💻 computer science

Modeling Failure Dynamics in Time-Constrained Authentication Systems: Evidence of a Success Cliff in USSD Workflows

This paper proposes a simulation-based framework to model failure dynamics in time-constrained USSD authentication systems, revealing a non-linear "Success Cliff" phenomenon where session completion rates sharply decline once authentication complexity exceeds a critical threshold due to the interplay of user response time and network latency.

Original authors: Aklile Seyoum Mamo, Amanuel Kebede, Anny Christelle Irakoze, Jema Ndibwile

Published 2026-07-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Aklile Seyoum Mamo, Amanuel Kebede, Anny Christelle Irakoze, Jema Ndibwile

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to fill out a very strict, timed form on a bus that is moving very fast. You have exactly two minutes (120 seconds) to finish the whole thing before the bus drives away and the form disappears forever. This is how USSD works—the text-based menus you use on basic phones to send money or check balances.

The paper investigates what happens when you try to make this form more secure by adding extra steps, like asking for a PIN code or waiting for a text message code.

Here is the story of their discovery, broken down into simple concepts:

1. The Setup: The Ticking Clock

Think of the USSD session as a sand timer. Once you start, the sand starts falling.

The Goal: Finish the transaction before the sand runs out.
The Problem: Every time you add a security step (like typing a password), you use up more sand.
The Variables:
- You: How fast you type and think.
- The Bus: How fast the network sends messages (sometimes it's a smooth highway, sometimes it's a traffic jam).
- The Wait: Sometimes, the system has to pause and wait for an outside signal (like an SMS code) before letting you continue.

2. The "Success Cliff" Discovery

The researchers found something surprising. They expected that if you added more security steps, the number of people failing would go up slowly and steadily, like a gentle slope.

Instead, they found a cliff.

The Gentle Slope: If you just add more typing steps (like a 6-digit PIN instead of a 4-digit one), people fail a little bit more often, but it's manageable. It's like walking up a hill; it gets harder, but you can still make it.
The Cliff: However, if you add a step that requires waiting for an outside message (like an SMS code), the success rate doesn't just drop a little; it plummets.

The Analogy:
Imagine you are running a race with a strict time limit.

Scenario A (Just Complexity): You have to carry a slightly heavier backpack. You get tired faster, and you might finish a few seconds slower, but you still cross the finish line.
Scenario B (The Blocking Delay): Suddenly, halfway through the race, you are forced to stop and wait for a friend to walk over and hand you a key. You don't know how long they will take. They might take 5 seconds, or they might take 30.
- Because you are already running against the clock, that unpredictable wait eats up your remaining time.
- If the friend is slow, you miss the finish line entirely.
- The paper calls this the "Success Cliff." It's the point where adding that one "waiting" step causes the number of successful finishes to crash from 98% down to 75% almost instantly.

3. Why Does the Cliff Happen?

The paper explains that the "cliff" isn't caused by the difficulty of the task itself, but by the uncontrollable wait.

The Timer Doesn't Stop: Even while you are waiting for the SMS code to arrive, the sand in your timer is still falling.
The Double Whammy: When the SMS is slow, it uses up your "thinking time" budget. If you then take a normal amount of time to type the code, you accidentally run out of time before you can even finish typing.
The Result: The system doesn't just fail a little; it fails catastrophically because the "wait" and the "typing" happen at the same time, draining the clock twice as fast.

4. The "Two Failure Modes"

The researchers identified two different ways the system breaks:

The Slow Leak (Gradual Degradation): If you just make the steps harder to type (more numbers, more menus), people fail a bit more, but it's predictable. It's like a slow leak in a tire; you can still drive for a while.
The Cliff (Non-Linear Collapse): If you add a step that requires waiting for an outside message (like an SMS), the tire doesn't just leak; it bursts. This happens specifically when the network is slow or the SMS is delayed.

5. The Bottom Line

The paper concludes that for people using basic phones in areas with spotty internet (where USSD is most common), adding "stronger" security that requires waiting for text messages is dangerous.

It's not that the security is bad; it's that the timing is broken. By forcing users to wait for an outside message while their session clock is still ticking, the system creates a "Success Cliff" where a huge number of people simply cannot finish their transactions, no matter how hard they try.

In short: You can make a puzzle harder, and people will still solve it. But if you make them stop and wait for a stranger to bring them a piece of the puzzle, and the stranger is late, the whole game ends before they can finish. That is the "Success Cliff."

Technical Summary: Modeling Failure Dynamics in Time-Constrained Authentication Systems

Problem Statement
Time-constrained interactive systems, particularly Unstructured Supplementary Service Data (USSD) used for mobile financial services, operate under strict session limits (typically ~120 seconds) and sequential user interaction. While enhanced authentication mechanisms (e.g., multi-factor authentication, SMS OTPs) improve security, they introduce interaction complexity and time burdens. The core problem addressed is the lack of understanding regarding how authentication complexity, user response time, and network conditions interact to cause session failures. Existing research focuses on strengthening protocols or identifying vulnerabilities but fails to model the operational reliability of these workflows under real-world temporal constraints. Specifically, there is a gap in understanding whether small increases in complexity lead to gradual degradation or abrupt, non-linear collapse in session success rates.

Methodology
The authors propose a simulation-based framework to model failure dynamics in USSD authentication workflows. The methodology involves:

System Modeling: A USSD session is modeled as a sequence of $N$ steps (base navigation and authentication). The total session time ( $T_{session}$ ) is the sum of user response time ( $U_k$ ), network round-trip time ( $R_k$ ), blocking delay ( $H_k$ ), and retry delays ( $D_k$ ).
Parameterization:
- User Response Time ( $U_k$ ): Modeled using a modified Keystroke-Level Model (KLM) adapted for feature phone numeric keypads, incorporating cognitive processing, physical input, and context-switching times.
- Network Latency ( $R_k$ ): Modeled as a stochastic variable drawn from Gamma distributions representing low, medium, and high network conditions (stable urban to degraded rural).
- Blocking Delay ( $H_k$ ): A critical variable representing the time elapsed while waiting for external subsystem delivery (e.g., SMS OTP). This is modeled as a uniform distribution $U(5, 30)$ seconds, during which the session timer continues to count down without user interaction.
- Abandonment: Three probabilistic models (Time-Based, Event-Based, and Combined) simulate user disengagement prior to hard timeouts.
Complexity Metric: Authentication complexity ( $C_j$ ) is defined as a time-based proxy relative to a baseline 4-digit PIN.
Experimental Design: Four controlled experiments were conducted with 50,000 simulation runs each, varying authentication complexity (from single PIN to multi-step with OTP), network latency, and the presence of blocking delays.
Sensitivity Analysis: The robustness of findings was tested against variations in error probabilities and the detection thresholds for the "Success Cliff."

Key Contributions
The paper introduces the concept of the "Success Cliff," a non-linear failure phenomenon where session success rates sharply decline once authentication complexity exceeds a critical threshold.

Formal Definition: The Success Cliff ( $Q$ ) is defined mathematically as the complexity level where the drop in success rate ( $\Delta$ ) and the acceleration of that drop ( $\Gamma$ ) exceed specific thresholds.
Identification of the Necessary Condition: The study identifies blocking delay (out-of-band delivery wait) as the necessary condition for the emergence of the Success Cliff. Complexity alone results in gradual, manageable degradation, but the interaction between stochastic blocking delays and fixed session time budgets causes abrupt collapse.
Failure Taxonomy: The research distinguishes between two failure regimes:
1. Gradual Degradation: Caused by in-band complexity accumulation and network latency, leading to predictable declines in success rates.
2. Non-Linear Collapse: Caused specifically by the introduction of blocking delays, which exhaust the session budget through concurrent mechanisms (session timeout, application timeout, and user response timeout).

Results

Baseline Viability: Standard single-PIN flows (low complexity) maintain success rates above 98% across various network conditions.
Complexity Without Blocking Delay: Increasing authentication steps (e.g., multiple PINs or challenge-response) without external delays results in a linear, gradual decline in success rates (approx. 2.3–3.3 percentage points drop from low to high complexity). No Success Cliff is detected.
The Impact of Blocking Delay: When SMS OTP (blocking delay) is introduced:
- A Success Cliff is detected across all network conditions, including low latency.
- The transition from complexity level C3 (high complexity without OTP) to C4 (high complexity with OTP) triggers a sharp drop in success rates.
- Under high network latency, the success rate for the C4 configuration collapses to 74.88%–75.28%, well below the 98% operational viability threshold.
- The blocking delay introduces a secondary failure mode: User Response Timeout. Because the blocking delay consumes a portion of the per-step time budget, normal user response times that would otherwise succeed are pushed past the user response timeout limit ( $\tau_u$ ).
Robustness: Sensitivity analysis confirms that the location of the Success Cliff (at the C3-to-C4 transition) is invariant to error probability assumptions and detection thresholds when blocking delay is present.

Significance and Claims
The paper claims that the operational reliability of secure authentication workflows in resource-constrained environments is fundamentally threatened by the interaction between uncontrollable delivery waits and fixed session timers.

Architectural Threat: The primary finding is that blocking delay is the necessary condition for non-linear failure. Security mechanisms relying on out-of-band delivery (like SMS OTP) may render USSD workflows operationally non-viable in regions with degraded networks, even if the authentication steps themselves are logically sound.
Financial Inclusion Implications: For populations relying on USSD as their primary financial channel, particularly in rural areas with poor network conditions, the adoption of stricter authentication standards could lead to exclusionary failure rates. The study suggests that while complexity accumulation is a concern, the introduction of blocking delays poses a more severe, structural risk.
Design Guidance: The work provides a framework for designers to evaluate whether a proposed authentication configuration is operationally viable before deployment, highlighting that security enhancements must be weighed against the specific temporal constraints of the user's environment.

The authors maintain a modest stance, acknowledging that their simulation relies on KLM-derived estimates and conservative abandonment models. They note that real-world failure rates might be even higher due to factors like per-step retry limits and higher SMS variability in rural areas, suggesting their reported dynamics are likely conservative estimates of the actual problem.