radrootsd

JSON-RPC bridge for Radroots event publishing
git clone https://radroots.dev/git/radrootsd.git
Log | Files | Refs | README | LICENSE

audits.toml (57300B)


      1 
      2 # cargo-vet audits file
      3 
      4 [criteria.network-parser-reviewed]
      5 description = "The package's untrusted network parsing, canonicalization, size limits, recursion limits, and malformed-input behavior were reviewed for fail-closed operation."
      6 implies = "safe-to-deploy"
      7 
      8 [[audits.async-wsocket]]
      9 who = "Codex (automated single-owner source review)"
     10 criteria = "network-parser-reviewed"
     11 version = "0.17.0"
     12 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete native production modules, message/socket conversions, manifest and selected SDK WebSocket SPI, plus browser actor/stream ownership. Native errors and sink/stream polling delegate to Tokio/Tungstenite; no new unsafe Send wrapper is introduced. The Radroots adapter supplies its own DNS/address/TLS policy, exact wire-identifier admission, configured connection deadline, fixed frame bounds and metered ingress, so default proxy/SOCKS dialing is not admitted. Both native loopback suites exercise the actual composition. The new browser actor uses unbounded channels; it is not admitted as the bounded Radroots browser implementation. This focused automated native-use review does not qualify those browser queues or unrestricted default dialing."
     13 
     14 [[audits.autocfg]]
     15 who = "Codex (automated single-owner source review)"
     16 criteria = "network-parser-reviewed"
     17 version = "1.5.1"
     18 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete production sources and the published 1.5.0 -> 1.5.1 delta. This build-only helper accepts trusted build-script snippets and compiler/wrapper/flag inputs, invokes Command without a shell, checks the output directory, writes compiler IR there and removes only its generated probe path. Version/IO/process failures remain errors; it has no runtime network parser. The release changes the wrapper test harness, not production probing. Current Rust 1.99 compilation executes the selected build dependency. This automated review assumes governed compiler/build inputs and is not an independent audit or permission to compile untrusted network text."
     19 
     20 [[audits.base16ct]]
     21 who = "Codex (same-owner reuse of published RDP050 source review)"
     22 criteria = "safe-to-deploy"
     23 version = "1.0.0"
     24 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the no_std feature boundary, decoded-length and decode_inner implementations. Decoding requires even input length and a sufficiently large destination, accumulates invalid-nibble errors and returns a bounded slice. Identity uses its own exact-size public hexadecimal admission and calls k256 with bytes, not this package's text API. No key generation or entropy source is selected by that call. Constant-time descriptions remain best effort, with public input length observable. This is automated single-owner source review, not an independent full-crate audit."
     25 
     26 [[audits.bech32]]
     27 who = "Codex (automated single-owner source review)"
     28 criteria = "network-parser-reviewed"
     29 version = "0.12.0"
     30 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the manifest, top-level decode entry, HRP admission, character/separator checking and checksum-to-byte-iterator path. HRPs have a fixed 83-byte buffer; invalid characters, mixed case, missing separators, insufficient checksum and code-length violations return errors before collecting decoded bytes. Allocating decoding is proportional to admitted input, so callers still need input bounds. The private native Nostr path receives bounded frames and requires canonical hexadecimal EVENT/OK identities before SDK parsing; it does not accept Bech32 aliases as wire identity. This focused automated native-use review does not qualify Bitcoin witness policies, checksum selection by arbitrary callers or unbounded NIP-19 data."
     31 
     32 [[audits.bitcoin-consensus-encoding]]
     33 who = "Codex (automated single-owner source review)"
     34 criteria = "network-parser-reviewed"
     35 version = "1.3.0"
     36 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the manifest and feature boundaries, slice decoder drivers, fixed ArrayDecoder and byte/array encoders, and complete compact-size byte admission and minimal-encoding checks. The selected native dependency has no alloc/std/serde features; bitcoin_hashes uses fixed-size array codec traits, not a Bitcoin peer or transaction parser. Fixed buffers advance by the available remaining bytes and reject incomplete completion; exact slice decoding rejects trailing input. Arbitrary user-supplied decoders must make progress, and optional allocating collection/IO drivers are outside this admitted use. This focused automated native-use review does not qualify those optional APIs or provide a general resource bound for caller-defined codecs."
     37 
     38 [[audits.bitcoin-internals]]
     39 who = "Codex (automated single-owner source review)"
     40 criteria = "network-parser-reviewed"
     41 version = "0.6.0"
     42 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed all production modules and the build script/manifest: fixed-array/slice transformations, the Copy-only ArrayVec safety boundary, bounded serde insertion, error/storage macros and integer casts. ArrayVec tracks initialized elements within static capacity; safe insertion refuses excess capacity and its raw slice view covers only initialized elements. Slice chunking checks nonzero constant chunk sizes. Some caller APIs can panic on invalid ranges/capacity, and allocating errors can retain raw input. Native hash callers use fixed block/array sizes and public adapter errors normalize upstream causes. The build script runs the governed rustc directly and writes only OUT_DIR/rust_version.rs. This automated source review is not a general unsafe-code proof or redaction guarantee for arbitrary callers."
     43 
     44 [[audits.bitcoin-internals]]
     45 who = "Codex (automated single-owner source review)"
     46 criteria = "network-parser-reviewed"
     47 delta = "0.6.0 -> 0.7.0"
     48 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete production delta and manifest: shared array/slice safety code and build script remain unchanged, input-error formatting escapes debug characters, an integer widening cast is added, and U256 uses two fixed u128 words. U256 hex/byte serde requires exactly 32 bytes; decimal parsing returns overflow/invalid-input errors, formatting uses fixed buffers, and arithmetic/division retain documented panic conditions. Test-only serde dependencies move out of the production feature surface. Current native dependencies select no optional serde/alloc features on either internals version and do not expose U256 arithmetic as Radroots agreement values. This automated delta inherits the reviewed 0.6 source and does not qualify arbitrary arithmetic operands or monetary semantics."
     49 
     50 [[audits.bitcoin_hashes]]
     51 who = "Codex (automated single-owner source review)"
     52 criteria = "network-parser-reviewed"
     53 version = "1.2.0"
     54 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the selected feature manifest, fixed-hash construction/serde/hex macros, SHA256 buffering and finalization, normal software and CPU-dispatched block-processing paths, and native call sites. Fixed hash byte decoding requires exact array length; string decoding uses the reviewed fixed-size hex decoder. SHA256 buffers at most one 64-byte block and dispatches intrinsic implementations only after CPU feature detection; their four 16-byte block loads are reached with complete blocks. Native adapter ingress is bounded before hashing and its interoperation tests compare event identities across the independent codecs on macOS and Linux. This focused automated native-use review does not independently certify the cryptographic algorithm, every hash family, batch hashing API, unlimited incremental input, or another platform."
     55 
     56 [[audits.bitflags]]
     57 who = "Codex (automated single-owner source review)"
     58 criteria = "network-parser-reviewed"
     59 version = "1.3.2"
     60 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the manifest and complete production macro expansion and attribute filtering. Generated values contain a caller-selected integer; checked from_bits rejects unknown bits, truncate masks them, and remove uses bit-clear while toggle uses XOR. The unsafe constructor preserves bits and requires the caller's documented validity invariant. There is no byte decoder, allocation, filesystem or process access. This same-version package was already locked with a safe-to-deploy exemption; its newly inherited network criterion is evaluated for the current native composition, where the optional embedded logging edge is not selected. This is automated source review, not application authorization or optional-feature execution evidence."
     61 
     62 [[audits.byteorder]]
     63 who = "Codex (automated single-owner source review)"
     64 criteria = "network-parser-reviewed"
     65 version = "1.5.0"
     66 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed feature boundaries, fixed integer reads/writes, variable-byte-length admission and slice conversion macros used by the hash32 edge. The selected no_std feature set excludes I/O extension helpers. Primitive reads use exact bounded slices and native endian conversions; invalid caller buffer lengths panic, rather than supplying partial values. This edge is compiled through heapless, while the admitted faster-hex std/alloc runtime uses its allocating implementation and no hash32 parser. This focused automated native-use review qualifies no arbitrary-length reader, dynamic allocation policy or invalid-call panic freedom."
     67 
     68 [[audits.cmov]]
     69 who = "Codex (same-owner reuse of published RDP050 source review)"
     70 criteria = "safe-to-deploy"
     71 version = "0.5.4"
     72 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed backend selection and the aarch64 CSEL, x86 CMOV and portable mask-selection implementations. Native inline assembly uses register operands with nomem/nostack; the portable backend explicitly documents compiler-dependent branch limitations. This supports fixed-width selection in the selected public-key parser. It is not proof of compiler output, a universal side-channel guarantee, or qualification of secret operations on every architecture. This is automated single-owner source review, not an independent full-crate audit."
     73 
     74 [[audits.const-oid]]
     75 who = "Codex (same-owner reuse of published RDP050 source review)"
     76 criteria = "safe-to-deploy"
     77 version = "0.10.2"
     78 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the no_std/deny-unsafe boundary and bounded ObjectIdentifier construction, fallible byte admission and default 39-byte storage. Its new inherited criterion belongs to the SEC1/DER type graph; identity supplies a fixed 33-byte compressed point and never parses an OID or private-key document. No randomness or secret owner is introduced. General ASN.1/BER admission is outside this public-key use review. This is automated single-owner source review, not an independent full-crate audit."
     79 
     80 [[audits.cpubits]]
     81 who = "Codex (same-owner reuse of published RDP050 source review)"
     82 criteria = "safe-to-deploy"
     83 version = "0.1.1"
     84 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the macro selector and target overrides. It chooses fixed 32/64-bit arithmetic backends at compile time, including the explicit wasm32 promotion to 64-bit arithmetic; no runtime feature input, entropy or key material is involved. Governed builds do not supply a cpubits override. Native and wasm32 identity checks exercise the selected configurations. This does not qualify unsupported 16-bit or arbitrary overridden configurations. This is automated single-owner source review, not an independent full-crate audit."
     85 
     86 [[audits.crypto-bigint]]
     87 who = "Codex (same-owner reuse of published RDP050 source review)"
     88 criteria = "safe-to-deploy"
     89 version = "0.7.6"
     90 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the selected fixed-width comparison, limb subtraction/word primitives, fixed safegcd inversion and its constant-time versus explicitly variable-time entry points, plus the 0.7.6 correction scope. Comparison scans the public limb count; subtraction delegates to ctutils borrow primitives; fixed inversion runs a precision-derived iteration count and returns invertibility through CtOption. The release removes the affected 0.5.5/0.7.5 selections identified by RUSTSEC-2026-0387. Identity admits public points only; retained X448 remains behind its existing adapter and independent RFC7748 vectors. This focused use review is not a complete review of the 0.5-to-0.7 rewrite, arbitrary dynamic integers, all arithmetic methods or generated-code timing proof. This is automated single-owner source review, not an independent full-crate audit."
     91 
     92 [[audits.crypto-common]]
     93 who = "Codex (same-owner reuse of published RDP050 source review)"
     94 criteria = "safe-to-deploy"
     95 version = "0.2.2"
     96 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the no_std/forbid-unsafe boundary, fixed-size key/IV traits and Generate implementations. Generation requires a caller-supplied TryCryptoRng, propagates failure and only enables ambient SysRng behind getrandom. Identity's arithmetic-only parser invokes neither generation nor key/IV initialization and adds no entropy source. This same-version inherited criterion does not certify arbitrary caller-provided RNGs or optional encryption implementations. This is automated single-owner source review, not an independent full-crate audit."
     97 
     98 [[audits.ctutils]]
     99 who = "Codex (same-owner reuse of published RDP050 source review)"
    100 criteria = "safe-to-deploy"
    101 version = "0.4.5"
    102 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete 0.4.4 -> 0.4.5 Rust/manifest delta. Comparison helpers preserve bitwise predicates, borrow subtraction uses a full-width black_box barrier, and leading/trailing-zero operations use advertised native instructions or a fixed-stage fallback. Architecture workarounds are explicitly gated; addition and subtraction return Choice carry/borrow predicates. Boundary and differential regressions are present upstream. The selected identity path processes public values; no compiler-output constant-time proof or unexecuted architecture/runtime claim is made. This is automated single-owner source review, not an independent full-crate audit."
    103 
    104 [[audits.defmt]]
    105 who = "Codex (automated single-owner source review)"
    106 criteria = "network-parser-reviewed"
    107 version = "0.3.101"
    108 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete production source and manifest. This compatibility package reexports defmt 1.x macros, traits and encoding helpers and forwards its feature flags; it has no independent build script, runtime parser or network owner. Both selected native public graphs exclude it. This automated review admits that inactive reexport package for the current composition and does not qualify embedded logging, arbitrary global logger implementations or another platform."
    109 
    110 [[audits.defmt]]
    111 who = "Codex (automated single-owner source review)"
    112 criteria = "network-parser-reviewed"
    113 version = "1.2.0"
    114 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed manifest/module boundaries, the complete build script, logger safety contract and raw-frame/rzcobs encoder ownership. The script reads the packaged linker template and writes only its generated linker script under Cargo OUT_DIR; the encoder consumes caller slices with fixed run/zero counters, not a network decoding API. Global logger thread/interrupt safety remains the implementing host's obligation. This package and build script are absent from both selected native public graphs. This focused automated review admits the inactive dependency for that composition, not embedded runtime qualification, every formatter implementation or a global logger safety proof."
    115 
    116 [[audits.defmt-macros]]
    117 who = "Codex (automated single-owner source review)"
    118 criteria = "network-parser-reviewed"
    119 version = "1.2.0"
    120 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed manifest/build entry, macro dispatch and format-string argument parsing, log expansion/code generation, symbol construction and compile-time environment filtering. The build script emits only DEFMT_LOG rerun tracking; macros accept Rust token streams and literal format strings and generate typed Rust/linker metadata. The selected native graphs exclude this package. Format argument indices and generated code are trusted compiler inputs, not admitted network text; optional logger/panic attributes still require their host safety contracts. This focused automated inactive-use review does not certify all macro expansions or bounded parsing of arbitrary untrusted Rust/format strings."
    121 
    122 [[audits.defmt-parser]]
    123 who = "Codex (automated single-owner source review)"
    124 criteria = "network-parser-reviewed"
    125 version = "1.1.0"
    126 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed all production parsing modules: format fragments, type names, display hints, array/range integers, brace admission and argument consistency. Parsing follows UTF-8 character boundaries and reports malformed types/ranges/braces. Argument indices control vector resizing and index-plus-one arithmetic, so this parser is unsuitable for untrusted network input without separate input/index bounds. The only relevant dependency route is optional compile-time defmt macros with trusted literals, and both selected native graphs exclude it. This automated inactive-use review supplies no runtime parser/resource qualification or permission to route network strings through this API."
    127 
    128 [[audits.der]]
    129 who = "Codex (same-owner reuse of published RDP050 source review)"
    130 criteria = "safe-to-deploy"
    131 version = "0.8.2"
    132 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the feature boundary, Reader completion and SliceReader construction, nested limits, checked cursor advancement and failed-state handling. These bounds reject incomplete/trailing input rather than returning partial documents. Identity passes SEC1 point bytes and invokes no DER/private-key decoder; this same-version dependency criterion is scoped to that inactive decoding surface in the identity composition. It does not qualify arbitrary recursive documents, BER mode, encrypted keys or all workspace consumers. This is automated single-owner source review, not an independent full-crate audit."
    133 
    134 [[audits.elliptic-curve]]
    135 who = "Codex (same-owner reuse of published RDP050 source review)"
    136 criteria = "safe-to-deploy"
    137 version = "0.14.1"
    138 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete rc.35 -> 0.14.1 delta, checked public-key SEC1 admission and curve byte-order helpers. PublicKey construction decodes the point and rejects the identity; the selected compressed representation has a fixed size. The stable API uses explicit curve endianness rather than the former encoding trait, and distinguishes variable-time arithmetic entry points. Identity enables arithmetic/SEC1 only and invokes no secret generation, signing or PKCS8 decoder. This is a focused selected-use review, not blanket certification of every optional algorithm or secret lifecycle. This is automated single-owner source review, not an independent full-crate audit."
    139 
    140 [[audits.faster-hex]]
    141 who = "Codex (automated single-owner source review)"
    142 criteria = "network-parser-reviewed"
    143 version = "0.10.1"
    144 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed all production encoder, decoder, dispatch, error, serde and build-probe source and the actual Nostr fixed-array call sites. CPU dispatch checks x86 processor/OS state or aarch64 NEON support; vector reads and writes require complete source and destination chunks, with scalar tails. Encoding returns only its initialized ASCII prefix. Checked decoding rejects odd/nonhex input but permits a longer even source than the destination: it is not an exact-identifier validator. The bounded native adapter therefore checks canonical exact-length EVENT and OK identifiers before upstream normalization; real socket regressions retain the previously observed failures. Allocating helpers assume caller-sized inputs, and the admitted messages retain the adapter's frame/ingress limits. This automated native-use review does not claim general constant-time behavior, unrestricted allocation safety, optional embedded logging or browser qualification."
    145 
    146 [[audits.ff]]
    147 who = "Codex (same-owner reuse of published RDP050 source review)"
    148 criteria = "safe-to-deploy"
    149 version = "0.14.0"
    150 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the no_std/forbid-unsafe feature boundary and field trait definitions, random generation/error propagation, CtOption inversion/square-root contracts and constant-time versus variable-time method distinctions. Concrete curve implementations own arithmetic; identity's checked parser selects the reviewed k256 field implementation and calls no random sampler. This same-version dependency review does not certify arbitrary third-party Field implementations, derive-generated fields or every exponentiation caller. This is automated single-owner source review, not an independent full-crate audit."
    151 
    152 [[audits.group]]
    153 who = "Codex (same-owner reuse of published RDP050 source review)"
    154 criteria = "safe-to-deploy"
    155 version = "0.14.0"
    156 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the no_std feature boundary, group/curve/scalar traits, random-generation propagation and fixed input/output length requirement for batch normalization. The package supplies interfaces, with arithmetic delegated to concrete curve implementations. Identity's public-point parser neither samples a random group element nor invokes its scalar-multiplication/wNAF helpers. This same-version inherited criterion is limited to that composition; it does not qualify arbitrary group implementations or secret scalar algorithms. This is automated single-owner source review, not an independent full-crate audit."
    157 
    158 [[audits.hash32]]
    159 who = "Codex (automated single-owner source review)"
    160 criteria = "network-parser-reviewed"
    161 version = "0.3.1"
    162 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed all three production modules and the manifest: trait forwarding, stateless hasher construction, FNV wrapping accumulation and the Murmur3 block/tail implementation. This is a non-cryptographic hash dependency, not an event-identity or signature algorithm. The only selected reverse edge is heapless through faster-hex; the native faster-hex std/alloc implementation invokes neither those collections nor either hasher. Murmur3's potentially uninitialized buffer references are outside the admitted native call path and require separate review/proving inputs before any use. This automated use-scoped review does not certify that unsafe implementation, adversarial collision resistance or an embedded runtime."
    163 
    164 [[audits.heapless]]
    165 who = "Codex (automated single-owner source review)"
    166 criteria = "network-parser-reviewed"
    167 version = "0.8.0"
    168 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed module/feature and build-probe boundaries plus fixed-capacity Vec/String insertion, initialization and UTF-8 conversion paths referenced by faster-hex's non-alloc alternative. Checked push/extend refuse excess capacity and strings preserve UTF-8 after safe insertion; unsafe constructors retain caller initialization/encoding requirements. The build script compiles a fixed trusted probe under OUT_DIR with governed compiler/wrapper arguments and owns no runtime network parser. The only selected reverse edge is faster-hex, whose actual std/alloc implementation uses alloc Vec/String instead; optional embedded logging and non-alloc containers are not invoked by this native composition. This focused automated review does not qualify every pool/queue/map API or the hash32 Murmur3 implementation."
    169 
    170 [[audits.hex-conservative]]
    171 who = "Codex (automated single-owner source review)"
    172 criteria = "network-parser-reviewed"
    173 version = "1.3.0"
    174 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed manifest/feature boundaries, fixed/variable decoding entries, complete hex iterators and their initialized-output handling, Char representation/table conversion and the bounded buffer encoder. Fixed-array decoding checks exactly N*2 bytes before writing; invalid/odd/nonhex input returns typed errors, and decoded Vec length is published only after initialization succeeds. Unchecked ASCII conversions originate from the closed ASCII Char variants. Allocating helpers scale with caller input, so they are not standalone network size limits. The native graph supplies bounded frames and fixed-size public hash values. This focused automated source review does not qualify unbounded formatting/collection by arbitrary callers or every optional feature."
    175 
    176 [[audits.hybrid-array]]
    177 who = "Codex (same-owner reuse of published RDP050 source review)"
    178 criteria = "safe-to-deploy"
    179 version = "0.4.15"
    180 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete 0.4.12 -> 0.4.15 production/manifest delta and fixed-size array feature boundary. The delta adds supported type-level sizes, including retained SNTRUP buffers, without changing arithmetic, allocation or reference-conversion bodies. Arrays remain statically sized and the existing zeroize forwarding is preserved. The newly inherited crypto criterion covers these fixed buffers in the identity composition, not every unsafe reference conversion, optional serialization implementation or arbitrary array-size implementation. This is automated single-owner source review, not an independent full-crate audit."
    181 
    182 [[audits.k256]]
    183 who = "Codex (same-owner reuse of published RDP050 source review)"
    184 criteria = "safe-to-deploy"
    185 version = "0.14.0"
    186 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the arithmetic-only manifest selection, checked SEC1/public-key route, affine decompression and the selected field parsing, normalization, fixed square-root addition chain and 5x52 multiplication/reduction. Decoding rejects noncanonical field elements, verifies the square root, chooses the encoded parity and rejects the identity through PublicKey. Identity supplies only 32 public x-coordinate bytes with a fixed even compressed prefix; no private scalar, signer, ECDSA, Schnorr, entropy acquisition or PKCS8 API is introduced. Native/no_std/wasm checks and independent valid/invalid field-boundary vectors pass. This focused review does not certify all of k256 0.14, secret scalar multiplication or every architecture backend. This is automated single-owner source review, not an independent full-crate audit."
    187 
    188 [[audits.lru]]
    189 who = "Codex (automated single-owner source review)"
    190 criteria = "safe-to-deploy"
    191 delta = "0.16.3 -> 0.16.4"
    192 notes = "Automated single-owner review of the complete checksum-verified published 0.16.3 to 0.16.4 production and manifest delta. Four with-key insertion variants pass a borrowed key to their initializer before node replacement; existing no-key APIs forward through closures. Map/list ownership and raw pointer accesses are unchanged. Fallible initializers return before mutation and cache hits never invoke the initializer. Added regressions and unchanged dependencies/license are reviewed. This bridges the existing 0.16.3 exemption to the separately reused 0.16.4 to 0.18.5 review. It does not qualify either old version as free of known unsoundness; the direct legacy SDK and old LRU remain explicit 070/130 migration debt. No independent full-crate or unsafe-code proof is claimed."
    193 
    194 [[audits.lru]]
    195 who = "Codex (automated single-owner source review)"
    196 criteria = "network-parser-reviewed"
    197 delta = "0.16.4 -> 0.18.5"
    198 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete published release delta: the returned mutable reference now borrows the cache, pop detaches before a potentially panicking key destructor, allocation replaces unchecked NonNull construction, and sparse/find-and-promote/retain preserve list ownership. Reviewed the borrowed return and added retention, promotion and destructor regressions. Native SDK callers retain bounded capacities and immutable EventId keys. This delta inherits the existing 0.16.4 exemption; it is not an independent full-crate audit or a general unsafe-code proof."
    199 
    200 [[audits.negentropy]]
    201 who = "Codex (automated single-owner source review)"
    202 criteria = "safe-to-deploy"
    203 delta = "0.5.0 -> 0.5.2"
    204 notes = "Automated single-owner review of the complete checksum-verified published 0.5.0 -> 0.5.2 delta. All 21 retained current source/support hashes match the previously reviewed 0.5.2 snapshot. Reviewed fixed-array ID copying, modulo-2^256 accumulator carry propagation, equivalent duplicate sorting, bounded ID-list admission before allocation, private always-Ok varint cleanup, and retained regressions. Both manifests declare MIT; 0.5.2 adds the MIT license text absent from the 0.5.0 archive. No new dependency or default feature is introduced. This delta relies on the existing 0.5.0 baseline exemption; it does not claim an independent full-crate audit, full parser canonicality or release qualification."
    205 
    206 [[audits.opaquerr]]
    207 who = "Codex (automated single-owner source review)"
    208 criteria = "network-parser-reviewed"
    209 version = "0.2.0"
    210 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed both complete production modules and the manifest. The no_std error container and declarative macros construct typed kinds and optional boxed sources; they do not decode network input, invoke processes, access files or use unsafe code. Allocating construction caches the source Display text and Debug exposes the source, so this dependency supplies neither redaction nor an independent allocation limit. The native adapter maps upstream causes into its existing typed public errors, with ingress bounded before parsing. This is automated source review for that use, not a secret-handling or unrestricted-error-formatting claim."
    211 
    212 [[audits.primefield]]
    213 who = "Codex (same-owner reuse of published RDP050 source review)"
    214 criteria = "safe-to-deploy"
    215 version = "0.14.0"
    216 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the no_std/forbid-unsafe exports, fixed-size Montgomery field admission and explicit endian conversion traits. Canonical bytes are copied into a fixed integer representation according to trusted field parameters; concrete arithmetic delegates to the selected crypto-bigint implementation. K256 identity decompression uses its own concrete base-field implementation and introduces no generic-field random generation. This selected-use review does not certify arbitrary field parameters, every Montgomery operation or a new secret-key protocol. This is automated single-owner source review, not an independent full-crate audit."
    217 
    218 [[audits.primeorder]]
    219 who = "Codex (same-owner reuse of published RDP050 source review)"
    220 criteria = "safe-to-deploy"
    221 version = "0.14.0"
    222 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the generic curve-parameter boundary and affine-point implementation, including checked coordinate construction, curve-equation verification, identity state and conditional selection. K256 uses its own affine/field implementation for identity decompression; its new generic arithmetic support is not a new Radroots secret-key or signer boundary. Optional precomputed tables, arbitrary curves and variable-time multiplication are outside this focused public-parser use review. This is automated single-owner source review, not an independent full-crate audit."
    223 
    224 [[audits.rand]]
    225 who = "Codex (automated single-owner source review)"
    226 criteria = "network-parser-reviewed"
    227 version = "0.10.1"
    228 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the feature manifest, RNG extension interface and integer uniform sampling used by the selected SDK. Its actual features select std/alloc/sys_rng and the getrandom SysRng reexport; software StdRng/ThreadRng and their optional ChaCha generator are not selected. The SDK uses fixed ping bytes and a constant nonempty bounded reconnect-jitter range, with no network-controlled sampling range or collection allocation. Infallible wrappers can panic on OS entropy failure and invalid caller ranges; this review does not claim those are impossible. This focused automated native-use review is not a full distribution/cryptographic algorithm audit or another platform's runtime qualification."
    229 
    230 [[audits.rand]]
    231 who = "Codex (automated single-owner source review)"
    232 criteria = "safe-to-deploy"
    233 delta = "0.10.1 -> 0.10.3"
    234 notes = "Automated single-owner review of the complete checksum-verified published 0.10.1 -> 0.10.3 production/manifest delta. Core RNG/SysRng and fixed-byte filling remain unchanged. Checked Unicode and slice access replace unchecked access; sampler deserialization validates ranges. Weighted constructors/updates return overflow before mutation, float-inclusive range rounding handles finite overflow and NaN masks, and weighted sampling rejects non-finite progress. The private native SDK uses fixed arrays and a trusted bounded integer jitter range, not weighted, float or serde samplers. This delta inherits the exact reused Lib 0.10.1 source review; it adds no exemption and does not independently audit cryptographic algorithms, every optional feature or entropy-failure panic freedom."
    235 
    236 [[audits.rand_core]]
    237 who = "Codex (automated single-owner source review)"
    238 criteria = "network-parser-reviewed"
    239 version = "0.10.1"
    240 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the TryRng/Rng and crypto-marker trait interfaces, dereference forwarding, complete UnwrapErr adapter and integer/byte conversion helpers used by SysRng. These interfaces own no network decoder or variable-length packet queue; they forward bounded caller buffers and propagate fallible completion. The explicit UnwrapErr conversion panics on provider errors, and a crypto marker alone proves no algorithm properties. The selected native SDK supplies fixed-size arrays and trusted sampling bounds. This focused automated source review does not qualify optional block/seed generators, arbitrary provider implementations or secret generation policy."
    241 
    242 [[audits.sec1]]
    243 who = "Codex (same-owner reuse of published RDP050 source review)"
    244 criteria = "safe-to-deploy"
    245 version = "0.8.1"
    246 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed fixed-size point representation and from_bytes admission: a valid tag determines one exact length before copying into the statically sized buffer. The identity path always supplies 33 bytes with compressed-even tag; the concrete curve then checks field membership and curve validity. DER/private-key and optional serde paths are not called by identity. This same-version selected-use review does not treat SEC1 syntax alone as curve validation or certify unrelated key-document APIs. This is automated single-owner source review, not an independent full-crate audit."
    247 
    248 [[audits.secp256k1]]
    249 who = "Codex (automated single-owner source review)"
    250 criteria = "network-parser-reviewed"
    251 delta = "0.29.1 -> 0.30.0"
    252 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete published Rust release delta and manifest. Fixed-array key/signature constructors reject incorrect slice lengths before FFI; recovery IDs become a closed checked enum. Schnorr signing and verification now pass an explicit message slice length through the existing C backend, and the admitted Nostr path supplies a fixed 32-byte event digest. Secret Debug output remains hidden or an explicit feature-gated fingerprint, and removed no-context parsing paths no longer panic unconditionally. The resolved secp256k1-sys 0.10.1 source is unchanged. This automated delta review inherits the existing 0.29.1 network-parser exemption and does not claim a fresh C/cryptographic algorithm audit, arbitrary-message signing policy, or another platform's runtime qualification."
    253 
    254 [[audits.syn]]
    255 who = "Codex (automated single-owner source review)"
    256 criteria = "network-parser-reviewed"
    257 version = "3.0.2"
    258 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed package/features, parser entry/completion checks and token-buffer construction/cursor ownership, tracing the optional defmt procedural-macro route. It parses compiler-owned token streams and rejects unconsumed tokens; recursive token construction and syntax trees do not supply a network-input allocation/depth limit. Version 3.0.2 is absent from both selected native public graphs, including build dependencies. This focused automated review admits that inactive compile-time dependency for the current composition; it does not audit every Rust grammar/unsafe cursor operation or qualify an untrusted runtime source parser."
    259 
    260 [[audits.syn]]
    261 who = "Codex (automated single-owner source review)"
    262 criteria = "safe-to-deploy"
    263 delta = "3.0.2 -> 3.0.9"
    264 notes = "Automated single-owner review of the complete checksum-verified published 3.0.2 to 3.0.9 production and manifest delta. Parser changes distinguish grouped lifetimes and double-colon labels, reject ambiguous arrow operators, admit explicit safe foreign signatures and preserve literal error spans; retain uses existing safe Vec/Option traversal. Other changes document verbatim grammar and add regressions. The only new route is optional defmt macro syntax over compiler-owned tokens; actual native normal/build discovery excludes Syn 3.0.9. This delta inherits the exact Lib 3.0.2 use-scoped source review and adds no exemption. It does not audit every grammar/unsafe cursor, qualify arbitrary runtime source input or execute the optional macro feature."
    265 
    266 [[audits.tokio-happy-eyeballs]]
    267 who = "Codex (automated single-owner source review)"
    268 criteria = "network-parser-reviewed"
    269 version = "0.1.0"
    270 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the entire unsafe-free source and manifest, including empty resolution, IPv4/IPv6 partitioning, the 250 ms race, error completion and remaining-address iteration. The helper extracts the prior Async WebSocket dialing algorithm; it owns no protocol decoder, filesystem or credential input. Its caller owns total dialing deadlines and destination admission. The configured Radroots native connector bypasses this default dialer and enforces its own bounded DNS/address/TLS admission and deadline. This automated native-use review does not qualify unrestricted default dialing or another platform."
    271 
    272 [[audits.tokio-tungstenite]]
    273 who = "Codex (automated single-owner source review)"
    274 criteria = "network-parser-reviewed"
    275 delta = "0.26.2 -> 0.28.0"
    276 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete published release delta: into-inner/get-ref accessors, boxed TLS error conversion, the matching Tungstenite parent and example error adaptation. The native connector retains its verified TLS roots, original server name, pinned addresses and bounded frame configuration; real socket, exact-byte, cancellation, authentication and deadline controls execute separately. This automated delta review inherits the existing 0.26.2 exemption and does not claim an independent full-crate audit or browser/runtime-platform qualification."
    277 
    278 [[audits.tungstenite]]
    279 who = "Codex (automated single-owner source review)"
    280 criteria = "network-parser-reviewed"
    281 delta = "0.26.2 -> 0.28.0"
    282 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed the complete published release delta: boxed errors, HTTP-version admission and duplicate-header handling, frame buffer reservation/read bounds, borrowed UTF-8 hashing, fragment/control/close state transitions and TLS error conversion. The native Radroots connector supplies fixed message/frame limits and metered ingress before this decoder. Real malformed, oversized, continuation, exact-byte and deadline cases execute separately. This automated delta review inherits the existing 0.26.2 exemption; it is not an independent full-crate audit, unbounded-input guarantee or 32-bit/browser qualification."
    283 
    284 [[audits.typenum]]
    285 who = "Codex (automated single-owner type-level dependency review)"
    286 criteria = "safe-to-deploy"
    287 version = "1.20.1"
    288 notes = "Reviewed the no_std/forbid-unsafe source boundary, marker/operator exports and const-generics selection. hybrid-array0.4.15 requires typenum1.20; the provider already selects this exact version/checksum. Its selected use fixes trusted array dimensions at compile time; identity invokes neither a runtime number parser, secret-key owner nor entropy source in this package. Current native and wasm32 compilation and public field-boundary cases exercise those fixed representations. This focused use review is not a full-crate independent audit, arithmetic-algorithm certification or approval of arbitrary type-level implementations."
    289 
    290 [[audits.universal-time]]
    291 who = "Codex (automated single-owner source review)"
    292 criteria = "network-parser-reviewed"
    293 version = "0.3.1"
    294 notes = "Reused exact published Lib 3af7814ea4f76f2cca3d33d4c2bd8a854f3fb78f source review for the identical registry version/checksum and same private native transport composition. Rand 0.10.1 is the reviewed baseline for a separately recorded 0.10.3 delta. Original use restrictions remain binding; this is not a second independent audit. Reviewed all production source, feature gates and manifest: native std time uses a OnceLock monotonic origin and explicit wall-clock conversion; checked arithmetic returns Option and duration subtraction saturates, while duration operators can panic on overflow/underflow. There is no network parser or variable-sized input allocation. Native SDK connection durations come from validated adapter configuration, and packet timestamps do not supply Instant arithmetic. The no_std/WASM provider is a separately linked Sync static and is not admitted as the Radroots browser clock. This automated native-use source review is not independent audit, arbitrary-duration panic freedom or browser/no_std runtime qualification."
    295 
    296 [[audits.web-time]]
    297 who = "Codex (automated single-owner source review)"
    298 criteria = "safe-to-deploy"
    299 version = "1.1.0"
    300 notes = "Reused exact published RHI fa61581f3aaa6e6714288ba06e7e94f5b6a43b01 review for the identical archive and native std-clock composition. Original restrictions remain binding; this is not a second independent audit. Automated single-owner review of all seven production Rust modules, both manifests and exact registry archive SHA-256 5a6580f308b1fad9207618087a65c04e7a10bc77e02c8e84e9b00dd4b12fa0bb. The current daemon macOS/Linux composition selects only the cfg-gated std::time reexport; no build script, filesystem/network access, unsafe code or independent native clock implementation is present. Browser modules use fixed Duration arithmetic and thread-local Performance/Date bindings; missing Performance, negative wall time and unchecked duration operators can panic, and host clocks are not trusted authorization time. Optional Serde checks numeric epoch-offset overflow and duplicate/missing/unknown fields, but the caller must bound deserializer input and normalize errors. The daemon does not select those browser/Serde modules. This audit admits the exact native daemon dependency use; it is not an independent review, browser clock/resource qualification, a guarantee about hostile JavaScript globals or permission to deserialize unbounded network input."
    301 
    302 [[audits.wnaf]]
    303 who = "Codex (same-owner reuse of published RDP050 source review)"
    304 criteria = "safe-to-deploy"
    305 version = "0.14.1"
    306 notes = "Reused exact published Lib e55c1749902811732167a2ce5c099e22e964624f source review for the identical registry version/checksum and the same identity public-point parser composition. Provider crypto-reviewed implies the required safe-to-deploy criterion. Original use restrictions remain binding; this is not a second independent audit. Reviewed the no_std/forbid-unsafe feature boundary and bounded-window representation constructor. Representation generation has value-dependent control flow and must not be treated as a general constant-time secret-scalar primitive. Identity only invokes checked public-point decompression, which does not use wNAF generation or scalar multiplication. The package's inherited criterion is scoped to that unused arithmetic surface in the identity composition, with no new signer or private scalar API admitted. This is automated single-owner source review, not an independent full-crate audit."
    307 
    308 [[audits.zeroize]]
    309 who = "Codex (automated single-owner source review)"
    310 criteria = "safe-to-deploy"
    311 delta = "1.8.2 -> 1.9.0"
    312 notes = "Automated single-owner review of the complete checksum-verified published 1.8.2 -> 1.9.0 production and manifest delta, new barrier/stack modules, x86 delta and retained daemon/provider callers. Initialized fixed arrays and owned Vec/String Drop paths retain volatile erasure; native optimization barriers now use readonly inline assembly with the referenced address. Zeroizing gains transparent unsized support; daemon retains initialized owned buffers. The new opt-in stack API is not called. Edition/MSRV and unchanged dual MIT/Apache licenses are checked. This delta inherits the existing 1.8.2 safe-to-deploy baseline exemption; it does not prove erasure of prior copies, constant-time behavior, every backend, or an independent full-crate audit."