Old Voter List

• The site now leads with SIR -- the reason most people arrive -- in its page titles, its landing page, and the preview card a shared link produces.
• Added a site icon for the browser tab, and turned on compression: pages are a fraction of their previous download size, which matters most on a slow connection.
• The Usage page now loads instantly instead of counting logs while you wait, and pages back through earlier 30-day windows instead of only showing the last 30.

• Voter counts on the home and roadmap pages now read in crores, with the exact figure still available on hover.
• The roadmap and About pages now credit the contributors who wrote each state's connector.
• The landing page now shows a loading indicator while it starts up, instead of the search form appearing and then changing.

• Nineteen more states are now searchable, including Madhya Pradesh, Rajasthan, Telangana, Assam, Punjab and Delhi.
• West Bengal now covers 290 constituencies across the state, not only the 19 in Kolkata.
• A birth year later than the newest roll a state published now says so, and says which years your age tolerance does cover, instead of returning nothing.
• A search link with an unusable year of birth or age tolerance now explains the problem instead of failing.
• The landing page no longer shows the old one-page form for a moment before the step-by-step search loads.
• The copy-link summary under the results now updates when you run another search on the same page.
• A state that is already live no longer also appears under Planned on the roadmap.

• Credit the project's supporters, and the contributions that are coming
• Separate the promotion from the product, and make it reachable on a phone
• Add a landing-page showcase, from real screenshots rather than mockups
• Update release notes (v26)

• Take the debug view out of the bot gate
• Expose REPLACE_CATALOG= and ALLOW_CATALOG_SHRINK= on build-db-ac
• Re-point the pin to vsp main's West Bengal merge round
• Add make migrate-translit-ac: recompute Latin columns across a per-AC tree
• Ask for the catalogs, not the prefix that might hold them
• Establish the live bucket is listable before trusting an empty fetch
• Say we could not read the age, not that the roll did not have one
• Compare a state's catalog against its own files, not against other states
• Pin the pipeline at the column-boundary fix, and record the live-catalog rule
• Refuse an AC-db push that would move a state backwards
• Check the shipped catalog's coverage claim against its own index
• Point the cap test at the catalogs it reads its expectation from
• Pin the pipeline to vsp #49's merge commit
• An unreadable sex is unknown too, not just an unread one
• Hand the Latin bridge rows that need it, not states that might
• Correct a false claim about KarnatakaConnector in the fixture docstring
• Bridge every non-Latin state to Latin queries, not just Devanagari ones
• Pin the pipeline at 593b175 and correct the catalog-scoping gotcha
• Scope the constituency-selection cap to the states selected
• Thread ACS/PATCH/OUT/WORKERS through make build-db-ac
• An unread sex means "unknown", not "not your match"
• Write the catalog-scoping rule into the gotcha list
• Bump the pipeline pin to vsp#24 (build scope declared, not inherited)
• Re-pin the pipeline to vsp main after #21
• Re-pin the pipeline to vsp main after #19/#20
• Construct the no-roll_year catalog instead of relying on the pin
• Pin voter-search-pipeline to 8b7162f (the #14-#18 stack)
• Derive the birth-year ceiling per state instead of hardcoding 1984
• Update release notes (v25)

• The site now handles busy periods without turning visitors away.

• Traffic-statistics page now waits out temporary rate limits on its log reads rather than dropping a chart.

• Follow-up fix to the traffic-statistics page, which could wrongly report itself as unavailable while it was still computing.

• Internal maintenance to the site's own traffic-statistics page, which had been under-reporting visitor counts.

• Usage page: the "Min" column heading now sits over its own column instead of the one to its left.

• The usage page's search-time table now reads in seconds and is split into a summary group (min/max/average/median) and a percentile group (p10/p50/p90/p99).

• The usage page now reports the full spread of search times (p10 through p99, plus median) instead of just the fastest, average and slowest.

• New "Which site is this?" section on the About page, stating the one address this tool is served from and who operates it, with coverage figures generated live from the running index rather than quoted from memory.
• New /llms.txt: the same facts in plain text, for AI assistants and other tools that summarise this project.
• States & Covered page: the OCR sponsorship estimate is now a range (₹15,000 to ₹1,25,000) instead of a single flat per-state figure, because the cost scales with how many scanned pages a state actually has -- the old number was roughly right for the smallest state and several times too low for the largest.
• New section inviting pull requests that add a state we don't cover yet, linking to the open-source data pipeline.
• The interest form is now labelled "Get updates" rather than "Request a state", which is closer to what it does.
• A contact link on the About page.

• Minor wording change on the About page.

• Small update to the About page, plus internal maintenance.

• Bengaluru's constituencies are now grouped under one "Bengaluru Urban" district instead of being split three ways, so all 16 on the 2002 roll are visible in one place (the published data's own names are still shown alongside).
• Present-day constituency names created in the 2008 delimitation -- Mahadevapura, Bommanahalli and the rest -- now appear in the district selector and point to the 2002 district they fall within.

• Add a Cloudflare Turnstile bot gate on the expensive paths
• Stop the scroll wheel silently rewriting a number field
• Make the form's own errors visible, and its escape hatch respond
• Offer the unfiltered search whenever the birth year narrowed it
• Show an unusable age as "not recorded", not as a 500 or a fake year
• Fix birth-year filter silently zeroing valid searches
• Update release notes (v13)

• Re-check MIG-vs-Cloud-Run cost against real traffic figures
• Add data-freshness guards, batch the live self-check, fix source link scope
• Update release notes (v12)

• Enforce distorted-query pass/fail bar, add its pytest coverage
• Add distorted-query (typo-tolerance) diagnostic to /dbg self-check
• Add search-quality regression fixture + live /dbg self-check with rank stats
• Update release notes (v11)

• Add per-app UPI picker popup and GA4 analytics, env-scoped per deploy target
• Add per-state sponsorship, donate-box cost figures, and donate UX polish
• Add real "Load more" pagination to search results (#31)
• Update release notes (v10)

• Results table redesign: name+score, birth year, relative name+relation, and AC/part/serial are now more compact columns, with a combined source/status/remark column.
• Constituency code in results is now a direct link to download that AC's source roll.
• "Check 2025 status" link relabeled for clarity.
• Slightly smaller stats banner on the home page.
• Removed an unfilled placeholder line from the About page's Versions section.

• Promote-to-prod now rebuilds instead of reusing dev's digest
• Commit the fold-pending rewrite of release_notes_overrides.json
• Update release notes (v8)

• Fixed a bug where a release's own changelog entry could be missing from this page right after it shipped.

• Search form is now a step-by-step wizard, with a classic all-on-one-page toggle for anyone who prefers it.
• Relative's (father/husband/mother's) name is required again on the search form.
• Search form order: name, year of birth, gender, and relative's name now come before picking a state or constituency.
• States & roadmap page: a reload button to pick up newly published constituencies without waiting for a redeploy, cold-start times shown in IST, and clearer Partial vs Planned badges.
• States & roadmap page now shows per-constituency availability within each district.
• Reworded the about page's pipeline link.

• Moved the production database to Singapore, co-located with the app, fixing a cross-region cold-start reliability issue.
• Rolled out per-AC-file serving to production: searches now fetch just the constituency files they need instead of the entire multi-gigabyte database.

• Warm per-AC row cache during prefetch, not on first search after download
• Add bulk "Clear all" deselect + preserve constituency filter text across reloads
• Pin SCORING_WORKERS=4 on Cloud Run to match the real CPU quota
• Replace prefetch text-opacity feedback with a per-AC progress bar
• Quote APP_VERSION/APP_GIT_HASH placeholders to avoid YAML 1.1 float parsing
• Queue AC prefetch to 2 concurrent + enable Cloud Run session affinity
• Fix per-AC prefetch not firing on bulk district selection
• Fix 500 on search in AC_DB_DIR mode: create visitor DB dir before connect
• Reorder search-form fields: name, year of birth, gender, relative
• Move refine-search fields below constituency picker; auto-detect new state catalogs
• Add AC_DB_DIR dev deploy plumbing: GCS bucket target, per-AC service.yaml
• Phase 4: on-disk LRU eviction for cached per-AC files
• Phase 3.5: year-of-birth filter, gender as radio buttons
• Phase 3: lazy per-AC download + gray-to-black progress UX
• Phase 2: AC_DB_DIR catalog-driven loaders + per-AC search/verify path
• Stack on pipeline PR #6: pin to catalog-locality-index branch tip

• Skip Devanagari branch in array_build for non-Devanagari queries
• Cache fetched AC row sets, independent of search name
• Fix /usage min/avg: don't default untagged legacy log lines to uncached
• Narrow search fetch, district select-all toggle, cache-honest usage stats
• Add public /releases page; fix /livez ping for Cloud Run cold-start warmup
• Point the Cloud Run livenessProbe at a dedicated /livez, not /healthz
• Make the cold-start DB download resilient to transient GCS failures
• Show the /usage spinner explainer only while actually loading (#12)
• Show engine + pipeline versions, split across landing footer and About (#11)
• Compute the constituency-selection cap from real district sizes (#10)
• Add dev-suffixed Makefile targets for DNS/logs, fix DNS-record parsing bug
• Make dev vs. prod Cloud Run deploys structurally distinct
• Pin voter-search-pipeline by commit hash, not tag
• Note that /usage's loading spinner is expected, not a bug

• Align AC name/code/status into table-like columns in the picker grid
• Disambiguate district vs. same-named AC entries in the filter dropdown
• Replace single-select state tabs with a multi-select checkbox column
• Sort the constituency picker by district name, not state-major
• Add district/city/village picker: select-all-in-district, locality-aware search
• Route release tagging through a make target, not a direct script call
• Cut a version tag on every production deploy, display it in the footer
• Complete open-source repo split, add constituency-grouped concurrency
• Vectorize search scoring with rapidfuzz cdist
• Prep the open-source split: pip-installable package + attribution UI
• Vectorize search scoring with rapidfuzz cdist

Initial tagged release.