Why we publish this
There are more than 2,600 listed companies in Korea. No single person can read every filing and recompute every metric each day. Bong Stocks solves that scale problem with automation and AI.
We also believe readers should not have to guess that this is happening — least of all where money is involved. So the sections below set out plainly which parts are machine-produced and what a human is accountable for.
1. Fully automated, no AI involved — every number
All figures on the site are computed by program code. AI does not produce numbers.
- Price, market cap, volume — taken directly from the FSC public API
- P/E, P/B, P/S, ROE, dividend yield — computed from public financial statements and the closing price using fixed formulas
- EV, net debt, EV/EBIT, FCF yield, DCF, forward P/B, relative strength — all computed in code
- Sector median comparisons and four-axis diagnostic grades — assigned by fixed rules
This layer is deterministic: the same inputs always produce the same output. Every formula is documented in Methodology.
2. AI-drafted — the written analysis
The narrative sections of each stock page — business description, growth commentary, recent events, investment considerations — are drafted by AI. Specifically:
- The model reads the company's original DART filings together with the computed figures from section 1.
- It writes to a fixed protocol — an internal document specifying what must be verified and what may never be written.
- The current-year earnings outlook is produced in a dedicated step: operating profit is estimated first, quarterly trends are projected consistently, and the result is checked internally against the range of market expectations.
- Nothing is published unless it passes the automated checks in section 3.
The model used is a large language model. It is never instructed to recommend a stock — the instructions explicitly prohibit buy/sell language, price targets, and scores or rankings.
3. Automated verification before publication
AI-written text is not published as-is. Before it goes live, code checks the following, and any failure blocks publication.
- Figure cross-check — P/E, market cap, current price and similar values written in the text must match the computed values
- Source restriction — cited links must be DART filings, official company material, or the exchange
- Prohibited language — buy/sell wording, price targets, investment opinions, scores or rankings are rejected
- Required elements — a minimum number of filing-based events, cross-verification entries and peer comparisons
- Korean–English consistency — figures must agree across both language versions
4. What the human does
The parts automation cannot own are handled directly by the site's operator.
- Rule design — which metric leads for which sector, how forward earnings should be projected, what may never be written. These are human judgements.
- Sample review — after each update, major stocks are opened and checked by hand against filings.
- Root-cause fixes — when something is wrong, the fix is not limited to one page. If the rule was wrong, the rule is changed, applied to every company again, and a regression test is added so it cannot come back.
- Classification corrections — companies whose official industry code does not match their real business are reclassified by hand.
A concrete example: large-cap automakers trading below book value were once shown as “overvalued”. The cause was a sector-median adjustment rule. Rather than editing those pages, we added an exception to the rule, recomputed every company, and added a test that catches the same error on every future build. That is what we mean by human oversight.
5. What we do not do
- We do not copy or rewrite content from other sites.
- We do not republish brokerage research, analyst consensus figures, or price targets.
- We do not create pages purely for search visibility. Company pages without in-depth analysis are excluded from search indexing and kept for browsing only.
- We do not ask AI to form investment opinions.
6. Limits, stated plainly
- AI-written passages can contain errors or awkward interpretation. Automated checks catch wrong figures and prohibited language, but they cannot fully validate whether an interpretation is sound.
- Current-year outlooks are estimates and can be wrong, especially for companies with volatile earnings.
- Not every sentence of every company is read by a human. Review happens at the level of samples and rules.
For that reason our material does not replace primary sources. Please verify with the original filings before making any decision. If you find something wrong, tell us through Corrections — we will fix the page and, where needed, the rule behind it.