Numbers from
recorded observations.
Bring coin outcomes or completed bubble timestamps. Inspect which pairs yield bits, then consume that finite record one draw at a time.
Observations stay in this tab. No uploads, observation logging or browser storage. Export a transcript to keep your record.
Record the observations
Up to 1 MiB UTF-8 and 100,000 observations.
H = 0, T = 1. Spaces are optional. Pairs 01 → 0 and 10 → 1; equal pairs are discarded. Only ASCII whitespace is accepted.
Your source label and physical-recording classification are unverified. Neither establishes entropy.
Consume the finite record
No seed expansion. No replacement bits.
- Extracted bits
- —
- Cursor
- —
- Remaining bits
- —
Canonical decimal integer, 2…9223372036854775808. Values lie in [0, upper bound). Whole out-of-range bit blocks are discarded.
The lab never invents an output when observations run out.
Start with your own observations or load a clearly predictable fixture.
Extraction cannot validate a source.
Coins: the Von Neumann rule discards equal pairs. Independent, identically distributed tosses with stable bias justify fair extracted bits; observation errors, changing bias and correlations can break that argument.
Bubbles: swapping an interval pair must be equally plausible. Pairwise exchangeability can balance an individual sign; it does not establish independence across extracted bits. Resolution, missed events and correlated intervals require separate measurement work.
Bounded draws: uniformity conditional on success needs independent fair input bits and source-independent capture and stopping. A digest, a clock, a seed or a long recording does not establish those conditions.