我哋點樣搜集賽事

1. Our goal. To be the region's one stop race finder — one place that covers trail running, road running and triathlon across Hong Kong, Mainland China and the wider Asia-Pacific, so nobody has to open a dozen websites and reconcile a dozen calendars. Race information here is scattered across hundreds of organiser sites, social platforms and entry platforms in mixed languages, and runners face three problems: they have to check platform by platform to reconcile dates, distances and entry status; they cannot see the course before entering; and Chinese and English searches do not talk to each other. Two main functions follow from that goal. Function 1 — choose and plan races more easily: one search box plus a calendar and filters (year, month, discipline, region); a search that understands traditional and simplified Chinese, Hanyu Pinyin, Cantonese Jyutping, English place names and common event abbreviations; one page of facts per race (date, location, distance categories, elevation, entry status, official link); and side-by-side comparison so an athlete can plan a whole season — which race is the goal, which is a warm-up, which ones clash. Function 2 — GPS course analysis for effort insight and preparation: where a public route file exists we compute distance, total climb, major climbs, steepest sections and pace estimates (Minetti 2002 model), so you know what the race will actually demand of you rather than just a headline distance; that feeds training preparation (how much climbing to train, how to plan fuelling, what finish time is realistic, whether to arrive early to acclimatise) and trip preparation (where to stay relative to the start, transport connections, race-day timing) — the trip side is being added progressively. Function 3 (a further goal) — promote sport races around the region: a healthy race scene needs more than runners finding races; the races themselves need to be found and entered. So every race — international or small regional event — is listed by the same standard in the same place, and there is no paid promotion or paid placement; whether you are visible here is never decided by money. Each event gets its own page (/race/<slug>/) with structured data so search engines and AI assistants can read and index it, rather than being trapped inside a single JavaScript-only page. Cross-language search means a Cantonese-speaking runner, a Mandarin-speaking runner and an international runner all find the same race, so no event loses a whole group of potential entrants to a language barrier. We are most useful to smaller, independently organised races that would otherwise go unnoticed. What we will not do: sell rankings or placement, or alter data, ratings or ordering because of any partnership. It is not an entry platform, it charges nothing, it sells no data, and it does not try to be the authoritative record — the organiser's own page always wins.

2. Where the information comes from. Everything on this site is compiled from publicly available information: official organiser websites and official social accounts, public race calendars and public entry platforms, public web search results, and route files or corrections submitted by the running community. We have no partnership, licence, sponsorship or data agreement with any organiser. What we publish is factual event data — name, date, location, distance categories, entry status and a link to the official page.

3. How the collection works. An AI agent runs on a two-week cycle: it reads the public sources, proposes candidate events and normalises each one into a fixed schema. Candidates are then matched against the existing index — an event already present (same race, same date) is updated in place, never duplicated, even when public sources describe it differently. Region and discipline are assigned so events land in the right filters, and anything that cannot be classified or matched confidently is left out or flagged for manual review rather than guessed at.

4. How this website was built with an AI agent. The site — front end, data pipeline, course analysis, automation and ongoing maintenance — is built by an AI agent working with the site's owner. The division of labour is simple: the human sets direction, standards and decisions and gives final sign-off; the AI agent writes the code, runs the tests, deploys, investigates root causes and keeps the project log. In practice that means: a framework-free static front end on a global CDN; a structured data layer so search and filtering happen in your browser; scheduled jobs at different rhythms (race data every two weeks, traffic statistics hourly, a system health report each morning); course analysis computed from public route files; headless-browser testing on desktop and mobile viewports for every change, with a zero-JavaScript-error standard, plus human review for significant changes; automatic deployment with a cache version bump; and a written knowledge base so the same mistake is never made twice. The practical consequence for you: the data is machine-gathered, not hand-checked race by race, so it can lag or contain errors — and where a record cannot be matched or read confidently, we leave it out rather than invent a number.

5. What we deliberately do not do. We do not copy or re-host organiser content — no logos, photographs, article text, maps or route files; where a route file exists we link to the official file, which stays on the organiser's own server. We do not present entry fees or terms as if they were ours. We do not invent events that cannot be traced to a public source. We collect no personal data and set no tracking cookies.

6. Route profiles and course analysis. Where a public route file exists, the numbers are our own calculations: distance is summed point-to-point using the Haversine formula; elevation gain is computed after smoothing to suppress GPS noise; a major climb is a continuous ascent of at least 600 m horizontal length and at least 40 m of gain (both conditions required, a cumulative drop of more than 6 m ends the segment); pace estimates use the Minetti (2002) energy-cost model, with uphill capped at 60% grade and downhill at 25% faster. Because of smoothing and approximation, our figures can differ from the organiser's published numbers by roughly 1–3%.

7. How the index is organised. One page per event at /race/<slug>/. Search understands traditional and simplified Chinese, Hanyu Pinyin, Cantonese Jyutping, English place names, common event abbreviations, and both 10 km and 10K style distance writing. Filters by year, month, discipline and region.

8. Accuracy and limits. Please read this section carefully — we would rather set expectations honestly. Errors can occur at several layers, and each layer feeds the next: (1) the public source itself — race dates, courses, distances, entry status and fees can change, be postponed or be cancelled at any time, and an organiser's own public announcement can be wrong, out of date, or exist in several versions; (2) the route file (GPX) — most route files we use are submitted by the running community and are not necessarily the organiser's final version; the course may be re-routed, temporarily closed or adjusted, and the file itself may contain GPS drift, missing segments, duplicate logging, or even last year's route; (3) the course analysis — elevation profiles, distances, climb segments and pace estimates are all computed from that file, so any problem in the file carries straight into the numbers; (4) the AI agent and its tools — collection, matching, normalisation and analysis are performed by an AI agent, and automated processes and the tools they use can be wrong: parsing failures, matching a race to the wrong event, miscalculated values, or updates that were simply missed — "a machine did it" is not a guarantee of accuracy; (5) human review — significant changes get a human check, but a human can also miss something or judge it wrongly; human review is an extra layer, not a guarantee. We therefore cannot guarantee that any number, any route or any date is entirely correct. This index is a discovery and research tool, not an authoritative record. Always confirm on the official page before you enter, and treat the organiser's final published route as authoritative. Nothing here should be your only basis for a training or racing decision. We accept no liability for any loss, injury or inconvenience arising from use of this index. If you spot an error, please tell us (see the last section) — we will investigate and correct it.

9. Corrections and removal requests. Organisers and rights holders: email racingdaygpx@gmail.com and we will act within 24 hours. The same address is open to anyone who spots an error or wants to suggest an event.