Methodology

This section documents how The Coin Daily's market intelligence is actually produced — the substrate it reads, the deterministic steps, the LLM steps, and the gates every published artifact passes before it ships. It exists to be checked. If a tool returns a number, this section is where you confirm what that number is, where it came from, and what it is not.

Everything the MCP surfaces is editorial market commentary — see editorial scope for what that commits to and what it explicitly doesn't. The methodology pages below are the mechanism behind that commitment.

What we publish, and what we don't

These pages document the framework, the definitions, and the gates — enough to check what any number the tools return actually is, where it came from, and what it is not. The proprietary calibration — the exact component weights, scoring constants and thresholds, and the LLM prompts — is not published. The methodology is open; the dials are not.

The end-to-end flow

The publication runs one continuous pipeline, refreshed roughly every 30 minutes per cycle. Two tracks run in parallel and meet at the editorial layer.

Track 1 — the tape. A first-party market substrate — built in-house from raw trade and L2 order-book streams, from which the publication constructs its own candles — plus derivatives positioning (funding, OI, liquidations, long/short) computed in-house and complemented with licensed cross-exchange aggregates, feeds a per-ticker analyst. For each tracked asset the pipeline:

1. computes a deterministic direction signal — a pure-Python composite over trend, structure, positioning, and critical events (no LLM, same inputs → same output);
2. derives direction (long_bias / short_bias / neutral / no_signal) and a conviction band from that signal;
3. attaches regime classification and structural levels from the substrate;
4. has an LLM write the per-ticker brief — narrating only what the tools returned, and required to be coherent with the locked signal;
5. orders the universe into a conviction-ranked watchlist (the signal book) — by how clear and confident each read is, not by trade merit or recommended action. The exact ordering criterion is defined under the universe ranking.

The signal book and the per-ticker substrate then feed the Market Analysis Report (MAR) — a single cross-sectional synthesis of what the whole board is doing: observations, tensions, tone, and a confidence the data has to earn.

Track 2 — the corpus. In parallel, an article-enrichment pipeline tracks what the news corpus is saying about each asset: mention counts, article tone, attention velocity across trailing windows, and critical-event flags on elevated negative attention. This corpus-derived signal is distinct from the tape. It drives the daily news-analysis show, surfaces through the sentiment, news, trending, and critical-event tools, and provides the news context the per-ticker briefs weave in.

The gates. Nothing ships unchecked. Per-ticker briefs pass a tape-fidelity gate and a deterministic signal-coherence check. The MAR passes a grounding gate (every number must trace to a cited metric), a significance gate (every claim must map to a scored candidate), and an adversarial semantic challenge. Anything that reaches an on-air channel additionally passes the inducement gate — the line that keeps decision-support commentary from becoming a financial promotion or personalised advice.

The three detail pages

• Market analysis

  The tape track. The first-party substrate and licensed derivatives aggregates, the deterministic direction signal, conviction, regime, structural levels, the Fibonacci lookback windows, and the MAR cross-sectional synthesis.

• News & enrichment

  The corpus track. Entity classification, attention (mention velocity) versus article-tone, the trailing windows, critical-event flagging, and sourcing back to the primary publishers.

• Editorial gates & governance

  The gates. Tape-fidelity, grounding, significance, and challenge gates on briefs and the MAR; the channel-wide inducement gate; and the decision-support — not inducement, not personalised advice — line.

How this connects to the published standards

The pipeline is the mechanism; the publication's public policies are the commitment. The two should agree, and where this section describes a gate, the corresponding policy describes the standard it enforces:

• Editorial standards — the publication's editorial floor.
• AI disclosure — how AI is used in producing the analysis and coverage.
• Corrections — how errors are handled after publication.

Refresh cadence

The pipeline refreshes on a per-cycle basis of roughly 30 minutes. Each cycle recomputes the direction signal from scratch — conviction and direction do not trend or carry forward, they are re-derived from the current substrate every cycle. A read you fetch is at most one cycle stale. See conviction & direction for why a value can move sharply between cycles and why that is the system working, not failing.