📖 Qalarc Blog — Publish Protocol

Draft (readiness 0–40%)

The post exists in the data store with source files referenced. Either raw notes exist, or sub-agents have been briefed. No polish yet.

• Check that source files are referenced in posts-data.js
• Identify any credentials or sensitive info in source files before reading
• Flag which claims need citation

Write Both Versions (readiness 40–70%)

Every post exists in two versions: Technical (for developers, researchers, engineers) and Plain English (for anyone curious, no background assumed).

• Technical: include real numbers, code snippets, architecture decisions, citations
• Layman: use analogies, no jargon, explain why it matters to a non-expert
• Both versions must be honest — no overselling, no hiding limitations
• Both versions must link to each other via the version toggle

Reference Vetting (required before readiness > 70%)

Every factual claim must be traceable. See the Vetting section below.

• All academic claims cite author, year, venue
• All accuracy/performance numbers verified against source data
• Limitations stated explicitly alongside achievements
• No claims made about future plans as if they are current reality

Self-Review Using This System (readiness 70–90%)

Use the vote/comment system on this blog to review your own posts before marking them published.

• Read the post as if you'd never worked on the project
• Use "🔍 Fact Check" comment type for claims you're unsure about
• Use "🔧 Needs Fix" for structural issues
• Use "❓ Question" for things that need follow-up before publish
• Only vote 👍 Approve when all Fact Check and Needs Fix comments are resolved

Answer Pending Questions (required for publish)

Each post in posts-data.js has a pendingQuestions array. Every item must be resolved before the post can go from needs-review to published.

• Update pendingQuestions to [] when all resolved
• Change status to 'published' in posts-data.js
• Update readiness to reflect actual state

Push to qalarc.com

When a post is approved (status: published, 👍 voted, no open pending questions), it's ready for the live site. The online system will be built separately and will consume this same posts-data.js structure.

Academic Claims

• Format: Author et al., Year, Venue — e.g. "Karkkainen & Joo, WACV 2021"
• Include dataset license where applicable (CC BY 4.0, MIT, etc.)
• If citing a benchmark number, specify which split (train/val/test) and which metric
• If reproducing a result, state exactly what hardware/config you used

Performance Numbers

• Always include the baseline/comparison: "71% vs 14% random baseline vs 72.2% academic SOTA"
• Always state the conditions: hardware, dataset split, lighting, distance, etc.
• If a number is theoretical or estimated, say so explicitly
• Never round numbers up. Round down or state exact.

Sensitive Research (e.g. 4chan analysis)

• State the research question clearly upfront
• Describe data collection methodology and any ethical review
• Do not publish aggregate statistics that could identify individuals
• Frame findings with appropriate academic caution
• Get a second read before publishing anything in this category

The 7 Questions

• 1. What is this? One sentence. No jargon in the layman version.
• 2. Why does it exist? What problem was being solved? What was the motivation?
• 3. What makes it distinctive? What is genuinely novel or unusual about the approach?
• 4. Does it actually work? What are the real numbers? What doesn't work yet?
• 5. What are the limitations? Be honest. Readers trust posts that acknowledge limitations.
• 6. What does this connect to? How does it link to other projects? What's the broader context?
• 7. What's next? What's the current status and what would make this better?

AI Vision

Projects involving computer vision, face tracking, gesture recognition, or camera-based analysis. Includes demographic analysis, biomarker detection, gesture engines, and the smart glasses project.

Hardware

Physical builds, reverse engineering, RF/SDR work, wearables. Robot hand, HackRF, Colmi ring.

AI Systems

Operating systems, orchestration, memory architectures, voice interfaces. QalarcOS, gmux, mempalace, voice control, AI orchestrator.

Health & Biomarkers

Camera-based health measurement, rPPG, wearable integrations, health analytics platforms.

Games & Experiments

Interactive experiments, gesture-controlled games, creative coding. Castle Defense, gesture control effects.

Data & Research

Scraping, analysis, quantitative research. Requires extra ethical care before publishing.

Retrospectives

Hackathon write-ups, project post-mortems, lessons learned.

Technical Version

• Target reader: developer, researcher, engineer who wants the real implementation details
• Include: architecture decisions, code snippets, benchmark numbers, paper citations, model sizes, latency measurements
• Explain trade-offs in technical terms: "INT8 quantisation reduces model size by 8x at a cost of 1.2% accuracy"
• Link to: GitHub repos, papers, datasets

Plain English Version

• Target reader: curious non-technical person, journalist, potential collaborator, investor
• No acronyms without explanation. No assumed knowledge.
• Use analogies: "The model is smaller than most photo files on your phone"
• Still honest about limitations — don't oversell for the lay audience
• Focus on: why it exists, what it does, why it's interesting, what the limitations are
• Shorter: aim for 60–70% of the word count of the technical version

When a post is ready to publish live:

• Status in posts-data.js must be 'published'
• Vote must be 👍 (approved)
• All pendingQuestions must be resolved (empty array)
• readiness must be ≥ 85
• Both technical and layman versions must exist
• All credentials/secrets verified as NOT present in the post content

The online publish system will:

• Read from the same posts-data.js
• Deploy to qalarc.com/blog via Cloudflare Pages
• Include sitemap.xml and robots.txt for search indexing
• Support og:image and structured data for social sharing
• Link back to project pages on qalarc.ai
• Comments will migrate to a server-side solution (TBD: Cloudflare D1 or Giscus)