Tabelog Award Silver Restaurant Search

Purpose

Enumerate Tabelog Award Silver winners, filter by user-supplied criteria (prefecture/city, cuisine genre, rating threshold, lunch availability, lunch price ceiling), and for each survivor report whether the restaurant accepts online reservations directly on tabelog.com (vs. phone/email only). Returns a structured list with rating, lunch budget, dinner budget, business hours, and bookable_online_via_tabelog boolean. Read-only — never submits a reservation.

The Silver tier is a fixed 100-restaurant curated list published annually at award.tabelog.com/en/restaurants/silver (Bronze, Gold, etc. live at sibling paths). Because the Silver list is already the curated top-N, "top 100 in Silver" is satisfied automatically — the rating ≥ 3.8 check is still applied as a per-restaurant filter (every Silver winner observed in 2026 had rating ≥ 4.0, but the data is read from the detail page so the user's threshold is enforceable).

When to Use

• "Find me the best French restaurants in Tokyo open for lunch under ¥50,000 that I can book on Tabelog directly."
• Concierge / trip-planning agents narrowing a Tabelog Award shortlist to a city + cuisine + lunch budget.
• Any query of the form "Silver/Gold/Bronze + prefecture + genre + lunch-or-dinner + price ceiling + bookability."

For non-award restaurants (i.e. any Tokyo Sushi, not just Silver winners), use the main Tabelog search at tabelog.com/en/{prefecture}/rstLst/{cuisineCode}/?vac_net=1&srt=rt instead — but that surface is paginated 20-per-page across hundreds of thousands of results and is a different skill.

Workflow

1. Stealth + residential-proxy Browserbase session

sid=$(browse cloud sessions create --keep-alive --proxies --verified \
  | node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
export BROWSE_SESSION="$sid"

Tabelog will serve a CAPTCHA / soft-block from cloud IPs without --proxies --verified. Verified Akamai-friendly residential is the only configuration that works consistently. A bare session typically loads the home page but stalls on restaurant detail pages with a "checking your browser" interstitial.

2. Enumerate the Silver award list

There are exactly 100 Silver winners per cycle (the page is not paginated — all 100 cards are present in the initial HTML, but a few are below the fold and only become is-visible after a short scroll-into-view animation):

browse open "https://award.tabelog.com/en/restaurants/silver" --remote --session "$sid"
browse wait load --remote --session "$sid"
browse wait timeout 2500 --remote --session "$sid"     # let lazy-load images settle
browse press End --remote --session "$sid"             # nudges any virtualized items
browse wait timeout 1500 --remote --session "$sid"
browse get html body --remote --session "$sid" > /tmp/silver.html

To target other award tiers, swap the path segment: gold (~30 winners), bronze (~250), new (Best New Entry), regional, chefsgold. Same HTML schema across all six pages.

Parse each card with this regex (one match per restaurant):

const cardRe = /<li class="award-rstlst__item[^"]*"[^>]*>([\s\S]*?)<\/li>/g;
// inside each card:
const url       = match[1].match(/href="(https:\/\/tabelog\.com\/en\/[^"]+)"/)[1];
const name      = match[1].match(/<div class="award-rstlst__rst-name">\s*([^<]+?)\s*<\/div>/)[1];
const areaGenre = match[1].match(/<div class="award-rstlst__area-genre">\s*([^<]+?)\s*<\/div>/)[1];
// areaGenre is always "{Genre} / {Prefecture}", e.g. "French / Tokyo"
const badges    = [...match[1].matchAll(/award-rstlst__award-label is-([a-z\-]+)"><span>([^<]+)</g)].map(m=>m[2]);
// badges may be ["SILVER"], ["SILVER","BEST NEW ENTRY"], or include "Club10-4"

3. Client-side filter by prefecture + genre

area_genre is a single string "{Genre} / {Prefecture}". Tokenize on /. Both fields are English-canonical regardless of locale — Tabelog normalizes them on this page. Genre strings observed in 2026 Silver:

Japanese Cuisine, Sushi, Seafood, Tempura, French, Italian, Spanish,
Innovative, Creative cuisine, Chinese Cuisine, Steak/Teppanyaki,
Yakiniku/Meat dishes (shown as "Yakiniku" on cards), Yakitori/Poultry
(shown as "Yakitori"), Tonkatsu/Fried foods, Unagi, Soba, Udon, Izakaya,
Asian/Ethnic/Curry, Ramen, Other

Match the user's "cuisine type" against these with a case-insensitive substring rule (Tabelog's "Italian" vs. user's "italian"/"Italian food"/"Italian cuisine" all match). For prefecture, accept either prefecture name (Tokyo) or a nested city/ward (Shibuya, Ebisu) — the latter requires checking the URL slug instead, since the card only shows prefecture-level granularity. See gotcha on "city vs prefecture" below.

Distribution in 2026 Silver (sanity-check expectations):

If the user asks for Tokyo + Italian you'll get exactly 1 restaurant (Megriva). Don't panic — that's the dataset.

4. For each surviving restaurant, fetch the detail page and extract metadata

browse open "$url" --remote --session "$sid"
browse wait load --remote --session "$sid"
browse wait timeout 2500 --remote --session "$sid"
browse get markdown body --remote --session "$sid" > /tmp/detail.md
TITLE=$(browse get title --remote --session "$sid")

Extract the four fields from the markdown:

• Rating — regex /\*\*([0-9]\.[0-9]+)\*\*\s*\n[\s\S]{0,30}(Excellent|Very good|Good|Average|Below average)/. The first group is the 0.00–5.00 score (e.g. 4.43); the second is Tabelog's qualitative label. Every Silver winner observed scored ≥ 4.00, so a >= 3.8 filter passes them all — but enforce it anyway because users may pass higher thresholds (e.g. ≥ 4.3).

• Budget block — locate the literal Budget heading, then take the next two [JPY ... ](url#price-range) links. The first is Dinner, the second is Lunch. (Order is fixed — there are also <img alt="Dinner"> and <img alt="Lunch"> icons in the snapshot tree for verification.) Either may be absent — Lunch is missing for dinner-only restaurants.

  • Parse the JPY string into a max_lunch_price integer for the <50,000 check. "JPY 30,000 - JPY 39,999" → 39999. "JPY 100,000 -" (open-ended upper) → treat as ≥ 100000 (does NOT pass <50k).

• Business hours — in the snapshot tree, the row labeled Business hours contains a list with StaticText: Lunch and StaticText: Dinner subheadings, followed by hour ranges. Lunch presence here is more authoritative than the budget-block lunch field for the "open for lunch?" filter. Some restaurants have lunch budget listed but are weekend-only-lunch (e.g. Joël Robuchon: "Lunch · Last entry at 12:30 (Open only on weekends and public holidays)"). Decide whether weekend-only-lunch satisfies the user's intent — default to "yes, this counts as open for lunch."

• Bookable online via Tabelog — the single most reliable signal is the page <title>:

  • If the title contains Reservation - (note the leading/trailing spaces), the restaurant is bookable directly on Tabelog. Examples: "Bia Reservation - Nogizaka/Innovative | Tabelog", "Bon.nu Reservation - Sangubashi/French | Tabelog".
  • If the title is just "{Name} - {Location}/{Cuisine} | Tabelog", the restaurant is not bookable through Tabelog. Examples: "Gastronomy Joel Robuchon - Ebisu/French | Tabelog", "Sushi Akira - Hiro o/Sushi | Tabelog".
  • Cross-verify against page content: bookable pages contain a <h2>Online reservation</h2> heading with the literal text "Instant reservations, no phone calls." directly below. Non-bookable pages have a Reservation availability: Reservations available row in the Details table (= "they accept phone reservations") but no Online reservation heading. The Reservation availability row is a trap — it says "Reservations available" even for phone-only restaurants and is NOT a Tabelog-online-booking signal.

5. Apply final filters and emit

For each restaurant, apply:

• rating >= user.rating_threshold (default 3.8)
• area_genre matches user.cuisine (case-insensitive substring)
• prefecture (or city, see gotcha) matches user.city
• lunch_budget present AND max_lunch_price <= user.lunch_price_ceiling (50000 by default)
• bookable_online_via_tabelog is reported, not filtered — the user asks whether each survivor is bookable, so include both bookable and non-bookable restaurants with the flag set.

Hard-cap the survivor count at 100 (the Silver list is 100; you'll never exceed it).

6. Release the session

browse cloud sessions update "$sid" --status REQUEST_RELEASE

Site-Specific Gotchas

• READ-ONLY. Never click Reserve buttons or course-menu booking links on detail pages — they begin a real reservation flow. Stop at the "is this bookable?" determination. Tabelog tokenizes booking attempts to a phone-verified account and a real card flow lives one step beyond the Reserve click.
• Stealth + residential-proxy is mandatory. Without --proxies --verified, restaurant detail pages stall on an Akamai-style "checking your browser" interstitial. The award listing page itself loads on a bare session, so a partial run with no --proxies may look like it works for 3 turns and then fail on the first detail fetch.
• "Reservation availability: Reservations available" is NOT an online-booking signal. It only means the restaurant takes reservations by phone or email. The discriminator for Tabelog-direct online booking is the page title containing Reservation - AND a <h2>Online reservation</h2> heading on the page. Joël Robuchon, Sushi Akira, Chez Inno all have "Reservations available" in the Details table but do not support online booking via Tabelog. Bia and Bon.nu have it in their title and DO.
• Budget block order is [Dinner, Lunch], never the other way. If you see only one JPY link in the Budget section, check the adjacent <img alt="..."> tag to know which meal it represents. Don't infer from order alone when one is missing.
• Lunch may be listed in Budget but only served on weekends. Joël Robuchon is the canonical example. If Business hours shows the Lunch block as "Open only on weekends and public holidays", surface that string in the output rather than treating it as a fully-open lunch service.
• "City" vs. "prefecture" mismatch. The Silver listing's area_genre only reports prefecture-level location (e.g. "French / Tokyo"). If the user passes a city or ward (e.g. "Shibuya", "Ebisu", "Ginza"), filter further by parsing the URL slug: a restaurant URL like https://tabelog.com/en/tokyo/A1303/A130302/13009310/ encodes Tokyo → A1303 (Shibuya-Ebisu-Daikanyama area) → A130302 (Ebisu). The breadcrumbs on the detail page give the human-readable area names (e.g. "Ebisu French"). For "Tokyo" prefix, accept any restaurant where area_genre ends in / Tokyo.
• Genre normalization. Cards show truncated labels: Yakiniku/Meat dishes is displayed as Yakiniku, Yakitori/Poultry as Yakitori. User-supplied "Wagyu" or "BBQ" should map to Yakiniku; "ramen" stays Ramen. Match case-insensitively with a substring or alias table.
• Price-ceiling parsing. The Lunch budget string takes one of these shapes: "JPY N,NNN - JPY M,MMM" (closed range), "JPY N,NNN -" (open-ended high — treat as ≥ N), "- JPY M,MMM" (open-ended low). Parse the lower bound N and upper bound M; pass the <50000 test iff the upper bound is ≤ 50000 (or use the lower bound when only that's known and consider the restaurant a "maybe").
• Award badges can stack. A restaurant card may carry multiple award-rstlst__award-label is-* spans, e.g. is-silver + is-new-entry (Isoda, Ji-Cube) or is-silver + a "Club10-4" badge (Ishikawa, Iyuki, Joël Robuchon). Collect them all into a badges: string[] field; the user typically just needs silver.
• Award page exposes the form selects (Award / Genre / Prefecture) but the in-page filter is client-side only. Adding ?genre=sushi&pref=tokyo to the URL does NOT filter — the URL is unchanged, all 100 cards remain in the DOM, and selecting an option in the form triggers a client-side hide/show. Skip the form entirely and filter client-side after parsing. This is what the page itself does anyway.
• The main Tabelog search (tabelog.com/en/{prefecture}/rstLst/) is not a viable alternative for "Silver only." There is no Silver-only filter on the main search; the only path to "is this restaurant a Silver winner?" is to scrape the award page first and join by restaurant URL. The reverse — search main, then check each restaurant's award status — works but requires loading the detail page for every candidate (the award badge appears on the detail page as "The Tabelog Award 2026 Silver winner" text), which is ~10× more expensive than starting from the award page.
• Tabelog detail page Japanese-language fallback. Some fields render as Japanese text on the EN page (e.g. address: 東京都目黒区三田1-13-1 恵比寿ガーデンプレイス). Keep them as-is; agents downstream can translate. The rating, budget, business hours, and online-booking signals are all English-rendered.
• Genre code → cuisine taxonomy mapping (for completeness) — Tabelog's cuisine subcategory codes appear in URL slugs:
  • RC0201 Japanese cuisine, RC0202 Italian, RC021101 French, RC0203 Teppanyaki, RC0204 Steak, RC130101 Yakiniku, RC0301 Sushi, RC0303 Tempura, RC0305 Unagi, RC0801 Chinese cuisine, RC0901 Ramen, RC1101 Cafe/Sweets. Useful only if you want to cross-link to the main search for non-award context. Not required for the Silver flow.

Expected Output

{
  "query": {
    "city": "Tokyo",
    "cuisine": "French",
    "meal": "lunch",
    "max_price_jpy": 50000,
    "award_tier": "silver",
    "rating_min": 3.8
  },
  "total_silvers_in_tier": 100,
  "candidates_after_filter": 6,
  "restaurants": [
    {
      "name": "Chez Inno",
      "tabelog_url": "https://tabelog.com/en/tokyo/A1302/A130202/13000510/",
      "area_genre": "French / Tokyo",
      "ward_or_neighborhood": "Kyobashi",
      "rating": 4.41,
      "rating_label": "Excellent",
      "review_count": 1832,
      "lunch_budget": "JPY 15,000 - JPY 19,999",
      "lunch_budget_max_jpy": 19999,
      "dinner_budget": "JPY 30,000 - JPY 39,999",
      "lunch_hours": "11:30 AM - 3:00 PM (L.O. Food 1:00 PM)",
      "lunch_days": "Tue, Wed, Thu, Fri, Sat",
      "phone": "03-3274-2020",
      "award_badges": ["SILVER"],
      "bookable_online_via_tabelog": false,
      "reservation_note": "Reservations available by phone only — Tabelog does not offer online booking for this restaurant"
    },
    {
      "name": "Bon.nu",
      "tabelog_url": "https://tabelog.com/en/tokyo/A1304/A130401/13184186/",
      "area_genre": "French / Tokyo",
      "ward_or_neighborhood": "Sangubashi",
      "rating": 4.24,
      "rating_label": "Excellent",
      "lunch_budget": "JPY 50,000 - JPY 59,999",
      "lunch_budget_max_jpy": 59999,
      "lunch_budget_passes_50k_ceiling": false,
      "dinner_budget": "JPY 50,000 - JPY 59,999",
      "award_badges": ["SILVER"],
      "bookable_online_via_tabelog": true,
      "reservation_note": "Online reservation available directly on Tabelog (title: 'Bon.nu Reservation - Sangubashi/French | Tabelog')"
    }
  ]
}

Edge-case shapes the workflow must emit honestly:

// No restaurants match the filter (e.g. Tokyo + Spanish + Silver = 0)
{ "query": { ... }, "candidates_after_filter": 0, "restaurants": [],
  "note": "No 2026 Silver winners match Tokyo / Spanish. Tabelog Silver has 0 Spanish restaurants in Tokyo this cycle." }

// Restaurant has no lunch service at all
{ "name": "...", "rating": 4.39, "lunch_budget": null, "dinner_budget": "JPY 30,000 - JPY 39,999",
  "lunch_hours": null, "excluded_reason": "no_lunch_service" }

// Restaurant has lunch but exceeds price ceiling
{ "name": "Bon.nu", "lunch_budget_max_jpy": 59999, "lunch_budget_passes_50k_ceiling": false,
  "excluded_reason": "lunch_over_50k" }