feed-protocol 0.2.0__py3-none-any.whl

feed_protocol-0.2.0.dist-info/METADATA

@@ -0,0 +1,200 @@
+Metadata-Version: 2.4
+Name: feed-protocol
+Version: 0.2.0
+Summary: FEED — Format for Enforced Evidence-based Digestion. A self-bootstrapping document protocol that forces downstream LLMs to ground answers in cited evidence.
+Author: Aniku Gul IEng, MIET, IMechE, VCAT II, CAA
+License: MIT License
+        
+        Copyright (c) 2026 Aniku Gul IEng, MIET, IMechE, VCAT II, CAA
+        
+        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.
+        
+Project-URL: Homepage, https://github.com/mrjesters/feed-protocol
+Project-URL: Spec, https://github.com/mrjesters/feed-protocol/blob/main/spec/feed-spec-v0.2.md
+Keywords: llm,rag,grounding,documents,protocol,feed,ai
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Text Processing :: Markup
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: tagger
+Requires-Dist: anthropic>=0.40; extra == "tagger"
+Provides-Extra: dev
+Requires-Dist: pytest>=7; extra == "dev"
+Requires-Dist: anthropic>=0.40; extra == "dev"
+Dynamic: license-file
+
+# FEED — Format for Enforced Evidence-based Digestion
+
+> Make any document tell the AI reading it: **here's what matters, here's the
+> evidence, cite it or say you can't.**
+
+FEED is a plain-text convention you embed in a document so that any LLM — Copilot,
+ChatGPT, Claude, Gemini, a local model — grounds its answers in your evidence and
+cites it by ID. No install, no plugin, no provider support, no prior knowledge of
+FEED required on either side. It works on every model **today** because the
+document teaches the rules inline.
+
+It also has **teeth**: because evidence carries stable IDs and answers must cite
+them, you can *mechanically verify* that an answer is actually grounded.
+
+```
+┌─ Header ──────────────────────────────────────────────┐
+│ <!-- FEED:DOC version="0.2" grounding="strict" -->     │
+│ > AI INGESTION NOTICE … ground answers, cite [E###] …  │  ← teaches any LLM the rules
+├─ Tier 0: Claims & Decisions (front-loaded) ───────────┤  ← small-context-safe
+├─ Tier 1: Findings (narrative, references [E001]) ─────┤
+├─ Tier 2: Evidence (atomic key/value facts, IDs) ──────┤  ← the source of truth
+└────────────────────────────────────────────────────────┘
+```
+
+## Why it exists
+
+You generate AI reports and send them to people who paste them into their own AI.
+Every hop loses fidelity — the reader's AI skims headings and riffs. FEED fixes
+the **author** side of that loop: the document constrains how the downstream AI
+reads and answers. Nothing else does this at the document level (`llms.txt` is
+website-level and has no grounding contract; RAG/citation systems are all
+retrieval-side, controlled by the AI, not the author).
+
+## FEED is AI-to-AI (the library never needs its own API key)
+
+The AI **already in your loop** — the one that wrote the report, your assistant, a
+pipeline step — is what produces FEED. The library just renders, validates, and
+verifies; it never calls an LLM of its own. Authoring is self-bootstrapping, the
+same way ingestion is:
+
+```
+Reading side:   the document carries a notice that teaches any AI to ground & cite
+Authoring side: an authoring prompt + schema teach any AI to emit FEED, then
+                feed.build() renders it deterministically — no key, no network
+```
+
+### Primary flow: the AI emits FEED, you render it
+
+```python
+from feed import AUTHORING_PROMPT, FEED_JSON_SCHEMA, build
+
+# 1. Hand AUTHORING_PROMPT + FEED_JSON_SCHEMA to whatever AI is already in your loop.
+#    It returns structured JSON (no FEED tooling needed on the AI's side).
+# 2. Render that JSON into a validated FEED document — pure Python, no API key:
+doc = build(ai_json, grounding="strict", author="N. Gul")
+doc.write("report.md")
+```
+
+Or entirely from the shell:
+
+```bash
+feed prompt > authoring-kit.txt        # the prompt + schema to give any AI
+feed build ai_output.json -o report.md # render the AI's JSON into FEED (no key)
+```
+
+Manual additions are just edits: open `report.md` and add/adjust evidence and
+claims by hand — it's plain markdown. `feed validate` checks it's still conformant.
+
+> `feed tag draft.md` is an **optional** convenience that calls Claude directly,
+> for when you have a plain document and *no* AI already in the loop. It is not the
+> primary path and is the only thing that needs an API key.
+
+## Quick start
+
+### Build one in Python (manual / programmatic)
+
+```python
+from feed import FeedDocument
+
+doc = FeedDocument("Q2 Pump Health Assessment", grounding="strict")
+doc.add_evidence("E001", asset="XYZ-003", metric="vibration_rms",
+                 value="12.4 mm/s",
+                 threshold="11.2 mm/s (ISO 10816-3 Zone C)", confidence="high")
+doc.add_claim("C1", "XYZ-003 needs intervention", evidence=["E001"],
+              decision="Approve bearing replacement work order")
+
+doc.write("report.md")    # clean markdown, FEED in HTML comments
+doc.write("report.html")  # styled, opens in any browser
+```
+
+Your team opens `report.md` as a normal report. They upload it to whatever AI
+they use → it reads the notice, sees the evidence, and answers grounded.
+
+### Verify an answer was grounded (the teeth)
+
+```python
+from feed import FeedDocument, verify
+
+doc = FeedDocument.read("report.md")
+report = verify(ai_answer_text, doc)
+print(report.passed)   # False if it cited evidence that doesn't exist,
+                       # or (strict mode) didn't cite anything
+```
+
+### From the command line
+
+```bash
+feed prompt                                     # authoring kit for any AI (no key)
+feed build ai_output.json -o report.md          # render an AI's JSON into FEED (no key)
+feed validate report.md                         # is it well-formed FEED?
+feed verify --doc report.md --answer answer.txt # is this answer grounded?
+feed render report.md --to html -o report.html  # styled HTML
+feed tag draft.md --grounding strict -o report.md  # OPTIONAL: auto-tag via Claude (needs key)
+```
+
+## What's in this repo
+
+| Path | What it is |
+|------|------------|
+| `spec/feed-spec-v0.2.md` | The protocol definition (the constitution) |
+| `feed/` | The reference library — build, render, validate, verify, auto-tag |
+| `feed/verify.py` | The citation verifier — FEED's defensible edge, ~40 lines |
+| `feed/cli.py` | The `feed` command-line tool |
+| `examples/` | A complete worked example: a pump condition report (`.md` + `.html`) and the script that builds it |
+| `templates/blank.feed.md` | A hand-authoring starter |
+| `tests/` | Round-trip, validation, and verification tests |
+
+## Install
+
+```bash
+pip install -e .            # library + CLI, zero dependencies
+pip install -e ".[tagger]"  # adds the auto-tagger (needs anthropic + ANTHROPIC_API_KEY)
+```
+
+The core — authoring kit, build, render, validate, verify — is **pure Python with
+no dependencies and never calls an LLM**. Only the optional `tag` convenience
+calls Claude directly; it defaults to Claude Opus 4.8.
+
+## The three primitives
+
+- **Evidence** — atomic, ID'd, key/value facts. Never prose. The source of truth.
+- **Claim** — a short statement grounded in evidence IDs, optionally a decision.
+- **Header** — declares the grounding mode and carries the self-teaching notice.
+
+Plus **grounding modes** (`strict` / `standard` / `open`) — the author's dial for
+how strict the reading AI must be. In `strict`, no evidence means "Not supported
+by this document."
+
+## Status
+
+v0.2 — spec + reference library + verifier + CLI + auto-tagger + worked example.
+Deliberately out of scope for now: PDF/DOCX embedding, a hosted validator, and any
+provider-native "FEED mode".
+
+MIT licensed. Spec and tooling are open — adoption is the point.

feed_protocol-0.2.0.dist-info/RECORD

@@ -0,0 +1,16 @@
+feed/__init__.py,sha256=qBYxMQ9fx_-cqIRodTFbZT4gV7QQ1yW_UJMri8YhkJw,1215
+feed/authoring.py,sha256=uSimEMbwSwNf2BTV5VY9lqFvldBdES8819Dpo-WiZpI,6606
+feed/cli.py,sha256=rJTeJfc_Z5HqDNDF0KIa5O-zmRx5Swjf2xYuIJdG2rY,5842
+feed/constants.py,sha256=Ih1B6_uoi2eE0bbCtGxdi9GMj0InabD6tVTWneTxBn8,3979
+feed/document.py,sha256=GV9nkD8Zv-RV5s9h67ulsSX1-lY2EHTijldb6dep48Y,6372
+feed/parser.py,sha256=w6plJHCj0ZsKMf_naYcUJCerxpTdzGoXWm9iRMY-s1g,6022
+feed/render.py,sha256=dOlfX_iKwutCOj-p0UGziEay6wzAZbXjYAUp92Yv3xE,8342
+feed/tagger.py,sha256=auV4scqW5lVF3gCvM-JRETwZIZRO7O5s3_QNOiC6SVM,2310
+feed/validate.py,sha256=6qlpoJ8RddL-2LEt4dZizlBKb4iHgGxjEZuHyXdUHvM,3353
+feed/verify.py,sha256=7F90l4yr7578N67V_bd6sIFF2kI0VFH5rctAWJ2jzpI,4409
+feed_protocol-0.2.0.dist-info/licenses/LICENSE,sha256=jdvT554DEjSjpLR44eAFs5gLsUu5hhFbMOVu-bK2Y98,1099
+feed_protocol-0.2.0.dist-info/METADATA,sha256=VS65QWpxNcIqpqAD9In2wWp8LVzl1IBczXbw4DmvIjI,9196
+feed_protocol-0.2.0.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+feed_protocol-0.2.0.dist-info/entry_points.txt,sha256=qgY2fsXxYg4IW6WO9s22O-XTpQ03vVU1YqiPpk0CEhs,39
+feed_protocol-0.2.0.dist-info/top_level.txt,sha256=T5FmWMMaAHxdIAnMhJwZJYmhMYVtrKXp08U3jdTTGow,5
+feed_protocol-0.2.0.dist-info/RECORD,,

feed_protocol-0.2.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (82.0.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

feed_protocol-0.2.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+feed = feed.cli:main

feed_protocol-0.2.0.dist-info/licenses/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Aniku Gul IEng, MIET, IMechE, VCAT II, CAA
+
+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.

feed_protocol-0.2.0.dist-info/top_level.txt

@@ -0,0 +1 @@
+feed