btclib-org/btclib

What is wrong The two ratio tests in tests/script_engine/tapscript_scaling_test.py time five runs of the small script, then five runs of the large one, and compare the best of each. If the machine's load rises between the two blocks, only the large runs slow down, and the ratio goes past the limit of 10 although the code is linear. Where At origin/main 4c81666: test_four_times_the_script_takes_about_four_times_as_long test_four_times_the_signature_checks_take_about_four_times_as_long Both use min(... large ...) / min(... small ...) with time.perf_counter(). How it is known A local run of the whole suite on a shared 10-core Mac failed both tests: "14.1 times as long" and "10.9 times as long". The 1-minute load was 19.31 at the start of that run and 113.84 at its end. The next whole-suite run passed, at load 17.02 at the start and 80.46 at the end. The branch under test did not touch the script engine. Six runs of only these tests, with -n 0 -p no:randomly -k times_as_long, at loads be

Where. src/btclib/block/build.py:73 @ 341b061. Flipping the default to False leaves the whole suite green; with True, extra_nonce=b"x"*200 raises "Invalid coinbase script size". Why it matters. ASSURANCE_CASE.md (Fail-safe defaults) says tests/check_validity_test.py asserts the True default; build_coinbase is outside its walk. Fixed means. A test calling build_coinbase without check_validity and an oversize extra_nonce expects the refusal (as tests/block/block_test.py:270 does for BlockHeader). Found in the security review of 2026-09-30 (@pmazzocchi), commit 341b061.

Where (ASSURANCE_CASE.md @ 341b061). :46 — "os is there for os.environ alone": nothing under src/ imports os; the census script does not list it. Likewise :206 "the one environment variable btclib reads" — btclib_ecc reads it, btclib reads none. :51 — magic.py "takes three names" of bitcoin_core_rpc: it uses two, chain_from_network and magic_from_chain (src/btclib/p2p/magic.py:80-81). :214-217 — "The curve arithmetic does not import the schemes … tests/imports_test.py holds each of those edges": that file has no curve/scheme test; the arithmetic is btclib_ecc's. Why it matters. The assurance case is read as evidence; each sentence should match what the census and the tests show. Fixed means. The three sentences are corrected (the layering edge pointed at btclib_ecc's test or dropped). Found in the security review of 2026-09-30 (@pmazzocchi), commit 341b061.

Where. tests/script/sig_hash_taproot_test.py:364-370 @ 341b061 checks output_prvkey only for the input with no merkle root; the comment says the other six "are covered by the signature above", but that signature is made with the vector's own tweakedPrivkey, not btclib's. Why it matters. ASSURANCE_CASE.md claims the answers agree with the vectors the BIPs publish. btclib is correct today (7/7 on both arms, checked during the review); the suite would not notice if it stopped being. Fixed means. Assert output_prvkey_from_merkle_root(internalPrivkey, merkleRoot or b"") == tweakedPrivkey for all seven, and correct the comment. Found in the security review of 2026-09-30 (@pmazzocchi), commit 341b061.

Where. SECURITY.md, Limitations: "ellswift.xdh (src/btclib/ecc/ellswift.py:117) is the one of them that returns octets". At 341b061 line 117 is a comment; the delegated return is line 113, the Python return line 120. Why it matters. The security policy's citations are the evidence a reader checks. tests/security_citations_test.py passes because this citation is anchored only by the dotted name, so any line inside xdh satisfies it. Fixed means. Cite line 113 with a quoted fragment, so the citation test checks the exact line. Found in the security review of 2026-09-30 (@pmazzocchi), commit 341b061.