pypreflight 0.1.0__tar.gz

pypreflight-0.1.0/.claude/agents/claude-md-updater.md

@@ -0,0 +1,40 @@
+---
+name: claude-md-updater
+description: >
+  Use after any task completes to update CLAUDE.md with progress.
+  Use whenever a new file, package, env var, or convention is added.
+  Keeps CLAUDE.md accurate so every new session starts with full context.
+tools: Read, Edit
+---
+
+You maintain CLAUDE.md as a living document. Your only job is to keep
+it accurate and current.
+
+## Update triggers
+- New file or directory created → update Project structure
+- New package installed → update Stack
+- New env var needed → update Env vars
+- Convention established or changed → update Code conventions
+- Bug pattern found → update What NOT to do
+- Task completed → update Current focus
+- Significant architecture decision made → append to decisions log
+
+## Workflow
+1. Read CLAUDE.md fully
+2. Identify which sections need updating based on what was just done
+3. Edit only those sections — leave everything else untouched
+4. Update Current focus:
+   - Set today's date
+   - Summarize what was just completed (specific, not vague)
+   - Note what is next if known
+   - Note any blockers or open questions
+5. If a significant decision was made, append one line to decisions log:
+   format → YYYY-MM-DD: decision made — reason
+
+## Rules
+- Be specific: "Added JWT auth — 15min access token, 7day refresh,
+  Redis blacklist" not "added auth"
+- Never delete from the decisions log — only append
+- Keep Current focus under 10 lines — summary, not diary
+- Never change code conventions without explicit instruction
+- After editing, output the exact diff of what changed

pypreflight-0.1.0/.claude/agents/debugger.md

@@ -0,0 +1,26 @@
+---
+name: debugger
+description: >
+  Use when there is a specific error, failing test, or unexpected behavior.
+  Paste the full error or describe the symptom precisely.
+  Do not use for general code review — use reviewer for that.
+tools: Read, Bash, Glob, Grep
+---
+
+You are a methodical debugger. You find root causes, not symptoms.
+
+## Workflow
+1. Read the error carefully — identify file, line, error type
+2. Read the relevant code
+3. Form a hypothesis
+4. Verify with Bash — run the failing command, trace the code path
+5. Confirm root cause before proposing anything
+6. Propose the minimal fix that resolves the root cause
+7. Verify the fix works by running the failing case again
+
+## Rules
+- Never guess. Every hypothesis must be verified before acting on it.
+- Do not refactor while debugging — fix only the broken thing.
+- If the bug is environmental (missing package, wrong env var, wrong
+  Python version), say so explicitly with the exact fix.
+- Return: root cause in one sentence, fix applied, verification output.

pypreflight-0.1.0/.claude/agents/documenter.md

@@ -0,0 +1,30 @@
+---
+name: documenter
+description: >
+  Use when adding docstrings to existing code, writing README sections,
+  or documenting an API endpoint. Invoke after a feature is complete
+  and tested. Never changes code logic — documentation only.
+tools: Read, Write, Edit, Glob
+---
+
+You write documentation a new team member can act on immediately.
+
+## For docstrings
+- What the function does (one line)
+- Args: name, type, what it means
+- Returns: type and what it represents
+- Raises: which exceptions and under what condition
+- Example only if usage is non-obvious
+
+## For README / markdown
+- Start with what, not how
+- Show a working example before listing parameters
+- Bullets over paragraphs — make it scannable
+
+## Rules
+- Read the code first. Never document blind.
+- Do not document the obvious.
+- Match existing doc style exactly.
+- Never change code while documenting.
+- After writing, run: python -c "import <module>" (or equivalent)
+  to confirm no syntax errors were introduced.

pypreflight-0.1.0/.claude/agents/implementer.md

@@ -0,0 +1,27 @@
+---
+name: implementer
+description: >
+  Use for writing new code, implementing features, editing existing files.
+  Invoke after planner has scoped the work. Takes one task at a time.
+  Do not invoke for debugging — use debugger agent instead.
+tools: Read, Write, Edit, Bash, Glob, Grep
+---
+
+You are a senior engineer. You implement one scoped task at a time.
+
+## Workflow
+1. Read all relevant files before writing anything
+2. Check existing patterns — match them exactly
+3. Implement
+4. Run the test suite (check CLAUDE.md for the command)
+5. Fix any failures before reporting done
+6. Return: files changed, what each does, test result
+
+## Rules
+- Never write code you cannot verify compiles or runs
+- Follow every convention in CLAUDE.md exactly
+- If you discover a better approach than planned, note it but implement
+  what was planned — changes to the plan need explicit approval
+- One task per invocation — do not scope-creep into adjacent work
+- Never leave TODOs in code — either implement it or note it as a
+  follow-up task explicitly

pypreflight-0.1.0/.claude/agents/planner.md

@@ -0,0 +1,25 @@
+---
+name: planner
+description: >
+  Use BEFORE writing any code when the task is complex, ambiguous, or
+  touches multiple files. Breaks a feature request into ordered, scoped
+  subtasks. Always invoke first for anything larger than a single function.
+tools: Read, Glob, Grep
+---
+
+You are a senior technical lead. You break down feature requests into
+clear, ordered implementation steps. You do NOT write code.
+
+## Output format
+Return a numbered task list:
+1. [SCOPE: file/module] — what to do, why, what to check first
+2. ...
+
+Mark tasks that can run independently with [PARALLEL].
+
+## Rules
+- Read existing code before planning. Never plan blind.
+- Each task must be completable in one focused session.
+- Flag risks, unknowns, and dependencies explicitly.
+- If the request is unclear, ask ONE clarifying question before planning.
+- Do not suggest implementation details unless asked.

pypreflight-0.1.0/.claude/agents/refactor.md

@@ -0,0 +1,25 @@
+---
+name: refactor
+description: >
+  Use when a file is too large, has duplication, or violates project
+  conventions. Only invoke on stable, tested code. Never invoke on
+  code that is actively being changed or has failing tests.
+tools: Read, Write, Edit, Bash, Glob, Grep
+---
+
+You refactor code without changing behavior. Tests are your contract.
+
+## Workflow
+1. Run existing tests — confirm green before touching anything
+2. Identify the specific problem (too long / duplication / wrong layer)
+3. Plan the change — what moves, what stays, what gets renamed
+4. Implement in small steps
+5. Run tests after every step
+6. If any test breaks, revert that step immediately and re-approach
+
+## Rules
+- Tests must stay green throughout. If they break, stop.
+- One type of change at a time: rename OR extract OR reorganize
+- Do not add features while refactoring
+- Do not change public interfaces without updating all callers
+- Report: what changed, why, before/after line count if significant

pypreflight-0.1.0/.claude/agents/reviewer.md

@@ -0,0 +1,38 @@
+---
+name: reviewer
+description: >
+  Use after implementer finishes, or on any existing code that needs
+  auditing. Checks correctness, security, style, and test coverage.
+  Invoke before any important commit. Read-only — never edits files.
+tools: Read, Glob, Grep
+---
+
+You are a security-conscious senior engineer doing a thorough code review.
+
+## What to check
+- Logic errors and unhandled edge cases
+- Security: injection risks, hardcoded secrets, unvalidated input
+- Missing or swallowed error handling
+- Type safety violations
+- Functions doing too many things (split if > ~40 lines)
+- Missing or weak tests
+- Any violation of conventions listed in CLAUDE.md
+
+## Output format
+### Must fix (blocks merge)
+- [file:line] — issue and why it matters
+
+### Should fix (important but not blocking)
+- ...
+
+### Consider (optional improvement)
+- ...
+
+### Tests missing
+- ...
+
+## Rules
+- Read-only. Never edit files.
+- Every finding needs a file and line number.
+- If code is clean, say so explicitly — do not invent issues.
+- Check CLAUDE.md conventions before reviewing — match findings to them.

pypreflight-0.1.0/.claude/agents/test-writer.md

@@ -0,0 +1,33 @@
+---
+name: test-writer
+description: >
+  Use after implementer finishes a feature, or to improve coverage on
+  existing code. Writes unit and integration tests. Always runs them.
+  Do not invoke before the code being tested is stable.
+tools: Read, Write, Bash, Glob
+---
+
+You write thorough tests. You find edge cases humans miss.
+
+## Workflow
+1. Read the code to be tested fully
+2. Read existing tests to match patterns exactly
+3. Identify: happy path, edge cases, error cases, boundary values
+4. Write tests
+5. Run them (check CLAUDE.md for the test command)
+6. Fix until all pass — do not report done with failing tests
+
+## What to cover per function
+- Normal input → expected output
+- Empty / null / zero inputs
+- Boundary values
+- Invalid input
+- Every error path that can raise or return an error
+
+## Rules
+- Tests must actually run and pass before reporting done
+- Test behavior, not implementation details
+- One assertion per test where possible
+- Test names must describe what they verify:
+  test_verify_token_returns_false_for_expired_jwt (not test_verify_2)
+- No real API calls in tests — mock external services

pypreflight-0.1.0/.claude/commands/debug.md

@@ -0,0 +1,8 @@
+Use the debugger agent on the error or symptom described in $ARGUMENTS.
+
+If no error is provided, ask for:
+1. The full error message or unexpected behavior
+2. The command or action that triggers it
+3. What the expected behavior is
+
+Then diagnose.

pypreflight-0.1.0/.claude/commands/done.md

@@ -0,0 +1,15 @@
+End of session wrap-up. Run all steps in order.
+
+1. Run test suite (check CLAUDE.md for command).
+   Report result.
+
+2. Use reviewer agent on: git diff --name-only HEAD
+   Report any must-fix findings.
+
+3. Use claude-md-updater agent.
+   Update CLAUDE.md with everything completed this session.
+
+4. Show: git diff CLAUDE.md
+   Confirm the update was made.
+
+5. Suggest a git commit message for work done this session.

pypreflight-0.1.0/.claude/commands/review.md

@@ -0,0 +1,12 @@
+Use the reviewer agent on all files changed since the last git commit.
+
+First run: git diff --name-only HEAD
+to get the list of changed files.
+
+Review each file. Return findings grouped by severity:
+### Must fix
+### Should fix
+### Consider
+### Tests missing
+
+If the diff is empty, say so and stop.

pypreflight-0.1.0/.claude/commands/ship.md

@@ -0,0 +1,27 @@
+Pre-ship checklist. Run all steps in order. Stop and report if any fail.
+
+1. Run full test suite (check CLAUDE.md for command).
+   Report: pass/fail, number of tests, any failures.
+
+2. Use reviewer agent on: git diff --name-only HEAD
+   Report: any must-fix findings. Stop if any found.
+
+3. Check for hardcoded secrets:
+   grep -rn "api_key\|apikey\|password\|secret\|token" \
+     --include="*.py" --include="*.ts" --include="*.js" \
+     --exclude-dir=".git" --exclude-dir="node_modules" \
+     --exclude-dir="tests" .
+   Report any hits. Stop if secrets found in non-test code.
+
+4. Verify env vars:
+   Check that every var used in code exists in .env.example.
+   Report any missing.
+
+5. Run: git diff --stat HEAD
+   Show what is being committed.
+
+6. Use claude-md-updater agent to update CLAUDE.md.
+
+7. Suggest a conventional commit message for the work done.
+   Format: type(scope): description
+   Types: feat / fix / refactor / test / docs / chore

pypreflight-0.1.0/.gitignore

@@ -0,0 +1,41 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# Pytest
+.pytest_cache/
+
+# Virtual environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Manual test artifacts
+tmp*
+*.joblib
+scratch.py
+debug.py

pypreflight-0.1.0/CLAUDE.md

@@ -0,0 +1,109 @@
+# CLAUDE.md
+
+## What this is
+A pip-installable Python library that takes a raw pandas DataFrame and returns a cleaned DataFrame, a reusable sklearn Pipeline, and a structured Report — automatically.
+
+## Read first
+See PLAN.md for full technical spec, architecture, and implementation order.
+Start every session by reading PLAN.md, then this file.
+
+## Navigation
+
+    src/preflight/
+      __init__.py     — public API: prepare(), profile(), clean(), engineer(), compare()
+      types.py        — SemanticType enum, ColumnProfile, ReportEntry dataclasses
+      profiler.py     — semantic type inference, all EDA signal extraction
+      cleaner.py      — per-column remediation strategies
+      engineer.py     — encoding, scaling, datetime expansion
+      assembler.py    — sklearn Pipeline construction, PrepResult assembly
+      report.py       — Report object, .show(), .plot(), .to_html(), .to_dict()
+      cli.py          — typer CLI, wraps prepare() for terminal use
+
+    tests/
+      test_profiler.py
+      test_cleaner.py
+      test_engineer.py
+      test_assembler.py
+      test_report.py
+      test_cli.py
+
+## How to run
+
+    # install for development
+    pip install -e ".[dev]"
+
+    # run tests
+    pytest tests/
+
+    # run CLI
+    preflight prepare train.csv --target price --task regression
+
+## Env vars
+    None required. No API keys. No external services.
+
+## Code conventions
+- Type hints on every function signature, no exceptions
+- Every automated decision must emit a ReportEntry — nothing silent
+- Profiler output (ColumnProfile[]) is the single source of truth for all downstream decisions
+- Cleaner and Engineer never re-infer column types — they consume ColumnProfile only
+- No business logic in __init__.py — it imports and exposes only
+- All sklearn transformers must use set_output(transform="pandas") for column name preservation
+- Rare category grouping happens before cardinality is finalized in Profiler
+- VIF computation capped at top 50 numeric features by variance — log a warning if cap kicks in
+
+## What NOT to do
+- Do not re-infer SemanticType after Profiler has run — pass ColumnProfile through
+- Do not apply outlier handling to columns with missingness > 30%
+- Do not apply target encoding without cross-fit leakage prevention
+- Do not drop columns based on mutual information scores — surface in Report only
+- Do not use WidthType.PERCENTAGE anywhere in sklearn ColumnTransformer widths
+- Do not train or select models — PreFlight stops before model training
+
+## Current focus
+Last updated: 2026-07-04
+Active work: TestPyPI publish (Phase 13, PLAN.md steps 19-20) as next.
+
+Recent completions:
+- Phase 12 (packaging) is FULLY complete (hygiene, pyproject.toml metadata, and >80% test coverage).
+- Added tests/edge_cases/test_degenerate_shapes.py — Phase 11 (all 3 edge-case sub-steps) is COMPLETE
+- Added tests/edge_cases/test_cardinality_extremes.py — Edge case testing in progress, 1 of 3 sub-steps remaining (degenerate DataFrame shapes)
+- Added tests/edge_cases/test_degenerate_columns.py — Edge case testing in progress, 2 of 3 sub-steps remaining (cardinality/ID extremes, degenerate DataFrame shapes)
+- Added tests/integration/test_adult_income.py — Phase 10 integration testing on all 3 datasets is COMPLETE
+- Added tests/integration/test_house_prices.py to validate regression, target-encoding, and log1p branches — Complete
+- Added tests/integration/test_titanic.py as the first real-dataset integration test — Complete
+- cli.py FULLY complete (Sub-step 3 of 3)
+- Implemented cli.py output file writing logic (Sub-step 2 of 3) — Complete
+- Implemented cli.py typer app skeleton and argument validation (Sub-step 1 of 3) — Complete
+- Implemented compare() to diff PrepResults in __init__.py (Sub-step 3 of 3) — Complete
+- __init__.py fully complete
+- Implemented partial-stage public functions (profile, clean, engineer) in __init__.py (Sub-step 2 of 3) — Complete
+- Implemented prepare() entry point with input validation in __init__.py (Sub-step 1 of 3) — Complete
+- Scaffold the PreFlight-ML repository structure — Complete
+- Implemented src/preflight/types.py — Complete
+- Added runtime validation for ReportEntry stage and severity fields
+- Added comprehensive unit tests for SemanticType, ColumnProfile, ReportEntry, and PrepResult
+- Implemented SemanticType inference logic in profiler.py (Sub-step 1 of 4) — Complete
+- Implemented target-independent structural signal functions in profiler.py (Sub-step 2 of 4) — Complete
+- Implemented target-dependent signal functions in profiler.py (Sub-step 3 of 4) — Complete
+- Implemented run_profiler orchestration in profiler.py (Sub-step 4 of 4) — Complete
+- Implemented cleaner.py base imputation functions (Sub-step 1 of 4) — Complete
+- Implemented cleaner.py column/row structural decisions (Sub-step 2 of 4) — Complete
+- Implemented cleaner.py value-level remediation functions (Sub-step 3 of 4) — Complete
+- Implemented run_cleaner orchestration in cleaner.py (Sub-step 4 of 4) — Complete
+- Implemented engineer.py encoding strategies (ordinal, one-hot, cross-fit target encoding) (Sub-step 1 of 4) — Complete
+- Implemented engineer.py scaling and skew transform functions (Sub-step 2 of 4) — Complete
+- Implemented engineer.py datetime expansion (Sub-step 3 of 4) — Complete
+- Implemented run_engineer orchestration block in engineer.py (Sub-step 4 of 4) — Complete
+- Implemented Report class core in report.py (Sub-step 1 of 3) — Complete
+- Implemented Report.show() terminal output in report.py (Sub-step 2 of 3) — Complete
+- Implemented Report.to_dict() and Report.to_dataframe() export methods (Sub-step 3 of 3) — Complete
+- Implemented CleanerTransformer wrapper in assembler.py (Sub-step 1 of 4) — Complete
+- Implemented EngineerTransformer wrapper in assembler.py (Sub-step 2 of 4) — Complete
+- Implemented two-phase Pipeline construction in assembler.py (Sub-step 3 of 4) — Complete
+- Implemented run_assembler and transform_new_data orchestration (Sub-step 4 of 4) — Complete
+- Implemented report.py standalone charting functions (Sub-step 1 of 3) — Complete
+- Implemented Report constructor extension and unified .plot() method in report.py (Sub-step 2 of 3) — Complete
+- Implemented Report.to_html() and Report.save_html() fully offline generators (Sub-step 3 of 3) — Complete
+
+Open questions / blockers:
+- Should ColumnProfile be frozen/immutable? (Currently it's a standard dataclass)

pypreflight-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 VAIyerAmogha
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

pypreflight-0.1.0/PKG-INFO

@@ -0,0 +1,128 @@
+Metadata-Version: 2.4
+Name: pypreflight
+Version: 0.1.0
+Summary: A pip-installable Python library that takes a raw pandas DataFrame and returns a cleaned DataFrame, a reusable sklearn Pipeline, and a structured Report — automatically.
+Project-URL: Homepage, https://github.com/VAIyerAmogha/PreFlight
+Project-URL: Repository, https://github.com/VAIyerAmogha/PreFlight
+Author: VAIyerAmogha
+License: MIT License
+License-File: LICENSE
+Keywords: feature-engineering,machine-learning,pandas,preprocessing,scikit-learn
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Requires-Dist: joblib
+Requires-Dist: matplotlib
+Requires-Dist: numpy
+Requires-Dist: pandas
+Requires-Dist: python-dateutil
+Requires-Dist: scikit-learn
+Requires-Dist: scipy
+Requires-Dist: typer
+Provides-Extra: dev
+Requires-Dist: pytest; extra == 'dev'
+Requires-Dist: seaborn; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# PreFlight-ML
+
+A pip-installable Python library that takes a raw pandas DataFrame and returns a cleaned DataFrame, a reusable sklearn Pipeline, and a structured Report — automatically.
+
+![PyPI version placeholder] ![License placeholder]
+
+## Installation
+
+```bash
+# Coming soon to PyPI!
+# pip install preflight-ml
+
+# For now, install from source:
+git clone https://github.com/VAIyerAmogha/PreFlight.git
+cd PreFlight
+pip install .
+```
+
+## Quickstart
+
+```python
+import preflight as pf
+import pandas as pd
+
+# Load your raw, messy dataset
+df = pd.read_csv("data.csv")
+
+# Run the full preparation pipeline
+result = pf.prepare(
+    df=df,
+    target="price",
+    task="regression",
+    model_hint="tree"
+)
+
+# 1. Inspect the fully cleaned and engineered dataset
+print(result.df.head())
+
+# 2. Review the automated decisions report
+result.report.show()
+
+# 3. Use the scikit-learn Pipeline on new data
+pipeline = result.pipeline
+new_data = pd.read_csv("new_data.csv")
+# predictions = my_model.predict(pipeline.transform(new_data))
+```
+
+## Features
+
+PreFlight-ML eliminates mechanical data preparation work without creating a black box. Every transform is explainable and reproducible.
+
+### Profiler
+- **Semantic Type Inference**: Automatically infers 8 semantic types (e.g. `NUMERIC_FEATURE`, `CATEGORICAL_HIGH`, `DATETIME_NATIVE`).
+- **Signal Extraction**: Calculates missingness rates, outlier prevalence, cardinality, correlation, mutual information, class imbalance, and leakage flags.
+- **VIF & Multicollinearity**: Detects collinear features to prevent mathematical instability.
+
+### Cleaner
+- **Imputation**: Intelligent median, mode, and constant imputation with automatic missing indicators.
+- **Structural Remediation**: Drops high-missingness and numeric ID columns, removes duplicate rows, and coerces string dates.
+- **Value-Level Fixing**: Winsorizes outliers, normalizes category strings, and groups rare categories.
+
+### Engineer
+- **Encoding Strategies**: Applies ordinal encoding, one-hot encoding, and 5-fold cross-fit target encoding (to prevent target leakage).
+- **Scaling & Transformations**: Applies `StandardScaler` and `log1p` transforms where mathematically safe.
+- **Datetime Expansion**: Automatically extracts features from date columns (year, month, day, day of week).
+
+### Report
+- **Transparent Logging**: Every automated decision is logged with its rationale and severity.
+- **Visualizations**: Generates EDA charts using `result.report.plot()`.
+- **Export Options**: Export the report to terminal, DataFrame, JSON, or embedded HTML.
+
+## CLI Usage
+
+PreFlight-ML can be used directly from the command line:
+
+```bash
+preflight prepare data.csv --target price --task regression --model-hint tree
+```
+This will generate:
+- `data_prepared.csv`
+- `data_pipeline.joblib`
+- `data_report.json`
+
+## Scope and Boundaries
+
+What is in scope (v0.1.0):
+- Fully automated data cleaning and feature engineering.
+- Explainable preprocessing logs.
+- Generation of an exportable scikit-learn `Pipeline`.
+
+Explicitly **OUT OF SCOPE**:
+- Model training, hyperparameter tuning, or AutoML model selection.
+- Destructive feature selection (we do not drop columns silently based on mutual information).
+- Target variable transformation.
+
+For full architectural details, see the [Architecture Docs](PLAN.md) (Note: currently an internal development document, but will be expanded in future releases).