Resolving Structural Resource Imbalance
Through Information-Guided Routing

The University of Nottingham Student Service Centre operates across two campus nodes-University Park and Jubilee-with asymmetric staffing distributions and spatial positions. Under a deliberate stress condition of 50 arrivals per hour and a 60/40 geographic demand bias, the network continually collapsed. This project investigates whether structural resource imbalances can be resolved architecturally through information guided routing rather than operationally through headcount expansion. By diagnosing the root vulnerability as an informational deficit, the simulation proves that providing agents with dynamic, system-state information achieves load equilibrium without adding capacity.

7.31 Mean SLA Breaches down from 140
94.8% Systemic Failure Reduction
0 Physical Capacity Added
1000 Replications Per Phase

CASE STUDY ABSTRACT

// 1. HYPOTHESIS

The central design question: If pure proximity-based routing is replaced with cognitively realistic, information-guided routing, at what threshold does a structurally imbalanced service network stabilise, and what does that transition actually look like?

// 2. STRESS CONDITION

A baseline with a symmetric 50/50 geographic demand distribution accidentally balances campus loads through simple proximity, creating a null problem. To model real-world accommodation density and force structural failure, the environment was engineered with a 60/40 geographic demand bias weighted toward Jubilee, combined with a 50 agents/hr arrival rate.

// 3. ARCHITECTURE & METHOD

A hybrid architecture coupling Discrete Event Simulation (DES) for the operational pipeline with Agent-Based Modelling (ABM) for individual human cognition. The system was tested across 1,000 replications at 0%, 50%, and 80% app adoption rates to isolate how information availability alters system behaviour, holding staffing strictly constant.

// 4. OUTCOME

The system crossed a non-linear threshold. Cognitive routing achieved load equilibrium and broke the psychological "Friction Transfer" loop at its origin point. A subsequent 500-iteration constrained optimisation found a cost-minimal configuration (total objective value: £5,778/shift) that practically eliminated service failures while minimising operational cost.

BASELINE NETWORK STATE

// 1. DETERMINISTIC ROUTING - THE BLINDNESS WALL

  • Condition: At 0% digital visibility, agents lack dynamic system-state data (live queue lengths, resource availability). They execute a deterministic proximity heuristic, routing via straight-line Euclidean distance from an asymmetric (60/40) spatial spawn distribution.
  • Outcome: Demand systematically concentrates at the nearest physical node. The network structurally defaults to localised queue collapse (Jubilee) and simultaneous resource under-utilisation (University Park), yielding 140.07 mean SLA breaches per shift (σ=39.57).

// 2. LOCAL OPTIMISATION VS. GLOBAL EQUILIBRIUM

  • Condition: In an information vacuum, every agent independently minimises their own transit cost. This represents perfectly rational local optimisation. No individual agent behaves incorrectly.
  • Outcome: When an entire population executes uncoordinated local optimisation under a shared information constraint, the system fails to reach a global load equilibrium. The structural failure is not a stochastic anomaly; it is the mathematical inevitability of systematically misdirected demand.

// 3. SUPPLY-SIDE EXPANSION INEFFICIENCY

  • Condition: The standard operational response to severe queue degradation is linear resource expansion-treating the bottleneck as a deficiency in aggregate processing capacity relative to total arrival rates.
  • Outcome: The network already possesses sufficient aggregate capacity. Adding service units (staff) to a system constrained by asymmetric demand routing merely scales the financial cost of the queue without altering its structural geometry.

CONCEPTUAL MODEL DIAGRAM

I got my mates with me I have got work to get done Got Damn short term memory I ain't walking till Park Campus Got my own trauma bro I get mad waiting too much bro Agent Model & Mechanisms Door In Door Out Task Split Join Back Seize specific Process specific Release specific Seize General Process General Release General Discrete Event Procedure Flow
I got my mates with me I have got work to get done Got Damn short term memory I ain't walking till Park Campus Got my own trauma bro I get mad waiting too much bro Agent Model & Mechanisms Door In Door Out Task Split Join Back Seize specific Process specific Release specific Seize General Process General Release General Discrete Event Procedure Flow

AGENT BASED MODELLING & MECHANISMS

// 1. BLINDNESS WALL & BOUNDED INFORMATION

  • Mechanism: 60% of agents spawn in the Jubilee-proximate region of the environment, encoding a structural demand asymmetry that makes proximity routing systematically catastrophic regardless of aggregate capacity. Agents without the app (isAppUser == false) execute a deterministic proximity heuristic via straight-line Euclidean distance. For app users, queue perception is bounded by an uncertaintyRange calculated from digital literacy and system trust.
  • Rationale: The model generates independent stochastic noise (upNoise and jubNoise) for each campus. This prevents the simulation from acting as a symmetric error-correction system. Two agents standing in identical spatial positions with different literacy profiles will mathematically compute different routing decisions from the exact same underlying queue data.

// 2. ASYMMETRIC SOCIAL GRAVITY

  • Mechanism: Group transit applies two competing logarithmic forces. Social Buffering divides perceived queue load (multiplier: 0.25). Social Distraction subtracts from analytical patience (multiplier: 0.5).
  • Rationale: Because the distraction force is twice as strong as the buffering force, highly social agents experience mathematically eroded activePatience. Since distance is additive while queue pressure is multiplicative in the cost function, social distraction progressively shifts agents toward proximity-defaulting behaviour, even when queue data suggests a better alternative.

// 3. COMPOSITE COST & THINKING LOOP

  • Mechanism: The algorithm evaluates perceived load, transit distance (amplified by a live weatherMultiplier), campus skill efficiency, and reputation memory. If the lowest computed cost exceeds a psychological panic threshold (activePatience * 5.0), the Thinking Loop fires.
  • Rationale: The agent does not simply accept a bad route. The simulation scales their rageIndex, elevates cognitiveBias, and recalculates the entire cost function under a heightened stress state within the same decision cycle, modelling stress-induced analytical recalibration.

OPERATIONAL PIPELINE - DES PROCEDURAL FLOW

// 1. SHADOW QUEUE HANDSHAKE

  • Mechanism: At the moment of cognitive routing commitment, an agent increments a task-isolated predictive counter (upSpecShadowQueue, etc.). Upon physical arrival at TimeMeasure_Start, the exact recorded decision (agent.chosenNode) is used to decrement the counter.
  • Rationale: This "Honest Math Handshake" gives spatially distributed agents temporal awareness of a node before they arrive. Using the agent's internal routing string rather than physical block location ensures architectural precision and prevents the predictive signal from double-counting physical arrivals (phantom congestion).

// 2. STRUCTURAL ONE-WAY FUNGIBILITY

  • Mechanism: Processing human casework requires strict task-to-skill matching. The Seize_Specific block requests {Pool_Specific} for standard cases, or a simultaneous compound pair {Pool_Specific, Pool_Translator} when linguistic assistance is required. Conversely, Seize_General offers alternatives ({Pool_Specific} or {Pool_General}).
  • Rationale: This encodes specialist absorption of general demand as a hard structural constraint rather than a statistical probability weight. Specialists can step down to clear general queues, but general staff mathematically cannot seize complex workflows. Furthermore, an agent requiring translation cannot be served until both resources in the compound set are simultaneously available, creating a realistic operational bottleneck.

// 3. DUAL TIME STATE & FRICTION TRANSFER

  • Mechanism: The pipeline executes a dual measurement freeze. actualWait is captured upon resource seize to compute psychological escalation. If rageIndex exceeds 0.1, it transfers directly to the node as staffFatigue—acting as a live speed multiplier (1.0 + this.staffFatigue) on base service times. Conversely, pureWaitTime is frozen strictly upon seize exit.
  • Rationale: This is the critical bridge connecting the ABM and DES layers. Frustration slows throughput, generating longer queues and more frustration. Conflating psychological wait time (actualWait) with the £50 SLA audit metric (pureWaitTime) would have produced either an unfair operational penalty or an inaccurate human frustration model.

EXPERIMENTAL DESIGN & HYPOTHESIS VALIDATION

// 1. THE BASELINE CONFIGURATION & METHODOLOGY

  • Configuration: The system was stress-tested across 1,000 replications at three app adoption phases (0%, 50%, and 80%). To isolate the causal impact of information availability, staffing was held strictly constant across all three phases.
  • Baseline Allocation: University Park operated with 24 total staff (7 General, 13 Specific, 4 Translator). Jubilee operated with 16 total staff (6 General, 6 Specific, 4 Translator). This total allocation of 40 staff actively mismatched the 60/40 spatial demand bias, placing the vast majority of processing capacity at the node furthest from the demand origin.

// 2. NON-LINEAR STABILITY THRESHOLD

  • Result: The reduction in service failure was mathematically non-linear across the three phases.
    • Phase 1 (0%): The structurally imbalanced baseline produced 140.07 mean breaches.
    • Phase 2 (50%): Achieved an 82.9% reduction (23.91 breaches), but the standard deviation (26.90) proved the Friction Transfer cascade was still firing under stochastic pressure.
    • Phase 3 (80%): Crossed the stability threshold. The network achieved global load equilibrium, reducing SLA breaches by 94.8% (7.31 mean) with zero operational restructuring.

// 3. STOCHASTIC BRITTLENESS & SHOCK EXPOSURE

  • Condition: Analysing the statistical variance structure across the 1,000 replications at the 80% high-adoption phase.
  • Result: At 80% adoption, the standard deviation (σ=12.68) spiked to significantly exceed the mean (7.31), confirming a bimodal distribution. This variance is not simulation noise; it is a precise quantification of structural vulnerability. Severe stochastic weather events temporarily compress routing options, forcing even app users back into proximity-defaulting behaviour. This proves the information architecture solved efficiency under normal load, but remains mathematically brittle under environmental shock.

// 4. CONSTRAINED COST OPTIMISATION

  • Configuration: A 500-iteration OptQuest execution (100 replications/iteration) with adoption locked at 90%. The objective function minimised total operational cost: the sum of staff wages (General: £15/hr, Specific: £25/hr, Translator: £20/hr) plus a £50 penalty per SLA breach.
  • Constraint: Strict capacity limits (University Park max 25 staff; Jubilee max 20). Hard human wellbeing constraints were enforced: any configuration producing an average staff burnout or agent frustration > 0.3 was immediately discarded by the optimiser as an infeasible solution.

// 5. OPTIMAL CONFIGURATION - ITERATION 173

  • Result: The optimiser converged on a cost-minimal configuration yielding a total objective value of £5,778 per shift. Total wages accounted for £5,680, leaving a penalty component of just £98 (representing ~1.96 expected breaches).
  • The Staff Allocation Shift:
    • University Park (£2,680/shift): 2 General, 9 Specific, 4 Translator. (Total: 15)
    • Jubilee (£3,000/shift): 5 General, 8 Specific, 5 Translator. (Total: 18)
  • Strategic Rationale: The mathematical proof of the capacity fallacy. The optimiser achieved near-elimination of SLA breaches while reducing total system headcount from 40 down to 33. It resolved the failure by inverting the physical allocation to match the 60/40 demand reality (shifting weight from UP to Jubilee) and relying on the 90% app adoption rate to cognitively route overflow demand efficiently back to the leaner UP node.

STRATEGIC INSIGHTS

// 1. INFORMATION VISIBILITY IS AN OPERATIONAL INFRASTRUCTURE PROBLEM, NOT A COMMUNICATIONS PROBLEM

  • Insight : Providing students with a poster showing average queue times would not have produced these results. The intervention works because it surfaces real-time, task-specific, predictive queue state at the moment of routing decision - before the agent commits to a direction. The precision of the information architecture matters as much as its existence.

// 2. HUMAN COGNITIVE FRICTION IS A SYSTEM VARIABLE, NOT A SOFT FACTOR

  • Insight : While pure human emotion cannot be perfectly quantified, its operational friction dictates network stability. Digital literacy, anxiety, social context, and campus reputation are not cosmetic considerations to be handled after a system is built. They are the structural variables that determine exactly where demand lands in a service network. Ignoring their impact produces a model that is technically accurate but operationally meaningless.

// 3. ADOPTION RATE FUNCTIONS AS A CRITICALITY LEVER, NOT A GRADIENT

  • Insight : The improvement from 0% to 50% adoption is substantial. The improvement from 50% to 80% crosses a qualitative threshold - from unstable to stable. This means adoption strategy is not "more is better, incrementally". It is "identify the threshold and design an adoption programme specifically to cross it".

// 4. STAFF WELLBEING IS A SECOND-ORDER OUTPUT OF INFORMATION ARCHITECTURE QUALITY

  • Insight : The reduction in burnout is not a direct consequence of lower workload. It is a consequence of fewer frustrated agents reaching the desk, which means less emotional labour transferred, which means fatigue accumulates slower than recovery can discharge. Improving information architecture is a staff wellbeing intervention. That reframe has implications for how service network investments are justified and evaluated.

// 5. OPTIMISATION AND RESILIENCE REQUIRE SEPARATE DESIGN RESPONSES

  • Finding (Stochastic Brittleness) : The high standard deviation at 80% adoption (σ=12.68) is not simulation noise; it is a precise quantitative characterisation of the system's failure mode. Under severe weather events, increased transit friction (weatherMultiplier) and reduced analytical patience cause even app users to partially revert toward proximity-defaulting behaviour. This creates intense demand surges that initiate the Friction Transfer cascade, producing high-breach outlier runs.
  • Insight : The optimiser found the cost-minimal configuration for near-zero breach performance under normal conditions, but it cannot solve the brittleness problem because those stochastic environmental shocks sit outside the optimisation's control variables. Any organisation deploying this system needs to design for both-an efficiency layer for normal operations and a resilience layer for shock conditions. These are different problems requiring different interventions.

LIMITATIONS, ASSUMPTIONS & VULNERABILITIES

// 1. STATIC TRUST DECAY & ARTIFICIAL COMPLIANCE

  • Assumption : The simulation treats user trust (infoTrustScore) as a mathematical constant. The architecture models a user base with zero longitudinal memory; agents do not dynamically update their trust parameters based on the accuracy of previous app predictions.
  • Vulnerability : In reality, systemic trust is a highly elastic variable. A user who follows the app's guidance and still experiences a severe wait time will rapidly abandon the platform. By holding this variable static, the model mathematically enforces artificial user compliance during repeated systemic failures, potentially overestimating the long-term stability of the 80% adoption threshold.

// 2. EPISTEMOLOGY OF SOFT VARIABLES

  • Assumption : The cognitive engine or brain successfully translates fluid human states into programmable mathematical logic
  • Vulnerability : This creates an inherent epistemic risk: the over-quantification of soft variables. Human emotion is highly contextual, yet the simulation requires it to function as a rigid mathematical operator. While this produces internally coherent data, empirically validating these exact behavioural coefficients in a physical setting is impossible. The architecture risks mistaking internal mathematical consistency for empirical truth.

// 3. ASYMMETRIC HUMAN AGENCY & OPERATIONAL PASSIVITY

  • Assumption : The hybrid architecture creates a structural imbalance in human agency. The ABM layer grants student agents extreme cognitive complexity, while the DES layer reduces human staff to passive resource pools that degrade linearly via staffFatigue.
  • Vulnerability : This ignores adaptive human resilience. Real service staff possess operational agency-under acute pressure, they temporarily surge processing speeds, shed non-essential tasks, or actively deploy interpersonal skills to de-escalate hostility. By treating the service desk purely as a degrading asset, the model successfully captures systemic drag but entirely deletes human adaptive recovery from the network.

// 4. FALLACY OF POOLED RESOURCE EXHAUSTION

  • Assumption : Fatigue is calculated globally for the entire campus service node as a single, pooled variable, rather than tracking individual operator degradation.
  • Vulnerability : Generalising operational fatigue across an entire resource pool mathematically dilutes acute individual burnout. It prevents the model from capturing shift dynamics where one specific operator reaches cognitive collapse while others remain fresh. This masks micro-level bottlenecks within the macro-level throughput data.

// 5. THE LINEARITY OF COGNITIVE COLLAPSE

  • Assumption : The Friction Transfer mechanism-the critical bridge between the cognitive and operational layers-executes strictly linearly, transferring frustration to staff fatigue at a steady, continuous rate (rageIndex * 0.1).
  • Vulnerability : Empirical studies in cognitive ergonomics prove that human burnout is rarely a steady drip; it behaves as a step-function or a cliff. An operator might handle twenty highly hostile interactions with zero measurable degradation, only for the twenty-first to trigger complete cognitive collapse. The linear accumulation model successfully simulates general network drag, but underestimates the suddenness with which real-world service nodes fail.

// 6. STOCHASTIC BRITTLENESS & THE OPEN RESILIENCE DEFICIT

  • Assumption : The model's primary objective function was cost and efficiency optimisation under normal load conditions.
  • Vulnerability : The high variance (σ=12.68 at 80% adoption) quantifies a severe structural vulnerability. Weather events-or any stochastic shock that simultaneously compresses demand and routing options-will overwhelm the information layer. The architecture precisely diagnoses this environmental brittleness, but makes no attempt to design a resilience layer to solve it.

FUTURE ITERATIONS & DESIGN IMPLICATIONS

// 1. THROUGHPUT SIGNAL VS. QUEUE LENGTH SIGNAL

  • Implication: Queue length alone is an ambiguous signal. A queue of twenty agents being served at high throughput may resolve faster than a queue of eight agents at a fatigue-degraded node. The current shadow queue mechanism surfaces predicted load (how many agents are heading to a node) but not predicted throughput (how fast that node is currently processing).
  • Iteration: Incorporate a dynamic throughput signal into the application architecture. This makes the cognitive routing engine informationally honest, allowing users to distinguish between a long queue produced by a high arrival rate and a short queue stalled by severe service degradation.

// 2. DYNAMIC TRUST AS AN ADOPTION LEVER

  • Implication: The threshold finding-that the system transitions from unstable to stable between 50% and 80% adoption-was produced under static conditions where the infoTrustScore remained fixed. In real-world deployment, trust and adoption are dynamic variables shaped by perceived utility. If stochastic weather shocks produce poor routing recommendations during early deployment, user trust will degrade, pushing the effective adoption rate back below the critical stability threshold.
  • Iteration: Future architectural models must treat application adoption as a dynamic, feedback-driven output of system performance rather than a fixed experimental input.

// 3. PSYCHOLOGICALLY DISTINCT AGENT POPULATIONS

  • Implication: The model implicitly produces two behaviourally distinct psychological profiles. App users carry a pre-arrival emotional load shaped by anticipated queue pressure; their operational frustration begins before they leave their origin point. Conversely, blind (non-app) users arrive emotionally neutral. This maps precisely onto the documented cognitive distinction between anticipated waiting and experienced waiting.
  • Iteration: Make this cognitive distinction mathematically explicit by introducing an on-arrival visual queue perception trigger for blind users at Door_In. This will capture the acute, immediate frustration response of walking into a physically visible long queue before a single minute of wait time has elapsed.

// 4. RESILIENCE LAYER DESIGN

  • Implication: The initial optimisation successfully solved the efficiency problem (finding cost-minimal staffing under normal load conditions). However, the identified stochastic brittleness proves that efficiency and resilience are entirely separate design paradigms. Solving for one does not mathematically solve for the other.
  • Iteration: Architect a dedicated resilience layer utilising entirely different operational instruments. This requires emergency capacity triggers activated when the weatherMultiplier breaches a defined threshold, dynamic routing algorithms that increase the balancingBias penalty under severe weather to forcibly counteract proximity-reversion behaviour, and automated real-time staff reallocation protocols.

MACRO-SYSTEM SYNTHESIS & SCALABILITY

// 1. DIMINISHING RETURNS & STOCHASTIC VOLATILITY

  • Context: Going into this project, I expected a steady, linear relationship between app adoption and breach reduction. The 1,000-replication dataset revealed a much harsher reality: the brutal curve of diminishing returns. Getting just 50% of the population to use the app crushed the baseline failure rate, dropping mean SLA breaches from 140 down to 24 (an 83% reduction). However, at 50% adoption, the standard deviation (26.89) was actually higher than the mean.

STRATEGIC TAKEAWAY

In mission-critical service design, fixing the "sunny day" scenario is the easy part—50% adoption achieves that. But users don't remember average days; they remember catastrophic failures during environmental shocks. The true product challenge isn't the initial rollout; it is engineering the aggressive adoption strategies and resilience layers required to capture that final 30% of the user base, which is the only way to insulate the system against stochastic volatility.

// 2. INDIVIDUAL RATIONALITY VS. COLLECTIVE FAILURE

  • Context: This project fundamentally changed how I view user behaviour versus system outcomes. Every single "blind" agent in this simulation makes a perfectly rational, logic-based decision: route to the nearest campus to save travel time. No individual is behaving incorrectly. Yet, systemic failure emerges precisely from the aggregate of these individually rational decisions made under a shared information constraint.

STRATEGIC TAKEAWAY

You cannot fix a structurally broken ecosystem by trying to change individual human behaviour. You can only fix it by changing the information architecture of the environment itself. It is an infrastructure problem, not a behavioural one.

// 3. THE 50/50 SPAWN FAILURE - DESIGNING THE RIGHT STRESS TEST

  • Context: The biggest breakthrough in my thinking didn't come from a success - it came from a massive failure in my early experimental setup. In my first prototype, I had agents spawning evenly with a 50/50 geographic distribution. Under this baseline, running the simulation at 0% adoption and 90% adoption yielded the exact same result: near-zero breaches. I initially assumed my routing code was broken. But after a deep audit, I realised the logic was perfect; the problem was the environment. A symmetric 50/50 spawn accidentally balanced the campus loads perfectly through simple proximity. The structural bottleneck I built the app to fix didn't actually exist under those conditions.

STRATEGIC TAKEAWAY

Shifting to a skewed 60/40 demand distribution wasn't an arbitrary tweak to force an interesting result; it was the only way to accurately model the real-world conditions under which the problem actually occurs. It transformed the project from a hollow validation exercise into a genuine stress test of whether information architecture can override physical demand imbalances.

// 4. CROSS-DOMAIN GENERALISABILITY

  • Context: This framework proves that total system capacity is often secondary to the architecture of demand routing. The blueprint built here is directly generalisable to any high-stakes, resource-constrained service environment where demand reverts to physical proximity in the absence of real-time signals. Hospital emergency room triage, airport security lane optimisation, and logistics network routing all share the exact operational geometry engineered in this project.