How We Operate

Last updated: July 29, 2026

Transparency is core to our mission. This page documents exactly how Inbound Reports aggregates, processes, and presents news. Nothing here is aspirational — this describes the system as it currently functions.

1. Source Aggregation

We maintain a curated directory of 900+ active technology news sources — including RSS feeds, APIs, developer communities, and regional publications — scaling daily. New sources are added by editing the YAML configuration; no code or database changes are required. Each source is manually vetted for relevance. Our ingestion workers fetch new content periodically, parse article metadata, and store them for processing.

2. Story Clustering

When multiple sources cover the same topic, our system groups related articles into a single "Story Event." The cluster is formed by comparing headline and summary similarity across articles. Stories with a similarity score above a threshold are grouped together, and the cluster's source count reflects how many independent outlets are covering that story — a signal of importance.

3. AI Processing Stack

We use a multi-provider AI router with automatic fallback to ensure reliability while keeping the platform free. The chain is:

If all providers are unavailable, the system degrades gracefully by using pre-written fallback text rather than failing entirely.

4. The Hype-Reality Bar

Our Telegram bot pipeline can score linguistic hype by analyzing marketing superlatives (e.g., "revolutionary," "game-changing," "unprecedented") versus measured claims and primary sources. That AI path produces a score from substance to hype.

Site UI today: until those AI scores are stored and served to the web app, story cards and topic pages proxy with a coverage intensity heuristic based on source density (source_count). More outlets covering a story raises the bar; it is not a stored AI hype score. Per-article title heuristics on the story page remain a lightweight linguistic signal only.

Outlet roles (tech axes): each source is labeled Booster, Trade, Critical, Community, Research, or Corporate from a static domain/name map — coverage roles for tech outlets, not political left/right.

5. The Literacy Layer

Jargon Highlighting: Complex technical terms in article summaries are automatically matched against our glossary and highlighted in red. Tapping a highlighted term opens a popover with a plain-English definition and an analogy.

Reading tiers (ELI5 / Standard / Deep): Deterministic text processing — simplifying vocabulary for ELI5 and appending source-context notes for Deep when available. The tier toggle only appears when those variants actually differ from Standard; identical copy is not presented as three modes.

Language: The web product is English-first today. Khmer UI and full story translations are planned; glossary entries may include Khmer fields in data, but the interface does not currently switch languages.

6. The "Local Lens — Cambodia" Feature

On each story page, a sidebar box attempts to generate a paragraph explaining why that story matters specifically to a Cambodian audience. This uses a lightweight AI model (Groq / Llama 3.1 8B) with a prompt focused on Cambodian relevance. The feature is best-effort: if the AI is unavailable, a category-level fallback paragraph is shown and labeled as category context rather than a story-specific lens.

7. Hype Predictions (Hype Retrospective)

Registered readers can vote on whether a technology or trend is "Overhyped," "Will deliver," or "Too early to tell." These predictions are stored locally in your browser and are not currently synced across devices. This feature is experimental and used for personal reflection rather than aggregate data.

8. Updates to This Methodology

As our system evolves, this page will be updated to reflect changes. The "Last updated" date at the top of this page indicates when the methodology documentation was last reviewed.