How we know

How we observe prices

We read each product page directly on a fixed cadence — hot tier every 6 hours, daily tier every 24 hours. Smart-spread sharding distributes requests so we never hit Olive Young's servers in a burst. Every observation is stored with server time, DOM hash, and a subset of response headers.

Integrity (Merkle + TSA anchoring)

Each day we hash every observation into a Merkle tree and anchor the root to the freetsa.org RFC 3161 Time-Stamp Authority. Anyone can later verify cryptographically whether we tampered with historical price data after the fact.

The FAKE_SALE classifier

Answers the simple question: "did the list price suddenly jump right before the sale?" Specific rules (per E058):

• List price rose ≥15% above 30-day median within 7 days before the sale started
• That list-price jump is NOT seen on the same SKU at other merchants (e.g. Naver)
• Both conditions → display the FAKE_SALE badge

The classifier is a heuristic and can mislabel. To appeal a classification, email hello@glowmeter.beauty with the SKU link.

What data we store

• Observations: product ID, price, timestamp, DOM hash, response-header subset
• Derived: 30-day low, average, savings percentage
• What we do NOT store: visitor IPs, cookies, accounts (there is no signup)

Open-source plan

The full codebase will be published to GitHub immediately after launch (per D002). Until then, only this methodology page is public; the crawler source stays private to reduce anti-bot evasion exposure.