azaharizaman · Aug 17, 2026 · Aug 12, 2026 · Aug 12, 2026 · Aug 12, 2026 · Aug 12, 2026
diff --git a/AGENTS.md b/AGENTS.md
@@ -5,17 +5,17 @@
 **Branch:** feature/CURRENCY-capability
 
 ## OVERVIEW
-Paxman is a Python 3.11+ canonicalization library: takes ambiguous human input, returns what authoritative specs say it means, with full provenance. Deterministic, provenance-first. 9 capabilities (Country, Currency, Date, Email, IP, ISBN, Money, Phone, URL). Toolchain: uv + hatchling, ruff, strict pyright, import-linter, pytest at 95% coverage.
+Paxman is a Python 3.11+ canonicalization library: takes ambiguous human input, returns what authoritative specs say it means, with full provenance. Deterministic, provenance-first. 10 capabilities (Country, Currency, Date, Email, IP, ISBN, Money, Phone, SI Unit, URL). Toolchain: uv + hatchling, ruff, strict pyright, import-linter, pytest at 95% coverage.
 
 ## STRUCTURE
 ```text
 paxman/
 ├── api/            # canonicalize() — sole public entry
 ├── engine/         # run_capability() pipeline orchestrator
 ├── core/           # domain objects, Contract protocol, registry, errors
-└── capabilities/   # 9 self-contained capability packages
+└── capabilities/   # 10 self-contained capability packages
 tests/              # unit / capabilities/<cap> / integration / property / e2e
-tools/              # regenerate_isbn_range_data.py (only script)
+tools/              # regenerate_isbn_range_data.py, regenerate_si_prefix_data.py, regenerate_idna_uts46_data.py
 docs/               # adr/, report/, research/, superpowers/plans+specs
 ```
 
@@ -31,7 +31,7 @@ docs/               # adr/, report/, research/, superpowers/plans+specs
 | Recognition (per cap) | `paxman/capabilities/<Name>/grammar/` |
 | Validation (per cap) | `paxman/capabilities/<Name>/rules/` |
 | Presentation seam | `paxman/capabilities/<Name>/capability.py` → `format_value()` |
-| Regenerate ISBN data | `tools/regenerate_isbn_range_data.py` |
+| Regenerate generated data | `tools/regenerate_isbn_range_data.py` (ISBN range), `tools/regenerate_si_prefix_data.py` (SIUnit prefixed units), `tools/regenerate_idna_uts46_data.py` (URL IDNA mapping) |
 | Merge-blocking commands | `.github/workflows/ci.yml` (authoritative) |
 | Past implementation plans | `docs/superpowers/plans/` |
 
@@ -76,17 +76,19 @@ uv run ruff format --check paxman/ tests/             # format check
 uv run pyright                                        # strict typecheck
 uv run import-linter lint                             # layer boundaries
 uv run pytest                                         # all tests
-uv run pytest -m unit|capability|integration|e2e      # by marker (also: property, country, isbn, money)
+uv run pytest -m "unit or capability or integration or e2e"      # by marker (also: property, country, currency, isbn, money, url, si_unit)
 uv run pytest --cov=paxman --cov-report=term-missing --tb=short -q
-uv run coverage report --include="paxman/{core,capabilities,engine,api}/*" --fail-under=95
-uv run python tools/regenerate_isbn_range_data.py     # regenerate ISBN data module
+uv run coverage report --include="paxman/core/*,paxman/capabilities/*,paxman/engine/*,paxman/api/*" --fail-under=95
+uv run python tools/regenerate_isbn_range_data.py     # regenerate ISBN range message module
+uv run python tools/regenerate_si_prefix_data.py      # regenerate SIUnit prefixed-unit modules
+uv run python tools/regenerate_idna_uts46_data.py     # regenerate URL IDNA UTS #46 mapping
 ```
-Full pre-PR gate: `ruff check . && ruff format --check . && pyright && import-linter lint && pytest`
+Full pre-PR gate: `uv run ruff check . && uv run ruff format --check . && uv run pyright && uv run import-linter lint && uv run pytest`
 
 ## NOTES
-- `paxman/capabilities/__init__.py` exports all nine shipped capabilities (Country, Currency, Date, Email, IP, ISBN, Money, Phone, URL); export completeness is enforced by `tests/unit/test_capability_exports.py`. A 10th — SI Unit (BIPM SI Brochure) — is in development on `feature/si-unit-capability`; update counts and the export/enumeration lists when it lands.
-- CONTEXT.md is the domain glossary for the full shipped set (nine capabilities). It is kept in sync with the code; when adding a capability, update its Notation/table entries there too.
+- `paxman/capabilities/__init__.py` exports all ten shipped capabilities (Country, Currency, Date, Email, IP, ISBN, Money, Phone, SI Unit, URL); export completeness is enforced by `tests/unit/test_capability_exports.py`.
+- CONTEXT.md is the domain glossary for the full shipped set (ten capabilities). It is kept in sync with the code; when adding a capability, update its Notation/table entries there too.
 - No `pyrightconfig.json` — pyright config is inline `[tool.pyright]` in pyproject.toml. No `.editorconfig`.
-- Data modules live under `rules/data/` (Country, Currency, ISBN, Money, Phone, URL) and `grammar/data/` (Country, Currency, Money) — plain module-level tables separating data from logic, maintained in place. Only ISBN's range message is generated: XML snapshot → `range_message.py` via `tools/regenerate_isbn_range_data.py`; unmarked data files are edited directly.
+- Data modules live under `rules/data/` (Country, Currency, ISBN, Money, Phone, SI Unit, URL) and `grammar/data/` (Country, Currency, Money, SI Unit) — plain module-level tables separating data from logic, maintained in place. Only the ISBN range message, the URL IDNA UTS #46 mapping, and the SIUnit prefixed-unit and grammar token tables are generated (each via its `tools/regenerate_*_data.py` script); unmarked data files are edited directly.
 - Library only — no CLI, no `__main__.py`, no `[project.scripts]`. Version 0.2.0.
 - Coverage: global `fail_under = 95` + per-package 95% gates in CI.
diff --git a/CONTEXT.md b/CONTEXT.md
@@ -41,6 +41,7 @@ Capability-defined intermediate representation that Grammars must produce:
 - **Currency:** `CurrencyNotation(text, shape)` — `shape` is `"code"` / `"qualified_symbol"` / `"symbol"` / `"word"`; codes are grammar-folded to uppercase, words to lowercase, symbols keep exact casing
 - **Money:** `MoneyNotation(currency_part, amount_part, currency_shape, amount_shape)` — verbatim currency + amount tokens with grammar-assigned shape discriminators
 - **ISBN:** `ISBNNotation(shape, digits)` — `shape` is `"isbn10"` / `"isbn13"`, `digits` is the digit string (`X` only as final char of an isbn10 shape)
+    - **SIUnit:** `SIUnitNotation(text, shape)` — `shape` is `"symbol"` / `"name"` / `"compound"` / `"split_word_prefix"` / `"split_symbol_prefix"`; `text` is the unit expression as written (symbols keep exact casing, names are grammar-folded to lowercase, compounds keep the written form)
 - **IP / Phone / URL:** capability-defined shapes for address / number / URI components
 
 **Note:** Capabilities define Notation using frozen dataclasses for type safety and immutability. The `as_list()` method bridges the typed notation to the generic `list[str]` interface.
@@ -65,7 +66,7 @@ class EmailNotation:
 
 ## The Capabilities
 
-Paxman ships nine built-in capabilities, each wired to an authoritative specification:
+Paxman ships ten built-in capabilities, each wired to an authoritative specification:
 
 | Capability | Domain | Authorities |
 |------------|--------|-------------|
@@ -77,11 +78,12 @@ Paxman ships nine built-in capabilities, each wired to an authoritative specific
 | **ISBN** | ISBNs | ISO 2108, ISBN Users' Manual, ISBN Range Message |
 | **Money** | Money amounts | ISO 4217, CLDR |
 | **Phone** | Phone numbers | ITU-T E.164, RFC 3966, NANP |
+| **SI Unit** | SI unit expressions | BIPM SI Brochure, ISO 80000-1 |
 | **URL** | URLs | WHATWG URL Standard |
 
 Capability classes are exported from `paxman/capabilities/__init__.py` as acronym aliases (`EmailCapability as Email`, etc.); the export list is enforced by `tests/unit/test_capability_exports.py`.
 
-**In development:** the **SI Unit** capability (BIPM SI Brochure, 9th edition) is the planned 10th capability (MILESTONE row 23; branch `feature/si-unit-capability`). It will be the first capability whose canonical value is case-meaningful (`K` kelvin vs `k` kilo). This document will be updated again once it ships.
+**Note:** the **SI Unit** capability is the first whose canonical value is case-meaningful (`K` kelvin vs `k` kilo). It canonicalizes unit expressions only (symbols, names, product/quotient compounds) to the canonical symbol form; no quantities, no magnitudes, no name-compounds.
 
 ---
 
@@ -194,7 +196,7 @@ class SectionCode(Rule[CurrencyNotation]):
         return self.TABLE[notation.text]
 ```
 
-Authority-backed lookup tables live in `rules/data/` (e.g., `iso4217_list_one.py`, `cldr_currencies.py`), separated from rule logic; lexicon keys serving grammars live in `grammar/data/`. Only the ISBN range message is generated from a source snapshot (via `tools/regenerate_isbn_range_data.py`) — everything else is maintained in place.
+Authority-backed lookup tables live in `rules/data/` (e.g., `iso4217_list_one.py`, `cldr_currencies.py`), separated from rule logic; lexicon keys serving grammars live in `grammar/data/`. Seven data modules across three generators are produced via tools: the ISBN range message (`tools/regenerate_isbn_range_data.py`), the URL IDNA UTS #46 mapping (`tools/regenerate_idna_uts46_data.py`), and the SIUnit prefixed-unit and grammar token tables (`tools/regenerate_si_prefix_data.py`) — everything else is maintained in place.
 
 ### Parser Example
 ```python
@@ -619,7 +621,7 @@ for candidate in result.candidates:
 ### Capability Versioning
 - Each capability has its own version in `capability.py`
 - Capability version is independent of engine version
-- All nine built-in capabilities currently ship `version = "1.0.0"`
+- All ten built-in capabilities currently ship `version = "1.0.0"`
 - Example:
   ```python
   # capabilities/Email/capability.py
@@ -775,6 +777,14 @@ paxman/
     │   ├── grammar/               # e164, tel_uri, international_00, national_recognition (+ common.py LEGACY)
     │   ├── rules/                 # e164_ed2010, rfc_3966_ed2004, nanp_ed2024
     │   └── rules/data/            # e164_country_codes, nanp_tables
+    ├── SIUnit/                    # grammar/ (5) + rules/ (3) + grammar/data/ + rules/data/ — BIPM SI Brochure, ISO 80000-1
+    │   ├── capability.py          # SIUnitCapability
+    │   ├── contract.py            # SIUnitContract
+    │   ├── notation.py            # SIUnitNotation (text, shape)
+    │   ├── grammar/               # symbol, name, compound_recognition, split_word_recognition, split_symbol_recognition
+    │   ├── grammar/data/          # unit_symbol_tokens, unit_name_tokens, compound_tokens (+ GENERATED via tools/regenerate_si_prefix_data.py)
+    │   ├── rules/                 # bipm_si_brochure_ed2019, iso_80000_ed2022, split_prefixes
+    │   └── rules/data/            # si_base_units, si_derived_units, si_nonsi_units, si_prefixes, unit_names (+ GENERATED prefixed_units, prefixed_unit_names)
     └── URL/                       # grammar/ (1) + rules/ (1) + rules/data/ — WHATWG URL Standard
         ├── capability.py          # URLCapability
         ├── contract.py            # URLContract
@@ -785,8 +795,6 @@ paxman/
         └── rules/data/            # idna_uts46_mapping
 ```
 
-**In development (not yet in tree):** the SI Unit capability (branch `feature/si-unit-capability`, research at `docs/research/2026-08-09-si-unit-canonicalization.md`). Structure will follow the Currency LOOKUP_TABLE template with BIPM SI Brochure data under `rules/data/`.
-
 ### Package Responsibilities
 
 | Package | Responsibility |
@@ -813,7 +821,7 @@ tests/
 │   ├── test_capability_contract.py# CapabilityContract (output_format policy, defaults)
 │   ├── test_capability.py         # Capability ABC
 │   ├── test_capability_surface.py # Surface homogeneity across capabilities
-│   ├── test_capability_exports.py # __init__ export completeness (9 capabilities)
+│   ├── test_capability_exports.py # __init__ export completeness (10 capabilities)
 │   ├── test_version_stamp.py      # VersionStamp
 │   ├── test_discovery.py          # Registry register/freeze/reset
 │   ├── test_errors.py             # Exception hierarchy
@@ -830,6 +838,7 @@ tests/
 │   ├── isbn/                      # + test_contract, test_notation, test_data
 │   ├── money/                     # + test_contract, test_notation, test_data, test_parsing
 │   ├── phone/                     # + test_data
+│   ├── si_unit/                   # test_grammar, test_rules, test_capability, test_contract, test_notation, test_data, test_data_consistency
 │   └── url/                       # + test_contract, test_notation, test_data, test_parsing, test_rule
 ├── integration/                   # -m integration pipeline, ambiguity, temporal,
 │   │                              # feature gating, format_value seam, per-capability pipelines
@@ -874,7 +883,7 @@ def test_ambiguity_detection(): ...
 def test_canonicalize_email_success(): ...
 ```
 
-Per-capability markers are registered for `country`, `currency`, `isbn`, `money`, and `url` — run one capability's suite directly with `uv run pytest tests/capabilities/<cap>` (or `-m <cap>`). Capability dirs are lowercase (`isbn`, not `ISBN`).
+Per-capability markers are registered for `country`, `currency`, `isbn`, `money`, `si_unit`, and `url` — run one capability's suite directly with `uv run pytest tests/capabilities/<cap>` (or `-m <cap>`). Capability dirs are lowercase (`isbn`, not `ISBN`).
 
 ---
 
@@ -922,6 +931,7 @@ markers = [
     "isbn: isbn capability tests",
     "money: money capability tests",
     "url: url capability tests",
+    "si_unit: si unit capability tests",
 ]
 testpaths = ["tests"]
 ```

diff --git a/README.md b/README.md
@@ -50,7 +50,7 @@ If multiple specifications disagree on the canonical value, the status is `AMBIG
 
 ## Capabilities
 
-Paxman ships with nine built-in capabilities:
+Paxman ships with ten built-in capabilities:
 
 | Capability | Domain | Grammars | Rules | Description |
 |------------|--------|----------|-------|-------------|
@@ -62,6 +62,7 @@ Paxman ships with nine built-in capabilities:
 | **ISBN** | ISBNs | 2 (isbn13, isbn10) | 4 | ISO 2108, ISBN Users' Manual, ISBN Range Message |
 | **Money** | Money amounts | 3 (code, symbol, word) | 3 | ISO 4217, CLDR |
 | **Phone** | Phone numbers | 4 (E.164, tel-URI, 00-prefix, national) | 5 | ITU-T E.164, RFC 3966, NANP |
+| **SI Unit** | SI unit expressions | 5 (symbol, name, compound, split_word_prefix, split_symbol_prefix) | 7 | BIPM SI Brochure, ISO 80000-1 |
 | **URL** | URLs | 1 (absolute-uri) | 1 | WHATWG URL Standard |
 
 ### Email Capability
@@ -331,6 +332,46 @@ result = paxman.canonicalize("http://münchen.de", contract)
 # → "http://xn--mnchen-3ya.de/"
 ```
 
+### SI Unit Capability
+
+Recognizes SI unit expressions: symbols, names, and product/quotient compounds, canonicalizing to the canonical symbol form with BIPM SI Brochure (9th ed.) and ISO 80000-1 provenance. Identity-only: no quantities, no magnitudes, no name-compounds ("metre per second" does not resolve as a compound — its words are recognized separately, yielding AMBIGUOUS; "25°C" is MISSING).
+
+```python
+from paxman.capabilities import SIUnit
+
+register_capability(SIUnit())
+
+# Unit name resolves to its canonical symbol
+contract = SIUnit.create_contract()
+result = paxman.canonicalize("Kilogram", contract)
+# → "kg"
+
+# Prefixed name resolves to the prefixed symbol
+contract = SIUnit.create_contract()
+result = paxman.canonicalize("megahertz", contract)
+# → "MHz"
+
+# Compound expression canonicalizes to the symbol form
+contract = SIUnit.create_contract()
+result = paxman.canonicalize("m/s²", contract)
+# → "m/s2"
+
+# Spoken word-prefix form, merged only when opted in
+contract = SIUnit.create_contract(allow_split_word_prefixes=True)
+result = paxman.canonicalize("kilo gram", contract)
+# → "kg"
+
+# Symbol-prefix spacing is always rejected (no flag)
+contract = SIUnit.create_contract()
+result = paxman.canonicalize("k g", contract)
+# → Status: INVALID
+
+# Multi-solidus preserved only when opted in
+contract = SIUnit.create_contract(allow_multi_solidus=True)
+result = paxman.canonicalize("kg/m/s", contract)
+# → "kg/m/s"
+```
+
 ---
 
 ## Contract Configuration
@@ -365,6 +406,8 @@ Every capability provides a `create_contract()` factory method with common and c
 | Money | `output_format` | `str` | Output format (`"code_amount"` default, `"compact"`) |
 | Phone | `default_country` | `str` | ISO 3166-1 alpha-2 country code to resolve national numbers (e.g., `"US"`) |
 | Phone | `output_format` | `str` | Output format (`"e164"` default, `"rfc3966"`, `"national"`) |
+| SIUnit | `allow_split_word_prefixes` | `bool` | Merge a word prefix split from its unit by whitespace (e.g. `"kilo gram"` → `"kg"`) when True; default False rejects the spoken form (→ INVALID) |
+| SIUnit | `allow_multi_solidus` | `bool` | Preserve the legacy accept-multi-solidus behavior (e.g. `"kg/m/s"`) when True; default False rejects more than one top-level solidus (→ INVALID) per ISO 80000-1 §6.6.2 |
 
 ### Rule Pinning and Exclusion
 
@@ -399,7 +442,7 @@ result = paxman.canonicalize("2026-01-15", contract)
 
 ## Community Extensions
 
-Paxman ships with nine built-in capabilities, but a capability is closed for modification yet open for extension: you can add recognition and validation without touching the library. Register a `Grammar` subclass and the `Rule` subclass that validates it, then opt a contract into them by naming the grammar in `extra_grammars`:
+Paxman ships with ten built-in capabilities, but a capability is closed for modification yet open for extension: you can add recognition and validation without touching the library. Register a `Grammar` subclass and the `Rule` subclass that validates it, then opt a contract into them by naming the grammar in `extra_grammars`:
 
 ```python
 import re