations tolerated because they remain hidden. By organizing oversight around Pattern, Access, Visibility, and Escalation, we can at least conceptually identify when aid flows are drifting toward the overshoot dynamic — and take corrective action before legitimacy flips.

If donors choose to maintain humanitarian flows under such conditions, restructure the pipeline so the same money is less usable for coercion and more visible:

1. Two-key tranching: Each tranche auto-pauses unless both (i) independent last-mile verification and (ii) metered commodity reporting arrive on time.
2. Ring-fencing via non-cash rails: Replace cash where feasible with tokenized vouchers or e-cards redeemable only at pre-cleared vendors, with anonymous but auditable spend categories.
3. Third-party payroll attestation: Salary support requires independent headcount verification, rotating samples, and a challenge mechanism for contested names.
4. Dual-use choke points: For fuel/cement/etc., require in-line metering + geo-fenced delivery windows; missing telemetry = automatic tranche deferral.
5. Sunset & circuit breakers: Flows expire by default unless proactively renewed after a PAVE review; any Escalation flag trips a cool-off period.
6. Visibility premiums: Donors pay a small top-up only when reporting quality improves (shorter lags, better granularity) — directly incentivizing Vt↑.

These design moves raise the θ-crossing cost for any actor tempted to weaponize aid, and they provide a defensible rationale if someone asks, “Why did you slow/reshape funding?”

Appendix A — The Dark Pretence Game

 A toy repeated-game formalization of the Aid Overshoot Paradox

Purpose. This appendix gives a clean, game-theoretic sketch of the logic in the article. It is deliberately spare (“toy”) but disciplined enough to communicate incentives, steady states, and policy levers. It is not case-specific and makes no empirical claims.

A. Players, timing, information

• Sponsor S (dominant state/donor consortium).
• Local Power L (armed political actor with partial administrative control).
• Discrete time t = 0,1,2,… with common discount factor δ ∈ (0,1).
• Public state each period: visibility Vₜ ≥ 0. There is a threshold θ > 0. When Vₜ ≥ θ, a legitimacy flip occurs (public, legal, diplomatic costs).
• Stage game is publicly observed; strategies are Markov in Vₜ. Equilibrium concept: Markov Perfect Equilibrium (MPE).

B. Moves and technology

In each period t:

Sponsor S chooses:

• aid intensity xₜ ∈ [0,1] (0 = block, 1 = generous flow),
• monitoring/visibility effort wₜ ∈ [0, w̄] (audits, access, transparency).

Local Power L chooses:

• manipulation mₜ ∈ [0,1] (0 = clean implementation; 1 = aggressive gatekeeping/diversion).

Visibility dynamics (one-state Markov)

Vₜ₊₁ = ρ·Vₜ + ϕ·mₜ − σ·wₜ + εₜ₊₁

Intuition: manipulation tends to surface (ϕ), monitoring suppresses/hastens legibility (σ), and memory decays (ρ).

C. Payoffs (reduced form)

Let 𝟙ₜ = 1 if Vₜ ≥ θ, else 0.

Sponsor S:
u_S = α·xₜ + β·xₜ·(1 − mₜ) − λ·wₜ − η·𝟙ₜ

Local Power L:
u_L = τ·xₜ·mₜ − κ·𝟙ₜ

Interpretation. S likes aid (stability, efficacy) but dislikes exposure and paying for monitoring; L likes capturing aid but dislikes exposure once visibility crosses θ.

D. Baseline Markov strategies

Pre-threshold trigger v̄ ∈ (0, θ):

Sponsor S: (xₜ, wₜ) = { (x̄, 0) if Vₜ < v̄ ; (x̱, w̄) if Vₜ ≥ v̄ } with 0 ≤ x̱ < x̄ ≤ 1

Interpretation: tolerate some manipulation when visibility is low; escalate monitoring and/or throttle aid when visibility nears θ.

Local Power L: mₜ = { m̄ if Vₜ < ṽ ; m̱ ≈ 0 if Vₜ ≥ ṽ } with 0 < m̱ < m̄ ≤ 1 and ṽ ∈ (0, θ)

E. Sub-threshold toleration (steady regime)

Δ(V) := E[Vₜ₊₁ − Vₜ | Vₜ = V] = (ρ − 1)·V + ϕ·m̄

If ρ ∈ (0,1), there is a unique fixed point V* = (ϕ·m̄)/(1 − ρ).

Proposition 1 (Managed manipulation). If V* < v̄ < θ and monitoring is sufficiently costly (λ ≫ η), there exists an MPE in which S plays (x̄,0) and L plays m̄ whenever Vₜ < v̄; Vₜ fluctuates around V* and remains sub-threshold with high probability absent shocks.

Sketch. Raising w below v̄ trades current λ for a small reduction in future V; as long as V* ≪ θ, gains are second-order. Cutting m below ṽ forgoes τ·x̄·m̄ now to reduce distant exposure risk; dominated for small V. ■

F. Overshoot and legitimacy flip

Let T_θ = inf { t ≥ 0 : Vₜ ≥ θ }.

Proposition 2 (Overshoot risk). Even under the managed regime, if shocks εₜ have unbounded right tail, or if L occasionally deviates to mₜ = m̄′ > m̄, then P(T_θ < ∞) > 0. When V_{T_θ} ≥ θ, both players suffer a discrete loss (η, κ), inducing a regime shift (e.g., S: w = w̄ and/or x = x̱; L: m ≈ 0).

Sketch. Shocks or transient escalations can push V above θ; payoff discontinuity at θ alters continuation values. ■

G. Pre-commitment and policy triggers

Sponsor trigger policy. Fix v̄ ∈ (0, θ) and commit that if Vₜ ≥ v̄, then (xₜ, wₜ) = (x̱, w̄) with σ·w̄ large enough that

E[Vₜ₊₁ − Vₜ | Vₜ = v̄] ≤ −δ_V < 0

Proposition 3 (Trigger deters drift). For any ε > 0, there exist v̄, w̄, x̱ such that the MPE under this trigger has P(T_θ < ∞) < ε.

Policy mapping.

• Two-key tranching & sunset/circuit breakers → credible (x̱, w̄) above v̄.
• Visibility premiums → subsidize w (raise σ/λ).
• Dual-use choke points → lower effective ϕ (metering, geo-fencing).
• Third-party payroll attestation → raises σ (faster legibility).

H. Comparative statics (qualitative)

• ϕ ↑ → V* ↑ → overshoot more likely.
• σ ↑ or λ ↓ → easier to keep V below v̄.
• η, κ ↑ → both prefer pre-threshold restraint (w ↑, m ↓).
• ρ ↑ → visibility lingers; shocks accumulate; risk rises.
• Var(ε) ↑ → rare overshoot even with “good” policies.

I. Operational early-warning (model-consistent)

V̂ₜ₊₁ = ρ·Vₜ + ϕ·m̂ₜ − σ·ŵₜ

Issue Amber if V̂ₜ₊ₖ ∈ [v̄, θ) for small k; Red if V̂ₜ₊ₖ ≥ θ. (Backbone of the traffic-light in the main text.)

J. What is testable (without numbers here)

1. Patterned manipulation (m̂ ↑) with opaque reporting (ŵ ↓) predicts drift of V toward θ.
2. Pre-committed triggers reduce manipulation near v̄ (selection effect).
3. Visibility shocks (investigations, imagery) change policy (w ↑, x ↓) at first crossing.

K. Notation summary

x: aid intensity; w: monitoring effort; m: manipulation; V: visibility; θ: exposure threshold; parameters α, β, λ, η, τ, κ, ρ, ϕ, σ.