The Overview Stats Bar

At the top of the dashboard you'll find four headline numbers. These give you an instant pulse-check of the Mantle DeFi ecosystem right now:

The AI Recommendation Cards

Below the stats bar, you'll find the AI Recommendation Cards — three cards, one for each risk tier (Stable, Moderate, Aggressive). Each card shows you the AI's current top pick within that tier, including:

• The protocol and pool name — what it is and where the yield comes from
• The current APY — the annual percentage yield at the time of the last AI scoring run
• The AI's reasoning — a plain-English paragraph written by the AI explaining why this pool is the best pick right now within this risk tier, and any caveats you should be aware of
• An Invest link — takes you directly to the protocol's app where you can deploy capital
• A Simulate button — lets you open a paper trade in the Telegram bot without committing real funds

The Leaderboard Table

The main table shows every tracked pool ranked and filterable. Here's how to use it:

• Sort any column by clicking the column header — click once for ascending, again for descending
• Search by protocol name, pool name, or asset type using the search bar
• Filter by risk tier using the toggle buttons (All / Stable / Moderate / Aggressive)
• Filter by minimum TVL to exclude small pools with low liquidity
• Add to Watchlist by clicking the star icon — watchlisted pools appear in your personal watchlist tab
• Click the pool name to open its contract address on Mantlescan (the Mantle blockchain explorer)
• Click Invest to go directly to the protocol's DApp to deposit funds
• Click Simulate to run a paper trade through the Telegram bot

The APY History Charts

Below the leaderboard, you'll find interactive charts showing how the APY of each protocol has changed over the past 7 days (or 30 days — toggle between them). Select any protocol from the tabs above the chart to view its specific history. Hover over the chart to see the exact APY and timestamp for any point. These charts are useful for identifying whether a high APY is a stable trend or a temporary spike.

The On-Chain Proof Section

At the bottom of the main page (accessible via /#on-chain-proof), you'll find the complete historical log of every AI recommendation that has been committed to the Mantle blockchain. You can search this log by protocol name, pool name, contract address, or transaction hash. Each entry links directly to its Mantlescan transaction and its verification page — where you can independently confirm the recommendation was never altered.

What Information Does the AI Receive?

Every hour, after the fresh data has been fetched, the AI Scoring Engine receives a comprehensive snapshot of all currently active pools. For each pool, the AI can see:

• The current total APY and how it breaks down into base fees vs reward emissions
• The pool's TVL and whether it has been growing or shrinking
• APY trends over 1 day, 7 days, and 30 days
• Which tokens are in the pool and what rewards are offered
• The protocol name and its on-chain contract address
• Comparative data across all other pools (so it can contextualise)

The Multi-Model Cascade — Why YieldSage Never Goes Offline

YieldSage doesn't depend on a single AI provider. Instead, it uses a cascading provider system — a series of AI models from different companies, ordered by preference. When the Scoring Engine needs to generate recommendations, it tries them in this order:

1. 1
   Cerebras (Llama 3.1 70B)

   Primary pipeline — chosen for maximum speed, completing each scoring run in seconds.
2. 2
   SambaNova (Llama 3.1 405B)

   Fallback 1 — one of the largest open-source models available, with exceptional reasoning depth.
3. 3
   Groq (Llama 3.3 70B)

   Fallback 2 — ultra-low latency inference with consistently reliable outputs.
4. 4
   NVIDIA NIM (Llama 3 70B)

   Fallback 3 — enterprise-grade cloud compute with high reliability.
5. 5
   Google Gemini Flash 1.5

   Ultimate fallback — Google's proprietary model, always available as a last resort.

If the first provider is unavailable or slow, the system immediately tries the next one — automatically, without any interruption to users. This means YieldSage's recommendations keep running even if one AI provider has an outage.

What Does the AI Actually Think About?

The AI is given a structured prompt that instructs it to evaluate each pool across several dimensions. It is specifically guided to:

• Assess risk holistically — not just based on the tokens, but considering TVL depth, the proportion of reward vs fee yield, and trend stability
• Consider sustainability — a 500% APY that comes entirely from reward emissions that are dropping fast is flagged differently than a 15% APY that comes from real trading fees
• Rank within tiers, not globally — the Stable tier's top pick is the best option for risk-averse capital, not compared to Aggressive picks
• Write honest, nuanced reasoning — the AI is instructed to acknowledge risks and caveats, not just cheerfully recommend everything
• Account for comparative context — "18% APY is good" means something different if every other pool is at 30%

Dashboard AI Picks vs Personalised Alerts — What's the Difference?

YieldSage generates two types of AI output every hour:

The Problem: How Do You Know the AI Said That?

Imagine an AI system that tells you a pool is a great investment today, but then the pool performs badly. In a traditional system, the company could quietly edit or delete the recommendation — and you'd have no way to prove it ever existed. This is a fundamental trust problem. YieldSage solves it by making every recommendation impossible to alter after it's created.

Step 1 — Building the Canonical Payload

When the AI produces a recommendation, YieldSage assembles all of the important data into a single structured document called the canonical payload. "Canonical" means it's always assembled in exactly the same way — same field names, same order, same formatting rules — regardless of when or where it's created. This payload includes:

Step 2 — The SHA-256 Fingerprint

Once the payload is assembled, it's converted to a canonical JSON string — all fields sorted alphabetically, no extra spaces, UTF-8 encoded. This string is then passed through the SHA-256 hashing algorithm.

SHA-256 is a cryptographic function that takes any piece of data and produces a unique 64-character string called a hash or fingerprint. The most important property of SHA-256 is this:

The hash looks something like this: a3f6d192b5e8c041... (64 hex characters total). This is the recommendation's digital fingerprint.

Step 3 — Embedding the Hash in a Mantle Transaction

The hash is then prefixed with a YieldSage identifier: yieldsage:<hash>. This string is embedded as the data field of a transaction on the Mantle blockchain.

The transaction is a self-transfer — it sends 0 MNT from YieldSage's wallet to itself. This means no funds move, and the transaction's only purpose is to record the hash on the blockchain. Once the transaction is confirmed (which takes seconds on Mantle), the hash is permanently, publicly, and immutably recorded. The Mantle blockchain is a public ledger — anyone in the world can see this transaction, forever.

The system includes retry logic: if the first attempt to submit the transaction fails (due to network congestion or an RPC issue), it automatically retries up to 3 times with increasing delays. If the transaction still fails after all retries, the hash is still stored in the database — and a background job runs every 6 hours to retry any recommendations that haven't yet been committed to the chain.

Step 4 — Verifying in Your Own Browser

The Proof Verification page (/verify?tx=<transaction hash>) lets you independently verify any recommendation without trusting YieldSage at all. Here's what happens when you open it:

1. 1
   YieldSage fetches the original recommendation from the database

   This includes the complete canonical payload exactly as it was assembled before hashing.
2. 2
   The canonical JSON string is rebuilt locally in your browser

   The same serialisation logic runs client-side — sorted fields, no extra whitespace.
3. 3
   Your browser computes the SHA-256 hash independently

   Using the Web Crypto API — a standard browser feature. YieldSage's servers play no role in this step.
4. 4
   The computed hash is compared to the one stored in the database

   If they match: Perfect Match. The data is identical to what was originally fingerprinted. If they don't: the data has been altered.

Why This Matters

This system means that YieldSage cannot retroactively change or delete any recommendation. If it tried to alter a recommendation in the database, the hash would no longer match — and anyone who visits the verification page would immediately see "Hash Mismatch — Tampered!". The blockchain record is the source of truth, and the blockchain is controlled by nobody — it's a global public ledger that will exist for as long as Mantle Network exists.

This is what trustless verifiability means: you don't have to trust YieldSage. You can verify everything yourself.

Available Commands

The Simulate Feature — Paper Trading from the Web Dashboard

Any pool visible on the YieldSage web dashboard has a Simulate button. When you click it, a dialog appears asking how much USD you want to simulate investing. After you enter an amount and click Approve, your browser opens Telegram with a pre-filled command that looks like this:

The bot reads the pool contract address directly from this command, looks up its current APY, and records the paper trade instantly — no need to navigate any menus. The pool address is extracted from the command or from the full Mantlescan URL (either format works).

Personalised Hourly Alerts — What They Include

When alerts are enabled, you receive a message every hour that includes:

• The current best opportunity within your chosen risk tier, with the AI's reasoning
• A comparison of where you are vs where you could be (if you have active paper trades)
• Any significant APY changes across all pools since the last update
• A one-line market summary (e.g., "Stable yields rose 2% on average this hour as liquidity shifted")

The Data Source: Dune Analytics

YieldSage pulls all of its yield data from Dune Analytics — a professional on-chain data platform that allows developers to write custom SQL queries against raw blockchain data. This approach means the data comes directly from Mantle's on-chain records, not from any protocol's own self-reported API (which could be manipulated or delayed).

YieldSage maintains a custom Dune query (Query ID: 7595582) that extracts the following for every active Mantle liquidity pool: pool address, protocol name, asset pair, current APY, base APY, reward APY, TVL in USD, reward tokens, and 1D/7D/30D APY trends.

The Fetching Cycle — Every Hour, Without Fail

The fetching system is built for reliability. Every hour:

1. The scheduler validates the active API key's credit balance and selects the next available key if needed
2. The query is executed on Dune — Dune runs the SQL against the latest blockchain data
3. The system monitors the execution status and waits for completion
4. The results are downloaded as a CSV file
5. The CSV is parsed and each row is matched to an existing protocol in the database (or a new protocol is created if it's new)
6. A new yield snapshot is inserted for each pool — a timestamped record of all metrics at this exact moment
7. The active key is cycled forward to distribute load across the next scheduled run

If a query execution fails, the system does not give up. It retries the execution up to 30 times within the same trial, waiting 15 seconds between each attempt and continuously monitoring the query status until it either completes or exhausts all attempts. If all 30 attempts within a trial fail, the system resets and begins a fresh trial — up to 3 trials in total. This means the fetcher makes up to 90 total execution attempts before a fetch session is considered failed. At the start of each new trial, the system validates the active key's credit balance and rotates to the next available key if needed. If all 3 trials are exhausted, the previous data remains visible on the dashboard — it is never wiped — and the next scheduled hourly run starts the entire process fresh.

Protocol Auto-Registration

When a new protocol or pool appears in the Dune data for the first time, YieldSage automatically creates a record for it in the database. The system intelligently infers the initial risk tier: pools containing stablecoin assets (USDC, USDT, DAI, USDB, etc.) are initially classified as Stable, while everything else starts as Moderate. The AI Scoring Engine may override this classification based on further analysis.

The Database — Your Data, Always Available

All data is stored in Supabase — a managed PostgreSQL database with enterprise-grade reliability. The key tables are: