Solana Block Analysis

Ingest consecutive Solana slots over JSON-RPC, reconstruct a per-block transaction-conflict schedule, derive 1-minute and 5-minute OHLCV candles from token-balance metadata, and serve both from one process: HTTP API plus a vanilla-JS dashboard.

What it does

• Ingest — getBlock over Helius (or any JSON-RPC), client-side 10 req/s token bucket, retries, skip-slot handling, v0 / ALT account resolution.
• Contention — order-preserving schedule from account locks (heuristic; RPC does not expose the validator’s real timetable).
• OHLCV — 1m / 5m candles from preTokenBalances / postTokenBalances only. No DEX instruction decode.
• Serve — axum + DuckDB in one process. /api/health does not take the DB mutex. Dashboard is embedded in the binary (rust-embed).

Storage is DuckDB (Appender batch writes, delete-then-insert per slot). Ingest and the API share one connection behind a mutex.

Requirements

Rust 1.85+ (edition 2024). RPC URL in a gitignored .env:

HELIUS_URL=https://mainnet.helius-rpc.com/?api-key=YOUR_KEY

Copy .env.example if needed. Rate limiting is enforced in this process (ingest-core::TokenBucketLimiter), not by relying on the provider.

cargo build --release
cargo test

On Windows, if cargo test hits MSVC LNK1318, run crates one at a time or powershell -File scripts\verify.ps1.

Ingest window

Default range (finalized mainnet, 20 Aug 2026): 1,000 consecutive slots 440522383–440523382 (tip 440524383 minus 2,000 so the window stays in long-term storage).

Start	440522383
End (inclusive)	440523382
Count	1,000

Usage

PowerShell (do not use bash \ line continuation):

# ingest the default 1,000-slot window (loads .env)
powershell -File scripts\ingest.ps1

# ingest and keep the dashboard up (Ctrl+C to stop)
powershell -File scripts\ingest.ps1 -Serve

# serve an existing database
.\target\release\block-analysis.exe serve --db-path blocks.duckdb --port 8080

macOS / Linux:

cargo run --release -- ingest \
  --rpc-endpoint "$HELIUS_URL" \
  --start-slot 440522383 \
  --count 1000 \
  --rate-per-sec 10 \
  --max-concurrency 8 \
  --batch-size 25 \
  --db-path blocks.duckdb

cargo run --release -- serve --db-path blocks.duckdb --port 8080

Open http://127.0.0.1:8080. The token list is mints that already have candles. Contention defaults to slot 440522383.

A full 1,000-slot DuckDB is on the order of 9 GB and is not checked in. Rebuild it with scripts\ingest.ps1, or run cargo test without RPC. Benchmarks: FINDINGS.md. More commands: HOW_TO_TEST.md.

Architecture

cli (one binary)
  ingest ── fetch (RPC, rate limit, retry)
         ── parse (locks + schedule + token deltas)
         ── store (DuckDB Appender, batches of 25)
  serve  ── axum  /  /api/health  /api/contention  /api/tokens  /api/ohlcv

Fetch → parse → store uses a bounded mpsc (capacity 2, block on send). Parse and DB work run on spawn_blocking unless --cpu-inline.

Contention model

• Conflict: a write on an account conflicts with a later read or write on the same account. Two reads never conflict.
• Step: 1 + max(blocking step among earlier conflicting locks), or 0 if nothing blocks. Depth is 1 + max(step) (0 if no txs).
• Heuristic. The schedule preserves in-block order. It is not the validator’s actual schedule and not a graph-coloring minimum.

See contention::build_schedule. cargo test -p contention.

OHLCV model

• Intervals: 60 s and 300 s.
• Inputs: transaction metadata only (preTokenBalances / postTokenBalances).
• Price: opposing SPL mint vs wrapped-SOL (So111…112) in the same successful tx, decimal-adjusted. Volume in SOL.
• Dropped: wrap/unwrap-only, same-direction deltas (LP-like), no wSOL leg, dust below 0.0005 SOL, failed txs (meta.err).

Tests

powershell -File scripts\verify.ps1

Area	Command
Conflict detection	cargo test -p contention
v0 / ALT accounts	cargo test -p ingest-core v0_account_resolution
1m / 5m candles	cargo test -p ohlcv candle
Idempotent ingest	cargo test -p storage replacing_the_same_slot_twice_does_not_duplicate_rows
Full offline suite	cargo test or scripts\verify.ps1