Inside the Neural Engine: Trading From the Source

Every prediction system inherits the weaknesses of its inputs.

Traditional traders see markets through intermediaries: exchanges, brokers, terminals, index providers, research desks and delayed disclosures. Most crypto systems reproduce the same architecture under different names. Price arrives through an exchange API. On-chain metrics arrive from an analytics vendor. Sentiment arrives through an aggregator. A model is then trained on conclusions produced by software it neither controls nor fully understands.

By the time the signal reaches the strategy, it may be several interpretations removed from the event that created it.

ADPT begins beneath that layer.

The Neural Engine operates from a Bitcoin full node that independently validates blocks and transactions against Bitcoin's consensus rules. Bitcoin Core maintains the unspent transaction output set — the UTXO set — as the current state required to validate new transactions and blocks. With archival block data and the necessary indexes, the engine can reconstruct the economic history from which that state emerged.1

This does not give ADPT perfect foresight. It gives it something more basic and more valuable:

A cryptographically verified foundation on which a genuine forecasting edge can be built.

One source of settlement truth

A full node does not know everything.

It does not know why a person moved coins. It does not inherently know that an address belongs to an exchange. It does not see private exchange order books, derivatives positioning or trades executed internally by custodians. It also does not produce a US-dollar price by itself.

What it does know is whether a Bitcoin transaction is valid, when it entered the confirmed ledger, which outputs it created and which earlier outputs it consumed.

That distinction matters.

ADPT does not treat the node as an oracle. It treats it as the root of provenance. Every higher-order metric must be traceable to validated transactions, explicit price data and disclosed modeling assumptions.

A vendor metric is a published conclusion.

A node-derived metric is a reproducible computation.

The UTXO ledger as economic memory

Bitcoin does not maintain account balances in the conventional sense. It maintains discrete unspent outputs. Each UTXO carries an amount and a creation point in the chain. When it is spent, its age, acquisition context and realized value can be reconstructed.

This gives Bitcoin an unusual property: its ledger preserves a machine-readable approximation of the market's economic memory.

The same bitcoin may be worth the same spot price wherever it sits, but its history can be radically different. One output may have moved yesterday. Another may have remained dormant for eight years. If both move today, the economic information is not equivalent.

ADPT reads that difference.

Cost basis: Realized Cap, Realized Price and MVRV

Market capitalization values the entire circulating supply at the latest spot price. That is useful, but crude. It treats long-lost coins, dormant holdings and recently traded supply as though all were acquired at the same price and remained equally active.

Realized Capitalization instead values each unit of supply at the price prevailing when it last moved. Realized Price divides that aggregate realized value by circulating supply, producing an estimate of the network-wide on-chain cost basis.2

From that foundation emerge several of Bitcoin's most widely used valuation metrics:

• MVRV compares market capitalization with realized capitalization.
• MVRV Z-Score measures the distance between market value and realized value relative to the historical volatility of market capitalization.
• STH-MVRV and LTH-MVRV isolate the embedded profitability of newer and older supply.
• UTXO Realized Price Distribution, or URPD, maps how much current supply was last created within different price bands.3

These metrics allow the engine to ask questions that price alone cannot answer:

Where is the market's cost basis concentrated? Are recent buyers underwater? Is old supply sitting on extreme unrealized profit? How much supply was accumulated near the current price? Where might capitulation, resistance or forced reassessment emerge?

Price shows where Bitcoin is trading. Realized-value structure helps reveal where the network is economically positioned.

Profitability: NUPL and supply in profit

Net Unrealized Profit/Loss, or NUPL, estimates the aggregate unrealized profitability embedded in the supply. It asks how much theoretical profit or loss would exist if coins were valued against their last-moved cost basis at the current market price.4

The engine can decompose this further:

• percentage of supply in profit;
• percentage of supply in loss;
• short-term-holder NUPL;
• long-term-holder NUPL;
• cost-basis distance from current price.

These measures help distinguish a routine decline from a structurally dangerous one.

A market in which recent buyers are slightly underwater is not the same as a market in which nearly the entire active supply is trapped below cost. Conversely, a rally in which long-term holders remain dormant is not equivalent to one accompanied by broad profit realization.

The price movement may look identical. The internal state is not.

Realized behaviour: SOPR

Unrealized profit describes what holders could realize. Spent Output Profit Ratio — SOPR — describes what moving coins actually realized.

SOPR compares the value of spent outputs when they are moved with their value when they were created:

A reading above 1 indicates that the average moved output realized a profit. A reading below 1 indicates realized loss.5

Common variants sharpen the signal:

• aSOPR removes outputs held for less than one hour, reducing noise from short-lived relay and change activity.
• STH-SOPR isolates coins associated with short-term holders.
• LTH-SOPR examines spending by long-term holders.6

SOPR gives the engine a behavioural fulcrum around 1.

In constructive regimes, realized-loss periods may repeatedly recover around that threshold as buyers absorb supply. In destructive regimes, attempts to return to profitability may fail as holders use every recovery to exit.

The metric does not predict by magic. It reveals whether profit and loss are being accepted, resisted or recycled through the market.

Holder cohorts: STH and LTH structure

Bitcoin analysis commonly separates supply into Short-Term Holder and Long-Term Holder cohorts around a 155-day age threshold. Glassnode's current implementation uses a smoothed transition centered on 155 days rather than treating the boundary as perfectly abrupt. The threshold is based on the observed decline in the probability that a coin will be spent as it ages.7

The distinction is behavioural, not moral.

Short-term supply tends to be more sensitive to price, leverage and recent narrative. Long-term supply is generally less liquid, but becomes especially informative when it begins moving in size.

ADPT can therefore monitor:

• STH and LTH realized price;
• STH-MVRV and LTH-MVRV;
• STH-SOPR and LTH-SOPR;
• supply held in profit or loss by cohort;
• long-term-holder net position change;
• movement of old supply back into younger age bands.

This helps the engine determine not only whether coins are moving, but which economic constituency is moving them.

Time as signal: HODL Waves, CDD, Dormancy and Liveliness

Bitcoin's ledger allows time itself to become a measurable variable.

HODL Waves divide supply into age bands, showing the proportion that last moved within intervals such as one day, one month, one year or several years. Realized Cap HODL Waves apply economic weighting, revealing which age groups hold the network's realized value rather than merely its nominal coin count.8

Coin Days Destroyed, or CDD, weights spending by both the number of coins moved and the time they remained dormant. Ten bitcoin held for 1,000 days destroys far more coin days than ten bitcoin held for one day.9

Related measures extract different aspects of that behaviour:

• Dormancy measures coin days destroyed relative to transfer volume, approximating the average age of coins being spent.
• Liveliness compares cumulative coin-day destruction with cumulative coin-day creation, rising when old coins become more active.
• Reserve Risk compares Bitcoin's price with the accumulated conviction expressed by holders who continue not to spend.10

Together, these metrics help distinguish superficial activity from meaningful distribution.

A billion dollars of young coins circulating among active traders does not carry the same information as a sudden revival of supply dormant since an earlier market cycle.

Network use: NVT and settlement velocity

Network Value to Transactions, or NVT, compares Bitcoin's market capitalization with the value being transferred across the network. NVT Signal reduces daily noise by using a moving average of transfer volume.11

NVT is sometimes described as a network analogue to a price-to-earnings ratio, but the analogy should not be taken literally. Transaction volume is not corporate revenue, and raw on-chain volume may contain change outputs, internal transfers and non-economic movement.

Used carefully, however, NVT can help measure whether network valuation is expanding faster than settlement activity — or whether transactional activity is strengthening beneath price.

The engine treats it as a contextual variable, not a standalone verdict.

Miners: issuance, fees and the Puell Multiple

Every newly issued bitcoin begins in a coinbase transaction. A full node verifies those outputs directly.

This allows ADPT to calculate:

• daily issuance;
• subsidy and transaction-fee revenue;
• fee share of miner revenue;
• movement of coinbase-derived supply;
• changes in hash rate and mining difficulty;
• miner revenue relative to historical baselines.

The Puell Multiple compares the daily US-dollar value of issuance with its 365-day moving average. It is commonly used to estimate whether miner revenue is unusually compressed or elevated relative to its recent history.12

The node can verify issuance and coinbase movement. Identifying a specific mining company, pool treasury or exchange destination requires additional entity attribution. That distinction remains explicit inside the system.

The mempool: pressure before settlement

The confirmed chain describes what Bitcoin accepted. A node's mempool describes the valid unconfirmed transactions that particular node currently knows about.

Bitcoin Core exposes mempool transaction count, virtual size, total fees and the minimum fee rate required for acceptance. It can also expose individual pending transactions and their dependency structures.13

This creates a near-real-time layer of network pressure:

• mempool depth;
• fee-rate distribution;
• block-space congestion;
• transaction arrival velocity;
• replacement activity;
• changes in fee pressure before confirmation.

But a mempool is not a universal global object. Nodes may receive different transactions at different times and apply different local policies. ADPT therefore treats mempool state as a local sensor — not as a perfect view of every pending Bitcoin transaction.

Where external enrichment remains necessary

Some of the most popular "on-chain" indicators are not derived from consensus data alone.

Exchange inflows, exchange outflows, exchange reserves, miner-to-exchange flows and whale ratios require analysts to identify addresses believed to belong to exchanges, miners or other entities. CryptoQuant, for example, defines exchange netflow as inflows minus outflows, while its Exchange Whale Ratio measures the share of exchange inflows represented by the ten largest deposits.14

Those are useful signals, but address attribution is probabilistic and mutable. Providers themselves note that labelled balances are estimates and may not represent an entity's complete holdings.15

ADPT therefore separates its data into two classes:

This separation prevents an estimate from being mistaken for a protocol fact.

From observation to position

The full node is not the strategy. It is the ground beneath it.

ADPT transforms that ground into decisions through four layers:

The chain determines the regime. The market determines the entry. The risk system attempts to destroy the thesis before capital is allowed to express it.

Why this can become an edge

None of these metrics is proprietary by itself. MVRV is known. SOPR is known. HODL Waves are known. Running a node is not rare.

The edge lies in the integration:

1. The raw ledger is independently verified.
2. The feature construction is controlled and versioned.
3. Every transformation has known assumptions.
4. Historical signals cannot silently change because a vendor revised its methodology.
5. The engine can combine metrics across valuation, holder behaviour, network activity and execution rather than trading any one indicator mechanically.
6. Every forecast can be audited against the exact data visible when it was made.

That last property is decisive.

A model should not be judged by how persuasive its explanation sounds after the market moves. It should be judged by what it observed beforehand, what probability it assigned, what position it took and what happened next.

What ADPT holders receive

The Node Intelligence interface is not meant to become another dashboard crowded with borrowed indicators.

It exposes the analytical state used by the engine itself:

• realized-value structure;
• short- and long-term-holder positioning;
• realized profit and loss;
• dormant-supply movement;
• miner and issuance conditions;
• network settlement pressure;
• regime scores derived from their interaction.

The utility is direct:

ADPT reads Bitcoin from the settlement layer. Members see the same instrument panel the engine uses to reason about risk.

Future research notes will publish the regime model's live scoring, document prediction failures as rigorously as successes, and show which combinations of MVRV, SOPR, NUPL, cohort cost basis, dormancy and network pressure actually survive out-of-sample testing.

The promise is not omniscience. It is provenance.