When someone downloads a daily news briefing app, the first question they should ask is simple: Who decides which 10 stories I see, and how do I know the summaries aren't slanted?
Most news apps answer that question with marketing slogans like "100% unbiased AI" or arbitrary percentage meters. At Kepce Labs, we took a different engineering approach. No summary, whether written by a human editor or synthesized by a language model, is ever completely free of framing. Instead of pretending bias does not exist, we designed Lynto's serving pipeline around verifiable rules: multi-publisher RSS clustering, transparent publisher prestige multipliers, an explicit opposing-perspective block on every Deep Dive, and a hard cap of 10 cards per edition.
Here is the exact mechanics of how Lynto selects, scores, and assembles your morning and evening briefings.
1. The 4 : 4 : 2 Edition Allocation
Purely personalized feeds trap you inside a narrow topic loop; purely global feeds miss national economic and policy shifts that affect your daily life. When Lynto builds your 10-card edition, our serving engine partitions the 10 slots into three sequential buckets:
- Cards 01 to 04 (Local Developments, 40%): Stories from your selected regional edition (such as Türkiye or United States), prioritizing domestic economy, policy, and regional events.
- Cards 05 to 08 (Global Macro & Tech, 40%): Major international developments, science
breakthroughs, and global markets. (If you set your region preference to
GLOBAL, Local slots collapse to 0 and Global expands dynamically to 8 cards.) - Cards 09 to 10 (Your Niche Interests, 20%): Two slots reserved for your highest-weighted interest categories that did not already appear in the first eight cards.
2. The 4-Tier Publisher Prestige Multiplier (0.50x to 2.00x)
Not all RSS feeds carry the same editorial rigor. A wire dispatch from Reuters or AP News, or a peer-reviewed brief from Nature or NASA JPL, should not be scored identically to a sensationalist tabloid blog.
In Lynto's source registry, every publisher is assigned an explicit prestige multiplier ranging from
0.50x to 2.00x. Inside the app's Settings screen, you can inspect and mute individual
publishers across all four tiers at any time:
| Source Tier | Multiplier Range | Example Publishers in Lynto |
|---|---|---|
| Titans | 1.55x to 2.00x |
AP News, Reuters, Bloomberg, BBC News, The New York Times, WSJ, Financial Times, The Economist, Nature, Quanta Magazine, NASA JPL |
| Core | 1.05x to 1.50x |
NPR, The Guardian, The Washington Post, The Atlantic, Politico, ProPublica, Foreign Policy, DW, France 24, Anadolu Ajansı, Pew Research Center |
| Specialized | 0.75x to 1.00x |
Domain-specific trade journals, regional desks, and beat reporters |
| Sandbox | < 0.75x |
Newly indexed feeds undergoing quality and clickbait evaluation |
3. How Candidate Stories Are Scored
Within each bucket, candidate stories published within the active freshness window (24 hours by default, expanding to 48h or 72h only if regional volume is low) are ranked using a composite formula that combines your topic weights, your editorial balance dial, the event's importance level, and the publisher's prestige multiplier:
Three Hard Guardrails Against Feed Fatigue
- Max 2 Breaking News Cards per Edition: Even on chaotic news days, Lynto caps high-urgency
breaking alerts (
importance_level ≥ 4) at a maximum of 2 cards per brief so alarmist headlines never hijack your entire edition. - Category Diversity Cap: If you follow 5 or more categories, Lynto enforces a strict maximum of 2 cards per category inside a 10-card brief so one noisy beat (like election horse-race coverage) cannot crowd out science, finance, or culture.
- 7-Day Cross-Brief Deduplication & Title Similarity Check: Any story you have already read, bookmarked, or discarded in the past 7 days, or any story whose headline duplicates an already-picked card in the same brief, is automatically excluded.
4. Inside Every Deep Dive: Opposing Views & Direct Citations
Tapping any card in Lynto opens its Deep Dive sheet. Rather than hiding how the synthesis was constructed, every Deep Dive follows a strict top-to-bottom structure:
- Key Takeaways & 40s Audio Brief: Three factual bullet points and a spoken audio summary in English or Turkish.
- Executive Summary: A concise synthesis of the event across clustered reporting.
- Ask Lynto (ELI5 + Analytical Q&A): Pre-computed plain-English explanations (starting
with an
ELI5breakdown) and deeper analytical questions answered directly from the source reporting. - Perspective (Competing Framing): An explicit amber-highlighted block summarizing how critics, opposing political camps, or independent analysts challenge the main narrative.
- Historical Background & Macro Impact: Precedent, timeline context, and sector-level implications.
- Read More (Direct Publisher Link): A prominent button at the bottom of the sheet that opens the original article directly on the publisher's site. We never scrape behind paywalls.
5. Reader Control: Mute Sources & Flag Clickbait Instantly
Algorithms need human correction. If a publisher repeatedly lets you down, you can mute that source in two
taps from the Deep Dive menu or inside Settings. Muting a publisher immediately strips its cards from your
active feed and excludes its source_id from all future morning and evening editions. Flagging
a card as "Less of this category" dynamically lowers your personal weight for that topic so tomorrow's
edition recalibrates automatically.