Educational example · No investment advice

Evaluating Evidence Under Uncertainty: A Digital Asset Example

AmberJun · Educational example · No investment advice

Educational use only

This page demonstrates a method and provides no investment advice or price forecast. Digital asset prices are highly uncertain, and historical data does not predict future results. Readers must verify any judgments independently and take responsibility for their decisions.

Bitcoin prices are difficult to predict accurately. When information is incomplete, it helps to define what the analysis must deliver, record conditions and evidence separately, and check whether someone else can reproduce the judgment. This example applies three working principles: value, handoffs and evidence.

1. Value: define what the review needs

Start by identifying who will use the judgment and what decision it must support. Long-term allocation research may need to revisit an assumption. Short-term research may need to identify when an observation no longer holds. The two uses call for different observation windows, update frequencies and evidence.

In this example, the deliverable is a one-page note for a risk reviewer. It says which scenario the verified conditions fit, what data is missing and when to review it again. The reviewer uses it to decide whether to request more research. The note meets the acceptance standard if the reviewer can trace the sources, recompute the conditions as of the cutoff, and explain the classification.

2. Handoffs: make each condition testable

The windows and thresholds below are teaching assumptions. Agree on a cutoff time, time zone, source and rule version before reviewing any data. First check completeness and consistency. Then mark each condition as met, not met or pending verification.

  • ETF flows: use Farside's US spot Bitcoin ETF totals to check for a net inflow on each of the three most recent completed US trading days before the cutoff. Mark the condition as pending if complete data for any of those days has yet to be published.
  • Short-term holder cost: compare the BTC spot reference price with Glassnode's Short-Term Holder Realized Price at the same completed daily observation point, using a consistent data source. The illustrative condition is met when the reference price is at or above that estimate. The metric estimates on-chain cost basis.
  • Macro proxy: use FRED's WALCL values available at the cutoff to check whether the latest weekly Federal Reserve total assets figure exceeds the preceding week. This is one proxy for the exercise. Other factors also affect market liquidity.

At handoff, record separately whether the data and calculations pass review, and whether the next reviewer has the access, versions, samples and assigned responsibility needed to begin. An accepted deliverable may still have to wait for the next stage.

3. Evidence: keep four kinds of records

  • Execution: source, observation time, retrieval time, time zone, unchanged data snapshot and processing steps.
  • Acceptance: agreed thresholds, rule version, individual checks, missing data and reviewer.
  • Uptake: who received the note, how they actually used it and when.
  • Objective attainment: whether the reviewer completed the check by the agreed time and could trace the basis for the decision. This evaluates the review objective.

Ask someone who did not make the original judgment to repeat it independently using the same snapshot, cutoff and rules. Check whether they can reconstruct each condition and the scenario classification. If they need an explanation from the original analyst, document what was missing and repeat the test. Verifiable evidence of reuse is needed before treating the method as a transferable capability.

4. Classify scenarios by their conditions

These are teaching categories. Classify a scenario only after all three conditions have been checked.

  • Bear case: none of the conditions are met. Record the evidence for each unmet condition.
  • Base case: the signals are mixed. Record where they disagree and when to check again. For example, if flows and the macro proxy meet the rules but the cost condition does not, classify it as the base case.
  • Bull case: all three conditions are met. Record the evidence and what would invalidate each condition.

If any condition is pending, leave the scenario unclassified and identify the missing data and who will obtain it. If a condition or source version changes, record the change, keep the prior judgment and recalculate using the same rules. The example uses no live data, price targets or scenario probabilities.

Back to the three working principles

Value defines the result needed. Handoffs separate conditions that can be tested individually. Evidence records the process and the outcome separately. Together, they make it possible for a method to move from individual experience to a capability others can use. The same logic applies when rebuilding enterprise operating capability with AI: start with one real workflow and use evidence to test whether the method can be reused.

Related reading

Back to the three principles: value, handoffs and evidence

Start with one workflow: the five-step consultation process

Original research: Rebuilding Enterprise Operating Capability with AI, V2.0 (Chinese)

Example source: Farside US Bitcoin ETF flows

Metric definition: Glassnode Short-Term Holder Realized Price

Example source: FRED Federal Reserve total assets (WALCL)

All research publications