24 prompts to merge Best of Austin into every subdomain

Working set. Data lands in /opt/wholeatx/bestof/winners.json as records of {name, category, year, poll, url, place, notes}. The roster of real businesses is /opt/wtdirectory/osm-travis.json (via roster.records()), scores are in audit-austin.json, sector assignment is sectors.assign(), and the site generator is /opt/wholeatx/wholeatx.py. Every prompt below assumes attribution to and links back to the Austin Chronicle.

Best of Austin publishes at austinchronicle.com/best-of-austin/year:YYYY/poll:PPP/category:CCC/.
Enumerate every valid (year, poll, category) tuple for years 2024 and 2025, polls "readers"
and "critics", across all 8 sections (Food & Drink, Services, Shopping, Nightlife, Arts &
Entertainment, Sports & Recreation, Kids & Family, Politics & Media). Output a checklist of the
exact index URLs to crawl. Do NOT fetch yet — just produce the target list and confirm the URL
pattern against two live pages.

Fetch each index URL from prompt 01 with a 3-second delay between requests, an honest
User-Agent ("WholeATX-research/1.0 (+https://wholeatx.com; walhus@gmail.com)"), and on-disk
caching so a re-run never re-fetches. Save raw HTML to /opt/wholeatx/bestof/raw/. Stop and
report if you see 403/429 — do not retry in a loop.

Parse the cached Best of Austin pages into /opt/wholeatx/bestof/winners.json — one record per
award: {name, category (the exact Chronicle category label), section, year, poll, place
(1st/finalist), url (the specific poll page), blurb_excerpt (<=160 chars, for attribution
context only)}. Preserve the Chronicle's exact business name string verbatim in raw_name.

Extend the parser to also capture finalists / runners-up where listed, tagged place:"finalist".
A finalist with a broken website is nearly as good an outreach lead as a winner — we want them.

Add a canonical_name to every winner: lowercase, strip legal suffixes (LLC, Inc, Co), collapse
"&"/"and", drop location qualifiers ("- South Lamar"), normalize whitespace and punctuation.
Keep raw_name untouched. This canonical form is the join key against the roster.

Some winners are chains with many roster rows (e.g. a coffee brand with 12 Austin locations).
Flag winner records as multi_location and decide the rule: the badge attaches to the BRAND
(dedupe by canonical_name) not to each storefront. Document the rule in the file.

Load the roster via roster.records(osm-travis.json). For every winner, attempt an exact
canonical_name match against roster canonical names. Write matches to
/opt/wholeatx/bestof/matched.json with match_confidence:"exact" and the roster osm_id + website.
Report the exact-match rate per category.

For unmatched winners, run token-set fuzzy matching (rapidfuzz) against the roster, restricted
to the same sector where possible. Auto-accept >=92 similarity; queue 80–92 into
/opt/wholeatx/bestof/review.csv for a human yes/no; discard <80. Never guess a match into
production — a wrong badge on the wrong business is worse than no badge.

List winners that matched NO roster business. Split them: (a) genuinely missing from our
roster — queue for a targeted add; (b) present but under a different name — fix the alias.
Producing this "beloved businesses WholeATX doesn't even know about yet" list is a deliverable.

Build /opt/wholeatx/bestof/crosswalk.json mapping every distinct Chronicle category string to
one or more of the 85 WholeATX sector slugs (see the brainstorm's starter table). Where a
category is ambiguous ("Best Local Business"), map to a network-wide bucket, not a wrong sector.
Flag any category that maps to zero sectors for review.

Produce /opt/wholeatx/bestof/by_sector.json: for each sector slug, the list of its matched
winners with {business, category, year, poll, place, website_score (from audit-austin.json),
chronicle_url}. This single file is what the generator will read.

Join every matched winner to its website score. Emit /opt/wholeatx/bestof/beloved_but_broken.csv
sorted by lowest score: {business, award, year, score, top 3 failed checks, phone, sector}.
This is the outreach call sheet — an award-winner losing customers to a bad website is the
warmest lead we will ever have.

Compare 2024 vs 2025 winners per category. Tag each business: "hall_of_fame" (won both years),
"rising" (2025 only), "faded" (2024 not 2025). Add these tags to by_sector.json. They drive
badges and a genuinely interesting trends page.

Intersect the winner list with roster tier-1 (no website found). Any Austin favorite the crawler
couldn't find a website for is a headline. Output the list; it feeds both a press angle and the
highest-priority "get online" outreach.

For each matched winner, draft a 2-sentence WholeATX blurb: sentence 1 acknowledges the award
(our words, attributing the Chronicle, never copying their prose); sentence 2 gives the honest
one-line website read and the single highest-impact fix. Neutral, non-snarky, genuinely helpful.

In /opt/wholeatx/wholeatx.py, load by_sector.json and render a gold "🏆 Best of Austin YEAR —
CATEGORY" badge next to any matched business in the sector worksheet and at the top of its
/biz/ detail page, linking to the specific Chronicle poll URL. Idempotent; safe on re-run.

On each winner's per-business page, add schema.org "award" to its LocalBusiness node
("Best [Category], Austin Chronicle Best of Austin YEAR"). This makes WholeATX the structured,
machine-readable source for "who won Best X in Austin" — exactly what LLM answer engines quote.

Add an optional "Best of Austin" view/filter on each sector page that pins matched winners to the
top of the worksheet, badge visible, score visible. Default sort stays by score; this is a toggle.

Generate wholeatx.com/bestof/index.html: every matched winner, both years, filterable by
category and sector, each with live website score and Chronicle link. Plus wholeatx.com/bestof/
beloved-but-broken/ as a public teaser of the gap (names shown, full report gated). Rich
ItemList + Award JSON-LD.

For high-value categories (Best Coffee, Best Tacos, Best Plumber, Best Realtor...), generate a
focused page pairing the winner + finalists with their website scores and a short guide. These
target the exact high-volume searches and are prime AI-citation bait.

For each beloved-but-broken business, draft a short, warm outreach email: congratulate them on
the Best of Austin win (specific category + year), note one concrete thing their website is
missing, offer the free full report. No pressure, no pricing. Save as drafts per your rules
(never auto-send).

From the joined data, write a factual findings one-pager: "N of Austin's Best-of winners have
websites that score under 45," "M award winners have no website at all," the most surprising
gaps. Numbers only, attributed, with methodology. This is link-worthy and AI-quotable.

Draft a partnership proposal to the Austin Chronicle: they own the love and the brand; WholeATX
brings the website-readiness layer they don't have. Offer co-branded "Best of Austin — is your
site ready?" reports, proper licensing of the marks, and traffic back to them. Frame it as
additive to their franchise, not competitive.

Add a build-time check: every rendered badge and winner card must carry "Austin Chronicle Best
of Austin YEAR" attribution and a link to the source poll page; no Chronicle editorial prose may
appear verbatim beyond a short attributed quote. On each new annual poll, re-run 01–19 and mark
prior-year badges as "20XX" (historical), never silently dropping them. Fail the build if any
winner card is missing attribution.