The calculation layer: manseryeok first
Before interpretation, the eight characters have to be right. The month follows solar terms, not the civil 1st. A late-night birth can cross a day at the hour-branch boundary. Hand that layer to a language model and the same question yields different charts. Sajuping casts the chart with a calculator, then hands that result to the interpretation layer.
Where general chatbots usually slip is written up in why ChatGPT saju goes wrong. An eloquent wrong chart is still not myeongni.
The interpretation layer: talk, with memory
Once the chart exists, Sajuping holds those eight characters and the situation you already named, then answers the next question. Job change, relationship, today’s weather — the natal chart should not be re-asked from zero. Memory is not perfect. Still, not making you paste your palja again is what makes this a different product from a one-shot shop reading.
We do not claim to replace a human reader’s in-person intuition. Consistency of calculation, twenty-four-hour access, and a recorded conversation are Sajuping’s seat. The comparison lives in AI saju vs traditional saju.
Naming limits is part of the method
Without a birth time the hour pillar is missing and resolution drops. Near a solar-term boundary, software should warn. A model can write sentences; it cannot see your manager or this morning’s face. That is why Sajuping’s copy is aimed at a self-understanding tool, not a prophecy engine.