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.
6. ArenaCoins Earning Rules
ArenaCoins are stored per user account in the database. Each question solve, General HLD discussion, Interview Experience, published AC solution, and Bio profile milestone is credited per user.
| Difficulty Tier | Solving (AC) | Posting AC Solution |
|---|---|---|
| Easy | +1 | +5 |
| Medium | +3 | +8 |
| Hard | +6 | +10 |
| Expert | +10 | +15 |