Reader pass prompt — v1
You are a thoughtful reader making a second pass over a source about Swedish entrepreneur Martin Källström. A first pass already harvested atomic facts. Your job is everything the fact harvest structurally misses: the stories, the arguments, and the surprises. Read the ENTIRE source before writing anything.
Three registers
1. narratives — stories as indivisible units
A story with an arc (setup → turn → payoff) that loses its meaning when atomized. Capture the arc in the summary. The span should cover the whole telling so it can be played back as one clip.
2. theses — Martin's arguments, with their supporting structure
Not isolated opinions — arguments: a claim plus how he builds the case for it in this source. Quote his own formulation. Link to the narratives/examples he uses as evidence. A good thesis is one you could put as a section headline on a "Martin's thinking" page.
3. observations — the serendipity register
The surprising, memorable, revealing, or delightfully specific — about ANYTHING: Martin's character, the era, the Swedish tech scene, odd numbers, casual revelations, prescient or badly-aged predictions, moments of humor or vulnerability. No bucket obligation, no relevance test beyond "a curious person would remember this." Do not repeat what plainly belongs in the first-pass facts.
Traceability (same hard rules as the fact pass)
- Quotes are VERBATIM and CONTIGUOUS from the source, original language, ≤240 chars.
- All *_seconds values are COPIED from the
[hh:mm:ss]marker preceding the passage — never estimated. Articles: null. - Verify every quote with grep -F against the source before writing; fix paraphrases.
Output schema (STRICT JSON, no markdown fences)
{
"source_id": "
Quality bar
- Narratives: only real stories with an arc — typically 1-5 per source; 0 is valid.
- Theses: only claims Martin actually argues — typically 1-6 per source; 0 is valid.
- Observations: 5-15 for rich sources, fewer for thin ones.
- English prose, original-language quotes. Never invent.