Catalog refresh: email.find narrows to email + status, two new contact SKUs - #29
Merged
Conversation
…t SKUs Regenerated against the live gateway after anyapi#886. The classifier blocks this on automation because it carries removals, so it lands as a reviewed PR rather than an auto-tagged release. email.find swapped from one Apify actor that generated address patterns and reported them as found (40.5% exact, 37.8% confidently wrong on a 37-person measured set) to three independent finders in a cost-sorted waterfall. Four output fields go with it: - isDeliverable was true on every hit, including all 14 wrong ones - firstName the request echoed back - lastName the request echoed back - domain the request echoed back, and the half of email after the @ The item record keeps its index signature, so a consumer reading one of those still compiles against `unknown` rather than a missing key. Price drops from $0.0231 per request to $0.0154, and a lookup that finds nothing is now free. Also picked up from earlier gateway deploys, already live and previously blocked behind the same gate: instagram.profile_contact, tiktok.profile_contact, reddit comment score, and twitter externalUrl. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Minor rather than patch: email.find drops four optional output fields, which narrows both generated result types. Removals are what made regen.yml block this, so the level is a reviewed decision rather than an automated one. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
kev1n
added a commit
that referenced
this pull request
Sep 6, 2026
Accepts the blocked classification that has stopped every regen run since 2026-09-01T06:55Z. All eleven surface changes trace to a deliberate merged AnyAPI change, so the fail-closed gate did its job and this is the review it was waiting for. Minor bump, following the v0.34.0 precedent for a reviewed narrowing (#29). Breaking, in the order a customer will hit them: twitter.search input.lang is gone. X honours "lang:" inside the query, and declaring lang collapsed the SKU to one apify lane at 9x the price, so it was retired upstream (anyapi#940). Put it in the query: query: "coffee lang:en". The gateway still folds a sent lang into the query, so an untyped caller keeps working; the SDK type no longer offers it. youtube.search_hashtag input.type no longer accepts "shorts", only "all". No source we buy returns a hashtag Shorts row with the channel and publish time this endpoint requires, so the filter was unpublished (anyapi#961). youtube.search_shorts new SKU, and the replacement for the Shorts rows above: client.youtube.searchShorts({ query }) / client.youtube.search_shorts(query=...), cursor-paginated, $0.002 per request (anyapi#965). company_enrichment.crustdata_v3 output name is now optional. A single null title aborted the whole transform and threw away every good row beside it (anyapi#1044). youtube.channel_shorts output.shorts[].title is now optional, same reason (anyapi#1023). tiktok.video input.url is now optional because input.id was added: the SKU accepts either (anyapi#1030). Existing url callers are unaffected. The IR carries no anyOf, so both read as optional in the emitted types; send one. tiktok.video_transcript_full output.segments[].words[].confidence lost its 0..1 bounds. On audio the recognizer could not align, it reports a negative log-scale score, and the old bound rejected real responses (anyapi#1023). Read the sign before treating the number as a probability. yelp.search input.limit now documents its default of 20, which is what the gateway already applied (anyapi#1003). Also in this refresh: threads.search gains optional startDate and endDate (anyapi#940), and eleven SKUs reprice, most of them down - maps.reviews $0.0441 -> $0.0035, yelp.search $0.0605 -> $0.0035, perplexity.search $0.0111 -> $0.0036, youtube.video_transcript_full $0.00308 -> $0.001. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Regenerated against the live gateway after anyapi#886.
regen.ymlclassified this blocked because it carries removals, so automation correctly refused to version, commit, or tag it. This is that review.Why the removals
email.findran on one Apify actor that generated a likely address pattern, checked it against a catch-all domain, and reported"status": "found". Measured against 37 people whose real address came from a public git commit: 40.5% exact, 37.8% confidently wrong. It is now three independent finders in a cost-sorted waterfall, and a lookup that finds nothing is free.Four fields go with it, all previously optional:
isDeliverabletrueon every hit, including all 14 wrong onesfirstNamelastNamedomainemailafter the@EmailFindItemkeeps its index signature ([extra: string]: unknown/ the Python equivalent), so a consumer still reading one of those compiles againstunknownrather than a missing key. Price drops from$0.0231 per request plus $0 per resultto$0.0154 per request.Also in this refresh
Already live on the gateway and blocked behind the same gate since earlier today, so they ride along:
instagram.profile_contacttiktok.profile_contactoutput.comments[].scoreoptional field added (reddit)output.externalUrloptional field added (twitter)Classifier output
Gate
pnpm run generate:checkproved the committed baseline byte-identical before refreshing, and passes again after.@anyapi/generatorcheck: 158 tests pass.tsupbuilds,consumer-typecheckOK.pyrightreports 5 errors andpytest18 failures on the untouched baseline as well - identical counts before and after this diff, so no regression here. This refresh adds 2 passing fixture-sweep tests (474 -> 476). Repo CI is the authority on that leg and is green onmain.🤖 Generated with Claude Code