Verify seed phrase entropy
A phrase can be the right length and still be worthless if the randomness behind it was weak. This skill confirms the generator, the entropy budget, and the checksum, and rejects phrases whose source cannot be named.
Procedure
- Name the entropy source. Acceptable: an OS CSPRNG (
/dev/urandom, getrandom(2)) or a hardware TRNG. Unacceptable: time-based seeds, Math.random, or an unidentified website.
- Confirm the word count matches the claimed strength. BIP-39 word count is fully determined: 12→128 bits, 18→192, 24→256. Any 15-word phrase is not generated by a conforming tool and is suspect.
- Validate the checksum offline:
from mnemonic import Mnemonic
m = Mnemonic("english")
phrase = open("candidate.mn").read().strip()
assert len(phrase.split()) in (12, 15, 18, 21, 24)
assert m.check(phrase), "checksum invalid"
print("checksum ok, entropy bits:", 32 * len(phrase.split()) * 11 // 33)
- Confirm the wordlist is the canonical 2048-word English list, not a lookalike:
python3 -c "from mnemonic import Mnemonic; print(len(Mnemonic('english').wordlist))" → must print 2048.
- Cross-generate: run two independent generators (e.g. the
mnemonic library and a hardware wallet) and confirm neither produces a phrase you have seen documented as a known-compromised vector. Reject any phrase whose first 4 words appear in public "vanity"/profanity-generation dumps.
- Only after all checks pass, fund the wallet.
Pitfalls
- A valid checksum does not imply good entropy; attacker-chosen phrases can carry a correct checksum. Checksum checks the typo, not the randomness.
- Twelve words from a strong CSPRNG are 128 bits and are fine; the danger is weak entropy, not short length.
- Regenerating until you "like" the address biases the sample and can strip entropy. Don't reject valid phrases for aesthetics.
- Copying a phrase from a tutorial or repo means it is public; treat any published phrase as already stolen.
Verification
python3 -c "from mnemonic import Mnemonic; m=Mnemonic('english'); p=open('candidate.mn').read().strip(); print(m.check(p), len(p.split()), m.to_entropy(p)[:2].hex())"
# expect 'True' and a word count in {12,15,18,21,24}
Report the checksum result, the word count, and the named entropy source, with the command behind each.
1---2name: seed-phrase-entropy-verification3description: Use when you must prove a generated mnemonic is cryptographically sound before funding it. Checks entropy source, word count, checksum validity, and known weak-RNG fingerprints.4metadata:5 source: https://skillmd.com/skills/jiayaoqijia/seed-phrase-entropy-verification6---78# Verify seed phrase entropy910A phrase can be the right length and still be worthless if the randomness behind it was weak. This skill confirms the generator, the entropy budget, and the checksum, and rejects phrases whose source cannot be named.1112## Procedure13141. Name the entropy source. Acceptable: an OS CSPRNG (`/dev/urandom`, `getrandom(2)`) or a hardware TRNG. Unacceptable: time-based seeds, `Math.random`, or an unidentified website.152. Confirm the word count matches the claimed strength. BIP-39 word count is fully determined: 12→128 bits, 18→192, 24→256. Any 15-word phrase is not generated by a conforming tool and is suspect.163. Validate the checksum offline:17 ```python18 from mnemonic import Mnemonic19 m = Mnemonic("english")20 phrase = open("candidate.mn").read().strip()21 assert len(phrase.split()) in (12, 15, 18, 21, 24)22 assert m.check(phrase), "checksum invalid"23 print("checksum ok, entropy bits:", 32 * len(phrase.split()) * 11 // 33)24 ```254. Confirm the wordlist is the canonical 2048-word English list, not a lookalike:26 `python3 -c "from mnemonic import Mnemonic; print(len(Mnemonic('english').wordlist))"` → must print `2048`.275. Cross-generate: run two independent generators (e.g. the `mnemonic` library and a hardware wallet) and confirm neither produces a phrase you have seen documented as a known-compromised vector. Reject any phrase whose first 4 words appear in public "vanity"/profanity-generation dumps.286. Only after all checks pass, fund the wallet.2930## Pitfalls3132- A valid checksum does not imply good entropy; attacker-chosen phrases can carry a correct checksum. Checksum checks the typo, not the randomness.33- Twelve words from a strong CSPRNG are 128 bits and are fine; the danger is *weak* entropy, not short length.34- Regenerating until you "like" the address biases the sample and can strip entropy. Don't reject valid phrases for aesthetics.35- Copying a phrase from a tutorial or repo means it is public; treat any published phrase as already stolen.3637## Verification3839 python3 -c "from mnemonic import Mnemonic; m=Mnemonic('english'); p=open('candidate.mn').read().strip(); print(m.check(p), len(p.split()), m.to_entropy(p)[:2].hex())"40 # expect 'True' and a word count in {12,15,18,21,24}4142Report the checksum result, the word count, and the named entropy source, with the command behind each.