dataforge-labs/bitcoin-fee-estimator-accuracy

Bitcoin fee estimator comparisons

Timestamped Bitcoin fee recommendations from public providers, paired with fee-rate percentiles from sampled transactions that were mined. The panel supports comparisons of provider recommendations across confirmation targets and network conditions.

Contents

Table	Record
bitcoin_fee_recommendations	A provider's fee recommendation at one observation time and confirmation target
bitcoin_fee_percentile_comparisons	A historical recommendation compared with sampled fee percentiles for a block

Interpreting the scores

sufficient indicates whether the recommendation met the sampled cleared_p10 fee-rate threshold. overpay_ratio divides the recommendation by that threshold. These are percentile comparisons: they do not prove that a transaction would have confirmed, or measure fees a customer could safely have avoided.

cleared_p10 is calculated from fee-observed transactions classified as standalone using their recorded ancestor count. cleared_p10_all_txs includes the wider fee-observed sample. Package structure can change after observation, and neither percentile is a miner's admission rule.

The historical matching rule selects the latest recommendation at or before the block's observation time minus target_blocks * 600 seconds. It approximates the target with elapsed time, rather than tracking the next N actual blocks. Use lead_seconds to inspect the resulting timing.

Limitations

• Provider targets and confidence levels may differ. Comparisons require care even when targets have similar labels.
• Fee coverage is partial. Check n_standalone and n_priced; the percentiles do not describe every transaction in a block.
• Polling, network delays, changing package structure and private submissions limit conclusions about confirmation.
• Ratios are undefined where the comparison fee is zero. Missing values should not be replaced with large overpayment estimates.
• These public sample tables do not contain the separate transaction-outcome observations being collected for further benchmarking.

Files and access

Data is stored as Parquet files under table_name/YYYY/MM/, with partitions for collection windows. Each measurement table has a fixed 7-day sample beginning at its configured collection start date. The sample windows in this repository span 2026-08-26 to 2026-09-05. Availability within each window depends on successful collection. The public sample dates remain fixed as additional history accumulates privately. Contact DataForge through the discussions tab to enquire about additional history.

Load a table

Install datasets and pandas to run this example. The train split contains all observations in the selected table; it is not a predefined modelling split.

from datasets import load_dataset

data = load_dataset("dataforge-labs/bitcoin-fee-estimator-accuracy",
                    "bitcoin_fee_percentile_comparisons", split="train")
df = data.to_pandas()

Coverage

collection_runs records collection windows, poll counts and failures. It is published in full and may cover dates beyond the fixed data sample. Collection gaps are not interpolated. Use this table together with measurement timestamps and error fields to assess coverage.

License and contact

The public sample is published under ODC-BY. Attribute it to "DataForge (dataforge-labs)". For questions about the data or access to additional history, open a discussion in this repository.

Earlier file paths

Each table is stored under a directory with the same descriptive name. The file contents and date partitions are unchanged. Scripts using an earlier directory name should use the corresponding table name below, or pin downloads to revision before-folder-rename-20260915 to access the original layout. Internal collector IDs are retained in the private archive and may appear in raw coverage records.

Current table and directory	Earlier directory
bitcoin_fee_recommendations	e15_fee_estimators/
bitcoin_fee_percentile_comparisons	e28_fee_estimator_accuracy/
collection_runs	e0_run_manifest/