synthCore
AIA market-data and paper-strategy research project. The frontend is retired until there is a substantially stronger replacement.
Successful failure
synthCore successfully failed. It established useful market-data, storage, candle-building, and paper-research foundations while also showing that the public frontend and available DEX coverage were not strong enough for the next build.
The work is not over. More DEX data and a stronger public inspection surface remain active needs; retiring the frontend does not turn TBMC into the whole project or close the underlying research.
Project boundary
synthCore joined four related but separate concerns: normalized market data, feed adapters, local storage, and paper-only strategy inspection. The public frontend was one view over that work; it was not the project by itself.
The repository contains no live-order path, custody path, or wallet execution path. Paper results were research outputs, not performance promises.
System map
| layer | implemented role |
|---|---|
| core | Canonical assets and pairs, source identity, normalized trades and quotes, deterministic interval buckets, OHLCV and VWAP, source-contribution metadata, and data-quality flags. |
| feeds | Adapter and runtime work for Binance, Bybit, Coinbase, Hyperliquid, Kraken, OKX, and Uniswap V3. An adapter's presence does not claim that its feed is currently live or healthy. |
| storage | SQLite WAL storage for candles and source health, plus daily append-only zstd JSONL archives for raw events. |
| control | Shared paper accounting and Donchian, low-volatility, and basis-arbitrage probes. |
| TBMC | Paper-only thesis and component comparisons across 1m, 5m, 15m, and 1h intervals, with starved-data rows identified rather than silently ranked. |
Data contract
synthCore normalized raw-ish trade, quote, and swap events before constructing its own candles. Venue candles could be used as a sanity control, but were not treated as the canonical market record.
- Pair, source, instrument kind, interval, source contribution, and data-quality state remained attached to the result.
- Spot, linear perpetual, inverse perpetual, and onchain pool observations were not silently merged.
- Native pairs were preserved. Derived views belonged above ingestion and retained their provenance.
- USD, USDC, USDT, and DAI quotes remained distinct observations rather than one collapsed asset.
That last rule describes market observation, not portfolio allocation. Retaining a quote source for research did not require taking direct exposure to its quote asset.
CEX and DEX boundary
Centralized and decentralized venue data informed the same market research without pretending every observed price was directly executable. Native pairs, derived views, spot instruments, perpetual instruments, and onchain pools were kept as distinct inputs.
Paper research layer
The control layer applied paper-only probes to the self-built candles and used shared accounting so comparisons could be inspected under the same rules. TBMC decomposed broader theses into components instead of hiding every decision inside a single score.
TBMC was one research surface over synthCore's data foundation, not the scope of synthCore itself.
The current TBMC variants were long-only research probes. Their tables were not trading systems, live strategies, or evidence that a result would survive fees, slippage, latency, changing liquidity, or future market conditions.
Review record
A read-only review on May 24, 2026 identified five material risks: late events could replace a candle; live arrival order could diverge from deterministic replay; directional accounting ignored quantity; missing-data and exposure behavior in a basket was unrealistic; and one JSON path escaped data manually.
A June 1 follow-up ran the existing suites successfully and found the system legitimate for research and inspection. It also found the public ranking language too authoritative for the maturity of the strategy probes. Passing the existing suites did not erase the earlier risks or turn an internal review into independent assurance.
Why the frontend was retired
The available frontend is not strong enough to keep as the public market surface. Retiring it removes a weak interface without discarding the underlying work or presenting FerroCore as a finished replacement.
What carries forward
The work on normalized events, self-built candles, CEX–DEX orderflow analysis, source health, data-quality boundaries, append-only archives, and paper comparison remains part of the project record. The current direction is the coherent conversion of the indicator suite and a narrower, execution-aware FerroCore design.
This page is a synopsis, not a restored synthCore application.