Every session, a language model reads that day's headlines from three Indonesian financial news feeds and files one number: how the market's mood reads, from −100 (panic) to +100 (euphoria), with 0 as neutral. It is a first-pass impression of the news, not a measurement of the market itself — the same headlines read twice, or read by a different model, would not always file the same number.
Reading today's edition…
The Reading, Over Time
7-day trailing average · −100 panic to +100 euphoria · 0 neutralThe line is a 7-day trailing average — one LLM read a day is noisy score-to-score, so the average is what carries the trend; the faint dots are each day's raw score. The live feeds only expose a rolling window, so the first three months here were filled in after the fact from search-engine results instead — from today it lengthens by one point a day, live. Green above the neutral line, red below.
Terms of the Reading
Three feeds · one model · one call a daySource. Three feeds — CNBC Indonesia's general market section, Kontan's investasi feed, and Detik Finance — each a mix of genuine market news and unrelated stories (disasters, consumer banking tips, retail promotions). The model reads each headline with its own summary, weighs market-wide news over sector news over a single company's news, and ignores what isn't market-relevant.
Judge. One call to a language model per day, not an ensemble and not a re-read for consistency. Treat the number as one reader's first impression, not a poll.
Window. From today onward, whatever the three feeds published that trading day, in Jakarta time — typically several dozen headlines. Further back, the feeds hold no archive at all, so the first three months were filled in by asking a search engine what each site published on that date instead — coverage that thins the further back it reaches, which is why the line doesn't yet stretch a full year.