We read every review so you don't have to.

Ownership in these hills turns over faster than aggregate ratings can track. A homestay that ran beautifully in 2020 may be under new hands — or a new caretaker, or a new cook — by 2025. The star average lags for years because every old five-star still counts. Guests write about what they actually experienced last month; the average number smooths that over.

The Kanchen Score is built by reading the recent reviews, not by trusting the running mean of every review ever written.

For every indexed property, we pull the last 24 months of guest reviews — up to 500 per property, in any language people write in. Older reviews are archived, not deleted; they stop counting toward the score but stay in the record. When a new review comes in, the oldest recent one drops out.

The effect is quiet but powerful: a property that has picked up in the last year gets credit fast, and a property that has slid gets flagged fast too.

The pillar framework wasn't invented in a workshop. It emerged from reading thousands of reviews across the two districts and noting the categories guests kept spontaneously returning to. What people complained about, praised, and warned each other about clustered into five stable groups:

• Shelter — rooms, cleanliness, warmth, upkeep.
• Water — hot water reliability, plumbing, supply.
• Hearth — food, hospitality, the warmth of the hosts.
• Path — road access, last-mile, arriving with luggage.
• View — what you actually see at first light.

The AI scores every review across all five pillars before the composite Kanchen Score is calculated. Because the pillars are scored independently, one strength can't paper over one weakness — a run of five-star View reviews won't rescue a property whose Water pillar collapses.

Twenty solid recent reviews and four solid recent reviews are not the same evidence, and pretending they are is how bad rankings get made. Every property card on Kanchen carries a confidence tier so the strength of the signal is visible next to the score itself:

• Deep review — enough recent reviews for a stable, high-confidence composite.
• Limited reviews — a real signal, but a narrower one; use with care.
• Summary only — very few or older reviews; the score is directional at best.

The default ranking respects this. A score built on thin evidence never leapfrogs a slightly lower score built on deep evidence. Properties with lots of old reviews but nothing recent are flagged as stale — a warning, not a rank.

• Take paid placement. No property can buy a rank, ever.
• Manufacture a decisive-looking score from four reviews.
• Hide the confidence tier behind a marketing star rating.
• Rewrite or edit guest reviews. The AI reads them; it doesn't republish them.

The methodology page has the exact rules the engine follows, including how limited-evidence scores are handled and what disqualifies a property from the index. The guide has every homestay we have indexed so far, with its score, tier, and the reviews it was built from.