correct: Say why the entry was invalid when suggesting a repair - #101

core_descriptors(private=False) asked a named Bitcoin Core wallet for its
HD key and built public descriptors from it, but only type-checked the
seed it was given, so a wallet holding a different seed returned that
wallet's descriptors. Nothing calls it: ms32 wallet hands Core the root
xprv with addhdkey and Core builds the descriptors itself with
createwalletdescriptor, and Core's listdescriptors and
exportwatchonlywallet already provide public descriptors.

We are pre-1.0, so delete public API that no command or tool calls
instead of maintaining it:

- core_descriptors, the WalletPublicDeriver protocol, the descriptor
  checksum (DESCSUM) and record helpers, and the Core adapter's
  public_descriptors, gethdkeys and derivehdkey calls;
- generate_core_lightning_secret and CreationCeremony.core_lightning,
  which created fresh Core Lightning secrets that no command creates,
  and the CL padding check only fresh CL generation reached. Parsing,
  recovering, deriving and re-sharing an existing cl secret are
  unchanged.

The real-Core fixture and installed-wheel smoke test now check xprv and
tprv serialization through master_xprv. The package exports 22 names.
The installed package drops from 5,196 to 4,965 logical lines.

Fixes #128
Refs #63

Claude-Session: https://claude.ai/code/session_01T233rKgZqE5wzDm3EVTHL1

Fresh Core Lightning generation was removed from the public API, but the architecture, capability table, identifier policy, and security model still described it as supported. Align those contracts with the remaining behavior: existing CL secrets may still be parsed, recovered, corrected, derived, and re-shared.\n\nRefs #128

Core Lightning imports a codex32 HSM secret only when its threshold is 0
and its index is S; lightningd rejects anything else with "This is only
one share of codex32!". The cl profile still parsed shares and shared S
strings, CreationCeremony.from_secret split CL secrets, and the generic
façade recovered and derived CL shares, so every CL share this library
made or accepted was unusable in Core Lightning.

Reject a nonzero threshold in the cl profile, which closes CL recovery,
derivation and splitting at the parse boundary. The profile hook now
receives the whole header instead of only the share index. Ceremonies
accept only MasterSeed, so their profile plumbing goes away. Tests that
only needed a valid CL secret use the BIP-93 cl10peev vector; the CL
sharing fixtures and tests are removed, and new tests cover the
rejections. The docs no longer promise CL sharing.

Package size goes from 4,965 to 4,959 logical lines.

Claude-Session: https://claude.ai/code/session_01T233rKgZqE5wzDm3EVTHL1

The identifier policy said a random identifier does not make a weak
supplied seed safe. Nothing suggests it would; the sentence only invites
the question.

Claude-Session: https://claude.ai/code/session_01T233rKgZqE5wzDm3EVTHL1

The v1 rewrite dropped the license headers that master carries in
bech32.py and checksums.py. bech32.py still derives convertbits from
Pieter Wuille's segwit_addr.py, and checksums.py keeps his polymod
arithmetic and Leon Olsson Curr and Pearlwort Sneed's codex32
generators. MIT and BSD-3-Clause both require the notice to travel with
the code, and CONTRIBUTING.md requires the original author's header on
derived work.

Restore master's headers, except that checksums.py now credits the
polymod arithmetic, since it no longer holds Bech32/Bech32m constants.
Each credit also names its upstream source, sipa/bech32's
ref/python/segwit_addr.py or BIP-93, as CONTRIBUTING.md asks. Comments
do not count toward the package line budget.

Claude-Session: https://claude.ai/code/session_01T233rKgZqE5wzDm3EVTHL1

The one-shot signature in api.md omitted generate_master_seed's
fingerprint argument. A fresh call that leaves identifier unset raises
ValueError unless that callable is supplied, so the documented defaults
could not generate a seed. Show the argument and its contract.

Claude-Session: https://claude.ai/code/session_01T233rKgZqE5wzDm3EVTHL1

WorksheetCorrection had the same repr leak as CorrectionEdit: for an
erasure the worksheet filled with q, the addend is the recovered
character itself, and for any other position it is the true character
XOR the written one. Leave addend out of the generated repr; field
access, equality and the CLI's explicit display are unchanged.

Tighten the regression to compare exact reprs and to check that the
repaired payload text does not appear.

Refs #129

Claude-Session: https://claude.ai/code/session_01T233rKgZqE5wzDm3EVTHL1

The per-case CSV and kernel JSON (~9,900 lines) are generated records of
an interrupted run that promoted no timing guarantee, and nothing in the
package, tests or tools reads them. Keep the summary in
alignment-benchmarks.md as informational guidance for setting correction
limits on slower devices, drop its run-condition and source-hash details,
point to the history for the raw JSON, and stop shipping docs
*.csv/*.json in the sdist.

Claude-Session: https://claude.ai/code/session_01MDsbGqZmvVThoJrs4vzc4K

Remove code from the substitution and erasure decoder that cannot change
a result.

Every BCH correction was re-checked against the full residue. The check
cannot fail. The generator is the product of the distinct minimal
polynomials of the 8 consecutive roots. GF(32) values that clear those
syndromes also clear the conjugate roots, so the generator divides the
corrected word. The linear erasure solver already rejects inconsistent
systems. Delete _corrections_reach_target.

The linear solver ran after a failed BCH decode even with at most 8
erasures. In that case an erasure-only solution leaves the Forney
syndromes zero, so BCH would already have found it. Call the solver only
when the erasures exceed the BCH roots. The worksheet path now makes the
same choice inline, so _error_corrections is gone. Its word-length and
substitution-limit parameters were unused.

The precomputed erasure-product tables were never built. Remove that
plumbing from _locator_poly, _bch_syndrome_corrections, _repair_body and
_FixedCorrector. Also drop the no-op "except TypeError: raise" clauses.

Keep the direct small-locator solve. Replacing it with Berlekamp-Massey
for every call returned identical results but made the bounded indel
search about 15% slower, which could change truncation under the ten-
second deadline.

Validation: full pytest in normal and -O modes; differential_correction
and differential_wallet --verify; verify_correction_constants; ruff
check, ruff format --check and mypy. An offline comparison against the
previous module found identical results for random BCH syndromes,
damaged ms strings with up to 14 errors or erasures, and worksheet
residues with up to 15 erasures.

Old topic branches can retain superseded contribution rules, as happened while addressing PR #144 after #152 merged. Require agents to consult the current target-branch contributor guide and AI policy before acting on a PR.

State that the tool-neutral policy permits AI-written descriptions and review replies while keeping publication within the user-authorized scope. The maintainer explicitly approved this AGENTS.md patch.

Validation: reviewed the exact approved patch, checked referenced local paths, and ran git diff --check. Documentation only; application tests are not applicable. Refs #144 and #152.

`correct` and the "Possible correction" prompt in secret, share, wallet
and create --existing showed only the repair. After the YES gate, print
one line to stderr first, "Invalid: <reason>", using the first parse
error the caller already caught, so nothing is parsed twice. The
`reason` argument is keyword-only so it can't be swapped silently.

A mixed-case string now says codex32 strings are all uppercase or all
lowercase, and only that case alone doesn't change the encoded data: the
same entry may also hold a wrong symbol, so the reason is not a full
diagnosis. create --existing uses the same wording as the other commands.

Closes #85

Claude-Session: https://claude.ai/code/session_015CuLXqAvAovfoVcUmmogwa
Claude-Session: https://claude.ai/code/session_01Q1TNm1YeYW8K9W2eBvxmCD