ContentForge — Deterministic Headline Scorer and Content Quality Gate

Start with the headline doctor · 50 endpoints behind it · Deterministic scoring in <50ms

Score content before you post. ContentForge is a deterministic headline scorer and before-publish quality gate that grades headlines, tweets, LinkedIn posts, and ad copy with an explainable A–F score, actionable suggestions, and a PASSED | REVIEW | FAILED verdict in under 50ms.

Think of it as a digital ruler for content quality. A ruler doesn't need a dataset to tell you something is 12 inches long — it just needs to be correctly calibrated. ContentForge's heuristic engine is that ruler: zero variance on the same input, zero hallucinations, fully auditable open-source logic. AI (Ollama or Gemini) kicks in only for generation endpoints like rewrites, hooks, and subject lines.

import requests

HEADERS = {"X-RapidAPI-Key": "YOUR_KEY", "X-RapidAPI-Host": "contentforge1.p.rapidapi.com"}

r = requests.post("https://contentforge1.p.rapidapi.com/v1/score_tweet",
    headers=HEADERS,
    json={"text": "I'm working on a new project."})
# → {"score": 32, "grade": "C", "quality_gate": "FAILED", "suggestions": [...]}

r = requests.post("https://contentforge1.p.rapidapi.com/v1/score_tweet",
    headers=HEADERS,
    json={"text": "Got 100 signups in 24 hours 🚀 Here's the copy that converted: #buildinpublic"})
# → {"score": 91, "grade": "A", "quality_gate": "PASSED", "suggestions": []}

→ Start free on RapidAPI — no credit card required, 300 requests/month on BASIC.

---

Current Status (v1.8.0)

Component	Status	Notes
ContentForge API	✅ Live	https://contentforge-api-lpp9.onrender.com
RapidAPI Listing	✅ Public	50 endpoints, 4-tier pricing
Keep-warm cron	✅ Active	cron-job.org pings /v1/status every 10 min (no LLM call)
Gemini backend	✅ Configured	gemini-2.0-flash on Render (1500 RPD free tier)
Ollama local	✅ Running	Scoring uses zero AI calls — pure heuristics
Twitter bots	✅ Active	Multi-account state machine, health scoring
Legal docs	✅ Done	docs/TERMS_OF_USE.md, docs/TERMS_AND_CONDITIONS.md

---

Calibration Status

ContentForge is deterministic today. That means the same input always produces the same score and audit trail.

It is not the same thing as saying the engine is fully validated against outcome data yet. Calibration is still in progress, and the public log lives in docs/validation.md.

Current practical framing:

• Use it as an explainable pre-flight quality gate.
• Treat scores as heuristic guidance, not guaranteed performance prediction.
• Expect weights to keep improving as blind test data comes in.

Calibration tooling now lives in:

• docs/calibration_dataset_template.csv
• docs/calibration_dataset_template.json
• scripts/calibrate_content.py
• docs/chrome-extension-readiness.md
• docs/reddit-launch-notes.md

Quick start:

python scripts/calibrate_content.py \
  --input docs/calibration_dataset_template.csv \
  --report-json docs/calibration_report.json \
  --report-md docs/calibration_report.md \
  --examples-json docs/calibration_examples.json

---

Launch Feedback So Far

Early launch feedback on Reddit pushed the positioning in a clearer direction:

• The calibration challenge got more useful engagement than the broad feature pitch.
• People understood "same input, same score" quickly, but trusted it more once the proof story was explicit.
• Before or after comparison is easier to believe than a giant endpoint list.
• The Chrome extension and headline workflow are easier to grasp than the full API surface on first contact.

The current landing page and docs now reflect that narrower first impression. Full notes live in docs/reddit-launch-notes.md.

---

Why Deterministic Scoring?

Every LLM-based scorer has the same flaw: ask it to score the same tweet twice and you'll get two different answers. For a professional content workflow, that's not a tool — that's a vibe check.

ContentForge's scoring layer is pure Python heuristics. Same input → same output, every time. The logic is open source; you can read exactly why a post scored 74 and not 83. This is the Deterministic Advantage:

	ContentForge	LLM-based scoring
Response time	<50ms	1–5 seconds
Variance on same input	0%	~15–30%
Explainability	Full — every deduction itemised	Black box
Cost per call	Free (heuristics)	$0.001–0.01 per call
Self-hostable	✅ (python scripts/api_prototype.py)	Depends on provider

---

All 50 Endpoints

Instant Scorers (no AI, <50ms)

Endpoint	What It Does
POST /v1/score_tweet	Score a tweet 0–100 with grade + quality gate
POST /v1/score_linkedin_post	Score a LinkedIn post for professional engagement
POST /v1/score_instagram	Score an Instagram caption for saves and reach
POST /v1/score_youtube_title	Score a YouTube title for CTR and SEO
POST /v1/score_youtube_description	Score a YouTube description for watch time
POST /v1/score_email_subject	Score an email subject line for open rate
POST /v1/score_readability	Flesch–Kincaid + grade level + suggestions
POST /v1/score_threads	Score a Threads post
POST /v1/score_facebook	Score a Facebook post
POST /v1/score_tiktok	Score a TikTok caption
POST /v1/score_pinterest	Score a Pinterest pin description
POST /v1/score_reddit	Score a Reddit post/title
POST /v1/analyze_headline	Headline power word detection + CTR scoring
POST /v1/analyze_hashtags	Hashtag strategy audit across platforms
POST /v1/score_content	Single unified endpoint — pass platform param
GET /v1/analyze_headline	GET variant for quick headline scoring

Multi-Content & Comparison

Endpoint	What It Does
POST /v1/score_multi	Score one post across all platforms simultaneously
POST /v1/ab_test	Head-to-head score comparison of two drafts

AI Generation (Ollama → Gemini fallback)

Endpoint	What It Does
POST /v1/auto_improve	Score → if not PASSED → AI rewrite → re-score loop (up to 5 iterations) — returns best version + full iteration history
POST /v1/compose_assist	Generate 2–5 rewrite variants, score each, return ranked with quality gates
POST /v1/improve_headline	Rewrite a weak headline N times, sorted by score
POST /v1/generate_hooks	Scroll-stopping openers for any topic/style
POST /v1/rewrite	Rewrite for Twitter, LinkedIn, email, or blog
POST /v1/tweet_ideas	Tweet ideas for a niche with hashtags
POST /v1/content_calendar	7-day content calendar with ready-to-post drafts
POST /v1/thread_outline	Full Twitter thread: hook + body + CTA close
POST /v1/generate_bio	Optimised social bio, auto-trimmed to platform limits
POST /v1/generate_ad_copy	Google/Meta ad copy with CTA and compliance signals
POST /v1/generate_subject_line	AI email subject line with open-rate optimisation

Quality Operations (QOps)

Endpoint	What It Does
POST /v1/quality_gate	Batch PASSED/REVIEW/FAILED verdict for up to 10 posts
GET /v1/platform_friction	Real-time platform state (rate limits, algo signals)
POST /v1/proof_export	Export scored posts + engagement delta as proof report

Utility

Endpoint	What It Does
GET /health	Service health: LLM backend, usage stats
GET /v1/status	Lightweight ping — version, endpoint count

(Full 50-endpoint list with request/response schemas: RapidAPI docs)

---

Self-Hosting

ContentForge runs fully locally with Ollama. No external AI calls needed for scoring.

git clone https://github.com/CaptainFredric/ContentForge.git
cd ContentForge
pip install -r requirements.txt
python scripts/api_prototype.py
# → Listening on http://localhost:5000

What runs locally with zero external calls:

• All 12 platform scorers (deterministic, <50ms)
• Quality gate evaluation (PASSED / REVIEW / FAILED)
• Rate limiting and proof dashboard

What uses Ollama locally or falls back to Gemini:

• Hook generation, rewrites, bio generation, subject lines, ad copy

LLM chain: Ollama first → Gemini 2.5 Flash if Ollama unavailable → model rotation. If Ollama is running locally, nothing leaves your machine for AI calls.

License: AGPL-3.0

---

Pricing (via RapidAPI)

Plan	Price	AI calls/mo	Requests/mo
BASIC	Free	50	300
PRO	$9.99/mo	750	1,000
ULTRA	$29.99/mo	3,000	4,000
MEGA	$99/mo	18,000	20,000

All plans include every endpoint. Heuristic scoring calls don't count against your AI quota.

→ Get your free API key

---

Architecture

scripts/
└── api_prototype.py         # ContentForge Flask API — all 50 endpoints (incl. /v1/auto_improve)
extension/
├── manifest.json            # Chrome extension (Manifest V3)
├── popup.html / popup.js    # Score, compare, rewrite from the toolbar
├── content.js / content.css # Real-time scoring badge on X, LinkedIn, etc.
└── background.js            # Service worker — API calls + offline fallback
deploy/
├── render.yaml              # Render Blueprint
├── openapi.json             # OpenAPI 3.0.3 spec (50 paths)
└── Procfile                 # Gunicorn start command
docs/
├── ContentForge_API_Documentation.md
└── RapidAPI_GettingStarted.md

---

Contributing

PRs against main. One feature/fix per PR. Open an issue first. See CONTRIBUTING.md.

License

Affero General Public License v3.0. See LICENSE.

Acknowledgements

Early development scaffolding adapted from MoneyPrinterV2 by @DevBySami.