Evaluation Rules, Marks Distribution & Costing Price
Every submission on HLDArena is graded by a 100% deterministic compiler and discrete-event simulator — zero LLM randomness. Every question splits marks into 60% Functional Requirements & Plain-English Write-up and 40% Non-Functional Diagram Simulation.
1. Marks Distribution Across Difficulty Tiers
Harder tiers carry higher maximum marks while preserving the exact 60% Write-up / 40% NFR ratio across all 110 questions.
| Scoring Component | Share | Easy | Medium | Hard | Expert |
|---|---|---|---|---|---|
| Total Question Marks | 100% | 500 pts | 1,000 pts | 1,500 pts | 2,000 pts |
| Part A — Functional Requirements & Plain-English Write-up | 60% | 300 pts | 600 pts | 900 pts | 1200 pts |
A1. Functional Requirements (FR-1..FR-N) Coverage Address every Functional Requirement bullet on the left end-to-end (request flow, state updates, and edge cases). | 22.5% | 113 | 225 | 338 | 450 |
A2. Plain-English HLD Explanation Quality Write complete, natural English sentences (15+ words) explaining which component acts and why. Keyword lists or copied prompt text fail. | 10.5% | 53 | 105 | 158 | 210 |
A3. Data Model, Keys & Partitioning State your record keys, identifier approach, and how data is partitioned as volume grows. | 10.5% | 52 | 105 | 157 | 210 |
A4. Consistency & Failure Handling State how reads and mutations maintain the required consistency model and how component failures are handled. | 9.0% | 45 | 90 | 135 | 180 |
A5. 1:1 Diagram <-> Write-up Alignment & Sizing Explain and state capacity sizing for every box drawn on your whiteboard, and draw every component you mention in text. | 7.5% | 37 | 75 | 112 | 150 |
| Part B — Diagram Simulation (Non-Functional Requirements) | 40% | 200 pts | 400 pts | 600 pts | 800 pts |
| B1. Latency SLO Adherence (weighted p50 / p99) | 11.0% | 55 | 110 | 165 | 220 |
| B2. Throughput & Burst Goodput Capacity | 11.0% | 55 | 110 | 165 | 220 |
| B3. Resilience, Failover & Availability | 9.0% | 45 | 90 | 135 | 180 |
| B4. Monthly Cloud Cost Efficiency | 6.0% | 30 | 60 | 90 | 120 |
| B5. Topology Clarity & Simplicity | 3.0% | 15 | 30 | 45 | 60 |
2. Standardized Monthly Costing Price Schedule
Every question computes monthly cloud spend deterministically from your drawn components and sizing specifications: Total Monthly Cost = Σ(Component Cost) + Network Egress. Exceeding a question’s budget cap triggers MLE_OVER_BUDGET.
| Infrastructure Component Role | Small ($/mo) | Medium ($/mo) | Large ($/mo) | Unit Scaling Formula |
|---|---|---|---|---|
| Client / External Origin | $0 | $0 | $0 | External traffic source (free) |
| Anycast DNS | $25 | $45 | $85 | flat per DNS node |
| Load Balancer | $35 | $80 | $170 | unit × instances |
| Object / Blob Store | $35 | $85 | $180 | unit × instances |
| Compute / App Service | $45 | $95 | $195 | unit × instances |
| API Gateway | $50 | $115 | $240 | unit × instances |
| Coordination / Lock | $55 | $115 | $220 | unit × instances |
| Message Queue / Stream | $60 | $135 | $290 | unit × max(instances, ceil(partitions/8)) × (1 + min(2, replicas)) |
| In-Memory Cache | $65 | $145 | $310 | (unit × nodes + extra RAM) × (1 + replicas) |
| Edge CDN | $75 | $160 | $340 | unit × instances (1.0× global edge) |
| Key-Value / NoSQL DB | $95 | $210 | $450 | unit × shards × (1 + replicas) × instances |
| Search / Index Store | $110 | $235 | $490 | unit × shards × (1 + replicas) × instances |
| Relational DB | $115 | $245 | $520 | unit × shards × (1 + replicas) × instances |
Client origin = $0. Unspecified boxes default to 1 instance, small tier, 1 shard, 0 replicas, single-AZ (8 GB RAM on cache).
Includes 8 GB RAM/node (+$3/GB above 8 GB×nodes). NVMe/SSD-tiered cache includes 64 GB/node (+$0.10/GB extra).
Single-AZ: 1.0× · Multi-AZ: 1.05× with redundancy (instances > 1 or replicas ≥ 1), 1.30× standalone · Multi-Region: 2.35×.
Auto-scaling (+25% burst pool). Serverless / on-demand DB or KV store (0.55× unit factor).
$55 per 10k avg goodput RPS ($32 per 10k RPS with Edge CDN) + $75 per cross-region link per 10k RPS.
3. Plain-English Write-up & 1:1 Diagram Alignment Rules
How to structure your Design Write-up so the deterministic compiler credits your design accurately.
Each question lists standardized Functional Requirement bullets (create/write path, read/fetch path, update/delete or expiry lifecycle, and consistency/fault behavior). Explain end-to-end how your architecture handles every bullet.
Write complete English sentences connecting which component acts, what operation it performs, and why. Bare keyword dumps, repetitive filler, unfilled [...] template prompts, or copied prompt text fail with WA_FUNCTIONAL.
The simulator enforces strict 1:1 topology lock: components mentioned only in text without being drawn on the whiteboard are ignored in simulation, and drawn boxes left unexplained in text receive uncalibrated single-node defaults.
Specify hardware tier (small, medium, large), instance/shard/replica counts, memory capacity, partitioning keys, consistency rules, and AZ/region layout in either box labels or your write-up.
4. Judge Verdict Codes
Verdicts are evaluated in priority order across 3 sample waves and 4 hidden stress/fault waves.
5. Competitive Architect Rating & Rank Tiers
Every account starts at 1200. Solving new questions with AC (or improving your best score on solved questions) increases your Architect Rating; experimenting with failed runs never penalizes your rating.