Lightspeed Trainer

1. Read the fast notes below. 2. Run the flashcards until the rules feel automatic. 3. Take the Practice Test for explanations. 4. Use Timed Mastery as your main mode: the timer runs behind the scenes while you answer, so there is no countdown panic. After each answer it tells you your time. A question only counts as mastered when you get it right, fast enough, twice in a row. Missed or slow questions keep coming back until they are mastered. That is the whole point: accurate AND fast, like the real work.

Start Timed Mastery Start Flashcards

Honest note: this is built from the older Guidelines version Jeremy found (Search Satisfaction v2.3.1, plus Overall Preference v2.2). OneForma says Lightspeed is the follow on to Milky Way and uses certifications to train you, so expect the live project to have its own current training too. Use this to learn the core grading brain, then trust the newest official training inside OneForma if anything conflicts.

You see a query a user typed on an Apple device (Safari web search or Spotlight on device search), plus their location and the date. You judge ONE result at a time: did this result satisfy what that user actually needed?

Can never be HS: advice and recommendation queries like how to lose weight, a recipe, best Beatles song, thai restaurant. One opinion can never be what almost all users would agree on. Also: a result for a secondary interpretation caps at SS, and if there is no dominant interpretation a would be HS result caps at S.

TAG vs Try Rating: In TAG, flag Wrong Language or Inappropriate and submit, done. Flag CU, add a comment, submit. In Try Rating, flag it AND still rate it, and every flagged result rates Not Satisfying.

Missing context rule: If query context you NEED is missing (location for weather outside, date where date decides the grade) flag CU. If the missing info clearly would not change the grade (date missing on a dictionary definition) just grade it. If RESULT context required for that type is missing, flag CU.

• Dominant interpretation exists: grade the dominant one normally. A secondary interpretation result caps at SS even if it would be HS for that meaning. Example: michael jordan result about Michael B. Jordan = SS, the NBA player is dominant.
• No dominant interpretation: grade normally, except a would be HS becomes S, because you cannot say almost all users wanted it. Example: um athletics could be Michigan, Miami, Maryland.
• Explicitly locale sensitive (query names the place, like amazon france): wrong locale = NS automatically.
• Implicitly locale sensitive (ticketmaster, visa rules, local law, a nearby business): wrong locale = NS automatically, even in the right language.
• Mildly locale sensitive (vaccine recommendations): foreign result in the right language drops ONE level. HS to S, S to SS, SS to NS.
• Not locale sensitive (tennis news): grade without regard to locale.
• English in a non English locale: where most users are fluent, grade normally. Where many are fluent (much of Western Europe), drop one level. Where few are fluent, NS.
• Spelling and autocomplete: grade as if the user typed the corrected query. fac = Facebook, ted cruise = Ted Cruz.
• Redirects: grade the page you land on, not the URL shown.

Lightspeed is local search, so Maps grading matters more here than anywhere else. Distance, intent and closeness decide the grade.

• No distance shown on a Maps card = Content Unavailable (and NS in Try Rating).
• Grade Maps on what is visible in the title and description only. Do not grade NS just because clicking goes nowhere. If multiple Maps results appear, grade on the FIRST one only.
• Permanently closed: drop one level ONLY if a similar or same business is open and nearby. Otherwise no penalty, knowing it is closed is useful.
• Distance is relative: a Lexus dealer 30 miles away can be S or even HS if it is the closest, people travel for expensive rare purchases. Donuts 30 miles away = NS. Rural users travel farther: restaurants 39 miles from a town of 237 people can be S, the same distance from New York City = NS.
• No map intent = NS: if the user was not looking for a place, a far Maps result is a tell. Prime Video query returning a Prime Time Video store 529 miles away = NS. Lakers query returning Great Lakes Brewing 2,165 miles away = NS.
• Local intent words matter: nearest, closest, near me, plus businesses you must physically visit, gas stations, cinemas, coffee.

• Skipping web research: misunderstanding the query, misunderstanding which interpretation is dominant, or inventing a dominant interpretation that is not there.
• Grading from the snippet: never click the destination and you miss dead pages, redirects to ads, and the wrong company with the same name (American Eagle web design company, not the clothing brand).
• Ignoring time and place: Harold Kitchen menu page that is actually in British Columbia, 3,000 miles from the Virginia user = NS. A 2022 election query answered with 2016 results = wrong date, wrong grade.
• Ignoring conceptual distance: Yelp reviews for Chez Panisse are useful but one degree away, so not HS. Too high in a site hierarchy (whole State Department site for passport info) and too low (lost passport page when the user never said lost) both hurt.
• Matching words instead of meaning, and using search engine ranking as your grade. Both are named mistakes in the guidelines.
• When no scenario fits, go back to definitions: is this what almost all, many, some, or no users would want?

Separate task from Search Satisfaction. Do not mix them. You get TWO result sets side by side for the same query. Rate every result using the same satisfaction scale, then pick: About the Same, Slightly Better, Better, or Much Better, for Left or Right.

• Prefer the side with higher satisfaction grades.
• Prefer the side where the higher graded results rank higher up. Position matters, position 1 differences outweigh position 4.
• Prefer a more varied set (maps, apps, web, different meanings) when grades are close.
• More results does NOT automatically mean better.
• Truly stuck? Choose About the Same.

• Long term project. Lightspeed is presented as the next step after Milky Way.
• Work: evaluate tasks made by other users, varied nature, from user intent to authentication of data accuracy. That is exactly this grading work.
• Requirements: living in your location at least 5 years and know local businesses, points of interest and colloquialisms; native or fluent in the local language and very fluent in English; good internet and a computer; at least 10 hours per week; willing to take the project certifications.
• A CV must be uploaded to your OneForma profile before applying, or processing can stall.

Application page: oneforma.com/projects/local-search-quality-evaluator - Jeremy, the link after your colon did not come through in that message, so this is the project link you sent first.