News that has nothing to sell you.
Know enough. Live more.I stopped trusting the news. Too loud, too agenda-driven, too much of nothing. RumboRumbo — Spanish and nautical for heading, course, direction. Where things are going. is what I made instead.
A few global items a day, picked for signal not volume. What is actually shifting in the world.
No sides. No country, no bloc, no ideology. Things are described by what they do, not whether they are right.
Each edition is AI-generated by a language modelCurrently Google's Gemini 2.5 Pro. Mistakes happen, like any news source. No spin, no agenda, no ads. The prompt is public.
Add a regional focus for local coverage when something significant happened.
↗ opens a search for that storyUses DuckDuckGo — a privacy-focused search engine that doesn't track you..
Errors happen. It's a starting point, not the last word.
- ► Sign up for the newsletter
-
Your task: pick 3 to 4 world news items from the last 72 hours that change how large groups of people live. Search the web broadly across all regions. If fewer than 3 qualify, publish fewer. Never fill slots with weak or stale stories. Every item must pass all three rules, or it is dropped: 1. It already happened. Completed actions only, not announcements, plans, proposals, reviews, consultations, committees, or signed letters. A deal, ruling, or law that still needs another party to accept or ratify it has not taken effect. An advisory court opinion is not binding law. 2. It is fresh, and it is real. The event itself happened in the last 72 hours, and was found in today's search, not pulled from memory of past events. A recent article about an older event does not count. If a story feels familiar from a past year, it is treated as an old event resurfacing and dropped. 3. It has a consequence beyond itself. Each item must name who it changes things for, and how, with an effect reaching beyond the country where it happened. Loud domestic politics with no outside effect does not qualify. Writing: two short sentences per item, plain language, neutral in tone. The first says what happened, the second says why it matters. EVENT-DATE CHECK: each item records the date its event actually happened. Rumbo checks this automatically and drops any item whose event falls outside the last 72 hours, guarding against old stories slipping in as fresh news. Additional formatting and technical instructions are included in the full prompt for consistency.
-
A running log of what each edition revealed and what was changed in response. 12 June 2026: Simplified prompt holds well for global items. Some soft regional items. 10 June 2026: First day with simplified prompt; quality improved, no hallucinations. 9 June 2026: Three old stories appeared as today's news. Removed them, and rewrote the prompt to be much shorter and clearer. 5 June 2026: Some hallucination reported on top of real happenings, making stories inaccurate. 4 June 2026: Regional updates failed for Spain and Norway due to Gemini server overload. Can implement retry-logic if reccuring. 3 June 2026: Global edition showed up empty again: an over-strict check was throwing out good runs Fixed the check. 2 June 2026: Problematic edition, old news presented as recent again. Re-ran, got zero global items. 1 June 2026: Several items from previous years presented as today's news. Overhauled how editions are checked before publication. 30 May 2026: The global feed didn't find items meeting the editorial bar. Adding a note for when this happens, instead of showing yesterday's run as backup. 29 May 2026: One wrong geo-tag. Tightened the rule on how regions are assigned. 28 May 2026: Reworked freshness rule, to avoid reporting of previous years' events. 27 May 2026: Generally ok edition, but a couple of half-relevant regional items. 26 May 2026: Second global call failure in five days. Shortened the look-back window and added diagnostic. 25 May 2026: Global items ok, some weak regional updates. 24 May 2026: Same stories returning across editions with different framings. Adjusted deduplication rule. 23 May 2026: Global prompt failed, edition used yesterday's run as backup. Investigating. 22 May 2026: One item presented an event from two years back. Tightened year verification in the prompt. 21 May 2026: One regional item out of date. Tightening the regional prompt. 20 May 2026: A couple of repeating items, some long items, generally ok edition. 18 May 2026: Tightened the consequence test to require effects that reach beyond a single country. 17 May 2026: Some out-of-style headline formulations. Tweaked prompt to fix it. 16 May 2026: Solid selection. One item showed an inflated source count, fixed for future editions. 15 May 2026: Formatting issues and mixed selection. Edited prompt with cleaner separation between events and intentions. 13 May 2026: Overall fair quality, with a couple of not so relevant items. 12 May 2026: Tweaks seem to work; fairly good edition today. 11 May 2026: Added a verb test and an explicit exclusion list to the daily prompt. 9 May 2026: Generally good selection, but with a couple of questionable items. 8 May 2026: Mixed quality and malformed sources field. Tightened the prompt for tomorrow. 7 May 2026: Decent edition. First day with Gemini 2.5 Pro as researcher and editor, replacing Claude Sonnet 4.5. 5 May 2026: Several weak items, will transition to new LLM this week. 4 May 2026: Weak edition, including sports and repeated items. Current setup not sustainable. 2 May 2026: Inaccuracies across news items. Continuing background testing of alternative LLMs. 1 May 2026: Two distinct morning editions, unintended effect from testing LLMs. 30 April 2026: Inconsistent quality and selection recently. Will test other LLMs in parallel before deciding Rumbo's fate. 25 April 2026: Selection still inconsistent. Added concrete consequence test before any item is included. 22 April 2026: Meanwhile section has hallucinations. Removing for good now. 21 April 2026: Two items missing geo tag; tweaked code to avoid in future editions. 20 April 2026: Meanwhile section weak, all items miscategorised. 19 April 2026: Meanwhile section reintroduced and added to newsletter emails. 18 April 2026: Tightened sentence length further, added deduplication rule for regional items. Re-introducing Meanwhiles from tomorrow. 17 April 2026: Many sentences too long; tightening instructions on length. 16 April 2026: Regional items working again. One main item outdated; need to tighten freshness rule. 15 April 2026: Regional calls have been failing silently. Implemented a fix, expecting results from tomorrow. 14 April 2026: Good edition, with one regional item only. 13 April 2026: One item several weeks old slipped through freshness check, partly repeating. Refined rule on deduplication context. 12 April 2026: Future events recurring in output. Rewrote freshness rule as positive framing — completed actions only. Trimmed both prompts by ~50%. 11 April 2026: Meanwhile section shows inconsistent quality over time; removing from future editions. 10 April 2026: Future events included as news items, and sentences exceeding word limit. Tightened both rules in prompt. 9 April 2026: Ok edition. No regional items generated. 8 April 2026: Decent edition. Refining rule to allow individual names when central to the story. 7 April 2026: Hallucination of ECB rate cut in global feed. Refining prompt to be more careful in areas with high risk of hallucination. 6 April 2026: Weak selection of stories and repetitions. Refined geographic spread rule and extended deduplication to 3-day rolling window. Manually rerun. 5 April 2026: Culture item in Meanwhile section still weak, needs tighter definition. 4 April 2026: Indonesia earthquake item lacking real consequence. Consider tightening of prompt re natural disasters.