[sparkysolutions10-pixel]

Closes #84

Summary of Changes

• Implements depth and slippage estimation logic in src/pricing/depth.ts backed by constant-product AMM curves and order-book snapshots.
• Adds calculateSlippage(spotPrice, executionPrice) with mathematical verification against known constant-product formulas:
  $$dy = \frac{y \cdot dx \cdot (1 - \text{fee})}{x + dx \cdot (1 - \text{fee})}$$
• Updates getDepth and getAMMDepth to accept network parameter and filter pool snapshots per network.
• Passes active network from /price/:assetA/:assetB/depth route in src/api/rest.ts.
• Added unit tests in src/__tests__/depth.test.ts verifying monotonic slippage increase, zero/negative size handling, and database mocking.

[Miracle656]

Merged. #84 is unassigned so this is yours, and the core of it is a real correctness fix rather than a feature.

That NOTE: getDepth reads order-book data with no network column yet — see L048 was the bug. pool_snapshots and price_points both hold rows for every network the process indexes, so before this an unfiltered read could answer a mainnet depth query using testnet reserves — and nothing in the response would have said so. Threading network through from the route closes it.

The thing I liked most is one you might not have thought of as a contribution: you deleted the block of thinking-out-loud comments in generateAMMDepthLevels that argued itself into and out of inverting the bid price —

// Wait no: actually for pair A/B, bids are orders to buy A with B...

The code was already correct (price = amountB / amountA, unchanged by this PR) and that comment was the only thing suggesting it might not be. A comment that contradicts working code is worse than no comment, because the next person trusts it and "fixes" the code. Removing it is a genuine improvement.

Also good: guarding calculateAMMPrice against a non-positive amount rather than returning NaN or Infinity into a price field, and writing each formula above its function so the constant-product arithmetic is checkable without deriving it.

One change while merging

The new getAMMDepth branched on network and kept a second, unscoped copy of the query for the falsy case:

const query = network ? `...WHERE ps.network = $2...` : `...` // no network filter
const params = network ? [pairKey, network] : [pairKey]

That branch can't be reached — network: NetworkName = activeNetwork means passing undefined takes the default, and there's no other falsy NetworkName. So it was a duplicate of the SQL that could only ever drift from the one actually in use, and it's exactly the unscoped read this PR exists to remove. I collapsed it to the single scoped query with a comment stating that the absence of an unscoped path is deliberate.

Worth keeping in mind generally: a defensive fallback for a case the type system already rules out tends to become the stale copy.

Verified on the merge result rather than your branch: tsc --noEmit silent, 885 tests across 82 suites, nothing red.

Closes #84.