rttp 1.2.8__tar.gz → 1.3.2__tar.gz

{rttp-1.2.8 → rttp-1.3.2}/CHANGELOG.md

@@ -1,238 +1,287 @@
-# Changelog -- `rttp` (Python)
-
-All notable changes to the `rttp` package. Format follows
-[Keep a Changelog](https://keepachangelog.com/); versioning is
-[Semantic Versioning](https://semver.org/).
-
----
-
-## [1.2.8] -- 2026-09-21
-
-**Version-only re-publish.** The code, the vector file and every self-test result
-are identical to `1.2.7`; only the version string moves, so the final PyPI release
-carries the stack number. `1.2.7` stays installable -- pin `1.2.8` for new work.
-
-* No wire-format, URI-grammar, public-API or vector change.
-* `python -m rttp.selftest` still reports 83 checks in full, and 80 passed with
-  3 skipped on a zero-dependency install.
-
-## [1.2.7] -- 2026-09-21
-
-**Conformance surface grown; wire format untouched.** The shipped
-`rttp/vectors/rttp-conformance-v1.2.6.json` is now 13,460 bytes
-(`sha256 4b743e63...`) and the self-test replays every added case: **83 checks**
-in full, and **80 passed with 3 skipped** on a zero-dependency install -- a check
-that cannot be performed is reported as skipped, never as passed.
-
-### Added
-
-* **Nine fail-closed frame cases, one positive frame.** Each of the nine frames
-  must be rejected; the positive frame must be accepted with an empty action.
-* **The published Ed25519 envelope vector**, replayed three ways: the canonical
-  signing input byte for byte, `AID == SHA-256(public key)`, and the signature
-  itself (the third one skips without the optional primitive).
-
-### Changed
-
-* **Skipped checks are now counted and printed**, so a default install can no
-  longer look like a full pass.
-* Per-section counts are reported: 38 conformance, 16 envelope rules, 22 envelope
-  flow, 2 RFC 8032, 5 zero-dependency.
-
-### Note
-
-* PyPI and npm carry `1.2.7`; crates.io carries the same code as the pre-release
-  `1.2.8-alpha`. The registry version strings differ by publishing policy, not by
-  code -- the vectors and their `sha256` are identical everywhere.
-
-## [1.2.6] -- 2026-09-19
-
-**Version alignment -- no code change.** Every module, byte layout and
-conformance vector is identical to 0.1.2, and the self-test still reports the
-the same 53 checks (full set, i.e. with the optional Ed25519 backend). Only the version string and the maturity classifier change.
-
-### Changed
-
-* **The version number now mirrors the stack version.** PyPI and npm carry
-  `1.2.6`, so a single number identifies this code on every distribution
-  channel; crates.io carries the same code as the pre-release `1.2.6-alpha`.
-  The old `0.1.x` line was a package-maturity number that corresponded to
-  nothing in the specification, and it made the same package look like two
-  different releases on two registries.
-* **`Development Status` classifier: `3 - Alpha` -> `4 - Beta`.** A
-  non-pre-release version must not ship with an "Alpha" classifier; leaving
-  the two in conflict invites the question "which one is true?".
-* Install command note: on crates.io every published version of this crate is
-  a pre-release, so Cargo will not select it from a plain `rttp` requirement.
-  Write `cargo add rttp@1.2.6-alpha` (or pin `rttp = "1.2.6-alpha"`).
-
-## [0.1.2] -- 2026-09-18
-
-**Metadata correction -- no code change.** Every module, byte layout and
-conformance vector is identical to 0.1.1, and the self-test still reports the
-the same 53 checks (full set, i.e. with the optional Ed25519 backend).
-
-### Fixed
-
-* **The distribution identity now matches the IANA registration.** The package
-  was authored as `RTTP Working Group` -- a name that appears in no filing. The
-  `rttp` URI-scheme registration on file with IANA (ticket `[IANA #1459939]`)
-  names **`RTTP.COM Organization`** as the `Author/Change controller`, and every
-  other public surface uses that same name. Both package manifests now say it
-  too:
-
-  | Field | Before | After |
-  |:---|:---|:---|
-  | `pyproject.toml` -> `authors` (PyPI) | `RTTP Working Group` | **`RTTP.COM Organization`** |
-  | `package.json` -> `author` (npm) | `RTTP Working Group` | **`RTTP.COM Organization`** |
-
-  Package metadata cannot be edited after publication, so 0.1.0 and 0.1.1 keep
-  the old string inside their own metadata; this is the version that carries the
-  correction. Project pages display the latest version, so the visible author
-  name becomes correct once this release is out.
-
-## [0.1.1] -- 2026-09-18
-
-**Documentation only -- no code change.** Every module, byte layout and
-conformance vector is identical to 0.1.0, and the self-test still reports the
-the same 53 checks (full set, i.e. with the optional Ed25519 backend).
-
-### Fixed
-
-* **A distribution claim that was wrong, not merely outdated.** The 0.1.0 entry
-  below said the JavaScript side "will publish under the `@rttp` scope". That
-  scope **cannot be created at all**: npm already has a personal account named
-  `rttp`, and an organization name may not collide with a user name. The
-  JavaScript side is published as **`@aicent/rttp`** -- this project's own
-  organization, with the package name itself still just `rttp`.
-* **README distribution table** now names the npm package exactly
-  (`@aicent/rttp`), and gives `pip install rttp` for this package, matching the
-  table in the JavaScript package's README.
-
-### Distribution status at this release
-
-| Registry | Name | State |
-|:---|:---|:---|
-| PyPI | `rttp` | this package -- `pip install rttp` |
-| npm | `@aicent/rttp` | **published 0.1.1** -- 2026-09-18 |
-| crates.io | `rttp` | held by this project |
-
-## [0.1.0] -- 2026-09-17
-
-First release. Scope is deliberately narrow: **framing, addressing and sealing**,
-with zero dependencies and no transport.
-
-### Added
-
-* **`rttp.pulse_header`** -- `PulseHeader128` codec for the 128-byte
-  hardware-aligned frame header (RFC-002 sec. 4.1), including the v1.2.6 extension
-  block:
-  * `SPEC_REV` `0x66` - `FLAGS` `0x67` - `ACTION_LEN` `0x68` - `ACTION` `0x69`(16B)
-    - `RESERVED` `0x79`(7B)
-  * bytes `0x00`-`0x65` are **not touched**, `VERSION_ID` stays 130 -- readers
-    that only understand the original layout remain conformant
-  * `build()` - `build_for_uri()` - `parse()` - `verify()` - `to_hex()` /
-    `from_hex()`
-  * strict verification (R1-R7): unknown `SPEC_REV` rejected, `SPEC_REV=0` with a
-    non-zero block rejected, non-zero `RESERVED` rejected, unknown `FLAGS` bits
-    rejected, `ACTION` padding must be zero, `ACTION_LEN` must agree with the
-    payload
-* **`rttp.rttp_uri`** -- validation, canonicalisation and `ROUTE_SHARD`
-  derivation for `rttp://` URIs (RFC-002 sec. 10.2 / sec. 10.3):
-  * `ROUTE_SHARD = SHA-256(ASCII(canonical_authority))[0:16]` -- deterministic,
-    pure computation, **no DNS** (sec. 10.5)
-  * hash-intent (`8lowhex`) and name-intent forms; open-set `action` vocabulary
-  * fail-closed rejection of uppercase, `userinfo`, `port`, `query`,
-    `fragment`, multi-segment paths, wrong segment counts, leading/trailing `-`,
-    whitespace, and non-`rttp` schemes
-  * accepts both `rttp://` and `web+rttp://` -- the second as a **tolerance** for
-    the browser protocol-handler form, which RFC-002 no longer defines
-    (2026-09-17). Not a conformance claim; see "Known gaps" below.
-* **`rttp.seal`** -- managed profile: HMAC-SHA256 over a pre-shared key table,
-  bidirectional switch<->agent signing, `ts|nonce` with a 120-second skew window.
-* **`rttp.seal_asym`** -- **sovereign profile**: Ed25519 seals with
-  `AID = SHA-256(public_key)`.
-  * self-certifying: the envelope carries the public key, so a verifier
-    recomputes the identity instead of trusting a claim -- **no issuance, no
-    registry, no directory**
-  * `ts` and `nonce` are **inside** the signed input, so they cannot be rewritten
-    without breaking the signature
-  * freshness enforced by default (`|now - ts| <= 120 s`); archival verification
-    available explicitly via `check_freshness=False`
-  * lowercase-hex only -- a case variant is rejected, not normalised
-  * `generate_keypair()` - `seal()` - `unseal()` - `verify_envelope()` -
-    `verify_claims()` - `public_bundle()` - `signing_input()`
-  * requires the `ed25519` extra; without it every call raises `SealError` with
-    the install instruction rather than degrading to an unsigned path
-* **`rttp.aid`** -- Autonomous Identity derivation (RFC-001), shared by both
-  profiles so an AID means one thing across the stack.
-* **`rttp.selftest`** -- `python -m rttp.selftest`, **53 checks with the optional Ed25519 backend (29 by default)**:
-  * 25 conformance vectors (6 positive URIs - 15 rejections - 3 deterministic
-    frames - 1 `SPEC_REV=0` compatibility vector)
-  * RFC 8032 sec. 7.1 TEST 1 / TEST 2, proving the signature backend is standard
-    Ed25519 -- so a failure identifies *which* half broke
-  * 22 envelope checks, including a 14-case tamper matrix in which every single
-    alteration must be rejected, and an explicit replay-window test
-  * 4 static-scan checks proving the core modules import nothing outside the
-    standard library
-  * `--vectors` prints the deterministic envelope vector for the specification;
-    `--quiet` prints the summary line only
-* **`rttp/vectors/rttp-conformance-v1.2.6.json`** -- the published vectors, shipped
-  as package data so the self-test runs from an installed wheel, offline.
-* Console entry point `rttp-selftest`.
-* **`MANIFEST.in`** -- the source distribution carries `README.md`,
-  `CHANGELOG.md`, `LICENSE`, `pyproject.toml` and the published vectors, and
-  excludes virtual environments and build noise. Both the wheel and the sdist
-  pass `twine check`.
-
-### Notes
-
-* **Zero dependencies** in the core, and that is a machine-checked claim rather
-  than a promise -- see the static-scan section of the self-test.
-* The reference implementation evaluates the
-  **`iqa`** namespace nowhere: this package is the `rttp` protocol only.
-* ~~Distribution: PyPI `rttp`. The npm name `rttp` is held by an unrelated,
-  unmaintained project, so the JavaScript side will publish under the `@rttp`
-  scope.~~ -- **corrected 2026-09-18.** PyPI `rttp` is now **published** (0.1.0,
-  2026-09-17). The npm name `rttp` is indeed held by an unrelated, unmaintained
-  project -- but the **`@rttp` scope cannot be created at all**, because a
-  personal account already holds that name. The JavaScript side is therefore
-  published as **`@aicent/rttp`**: this project's own organization, with the
-  package name still just `rttp`.
-
-### Known gaps
-
-Tracked in `SPEC/RTTP-SEAL-ENVELOPE-v1.2.6.md` sec. 8 and sec. 10. Deliberately listed
-here so nobody has to discover them:
-
-* No transport and no client. The core encodes, addresses and seals; sending is a
-  separate package's job.
-* No revocation, and no key rotation story. Both belong to RFC-003.
-* No replay **cache**: freshness (sec. 3.4) bounds the window, but within it a replayed
-  envelope still verifies unless the verifier keeps a short-lived nonce store.
-* No transport-origin binding: the seal proves who signed, not which connection
-  delivered it.
-* Rust parity: the published `rttp` crate has no transport layer and no envelope.
-  Python and Node are the conforming implementations so far.
-* ~~the browser-registration prefix is accepted because RFC-002 sec. 10.6 requires
-  it...~~ -- **superseded 2026-09-17.** That form was removed from both
-  specifications and will never be submitted to IANA. `rttp_uri` still accepts it
-  as a **tolerance** -- a browser hands the handler that string, and refusing it
-  would break the protocol-handler path -- but the specification defines no such
-  form, so the acceptance is **outside the specification** and a conforming
-  third-party implementation is not required to reproduce it.
-* ~~**Open, and ours to fix:** the published vectors asserted things the
-  specification no longer said.~~ -- **resolved 2026-09-18.** The generator now
-  classifies every forward case explicitly:
-  * `positive_uris` -- each entry carries `tolerance` (boolean) and `note`. The
-    case whose `uri` carries the browser-registration prefix is `tolerance: true`,
-    with its reason recorded: it is accepted only because that is the scheme name
-    under which a handler is registered (RFC-002 sec. 10.6) -- never because the
-    specification defines it. It is **not** a conformance claim, and the 25-check
-    total is unchanged.
-  * `negative_uris` -- the two case-clause citations were corrected from
-    `(RFC-002 sec. 10.2)` to **`sec. 10.3`**, the clause that actually carries the
-    lowercase rule. sec. 10.2 now appears nowhere in the vector set.
-  Regenerated (`SPEC/tools/conformance.py --generate`), replayed (`--check`,
-  25/25), and mirrored byte-identically into both packages. After the change:
-  Python self-test 53/53, Node runner 25/25.
+# Changelog -- `rttp` (Python)
+
+All notable changes to the `rttp` package. Format follows
+[Keep a Changelog](https://keepachangelog.com/); versioning is
+[Semantic Versioning](https://semver.org/).
+
+---
+
+## [1.3.2] -- 2026-11-09
+
+**Vector re-anchoring release -- no code change.** The published conformance
+vectors are re-anchored to `draft-li-rttp-iqa-addressing-02` (TXT sha256
+`02491c5a...`, 40 pp., 2026-11-01): every `applies_to`, `spec_ref`, `reason`
+and `expected` field now cites the I-D sections directly (`§§4, 7.1` set-level;
+per-row `§4.1` / `§4.2` / `§4.5.2` / `§7.1.1` / `§4.3`; the seal envelope stays
+under `AICENT-002 (RTTP-SEAL-ENVELOPE-v1.2.6 sec. 7)`, beyond -02). Vector
+bytes are unchanged -- only anchor metadata and citation text moved; every
+replay is byte-identical.
+
+* **New vector-set digest:** `sha256 8640e390...` (13,874 bytes, was
+  `4b743e63...` at 13,460 bytes). Shipped byte-identically in all three
+  packages; manifest regenerated.
+* **Documented annotations:** the `-lead` rejection vector carries a CONFLICT
+  note against the -02 ABNF (the I-D ABNF literally allows a leading hyphen;
+  the implementation is stricter and -03 will tighten it); the JS metadata
+  test now asserts the re-anchored `applies_to`.
+* **Docs:** specification status updated to revision -02 with the TXT digest;
+  conformance records now point at https://iqa.org/conformance/ and
+  https://iqa.org/challenge/ .
+* Self-test counts unchanged: **80 passed with 3 skipped** (83 with
+  `[ed25519]`); node --test all pass; the independent JS replay stays at 35.
+
+---
+
+## [1.3.1] -- 2026-10-05
+
+**Synchronization release -- no code change.** The codec, the pulse frame, the
+seal envelopes, the conformance vector file and every self-test result are
+identical to `1.2.8` (80 passed with 3 skipped; 83 with the `[ed25519]`
+extra). What moved:
+
+### Changed
+
+* **Version alignment:** the PyPI, npm and crates.io lines of the dual pillars
+  now all read 1.3.1. This release rides the crates.io side's first formal
+  (non-pre-release) publication of the `rttp` crate.
+* **Series designation:** `RFC-002` is now `AICENT-002` throughout the docs,
+  docstrings and error messages, aligning with the public specification line
+  (`rttp.com/AICENT-002/`). Historical entries below keep the designation they
+  were written with. The shipped conformance vector file is byte-frozen (its
+  sha256 is published) and still carries the old designation inside its labels;
+  renaming those labels is a coordinated vector-set regeneration, not an edit.
+* **Specification status refreshed:** the `rttp` URI scheme **is registered** --
+  IANA "URI Schemes" registry, Provisional, CRI 27 (2026-09-22) -- and the
+  combined Internet-Draft `draft-li-rttp-iqa-addressing` (both schemes, one
+  document) is under IETF review.
+
+## [1.2.8] -- 2026-09-21
+
+**Version-only re-publish.** The code, the vector file and every self-test result
+are identical to `1.2.7`; only the version string moves, so the final PyPI release
+carries the stack number. `1.2.7` stays installable -- pin `1.2.8` for new work.
+
+* No wire-format, URI-grammar, public-API or vector change.
+* `python -m rttp.selftest` still reports 83 checks in full, and 80 passed with
+  3 skipped on a zero-dependency install.
+
+## [1.2.7] -- 2026-09-21
+
+**Conformance surface grown; wire format untouched.** The shipped
+`rttp/vectors/rttp-conformance-v1.2.6.json` is now 13,460 bytes
+(`sha256 4b743e63...`) and the self-test replays every added case: **83 checks**
+in full, and **80 passed with 3 skipped** on a zero-dependency install -- a check
+that cannot be performed is reported as skipped, never as passed.
+
+### Added
+
+* **Nine fail-closed frame cases, one positive frame.** Each of the nine frames
+  must be rejected; the positive frame must be accepted with an empty action.
+* **The published Ed25519 envelope vector**, replayed three ways: the canonical
+  signing input byte for byte, `AID == SHA-256(public key)`, and the signature
+  itself (the third one skips without the optional primitive).
+
+### Changed
+
+* **Skipped checks are now counted and printed**, so a default install can no
+  longer look like a full pass.
+* Per-section counts are reported: 38 conformance, 16 envelope rules, 22 envelope
+  flow, 2 RFC 8032, 5 zero-dependency.
+
+### Note
+
+* PyPI and npm carry `1.2.7`; crates.io carries the same code as the pre-release
+  `1.2.8-alpha`. The registry version strings differ by publishing policy, not by
+  code -- the vectors and their `sha256` are identical everywhere.
+
+## [1.2.6] -- 2026-09-19
+
+**Version alignment -- no code change.** Every module, byte layout and
+conformance vector is identical to 0.1.2, and the self-test still reports the
+the same 53 checks (full set, i.e. with the optional Ed25519 backend). Only the version string and the maturity classifier change.
+
+### Changed
+
+* **The version number now mirrors the stack version.** PyPI and npm carry
+  `1.2.6`, so a single number identifies this code on every distribution
+  channel; crates.io carries the same code as the pre-release `1.2.6-alpha`.
+  The old `0.1.x` line was a package-maturity number that corresponded to
+  nothing in the specification, and it made the same package look like two
+  different releases on two registries.
+* **`Development Status` classifier: `3 - Alpha` -> `4 - Beta`.** A
+  non-pre-release version must not ship with an "Alpha" classifier; leaving
+  the two in conflict invites the question "which one is true?".
+* Install command note: on crates.io every published version of this crate is
+  a pre-release, so Cargo will not select it from a plain `rttp` requirement.
+  Write `cargo add rttp@1.2.6-alpha` (or pin `rttp = "1.2.6-alpha"`).
+
+## [0.1.2] -- 2026-09-18
+
+**Metadata correction -- no code change.** Every module, byte layout and
+conformance vector is identical to 0.1.1, and the self-test still reports the
+the same 53 checks (full set, i.e. with the optional Ed25519 backend).
+
+### Fixed
+
+* **The distribution identity now matches the IANA registration.** The package
+  was authored as `RTTP Working Group` -- a name that appears in no filing. The
+  `rttp` URI-scheme registration on file with IANA (ticket `[IANA #1459939]`)
+  names **`RTTP.COM Organization`** as the `Author/Change controller`, and every
+  other public surface uses that same name. Both package manifests now say it
+  too:
+
+  | Field | Before | After |
+  |:---|:---|:---|
+  | `pyproject.toml` -> `authors` (PyPI) | `RTTP Working Group` | **`RTTP.COM Organization`** |
+  | `package.json` -> `author` (npm) | `RTTP Working Group` | **`RTTP.COM Organization`** |
+
+  Package metadata cannot be edited after publication, so 0.1.0 and 0.1.1 keep
+  the old string inside their own metadata; this is the version that carries the
+  correction. Project pages display the latest version, so the visible author
+  name becomes correct once this release is out.
+
+## [0.1.1] -- 2026-09-18
+
+**Documentation only -- no code change.** Every module, byte layout and
+conformance vector is identical to 0.1.0, and the self-test still reports the
+the same 53 checks (full set, i.e. with the optional Ed25519 backend).
+
+### Fixed
+
+* **A distribution claim that was wrong, not merely outdated.** The 0.1.0 entry
+  below said the JavaScript side "will publish under the `@rttp` scope". That
+  scope **cannot be created at all**: npm already has a personal account named
+  `rttp`, and an organization name may not collide with a user name. The
+  JavaScript side is published as **`@aicent/rttp`** -- this project's own
+  organization, with the package name itself still just `rttp`.
+* **README distribution table** now names the npm package exactly
+  (`@aicent/rttp`), and gives `pip install rttp` for this package, matching the
+  table in the JavaScript package's README.
+
+### Distribution status at this release
+
+| Registry | Name | State |
+|:---|:---|:---|
+| PyPI | `rttp` | this package -- `pip install rttp` |
+| npm | `@aicent/rttp` | **published 0.1.1** -- 2026-09-18 |
+| crates.io | `rttp` | held by this project |
+
+## [0.1.0] -- 2026-09-17
+
+First release. Scope is deliberately narrow: **framing, addressing and sealing**,
+with zero dependencies and no transport.
+
+### Added
+
+* **`rttp.pulse_header`** -- `PulseHeader128` codec for the 128-byte
+  hardware-aligned frame header (RFC-002 sec. 4.1), including the v1.2.6 extension
+  block:
+  * `SPEC_REV` `0x66` - `FLAGS` `0x67` - `ACTION_LEN` `0x68` - `ACTION` `0x69`(16B)
+    - `RESERVED` `0x79`(7B)
+  * bytes `0x00`-`0x65` are **not touched**, `VERSION_ID` stays 130 -- readers
+    that only understand the original layout remain conformant
+  * `build()` - `build_for_uri()` - `parse()` - `verify()` - `to_hex()` /
+    `from_hex()`
+  * strict verification (R1-R7): unknown `SPEC_REV` rejected, `SPEC_REV=0` with a
+    non-zero block rejected, non-zero `RESERVED` rejected, unknown `FLAGS` bits
+    rejected, `ACTION` padding must be zero, `ACTION_LEN` must agree with the
+    payload
+* **`rttp.rttp_uri`** -- validation, canonicalisation and `ROUTE_SHARD`
+  derivation for `rttp://` URIs (RFC-002 sec. 10.2 / sec. 10.3):
+  * `ROUTE_SHARD = SHA-256(ASCII(canonical_authority))[0:16]` -- deterministic,
+    pure computation, **no DNS** (sec. 10.5)
+  * hash-intent (`8lowhex`) and name-intent forms; open-set `action` vocabulary
+  * fail-closed rejection of uppercase, `userinfo`, `port`, `query`,
+    `fragment`, multi-segment paths, wrong segment counts, leading/trailing `-`,
+    whitespace, and non-`rttp` schemes
+  * accepts both `rttp://` and `web+rttp://` -- the second as a **tolerance** for
+    the browser protocol-handler form, which RFC-002 no longer defines
+    (2026-09-17). Not a conformance claim; see "Known gaps" below.
+* **`rttp.seal`** -- managed profile: HMAC-SHA256 over a pre-shared key table,
+  bidirectional switch<->agent signing, `ts|nonce` with a 120-second skew window.
+* **`rttp.seal_asym`** -- **sovereign profile**: Ed25519 seals with
+  `AID = SHA-256(public_key)`.
+  * self-certifying: the envelope carries the public key, so a verifier
+    recomputes the identity instead of trusting a claim -- **no issuance, no
+    registry, no directory**
+  * `ts` and `nonce` are **inside** the signed input, so they cannot be rewritten
+    without breaking the signature
+  * freshness enforced by default (`|now - ts| <= 120 s`); archival verification
+    available explicitly via `check_freshness=False`
+  * lowercase-hex only -- a case variant is rejected, not normalised
+  * `generate_keypair()` - `seal()` - `unseal()` - `verify_envelope()` -
+    `verify_claims()` - `public_bundle()` - `signing_input()`
+  * requires the `ed25519` extra; without it every call raises `SealError` with
+    the install instruction rather than degrading to an unsigned path
+* **`rttp.aid`** -- Autonomous Identity derivation (RFC-001), shared by both
+  profiles so an AID means one thing across the stack.
+* **`rttp.selftest`** -- `python -m rttp.selftest`, **53 checks with the optional Ed25519 backend (29 by default)**:
+  * 25 conformance vectors (6 positive URIs - 15 rejections - 3 deterministic
+    frames - 1 `SPEC_REV=0` compatibility vector)
+  * RFC 8032 sec. 7.1 TEST 1 / TEST 2, proving the signature backend is standard
+    Ed25519 -- so a failure identifies *which* half broke
+  * 22 envelope checks, including a 14-case tamper matrix in which every single
+    alteration must be rejected, and an explicit replay-window test
+  * 4 static-scan checks proving the core modules import nothing outside the
+    standard library
+  * `--vectors` prints the deterministic envelope vector for the specification;
+    `--quiet` prints the summary line only
+* **`rttp/vectors/rttp-conformance-v1.2.6.json`** -- the published vectors, shipped
+  as package data so the self-test runs from an installed wheel, offline.
+* Console entry point `rttp-selftest`.
+* **`MANIFEST.in`** -- the source distribution carries `README.md`,
+  `CHANGELOG.md`, `LICENSE`, `pyproject.toml` and the published vectors, and
+  excludes virtual environments and build noise. Both the wheel and the sdist
+  pass `twine check`.
+
+### Notes
+
+* **Zero dependencies** in the core, and that is a machine-checked claim rather
+  than a promise -- see the static-scan section of the self-test.
+* The reference implementation evaluates the
+  **`iqa`** namespace nowhere: this package is the `rttp` protocol only.
+* ~~Distribution: PyPI `rttp`. The npm name `rttp` is held by an unrelated,
+  unmaintained project, so the JavaScript side will publish under the `@rttp`
+  scope.~~ -- **corrected 2026-09-18.** PyPI `rttp` is now **published** (0.1.0,
+  2026-09-17). The npm name `rttp` is indeed held by an unrelated, unmaintained
+  project -- but the **`@rttp` scope cannot be created at all**, because a
+  personal account already holds that name. The JavaScript side is therefore
+  published as **`@aicent/rttp`**: this project's own organization, with the
+  package name still just `rttp`.
+
+### Known gaps
+
+Tracked in `SPEC/RTTP-SEAL-ENVELOPE-v1.2.6.md` sec. 8 and sec. 10. Deliberately listed
+here so nobody has to discover them:
+
+* No transport and no client. The core encodes, addresses and seals; sending is a
+  separate package's job.
+* No revocation, and no key rotation story. Both belong to RFC-003.
+* No replay **cache**: freshness (sec. 3.4) bounds the window, but within it a replayed
+  envelope still verifies unless the verifier keeps a short-lived nonce store.
+* No transport-origin binding: the seal proves who signed, not which connection
+  delivered it.
+* Rust parity: the published `rttp` crate has no transport layer and no envelope.
+  Python and Node are the conforming implementations so far.
+* ~~the browser-registration prefix is accepted because RFC-002 sec. 10.6 requires
+  it...~~ -- **superseded 2026-09-17.** That form was removed from both
+  specifications and will never be submitted to IANA. `rttp_uri` still accepts it
+  as a **tolerance** -- a browser hands the handler that string, and refusing it
+  would break the protocol-handler path -- but the specification defines no such
+  form, so the acceptance is **outside the specification** and a conforming
+  third-party implementation is not required to reproduce it.
+* ~~**Open, and ours to fix:** the published vectors asserted things the
+  specification no longer said.~~ -- **resolved 2026-09-18.** The generator now
+  classifies every forward case explicitly:
+  * `positive_uris` -- each entry carries `tolerance` (boolean) and `note`. The
+    case whose `uri` carries the browser-registration prefix is `tolerance: true`,
+    with its reason recorded: it is accepted only because that is the scheme name
+    under which a handler is registered (RFC-002 sec. 10.6) -- never because the
+    specification defines it. It is **not** a conformance claim, and the 25-check
+    total is unchanged.
+  * `negative_uris` -- the two case-clause citations were corrected from
+    `(RFC-002 sec. 10.2)` to **`sec. 10.3`**, the clause that actually carries the
+    lowercase rule. sec. 10.2 now appears nowhere in the vector set.
+  Regenerated (`SPEC/tools/conformance.py --generate`), replayed (`--check`,
+  25/25), and mirrored byte-identically into both packages. After the change:
+  Python self-test 53/53, Node runner 25/25.