Method

Written by the generator on each run, so the list can never fall behind the code. Any number on the board traces back to one of these 19 rules.

1. Prices are never recomputed here: every baht is read from trips/roadtrip-atlas-2027/data/costed.json, which the Python cost model wrote. curve[days][0] is one traveller, [11] is twelve.
2. Songkran rows are a fixed week (Sat 10 – Sun 18 Apr 2027) at a fixed per-person price: party size and trip length do not reprice them.
3. Safety verdict: the row's CANONICAL countries (§1.2) are looked up in solo-safety-2026.json first — the table is indexed under its own spelling, its parenthetical-free name and every canonical alias, so "Turkiye" and "Türkiye (Turkey)" are one country — then the raw survey string, then the worst verdict among the parts of a compound. Duplicate entries for one country resolve to the worst verdict; a compound entry only answers for a part that has no row of its own. No match stays 'unknown'.
4. Thailand rows read 'safe' (home country, no national advisory) except where the trip NAME or its ROUTE names, as a whole word, one of the 2026 advisory bands — the 50 km Cambodia border strip (Sa Kaeo, Buriram, Si Saket, Surin, Ubon, Chanthaburi, mainland Trat incl. the Ko Chang piers) or the deep south (Yala, Pattani, Narathiwat, the named Songkhla districts and the Hat Yai–Padang Besar line) — which read 'caution'. The survey's batch title is NOT searched (it is shared by seven unrelated trips), and Mu Ko Surin / the Surin Islands are cut out of the text before the scan so the Andaman park does not match the border province.
5. Nepal rows touching Langtang / Syabrubesi / Rasuwa read at least 'caution' (FCDO flood damage plus a US Level 4 carve-out on that corridor).
6. Visa kind: 'free' when the notes say visa-free or the fee is 0 without a 'visa required' note; then e-visa, then on arrival; 'required' whenever a fee is charged; else 'unknown'.
7. Visa lead time takes the TOP of any range — the number to plan against is the slow case.
8. Geography (SPEC §1.2) is derived from the COUNTRY string only. The survey's own `region` is a research-batch title ("Europe at 13-15 days"), not a place, so it is kept as `batch` and never faceted.
9. Country names are normalised to one spelling each (USA -> United States, Turkey/Turkiye -> Türkiye, Korea -> South Korea, Czech Republic -> Czechia, UAE -> United Arab Emirates, Hong Kong SAR -> Hong Kong). A compound country is split on / & + , and " and "; every part is kept in `countries`, so a France/Italy/Switzerland row is findable under all three, and `country` is the first.
10. A bracketed qualifier becomes `area` ("China (Tibet)" -> China + Tibet) unless it names the sovereign ("Faroe Islands (Denmark)") or the approach ("Italy (crossing from Switzerland)").
11. Region is one of 22 canonical values keyed by country, with two overrides: a row touching the Alps goes to Europe · Alps whatever its countries, and a Thai row is placed into one of six sub-regions by a majority vote of place-name keywords over its name, batch title and route.
12. In that Thai vote a keyword in the trip NAME counts three times one in the route, because a route lists everything passed on the way: Phetchaburi's Tham Khao Luang cave and the Nan river at Phitsanulok would otherwise pull two Central & West weekends into the Gulf and the North.
13. The Alps override yields when the Alps are a minority leg — a five-country Western Balkans drive that starts in the Julian Alps is a Balkans trip — so it only re-files a row where at least half its countries are in the Alpine arc.
14. The Alps override needs a European country: read literally it would file the Japanese Alps, New Zealand's Southern Alps, the Transylvanian Alps and Norway's Lyngen Alps in Europe. Named Alpine places (Chamonix, Aletsch, Dolomites, Alta Via…) fire anywhere in Europe; the bare words Alps/Alpine only inside the Alpine arc (CH, AT, LI, FR, IT, DE, SI).
15. A multi-country row outside Thailand and the Alps takes the region of the majority of its countries; a tie goes to the first country written, which is the one the survey led with.
16. Continent is derived from region, never stored separately: Thailand | Asia | Middle East & Africa | Europe (incl. Russia) | Oceania | Americas.
17. packageFlat is read from data/solo-board.csv (app/price_solo.py), the only artefact that carries the survey's trek.package_thb_per_person flag. A package price is a FLOOR for a solo traveller: it does not fall with party size.
18. Only a holiday whose scope is in holidays_th.DEFAULT_SCOPES {national, bank_private} sets `off`; government and Bangkok-bank-only days are shown but cost leave. Two rows on one date are merged, never overwritten.
19. score() is build_pack.score() ported verbatim (tier×20 + capped value − unbounded over-budget penalty + ease), rounded the way Python rounds (correctly, on the exact binary value) rather than by multiplying by ten first — which moved 90 of 711 rows 0.1 above the pack's own board.

Trips

846

Road-trip atlas

711

Songkran 2027

135

With a price curve

711

With a written plan

752

With photos

164

Cut, with reasons

77

Countries verified

119

Dataset built 2026-09-16 · holiday years 2026, 2027, 2028.