Data and AI's role — who makes the numbers

Last updated: 2026-08-29

jini is very transparent about data and calculation. This guide explains where numbers come from and how jini handles them.

Data source: Yahoo Finance

All price and financial data comes from Yahoo Finance. U.S. stocks, Korean stocks — all of them.

Data reliability

Yahoo Finance is free and broad, but not perfect:

Delay: prices lag ~15 minutes (U.S. market), financials 24+ hours
Accuracy: rarely, but unrecoverable errors happen
Coverage: micro caps and IPOs may have incomplete data

jini does not hide this. The page footer always shows the reference date, and when data is missing, it honestly says "insufficient data".

Data cache intervals

Price: 15-minute cache — refreshed every 15 minutes
Financials: 24-hour cache — refreshed once a day
AI analysis: 7-day cache — same stock, same base numbers, no recharge

This balances real-time updates against cost and server load. Real-time every minute would work, but the server would crash. Investment decisions work fine on 15-minute intervals.

AI's role — interpretation, not invention

jini's AI works like this:

When you click [Generate AI analysis]:

All calculation outputs (scores, verdicts, fair values, etc.) become the input
AI generates an interpretation: why did these numbers appear?
AI invents no new numbers — it only explains the input

Examples:

✅ AI: "Earnings are declining, so growth score is low"
❌ AI never does: "By my calculation, fair value is $52"

What AI generates

WHY chain — why did the market price it this way, a 5-step causal chain
Undervaluation type (A–E) — what kind of mispricing is this
Moat assessment — how durable is this company's competitive edge
Verdict narrative — why the final verdict makes sense
Predicted risks — what could make this verdict wrong

Number validation gate

When AI quotes a number in its interpretation, that number is auto-validated against input data. Example:

AI says: "P/E is 12x, below the 18x market average"
Validation: is P/E actually 12x? Match, it shows. Mismatch, it does not.
If AI incorrectly said 14x, that whole sentence never reaches the user

This prevents AI from fabricating.

BYO (Bring Your Own) API key

jini's AI uses your own Anthropic API key:

Key lives only in your browser — jini's server never stores it
You pay per use — roughly tens to hundreds of yen per analysis (per Anthropic pricing)
Privacy — the server does not store analysis history, so no record remains on the server

If you do not provide a key:

All calculation features (scores, verdicts, screener, expected price) work fine
AI analysis only is unavailable
Cost: zero

Why make it so complex?

It would have been simpler to say "jini's server calls AI for you". But:

Cost: if jini paid for every user's analysis, the service would be unsustainable
Scale: thousands of concurrent AI calls would hit API rate limits
Responsibility: jini storing and validating AI results adds no value

So the policy is "bring your own API key". This is transparent.

What "insufficient data" means

jini labels these as "insufficient data":

Yahoo lacks a required financial item (e.g., quarterly cash flow)
The company went public too recently to have enough history
Special situation (SPAC, startup) where standard metrics do not apply
Micro cap without Yahoo coverage

In these cases jini:

❌ Does not shrug and skip a verdict
❌ Does not guess or estimate from small samples
✅ Explicitly says "insufficient data" and states the reason

This is jini's principle: calculate only from data you have; if you do not have it, admit it.

Caching and refresh

Candidates look old — check the reference date. If 24+ hours have passed, manually re-run screening or wait for the next auto-refresh.
Numbers are not changing — caching. We store 7 days (AI) · 24 hours (financials).
AI analysis feels slow — API calls go directly from your browser to Anthropic, so 1–3 seconds depending on network and API state.

You understand data. Next, learn how jini verifies itself.