Vendors change tier requirements without announcing it. This is the record of what changed in this dataset, including the things we had wrong, and the date behind every claim on every page.
68records tracked
96log entries
27vendor changes
24corrections of our own
How this works
A verified date is only worth something if you can see what moved
Anybody can print "checked September 2026" on a page. This is the part that makes it mean something.
01
Every claim is fingerprinted on every build
The tier requirement, sync direction, cost band, object list, setup effort and every documented limit on all 68 integrations, plus the entry price, auth model and rate limits on all 23 tools. Rewording an explanation is not a change to a fact and is deliberately not tracked.
02
The build refuses to run if a fact moved silently
The fingerprint is compared against the previous build. If a claim changed and no entry here covers it, the build stops and names the record and the field. A change log that relies on somebody remembering to update it is wrong within a month.
03
Our own mistakes are entries too
24 so far. A change log that only ever shows other companies moving the goalposts is marketing. The corrections are here with what the page used to say, because the version you read last month is the one you may have acted on.
04
What this does not do
It watches our records, not the vendors. Nothing here polls HubSpot's documentation nightly and no page claims it does. Re-reading a source is a person opening the page, and the verified date in the table below is when that last happened for each record.
The log
Everything that has changed
Newest first. 96 entries since the directory was first checked on 28 October 2026.
Our correction
Six places kept publishing a 7-day trial after the constant had been corrected
Was
A 7-day trial in the Reverse Contact factsheet, in the setup step of the Apollo, LinkedIn and Slack pairs, in the hub lede, and as a seven-day free trial in the call to action on all eight workflow pages. 21 published files.
Now
500 credits free to try, no card required, and the trial runs until they are spent. Their pricing page, re-read on 2026-11-17, says "500 credits free to try first. No card." and names no number of days anywhere on it.
The vendor entry of 2026-11-11 recorded the better terms and moved G.TRIAL, which is what every call-to-action footer reads. It did not move the five records and the one generator template that carried their own copy of the terms, so 21 published files went on advertising the worse offer for six days. That is the third time this split has been recorded, and the instrument that catches it is a grep for the old value after changing a vendor fact rather than trust in the constant. A blind replace was not available: the two "seven-day job" mentions on the same records are the 100,000-profile daily cap turning a million-row backfill into a week, and they stay.
The free trial lost its 7-day limit and now runs until the credits are spent
Was
A 7-day free trial, no card required, and 500 credits on a new account.
Now
No time limit. The trial runs until its 500 credits are spent and pauses within two hours of the last one. Their pricing page states 500 credits free to try first, no card, and names no number of days.
Their change log dates this to v2.44.0 on 3 November 2026. It is the better offer and it was still being published as the worse one on every page here, because G.TRIAL is in the footer note of every call to action: 106 pages said 7-day. The constant did its job and the three places that wrote their own copy of the terms did not, which is the same split this directory has now recorded three times. The change-log entry from October that quotes the old terms keeps them, because a log entry is a record of what a fact used to be.
Reverse Contact now publishes 800M+ profiles and 70M+ companies, not 1B+ and 80M+
Was
A corpus of 1B+ profiles and 80M+ companies, on the Reverse Contact factsheet, in the hub title, h1, meta description, stats and lede, and in the two competitor coverage tables.
Now
Their own site states 800M+ profiles on the homepage and in the pricing comparison, and 70M+ companies on the pricing page; the homepage gives no company figure at all. Every other headline claim is unchanged and was re-read in the same pass: the 70%+ identity match rate, the 500ms P95, 200M refreshes a month, 2,000+ teams and 4.6 out of 5 from 107 G2 reviews.
A revision downwards is the one direction this directory cannot absorb quietly, because the pages are published in Reverse Contact's own brand: for some weeks they claimed more coverage for the vendor than the vendor claims for itself, which is the single failure mode the sourcing rules exist to prevent. Two derived sentences had to be recomputed rather than renumbered. The ZoomInfo table called 1B "twice" their 500M contacts and now says more people and fewer companies, because 800M against 500M is 1.6 and a ratio that precise is not what that row is for. The Apollo table called 1B against 240M "about four to one" and now says about three to one. Both rows argue that dividing the two figures produces nothing you can act on, which is why the arithmetic in them mattered. The gate could not see any of this: a tool's stats and credits blocks were outside gen-snapshot.js until this entry, so six headline vendor claims and eight published credit costs could move with nothing asking for a log entry. They are fingerprinted now.
The Reverse Contact hub was still claiming 15 pairs and a 16-tool MCP server
Was
The hub stats block read 15 pairs documented here and 16 tools on the MCP server.
Now
Twenty-two pairs, which is every other tool in the directory, and 37 tools.
Both figures had been corrected everywhere they are derived and left alone in the one place they are typed. The pair count went stale when the second roadmap batch took the hub from 15 pairs to 22, and the tool count when the MCP server went from 16 to 37 eight days ago: four other places in the dataset say 37, and this one said 16. A hub lede is hand-written, so t.count moving does not move it, and nothing in the build reads a number out of prose. This is the third time a count inside a sentence has been the count that went stale, and the honest reading is that grepping for the old value after changing a vendor fact is not a habit that has held. The four stats on a hub are the figures a reader sees first.
Thirty-two of the 186 cited documents had moved, and one had split in two
Was
Every citation on the site pointed at the URL it was read from, and 16 of them had started answering a 301 by the end of October.
Now
Thirty-two now do, and all 32 are re-pointed at the document that answers 200 directly. Half are cosmetic: n8n dropped the trailing slash from every node and credential page, and docs.slack.dev added one. The rest are real. Outreach migrated its whole help centre, so three articles moved from support.outreach.io/hc/en-us/articles/ to /support/solutions/articles/ with new ids. ZoomInfo moved its PDF library to tech-docs-library.zoominfo.com. Attio moved its workflows block library under legacy-workflows/. HubSpot renamed its Slack workflow actions article to "manage your notifications in Slack", Gong renamed two, ClickUp renamed its Salesforce article to name Connected Search, Zoho renamed its functions limits page, Google moved the Sheets API under /workspace/, and Slack replaced its Salesforce setup article with a new one at a different id. Reverse Contact split /guides/rate-limits-credits into /guides/credits and /guides/rate-limits, so the 22 pairs that cited the combined page now cite both halves, which is what they actually rest on.
Every renamed document was re-read before its citation moved, because a citation names the document somebody read and a rename can mean the content changed. All seven still support the claims on them, and the Attio one supports its claim better than before: the limit here says Attio labels those blocks legacy, and the page is now explicitly the legacy one. Five labels were reworded to the title the document now carries. The target of a 301 is not always canonical either: both Slack redirects land on an /intl/en-us/ URL that redirects again, so the citations name the locale-free form, and every one of the 32 was checked for a direct 200 rather than trusted.
Vendor change
The free Check endpoints need 2 credits, not a positive balance
Was
At a zero credit balance the free Check endpoints were paused too and answered 402 NO_CREDITS, because no enrichment could follow one.
Now
They are paused below 2 credits, and the reason is published with the threshold: the fetch a check precedes can itself cost 2. GET /v2/usage, GET /v2/flags, the Search available-fields endpoints, webhook polling and the task routes keep answering whatever the balance.
Their own change log dates this to v2.46.1 on 9 November 2026, which makes it a vendor change rather than a misreading here, and it is the kind that does not announce itself: a workspace with 1 credit left now fails at the step this directory tells you to run first. The same release note settles a second thing. Since v2.44.0 on 3 November a free trial with no credit left gets 402 on /v2/search/people/count and /v2/search/organizations/count as well, so "the search count endpoint is free on every plan" needed the trial qualification it now carries. The sentences that say GET /v2/usage survives what stops the checks were reworded to name no threshold at all, because this one has now moved twice and the durable fact is that the usage endpoint runs no credit check.
The MCP server went from 16 tools to 37, and a company domain now goes straight in
Was
An MCP server with 16 tools, and company lookup by domain deliberately not one of them, so an agent had to search companies on the companyDomain filter and enrich the profile URL that came back.
Now
Thirty-seven tools by their own count, 36 in their own tables. The 19 new ones are three families this directory had never recorded: Lists, which keep people and companies in the workspace and buy profiles, emails and phone numbers for them; Bulk and the task routes, which enrich up to 25,000 people in one request; and goals, which carry a long job across conversations with a credit budget and a 15-minute pilot lease. And enrich_company now takes a company domain, at 1 credit cached or 2 by domain when it finds one.
The domain tool is the half that mattered, because the old absence was not merely recorded on a page: it sat in API_REF, which build/skill-api.js renders into all 76 agent skills, so 76 published files told an agent to do a two-step that is now one call. That is this directory's own rule arriving on its own doorstep, the one that says an absence is the claim most likely to expire, because a vendor shipping something is exactly what makes it false. The documented failure "an agent could not look up a company from its domain" is removed rather than reworded, which takes the error index from 495 to 494: a failure that can no longer happen is not a failure. The 16 against 37 is also their own inconsistency again, in the same direction as the 16-says-17-lists reading in October, so both numbers are stated.
Gong ships its own enrichment waterfall, and it names six data partners
Was
Nothing in the directory about Gong and Reverse Contact, and no record anywhere that Gong sells contact enrichment at all.
Now
Gong Enrich retrieves missing or outdated email addresses and phone numbers through a configurable provider waterfall, with email and phone ordered independently. The supported data partners are LeadIQ, Wiza, RocketReach, Lusha, ZoomInfo and Findymail. Bring your own key attaches your own credential to a partner on that list rather than adding one, so Reverse Contact cannot be placed in the waterfall at all. A credit is consumed when a partner returns an email or a phone number, including one that then fails validation; account enrichment consumes none; and within 90 days Enrich skips partners it has already queried.
The decision-changing detail is what Enrich does at a job change. Before querying any partner it checks whether the contact still works at the company on their record, and if they do not, it stops and uses no credits. So Gong establishes that the record is wrong and does not say what it should be instead, which is exactly what a reverse lookup from the same address returns. Also recorded: the Gong API is 3 calls a second and 10,000 a day at company level rather than per integration, and enabling Enrich cannot be undone.
Change
The two competitor pages ship, and the finding is the meter rather than the coverage
Was
Neither pairing was in the directory. ROADMAP section 0.6 held them pending a decision on whether a directory published by Reverse Contact should compare coverage with its competitors.
Now
Both ship as full waterfall pages with the coverage comparison included, approved by Eugene on 2026-10-28. Apollo publishes 240M contacts and 30M companies and lists API Enrichment and Waterfall Enrichment from Basic at $49 per seat per month annually with 30,000 credits per seat per year. ZoomInfo publishes 100M+ companies, 500M+ contacts, 200M+ verified business emails, 135M+ verified phone numbers and 120M+ direct dials, prices by quote, and meters one credit per record per rolling 12 months with free re-enrichment inside that year.
Three things are worth recording beyond the pages. Apollo is itself a waterfall orchestrator, documenting run_waterfall_email and run_waterfall_phone with a required webhook_url, and it states that it cannot give one answer for what a miss costs because some configured sources charge on a lookup and others only on a result. ZoomInfo's documentation moved host, so docs.zoominfo.com now 301s to docs.gtm.ai, and its rate-limits page and its credit page disagree about whether rate-limit headers are returned. And ZoomInfo publishes figures for its competitors that disagree with theirs: it gives Apollo 270M contacts and 70M companies against Apollo's own 240M and 30M, which is why both pages quote a vendor only about itself.
Change
Four more Reverse Contact pairs, and two plan gates that do not line up
Was
None of the four was in the directory. The Slack, LinkedIn and Apollo hubs carried no Reverse Contact guide, and the matrix cells were empty.
Now
Full guides for all four. ClickUp moves to tier, because task webhooks are listed as unavailable on Free Forever and on Unlimited and Core and available from Business, while the API stays at 100 requests a minute per token until Business Plus. Slack and Google Workspace stay on usage: Slack documents no built-in Workflow Builder step that calls an arbitrary URL and caps workflow webhook triggers at 10 a minute, and Apps Script allows 20,000 UrlFetch calls a day on a consumer account and 100,000 on Workspace with 6 minutes per script execution and 30 seconds per custom function. LinkedIn stays on usage and needs no Sales Navigator edition at all.
The ClickUp pair is the clearest double gate in the dataset: the plan that turns the webhook on is not the plan that makes it fast. The LinkedIn pair adds a fourth category to a hub that had three, because resolving a public profile URL through a third-party database is neither an approved LinkedIn integration nor a browser extension, and that page states both that no rep session is involved and that section 8.2 separately addresses information obtained from the Services through third parties such as data aggregators or brokers. Totals move to 68 integrations, 28 paywalled and 27 two-way, and the Slack hub's claim that every one of its integrations posts every event type it supports is now stated as nine of ten.
Change
Two figures the 2026-10-28 re-read corrected in the dataset and left standing on pages
Was
The Reverse Contact hub lede said a new account starts with 100 credits, all eight workflow pages said the same, and the /mcp page said the Reverse Contact MCP server has 13 tools.
Now
500 credits and 16 tools, matching the tool record and the change-log entries written on the same day.
Our own correction, and the useful part is the mechanism. The re-read moved the TRIAL constant and the tool record and missed three places that carried their own copy of the figure rather than reading the constant: a hub lede in the dataset and two generator templates. A constant only prevents drift where it is actually used, and nothing in the build checks a number written into prose, so the fix was found by grepping for the old value rather than by any gate.
AffectsReverse Contact
Change
Outreach and Zapier is added, and it breaks the Zapier hub's own rule
Was
Not in the directory at all. The matrix cell was empty and the Zapier hub listed five pairs.
Now
A full guide, and the sixth Zapier pair. Zapier's listing gives the Outreach app six triggers and four actions: New Prospect, New Call, Updated Account, Updated Opportunity and Updated Task are instant, Completed Sequence Task polls and needs a Sequence ID, and the actions are Create Prospect, Find Sequence, Add Prospect to Sequence and a beta raw API Request. Metered rather than gated on either side.
Worth a change-log entry beyond the new page, because it falsifies a generalisation the Zapier hub was making. That hub said Zapier is the middle of a chain and never the beginning, which was true of the five pairs it had: the Reverse Contact app has no triggers and nor does the LinkedIn one. Outreach has five instant ones, so the hub now names it as the exception rather than asserting the rule. Totals move to 61 integrations, 27 two-way and 439 documented failures, and the Outreach hub goes to eight pairs while Zapier goes to six.
Change
Clay and LinkedIn was pointing the wrong way, and is now the integration it actually is
Was
A facet stub claiming a native inbound integration: build native, flow in, and a hook reading "Sales Navigator searches as a table source, subject to LinkedIn's own terms", requiring a Sales Navigator seat. In the usage cost band.
Now
Flow out, cost tier, and a page about what Clay documents: pushing people and account audiences into LinkedIn Campaign Manager. Clay lists no LinkedIn or Sales Navigator table source anywhere. Its documented sources are CRM systems, webhooks, CSV files and its own Find People, and its only LinkedIn document is LinkedIn Ads in Clay. The gate is Growth, because Push audiences to ad platforms reads not included on Free and on Launch.
Our own error and the worst kind, because a filled matrix cell reads as a checked claim and this one asserted an integration that does not exist. It survived because the pair had no page: a stub carries no sources, so nothing in the build could contradict it. The lesson is recorded rather than just the fix, in CLAUDE.md: when you promote a stub, re-derive every facet from the documents instead of inheriting them. Paywalled goes from 24 to 25.
Change
Clay and Attio is written up, and it joins the other Clay CRM pairs behind Growth
Was
A facet stub with no page, in the usage cost band, citing nothing.
Now
A full guide. Clay's plan comparison marks CRM integrations as not included on Free and on Launch and included from Growth, and Attio is a CRM for that purpose, so the band is tier. Two import sources capped at 50,000 records, four write actions of which only Lookup records is free, OAuth that only an Attio workspace admin can authorise, and a documented precedence where a saved View silently overrides a Filter. On Attio's side, 100 reads and 25 writes a second, where one request can spend more than one write.
The band moved for the same reason hubspot+clay and salesforce+clay did: a team on Launch cannot run any part of it. Paywalled goes from 23 to 24, which the homepage restates. The fact worth carrying to other pages is Attio's write accounting: creating a person with two new referenced companies spends 3 of the 25 writes a second, so the fan-out of a record rather than the row count is what hits the limit.
Change
Clay and n8n is written up, and it moves to the paywalled band
Was
A facet stub with no page: a filled cell in the matrix that routed a reader to the Clay hub, in the usage cost band, citing nothing.
Now
A full guide. Clay documents webhook automation and HTTP API integrations as Growth features, and the plan comparison row for "Automate any signal via webhooks" reads not included on Free and not included on Launch, so both directions of this pair are gated by the same plan and the band is tier rather than usage. The documented ceiling is 50,000 submissions per webhook source, persisting after rows are deleted, lifted only by Enterprise Auto-delete. Neither vendor publishes a node for the other.
The cost band moved because the test is whether the gate changes a decision, and this one ends the conversation for a team on Launch: there is no half version of the pair to run while they decide. That takes the paywalled count from 22 to 23, which is restated on the homepage. Our own correction rather than a vendor move, except for the part that is a vendor fact nobody here had read before: the webhook cap.
Vendor change
The rate limits are published per plan, and "unlimited" turns out to have a daily ceiling
Was
Published as 100+ requests a minute on credits and on all three Unlimited plans, and 1,000+ custom on Platform, with no published daily limit of any kind.
Now
A workspace minute limit across every key: 50 on the free trial, 100 on pay as you go and on Go, Plus and Max, a paid add-on of 200 or 500, and from 1,000 on Enterprise. Separately, the calls an Unlimited plan covers run against 100,000 profiles a workspace a UTC day, after which they return 429 DAILY_EXTRACTION_LIMIT_EXCEEDED until midnight UTC. Credit-paid calls are not capped.
This is the most decision-changing of the six entries dated today, because it changes what an Unlimited plan is. It does not remove a ceiling, it moves one: the covered endpoints cost 0 credits up to 100,000 profiles a day and then stop until UTC midnight. For a one-off backfill of a million rows that is a seven-day job rather than a budgeting question, and nothing in the plan name says so. The minute limit being the same 100 on every Unlimited tier is the other half: paying more buys unmetered families, not throughput, and throughput is a separate add-on.
The MCP server went from 13 tools to 16, and two of the three gaps we documented closed
Was
13 tools, with company lookup by domain and contact email or phone deliberately absent, so an agent wanting an email had to leave MCP for the API.
Now
The guide says 16 tools and its tables list 17, which is their inconsistency rather than ours and is recorded as found. The four that are new are search_persons_lookalike, count_companies, find_email and find_phone. Both contact tools are asynchronous and return a taskId to poll with check_task.
The gap that closed is the one that mattered to an agent build: find_email and find_phone mean a whole contact workflow can now run over MCP without dropping to the REST API for the last step. The gap that remains is company lookup by domain, which is still not a tool of its own, so a domain still goes through search_companies on the companyDomain filter before anything can enrich it. An agent skill written against the old shape is not wrong, only longer than it needs to be.
Finding an email costs 3, 5 or 8 credits depending on what comes back
Was
3 credits for a business email, charged only when one is found. 20 for a direct phone number on the same terms.
Now
3 credits for a professional address, 5 for a personal one and 8 when the lookup returns both, refunded when neither is found. Phone is unchanged at 20 and refunded. A live fetch is the exception to the refund rule: its 2 credits are charged when the job is created and kept even when it finds no profile.
The published matrix also names endpoints the dataset did not carry: a lookalike search, an agentic search that takes a prompt instead of filters and bills 1 credit a page plus 1 a result, company search moving to /v2/search/organizations, and the live and check variants for companies. The live-fetch exception is the one worth designing around, because it inverts the rule every other page here states: on this one endpoint a miss is billed.
A 7-day free trial, no card required, and 100 credits on a new account.
Now
The same 7-day trial with no card, and 500 credits to start. The trial also has its own minute limit of 50 requests, which is the first time one has been published for it, and no daily profile cap.
Five times the credits is the difference between testing a build and testing it on real data. 100 credits bought fifty reverse lookups; 500 buys 250, which is most of the way to the 500-row sample every page here recommends measuring your own match rate on before you buy anything.
The Platform plan stopped publishing a price, and /v2/search/persons is now 410 Gone
Was
Platform from $2,500 a month on an annual contract. /v2/search/persons deprecated in favour of /v2/search/people, with the old spelling live until 13 October 2026.
Now
Platform is quoted rather than listed, and the rate-limits guide calls the top tier Enterprise. /v2/search/persons returns 410 Gone, as the deprecation notice said it would, on the date it said.
The deprecation landing on its announced date is the useful half of this entry: it is evidence about how this vendor handles a breaking change, which is worth more to somebody choosing an API than any feature on the pricing page. The Platform price going behind a conversation is the ordinary direction of travel and the pages say quoted rather than guessing at what it was.
Every link we published to their documentation pointed at a redirect
Was
30 citations across the dataset on app.reversecontact.com/docs/public/, which now 301s to a new host.
Now
All of them on docs.reversecontact.com, which is where the documents actually live.
A citation is a promise that somebody can go and check the claim, and a 301 does not break that promise so much as put a step in front of it. It is recorded as our own correction rather than as a vendor change because the redirect works and nothing was wrong on their side: what was wrong is that nobody had followed one of our own links since they moved. The re-read that found it also found the five vendor changes above, seventeen days after the last pass, which is the argument for doing this on a schedule rather than when a page is being written.
The Unlimited plans stop metering different endpoint families, and which one you call decides the price
Read off the pricing comparison table, which prices each endpoint family against each plan rather than publishing one rate. A freshness check and the search count endpoint are free on every option. Cached lookups from a profile URL stop being metered on Go at $249 a month; people and company search on Plus at $449; reverse lookups from an email, a name or a domain on Max at $699. Live refreshes, business emails and direct phone numbers stay on credits on every plan including Platform, where they move to a volume rate. Recorded on all fifteen pairs because it changes the answer per pair rather than per customer: a route that starts from a profile URL is unmetered at $249 and the same route starting from an email needs $699, which is the same build and a $450 a month difference. Also on that table: the published rate limit is 100+ requests a minute on credits and on all three Unlimited plans and custom on Platform, and credit validity is 3 months everywhere except Platform's 12.
Errors cost nothing, and a zero balance pauses the free checks as well
From the rate limits and credits guide, and it is the fact the error list on every one of these fifteen pages now turns on. All 4xx and 5xx responses cost 0 credits, except a 404 which depends on the endpoint, and the Check and Usage endpoints are always free. So a lookup that matches nobody is free, which inverts the usual arithmetic: on Clay's own meter a miss costs an Action and on Reverse Contact's it costs nothing. The counterpart is that at a zero balance the free Check endpoints are paused too and answer 402 NO_CREDITS, because no enrichment could follow one, while GET /v2/usage, GET /v2/flags, the Search available-fields endpoints and webhook polling keep answering. Async calls are charged at request creation and refunded automatically on an API error, no search result, no email found or a 404 on a live endpoint, and those refunds are auditable in the Webhook Events dashboard. A failed webhook delivery is explicitly not refunded, because the data was fetched: the job reports succeeded with a deliveryStatus naming the problem, one of webhook-url-invalid, webhook-url-errored, webhook-url-timeout or webhook-url-unreachable, and the result is still there to poll for. Two rate limits apply per workspace, a minute limit and an Enterprise-only daily limit that is disabled by default; the request that creates an async job counts and polling does not; and a per-key rate limit can cap one integration below the workspace maximum.
A live job is documented as possibly tens of minutes, which decides where it can run
The getting-started guide now states the aim as under 15 seconds and the ceiling as tens of minutes depending on conditions, and says in as many words that the live endpoints are built to update databases and feed async pipelines rather than to block a synchronous request a user is waiting on. That is a sharper claim than asynchronous and it changes the recommendation on ten of these fifteen pairs: a HubSpot custom code action has 20 seconds, an Attio HTTP block has 20, a Salesforce transaction has 120 cumulative across every callout in it, and none of those is a budget for a job with that ceiling. The one CRM here that can is Zoho, where a scheduled function gets 15 minutes. Polling GET /v2/webhooks/:webhookId remains free and off the rate-limit clock, so the cost of the wait is wall clock rather than credits.
The Clay listing is a Clay-built integration, and the four actions return two different field sets
Our own correction. The record claimed Reverse Contact maintained the Clay app, listed since May 2025 by its own account. Clay's integration page classifies it as a Clay-built Integration, which is a different statement about who fixes it, and the claim about the listing date had no source. Both are now gone. Read from the same page and from Clay's own documentation for the integration: all four actions are marked Included in All Plans and each is offered as Clay Credits or Bring Your Own Account, so the enrichment runs on a Free workspace while the plan comparison puts HTTP API integrations and CRM integrations on Growth at $446 a month. The required inputs differ per action and are not interchangeable: Find Professional Profile takes any of Email, Full name, Company name or Company domain; Find company professional profile requires a Company domain; the two Enrich actions require a profile URL. The documented output lists differ too, and the gap is the finding: the action that starts from an email returns identity, the current position title and the current company, while experience, education and skills come from Enrich professional profile run on the URL the first one returned. Two actions and 3 credits for one complete person. Clay's action description and its documented field list disagree about the longer set, which the page says rather than resolving.
AffectsClayReverse Contact
Vendor change
Neither no-code app has a trigger, and only one of them can reach an endpoint it does not name
Read off both marketplace listings. The Zapier app publishes nine operations, seven Functions and two Queries, and no triggers: Check a Person Profile and Check a Company Profile at 0 credits, Enrich a Person and Enrich a Company at 2 credits if a profile is found and 0 if there is no match, Get LinkedIn Profile, Get LinkedIn Company, Get Workspace Usage at 0 credits, and Search Person and Search Company. The Make app is listed as Verified with Reverse Contact as the Official Vendor and publishes four modules, all actions and no triggers: Reverse Email Lookup, Get LinkedIn Profile, Get a LinkedIn Company and Make an API Call. So Reverse Contact never starts a run on either platform and is always a step in the middle, which is correct for a data layer and does mean a scenario or Zap always has at least two other steps. The two apps fail differently and the pages now say so: Zapier names nine operations and has no way to reach a tenth without a generic Webhooks step and a second copy of the key, while Make names four and one of them is an arbitrary authorised call, so the freshness check, company lookup from a domain and search are reachable on the same connection.
AffectsZapierMakeReverse Contact
Vendor change
A HubSpot custom code action may re-use variables declared outside exports.main
From HubSpot's own caveats on custom code actions, and it is a correctness problem rather than a performance one. HubSpot documents that variables declared outside exports.main may be re-used for future executions of the action to save memory, and that many objects can be enrolled in a workflow at the same time. For an enrichment action that means a response held in a module-level variable can be served to the next contact, and no single-contact test will show it. Two more from the same page: a rate-limited call is retried only if you throw the error in the catch block in Node.js or raise it in the except block in Python, so a 429 caught and turned into a no_match is a contact silently left un-enriched while the workflow reports success; and the total length of all secret values in one action must not exceed 1,000 characters. Confirmed unchanged: Data Hub Professional or Enterprise for both the webhook and the custom code action, 20 seconds and 128 MB per run, and axios pre-loaded at ^1.2.0 among twelve Node libraries.
AffectsHubSpotReverse Contact
Vendor change
Salesforce meters inbound API requests and an outbound callout is not one
Read from the developer limits and allocations quick reference, and it is the strongest argument on that page for doing the enrichment inside the org. Total inbound API requests per 24 hours are 15,000 on Developer Edition, and on Enterprise or Professional with API access enabled they are 100,000 plus the licence count times the per-licence allowance, which is 1,000 for a Salesforce licence, plus any purchased add-ons; Unlimited and Performance raise the per-licence figure to 5,000. Calls made from installed managed packages count against the same number. An outbound callout to Reverse Contact is not an inbound request and spends none of it, while every external route spends it twice per record, once to read and once to write. Separately, the concurrent limit on inbound requests lasting 20 seconds or longer is 25 in a production org or sandbox and 5 in Developer Edition or a trial, returning REQUEST_LIMIT_EXCEEDED, with no limit at all on concurrent requests shorter than 20 seconds. API Usage Notifications can email a named user at a percentage of the allocation.
AffectsSalesforceReverse Contact
Our correction
Attio has Delay and Loop blocks, and an empty path is not a parse failure
Our own correction. The record said there was no loop block to poll a live job with. Attio's block library documents a Delay block taking seconds through weeks, a Delay until block, and a Loop block that iterates a list with an optional iteration limit. So a bounded poll is constructible, awkwardly, because Loop walks a list rather than repeating until a condition holds, and the conclusion that live resolution belongs elsewhere survives for a better-stated reason. Read from the same page, two notes that sit a few lines apart and together decide how the workflow behaves on a miss: a path that evaluates to nothing returns an empty value and will not fail, while an Output Type that does not match what the path evaluates to will fail, with setting name.first as a Number given as the example. A confident null is therefore quiet and a mistyped field is loud. The three output types map to attributes as Boolean to Checkbox, Number and Number (Array) to Number and Rating, and String and String (Array) to Text, Email Address, Domain, Select, Multi-Select, Currency, Status and Phone.
AffectsAttioReverse Contact
Vendor change
A Pipedrive automation webhook offers POST, PUT and DELETE, and no GET
From the automations webhook article, and it is a second reason the automation cannot fetch anything: beyond having no documented way to read a response, the method list does not include GET. Three more specifics from the same page: creating an automation webhook is global-admin only, existing webhooks cannot be re-used for automations so it has to be a new one of that type, and the endpoint URL must conclude with a slash, an ampersand, a question mark or an equals sign. The article states automations and webhooks in automations are available on Advanced and higher plans, using the previous plan names, while the current pricing card sells Lite at $14, Growth at $39, Premium at $59 and Ultimate at $79 a seat per month billed annually and puts automations on Growth; the page says both rather than choosing. From the rate limiting page: the daily budget is 30,000 base tokens times a plan multiplier of 1, 2, 5 or 7 times seats, weighted per endpoint at 2 tokens to read one entity, 20 for a list, 10 for an update and 40 for a search, resetting at midnight in the server timezone. Burst limits are per token on a rolling 2-second window, 20 requests on Lite to 120 on Ultimate for an API token and 80 to 480 for an OAuth app, with Search fixed at 10 per 2 seconds on every plan and auth type. Continuing to send volume after a 429 earns a 403 returned as a Cloudflare HTML error page rather than an API response.
AffectsPipedriveReverse Contact
Our correction
A scheduled Zoho function gets fifteen minutes, so live resolution is not ruled out
Our own correction. The record listed live resolution under when not to use this pair, without qualification. Zoho documents function timeouts by category: 10 seconds for a button, related list, validation rule or REST API function, 30 seconds for workflow, blueprint and approval, and 15 minutes on a schedule, with anything over its category forcefully terminated. Fifteen minutes is enough to start a live refresh, wait and read the result, which makes Zoho the only CRM in this directory that can poll a live job inside itself. The entry is now scoped to a workflow rule, where 30 seconds genuinely will not cover it. Read from the same two pages: functions have full access on Enterprise, CRM Plus, Ultimate and Zoho One with Standard and Professional reaching them only through Extensions; a Deluge execution is one function credit while Java, Node.js and Python are charged 1 credit per 10 seconds of execution; per-invocation limits are a 10 MB response and 200,000 lines of execution, counting lines the runtime executes rather than lines of source; the org-wide Invoke URL quota is 5,000,000 requests a day; API credits are 50,000 plus 250 to 2,000 per licence in a rolling 24 hours, capped at 100,000 on Standard and 5,000,000 on Enterprise; Deluge integration tasks count as API credit consumption, which Zoho changed deliberately; concurrency is 20 per org per app on Enterprise with no per-minute limit at all; and sub-concurrency is 10 on every edition for seven expensive calls including search records invoked from a function, returning TOO_MANY_REQUESTS.
AffectsZoho CRMReverse Contact
Our correction
monday's top API tier is Enterprise rather than Ultimate, and three other budgets apply
Our own correction. The record named the 25,000-call tier Ultimate; monday's own rate limits table says Enterprise, and the 1,000-call tier is Free, Basic and Standard rather than Basic and Standard. The allowance is also shared across every personal token, private app and public app on the account, excluding marketplace apps and monday's own, which the record did not say. From the same page, three further budgets each with their own error: a minute limit of 1,000 queries on other tiers, 2,500 on Pro and 5,000 on Enterprise; a concurrency limit of 40, 100 and 250; and a per-IP limit of 5,000 requests per 10 seconds. A request that hits any rate limit counts 0.1 of a call against the daily limit and returns retry_in_seconds, while a request that simply fails counts in full. Responses carry RateLimit-Policy and RateLimit headers in the IETF format, reporting minuteRate, concurrency and complexityMinute with a remaining value and a reset. From the webhooks page: the URL must return monday's randomly generated token in a challenge field or the webhook cannot be created, the URL is capped at 255 characters, a webhook created with an integration app token carries a verifiable JWT and cannot be disabled by end users, and deliveries retry once a minute for 30 minutes.
Affectsmonday CRMReverse Contact
Vendor change
The five sequencers meter the same enrichment five different ways
All read from the vendors' own rate-limit documentation, and the differences are large enough to change whether a backfill is an evening or a negotiation. Outreach is 10,000 requests an hour per user, with Kaia separately at 3 calls a second and 6,000 a day at org level, and it states that it reserves the right to adjust limits at any time. Salesloft is 600 cost a minute for the whole team, says explicitly that one integration can affect another's limit, charges extra for page numbers above 100 at 3, 8, 10 and 30 points by range, recommends a cursor poller on updated_at instead, and reserves the right to change any endpoint cost because a cost change deprecates nothing. Smartlead is per API key at 60 a minute, 1,000 an hour and a 10 a second burst on Standard, doubling to 120, 3,000 and 20 on Pro, plan-wide across every endpoint combined, with three endpoints carrying their own limit of 10 requests per 60 seconds per account whatever the plan. lemlist is 20 requests per 2 seconds per key on all routes, with Retry-After as an integer and X-RateLimit-Reset as a human-readable date string. Instantly is 100 requests a second and 6,000 a minute for the entire workspace, shared between API v1 and v2 and across every key in it, with listing emails at 20 a minute, a test send at 10 a minute and the AI reply label test at 500 per 30 days.
Two fields in the response are reasons not to send rather than values to write
Read off the published person schema, and recorded because every page in this directory had treated an enrichment response as a set of values to write. On a sequencer it is also a list of reasons not to send, and the five sequencer pairs now carry that as the second table on the field section. isInRemembrance means the person has died, which is the one outbound failure that cannot be apologised for afterwards. isOpenToWork means they are looking to leave, which is a reason for a human to read the opener once rather than an automatic stop. The third signal is not a field but a comparison: currentPosition.companyName against the account or company already on the record, where a disagreement means the person has changed job and the cadence written for the old role is about to reach the new one. Alongside those, data arriving null with a confidence score is a documented outcome rather than a failure, and metadata.updatedAt is what distinguishes a record that is stale from one that is wrong.
Clay bands its integrations by plan in three steps, and the sequencers sit in the middle
Read off Clay's own integration catalogue, which states a plan band per integration. Smartlead, lemlist and Outreach all read "Included in Launch+", against Apollo's "Included in All Plans" and the pricing page's CRM integrations from Growth. Every action on all four is additionally marked "Bring Your Own Account Required". Recorded in `requires` on the three sequencer pairs rather than as a cost band change, because Launch is the entry paid plan and the gate changes almost nobody's decision, where the Growth gate on the CRM pairs changes everything. Also recorded, from Clay's Outreach documentation: twelve OAuth scopes are required and none can be disabled, all of them the .all read-and-write variants, including opportunities, opportunityStages, users and roles, none of which the five available actions touches. Sequences must be Active in Outreach for a lead to be added. From the Smartlead documentation: campaigns cannot be created from Clay, Allow Duplicate Leads Across Campaigns is off by default and Allow Leads from Community Bounce List is on, and the Lead ID required by Update and Remove is Smartlead's internal identifier rather than a Clay row ID.
AffectsClaySmartleadlemlistOutreach
Baseline
lemlist creates and never updates, and a refused lead still rewrites its company
From lemlist's own API reference, added as sources because Clay's three-action page documents none of it and the record previously cited a lemlist article about HubSpot. The create-lead-in-campaign endpoint creates and never updates: a contact that already has a lead in that campaign is refused with 400 LEAD_ALREADY_IN_CAMPAIGN and nothing is written, and with deduplicate on, a lead in another campaign is refused with 409 LEAD_ALREADY_IN_OTHER_CAMPAIGN. Only that parameter drives the check, not the campaign's own remove-duplicates setting. The lead is always new, and the contact and the company it matches are updated with the fields sent regardless, with updateStrategy deciding what they keep: overwrite is the default and an empty value clears a stored one. Company-prefixed fields reach the shared company record every other lead linked to it reads. Custom variable names keep only letters, digits, underscore, hyphen, space and hash, and any other character becomes an underscore. Authentication is HTTP Basic with an empty username and the key as the password, and the rate limit is 20 requests per 2 seconds per API key on all routes.
AffectsClaylemlist
Baseline
Domain-wide authentication is on every Slack plan, and the weekly digest is not
Read off Slack's own help centre, and the opposite way round from what was assumed. Domain-wide authentication for the Google Calendar and Drive apps is available on all plans, for Workspace Owners and Admins on Free, Pro and Business+ and Org Owners and Admins on Enterprise, and needs a Google Workspace Super Admin to authorise on the Google side. Once a domain is verified, every member with an email on it is connected automatically, members cannot disconnect, and they are automatically subscribed to event reminder notifications they can then turn off. Each app has its own Add domain step. Against that, Slack marks the daily or weekly schedule notification, both to a channel and as a personal digest, as a paid-plan feature. Also recorded: a channel receives nothing until the app is invited with /invite @google calendar, Slack's own fix for stale events is to Revoke under Your Authorization in the Marketplace Configuration tab and re-add, and the org-level install of the Drive app is an Enterprise organization feature.
AffectsSlackGoogle Workspace
Baseline
Gong can post to a public channel it never joined, and cannot see a private one
From Gong's Slack scopes page, added as a source. chat:write.public lets Gong post to public channels it is not a member of, while groups:read only shows private channels Gong has been added to, so an uninvited private channel is absent from the picker rather than merely unpostable. groups:write is listed and Gong states it is not used at the moment. im:history exists to carry comments made in Slack back onto the call in Gong. Also recorded: the three personal notification types, a comment on a followed call, a reply to your comment and a mention, go to a user's personal Gong app and cannot be routed to a channel; channel alerting requires a stream whose type is Standard; a stream needs at least one filter; and a filter on an inactive AI tracker, or one on an account short of credits, collects nothing while appearing configured. A Slack admin must install the app before any Gong admin can connect it.
AffectsSlackGong
Baseline
The three events everyone wants in Slack are browser notifications instead
Read off Outreach's three Slack support articles. Outreach states that not all notifications support all channels, and the split is clean: a reply from an unexpected address, meeting updates, new voicemails, bounce limit warnings at 50%, 80% and the final 24-hour cutoff, plug-in and sync errors, mentions and urgent tasks can all go to Slack, while a prospect opening an email, clicking a link or replying, and an incoming call, are browser notifications requiring OS-level permission. At account level only the Kaia meeting summary is supported, each account links to a single public Slack channel, and Outreach notes its app does not need to be invited to that channel. Enabling the account-channel feature carries a warning that account notifications may be visible to all channel members. The admin connects once at instance level over Slack OAuth and Outreach then maps users automatically; a user who cannot see Slack in their personal settings has not been mapped.
AffectsSlackOutreach
Our correction
Apollo in Slack stopped being an alert feed and became a place to work
Our own correction, and the largest in this dataset. The record described a notification integration whose only real decision was which two events to route, and cited the Apollo knowledge base as a bare domain, which was the weakest citation in the directory. From Apollo's own articles: the Slack app lets a user find people and companies, refine searches, research, enrich with verified emails and phone numbers, add contacts to lists and sequences, export to CSV, ask about sequence performance and take simple sequence actions such as pausing one, all in natural language, with the conversation saved back into Apollo's AI assistant. It is a team-wide integration only an Apollo admin can connect, after which every user can act within the plan, permission and credit limits they already have. The app is on all Apollo plans and agents, which run on a schedule and recommend unprompted, need professional or organization. Notifications are configured in two separate places, system activity and tasks under Settings and AI assistant notifications under AI Assistant then Channels, and email opens and clicks are in neither, coming from the Chrome extension or a workflow. The flow moves from event to two-way to match.
AffectsApolloSlack
Vendor change
SmartAgents is a second route, and it does what the campaign-level sync cannot
From Smartlead's SmartAgents integration guide. The limitation that defined this pair, that HubSpot can only be enabled while a campaign is being created and never midway through an active one, still holds for the campaign-level sync and now has a documented way around it. A SmartAgent connects HubSpot once through OAuth under SmartAgents then Integrations, and runs on a schedule or on a campaign event against campaigns that already exist, searching HubSpot by email and creating or updating the contact. Two templates ship: Scheduled CRM Sync for replied leads, daily at 9am, processing only unsynced leads, and Inbox Reply Sync with Upsert. Four fields cross, email, first name, last name and company name, and an empty Smartlead field will not overwrite a populated HubSpot one. Event-based triggers cannot be edited after an agent is created. An agent whose workflow has a Slack node cannot be deployed until the SmartAgents app is in the target channel.
AffectsHubSpotSmartlead
Baseline
Three settings on this pair cannot be changed in place, only replaced
Read off Make's HubSpot CRM app documentation. Scopes added to a connection cannot be removed and the only remedy is a new connection. The Subscriptions field on the Watch Notifications module cannot be updated at a later time, so every subscription has to be added when the module is built. Each HubSpot custom application can be associated with only one webhook URL, and Make defers to your HubSpot subscription for how many custom applications you may create, which makes that allowance the real ceiling on separate webhook scenarios. Also recorded: instant modules need a HubSpot developer account, a custom app and a developer API key, with the Redirect URL still set to the pre-rename integromat.com callback; Make requests scopes automatically except on Make an API Call, where you choose them yourself; and a webhook subscription needs its scope named in two places, on the custom app and again as a connection custom scope.
AffectsHubSpotMake
Baseline
Two identifiers and two validation rules decide what survives the migration
From Attio's own migration and import documentation. Companies are keyed on Domain and people on Email address, and each value can belong to only one record, so HubSpot duplicates merge on the way in. Attio validates phone numbers, expecting a country code followed by the subscriber number including the area code and checking that country and area codes are real, and validates domains, storing only the root rather than a full URL. Some objects, Deals among them, have required attributes without which the import fails. Migrate CRM is available on plus, pro and enterprise plans and Import2 on any paid Attio plan. The documented reset for a bad migration is deleting every person and company record in bulk, and errors during the process go to help@import2.com with Attio support copied, quoting the migration ID.
AffectsHubSpotAttio
Our correction
Clay gates CRM integrations behind Growth, and its HubSpot documentation never says so
Read off Clay's own pricing page. The plan comparison marks "CRM integrations", "HTTP API integrations" and "Automate any signal via webhooks" as not included on Free or Launch and included from Growth, listed at $446 a month in the page header, while "Schedule recurring enrichment runs" arrives a tier earlier on Launch. The HubSpot integration documentation names no plan at all, which is why this was recorded as metered rather than gated. Our own correction: the cost band moves from usage to tier and `requires` now names the plan, because a team on Free or Launch cannot run this pair at any volume. The HubSpot side of the gate is narrower than it looks: the three custom object scopes can only be granted on an Enterprise portal and authorisation stops rather than continuing without them, which HubSpot's own custom objects article confirms as an Enterprise feature. Added as a source.
AffectsHubSpotClay
Our correction
The licence that cannot authorise is the one a Salesforce admin would pick first
From Clay's Salesforce integration documentation, read for the first time on this date and added as a source. Integration User and API Only licences cannot complete the OAuth sign-in flow; the documented routes are a full user licence or the Client Credentials flow with Run As set to the integration user. Four separate causes produce the same OAUTH_APPROVAL_ERROR_GENERIC, including a profile missing Approve uninstalled connected apps. Also recorded: list views that are not SOQL-compatible and all reports cap at 2,000 records, only tabular and matrix reports are supported, a report import without uniqueness fields identifies a row by the whole row, and Upsert requires an external ID. Our own correction: the cost band moves from usage to tier for the same reason as the HubSpot pair, since CRM integrations start on Clay Growth.
AffectsSalesforceClay
Vendor change
Bringing your own Apollo key stopped removing the platform half of the bill
From Clay's actions and data credits documentation and its credit conservation guide, both added as sources. Clay now runs two meters: each record enriched or exported counts as 1 Action regardless of data source or provider, and Data Credits buy the data itself. Connecting your own API key eliminates the Data Credit cost and an Action is still consumed. Actions do not roll over and cannot be bought as top-ups, while unused Data Credits accumulate to twice the monthly limit on a monthly plan and top up at a 30% premium. Our own correction: this record previously framed the choice as Clay credits or your own key with different costs, which is no longer the shape of it. Clay's integration catalogue also marks all fourteen Apollo actions as requiring your own Apollo account while listing the integration as included on all plans, and the two Apollo sequence actions carry prerequisites: the contact must already exist in Apollo, and Update Contact Status in Sequence additionally needs it enrolled.
AffectsClayApollo
Baseline
HubSpot allows one webhook, so a second n8n trigger silently stops the first
From n8n's HubSpot Trigger node documentation, added as a source. The trigger registers a webhook when it is activated and HubSpot allows only one, so activating a second trigger stops the previous one working. The events it covers are company, contact, conversation, deal and ticket created, deleted and property changed, plus privacy deletion on contacts and conversations and new message on conversations. Also recorded from HubSpot's own limits page: privately distributed apps get 100 requests per 10 seconds per app on Free and Starter and 190 on Professional and Enterprise, against a daily cap shared across every app in the account of 250,000, 625,000 and 1,000,000; the API Limit Increase add-on raises the burst to 250 and adds 1,000,000 a day, and a maximum of two can be bought. A custom code action's string output caps at 65,000 characters with OUTPUT_VALUES_TOO_LARGE beyond it, 50 properties can be passed in, and a rate-limit error is only retried if the code throws or raises it.
AffectsHubSpotn8n
Baseline
The JWT and OAuth2 setups need the same three checkboxes set opposite ways
Read off n8n's Salesforce credentials page in full. Under OAuth Policies on the external client app, JWT requires Require Secret for Web Server Flow, Require Secret for Refresh Token Flow and Require PKCE Extension all unchecked, and OAuth2 requires all three checked. JWT additionally needs the JWT Bearer Flow enabled and a digital certificate uploaded; OAuth2 needs the Authorization Code and Credentials Flow. The documented callback URL is http://localhost:1717/OauthRedirect or the instance URL when self-hosting, and Salesforce permits plain HTTP only on localhost. Salesforce's own recommendations on the app are a 90-day refresh token expiry, a 15-minute session timeout, and Permitted Users set to admin approved users are pre-authorized. The node has no lead conversion operation, and the Search resource returns every result in one response with no paging.
AffectsSalesforcen8n
Baseline
Posting as the company page needs a LinkedIn App Review, not a scope
From n8n's LinkedIn credentials documentation, added as a source. Two credential types exist: Community Management OAuth2 for new accounts and apps, and OAuth2 for older ones. A LinkedIn Company Page is a prerequisite for creating the developer app at all. The self-service products are Share on LinkedIn and Sign In with LinkedIn using OpenID Connect; the Organization Support option requests w_organization_social and requires completing LinkedIn's Community Management App Review. A Legacy option exists to request r_liteprofile and r_emailaddress instead of profile and email, and Lead Sync needs r_marketing_leadgen_automation and separately granted access. Also recorded from LinkedIn's rate limiting page: limits are per application and per member per application, reset at midnight UTC, are not published in documentation, and are visible in the Developer Portal Analytics tab only for endpoints called today in UTC.
AffectsLinkedInn8n
Baseline
Three of Gong's four export streams have a documented duplicate mode
Read off Gong's HubSpot export article and its required fields article. Exporting meeting invites and conference calls as HubSpot meetings may duplicate them when HubSpot also syncs Google Calendar, and the documented alternative is calls or notes. All calls duplicates telephony calls when another application also exports them, where Gong Dialer calls only does not. All emails behaves the same way against Gong composer emails only. Engage activity has no second source. Also recorded: a meeting invite needs at least one external participant matching a HubSpot record; seven calendar email tags are excluded from export; the activity update period governs how many months Gong keeps updating an exported activity; history is only exported when the CRM integration was connected before the email or telephony one; multiple checkboxes, file and HubSpot user field types are not imported at all; single-line text and dropdown values display to 120 characters; date fields must be epoch in milliseconds; and the Engage opt-out field needs a second authorisation as the integration user.
AffectsHubSpotGong
Vendor change
Existing Apollo Salesforce connections must move to the managed package by 31 December 2026
From Apollo's Salesforce integration article. Apollo states that all existing Salesforce connections must upgrade to the managed package by 31 December 2026 and recommends upgrading proactively, that the upgrade preserves existing field mappings and push and pull settings, and that the migrate option is being rolled out over time so a connection without it is waiting for an email rather than misconfigured. Also recorded from the push settings article: the lead-or-contact choice governs only records that have never been pushed, since existing leads keep being updated as leads and existing contacts as contacts, with Salesforce's own lead conversion the only way across and no way back; a contact matching more than one Salesforce account is assigned one at random; stage is the only push filter and contacts with no stage need No stage selected explicitly; a manual push overrides every sync setting; and records created through the team sync connection normally carry the integration user as Owner.
AffectsApolloSalesforce
Baseline
The call recording link lemlist writes into HubSpot is public and cannot be revoked
From lemlist's activity mapping article. Anyone who opens the recording link from the HubSpot record can listen to the call with no lemlist account, the link never expires and it cannot be revoked. Calls recording is one of four toggles that are on by default, alongside Calls note, Calls transcript and Calls summary. The note is rebuilt rather than merged on every refresh, so a manual HubSpot edit is replaced, and it caps at 65,000 characters with longer content cut from the end. The AI summary is pushed only if the call already has one and syncing never generates it, and calls arriving through Aircall or Ringover carry no recording, transcript or summary. Also recorded: the Connected fallback applies not only to custom lemlist statuses but to any portal whose HubSpot call outcomes were deleted and recreated, because the default mapping targets the factory outcomes by id. Mandatory property rules in Advanced settings gate record creation, default property values can be set per object, and field changes propagate in two to ten minutes.
AffectsHubSpotlemlist
Baseline
Apollo shows every Outreach mailbox to every rep until an admin narrows the profile
Read off Apollo's Outreach integration article and its add contacts to a sequence article. By default, send emails from shows all mailboxes in Outreach. An Apollo admin changes this under Settings, Users and teams, Permission profiles, where Can send emails from takes All users, Self only, or Other users on the same team. Also recorded: Apollo warns before enrolment on contacts already in another sequence, on unverified or missing emails and on information it considers outdated, and will not enrol anything blocked by a workspace sending policy; an inactive sequence does not appear in the list at all and neither does one you lack edit access to; mailbox rotation cannot be selected when adding contacts by CSV; credits may be checked for the whole batch before enrolment starts; and contacts enrolled by a workflow or the outbound copilot appear on that workflow's own trigger and schedule rather than immediately. Apollo's LinkedIn article describes sequence steps that remind a rep to complete them and states that only LinkedIn tasks created in Apollo are tracked, which is a task list rather than automation; added as a source.
AffectsApolloOutreach
Baseline
A node upgrade changes what authentication a bot identity needs
From n8n's Slack node documentation. From node version 2.6, the Custom Bot Profile Photo option on Message > Send works only with Access Token authentication, because it relies on chat:write.customize, a bot-token-only scope the OAuth2 user-token path cannot use. On node version 2.5 and below the option is still shown with OAuth2 and Slack ignores it, so the same workflow behaves differently across an upgrade and fails silently in one direction. Two further details recorded: an event n8n does not list can be added as a bot event in Slack and caught with Any Event in the node, and after Slack verifies the Request URL the subscriptions can be managed in the app manifest under settings.event_subscriptions. The search operation's new Channel Types, date filters and Keyword Search Only options sit alongside the 20-results cap already recorded.
AffectsSlackn8n
Baseline
Only personal connections are supported, so an admin cannot turn this on for a team
From ClickUp's own Salesforce integration article. A Salesforce admin installs the managed package once, and only Salesforce admins can install or upgrade it, with an upgrade keeping existing data. After that every ClickUp user has to authorise their own connection, because only personal connections are supported. The managed package being installed is necessary and not sufficient, which is why one person sees link previews and the next sees a bare URL. Both halves of the integration, Connected Search and Salesforce Sync, sit on ClickUp Business Plus and Enterprise, and ClickUp notes that pricing and packaging are subject to change. Link previews render in task descriptions, comments, Docs and Chat messages, and record creation happens through typed AI Command Bar commands rather than buttons.
AffectsSalesforceClickUp
Baseline
No triggers and no searches, which is the whole answer
Zapier's LinkedIn article states that no triggers are available and no search actions are available. The integration is two write actions: Create Share Update, which posts to a profile with text, URLs, images and visibility settings, and Create Company Update, which requires an administrator role on the page. Nothing on LinkedIn can start a Zap and nothing can be looked up. The record already said "no triggers" as a limit; what is new is that the absence of searches is documented too, and that both absences are now stated as the first thing a reader meets rather than as a footnote, because the people arriving at this page are overwhelmingly looking for the half that does not exist.
AffectsLinkedInZapier
Baseline
Connecting this needs Super Admin, and direction is set per object
From HubSpot's data sync article, which is what this pair actually runs on. Connecting the app needs Super Admin or App Marketplace permissions, and it can be done from the App Marketplace or from the imports page on any object. Sync direction is chosen per object through a dropdown, so contacts can run two-way while everything else runs one way, and more than one instance of the same third-party app can be connected to a single HubSpot account. The four limits already recorded here, the one-directional and narrow email sync, custom field mappings needing Data Hub, records without an email being excluded rather than failed, and re-saving the configuration restarting the initial pass, are now attached to the setup steps and errors they belong to.
AffectsHubSpotOutreach
Baseline
Two permission set licences, each carrying a different half
From Gong's own setup documentation. The integration user needs permission sets with both the Salesforce API Integration licence, which carries the system permissions required to create tasks and events, and the Salesforce Integration licence, which carries the field-level permissions required to import Salesforce data into Gong. Assigning one produces an integration where half works and the failing half is silent. API Enabled must also be ticked on that profile, because it is how Gong checks your API limits and keeps its own requests underneath them. Two more details recorded: View All Records on an object overrides the limit the role hierarchy would impose, and Gong trims picklist and string values to 120 characters on display. The destructive one already recorded, that removing a field from the configuration deletes the data imported through it, now sits next to the setup step it belongs to.
AffectsSalesforceGong
Baseline
One instant trigger, and two error messages worth recognising
From Zapier's Salesforce article. Every Salesforce trigger Zapier publishes polls, except New Outbound Message, which is instant and is configured on the Salesforce side rather than bought with a Zapier plan. Two error strings are now recorded verbatim with their causes: "This connection to Salesforce appears to no longer be working" means the five-token connected-app limit has been exceeded, and "Please check your username and password" on correct credentials means sandbox was selected on an account with no sandbox. Three further behaviours: newly created custom fields and objects can take time to appear in the Zap editor; Zapier supplies test data when an org has none, and the Updated Field on Record trigger returns the most recently updated record without filtering on the object chosen, so a Zap can test clean and behave differently on the first real record.
AffectsSalesforceZapier
Baseline
Two integrations behind one button, and a manual push ignores every rule you set
From Apollo's own HubSpot articles. Authorising this integration asks you to choose between HubSpot CRM, which is the full two-way sync, and HubSpot Data Enrichment, which enriches records already in the CRM without pushing Apollo's prospecting data in. That choice is made once, in a dropdown, and most teams do not know they made it. Three further behaviours are now recorded. A manual push from a contact profile overrides the configured sync settings entirely, including contact-stage restrictions and the other push limits. Ownership of a pushed record depends on whether the person pushing has linked their own HubSpot credentials in Apollo, so without that the records land under one shared account. And Apollo writes a source value onto every contact it pushes, defaulting to Apollo, selectable to another field and requiring a picklist value where that field is a picklist. The six-hour configuration window after enabling, during which manual pulls are blocked and after which syncing turns itself on, was already recorded and now has the settings it is a window for.
AffectsHubSpotApollo
Baseline
A free Slack workspace holds ten apps in total, and that is the real gate
Recorded from Zapier's own Slack article. Free Slack plans allow ten app installation slots across the whole workspace, so on a workspace that has been collecting integrations the blocker is a slot rather than a licence, and no Zapier plan changes it. Three behaviours that fail silently were recorded with it: a plain mention in a Zapier action posts as text and notifies nobody unless you use Zapier's special formatting; a slash command entered in an action posts as text rather than running; and test data for a new Slack message always carries a thread timestamp field even when the message is not in a thread, which is the documented way to filter a Zap down to top-level messages. Which triggers are instant and which poll is also now on the page, because that choice alone decides whether a Zap feels immediate.
AffectsSlackZapier
Baseline
Outreach publishes its own API arithmetic, and a recommended budget
From Outreach's 53-page end-to-end Salesforce guide, read in full for the setup and error sections. Inbound polling detects up to 2,000 new or updated Salesforce records in one API call and then imports 200 per call, so 2,000 records costs eleven calls; the 200 is a Salesforce limit rather than an Outreach choice. Outbound changes are usually visible in Salesforce within 30 to 45 seconds. An admin can cap Outreach's daily API usage, and Outreach recommends setting global usage to 70 to 80 per cent of the org total. Conflict resolution scans field by field rather than taking a row lock, which means a field both systems write to flips quietly instead of erroring. The integration user needs create, read and edit on every object shown, field-level security on every mapped field, and API Enabled under System Permissions. Outreach also lists its five common errors and names sandbox session timeouts as the cause of connections breaking every 24 to 48 hours.
AffectsSalesforceOutreach
Baseline
Three Zapier limits on this pair that produce silent wrong answers
From Zapier's own HubSpot article. Only one Zapier app connection to a HubSpot account can exist at a time, and adding a second invalidates the first, so a colleague reconnecting breaks every Zap in the account. Date and date-time values sent through Zapier to HubSpot are rounded down to UTC midnight, because HubSpot takes dates as UNIX timestamps. The Add Contact to List action works only with static lists, and active lists do not appear in the dropdown at all. Two plan facts recorded alongside them: Zapier gates polling-interval control off its Free plan and offers it on Professional, Team and Enterprise, with the available intervals depending on the trigger app; and HubSpot documents 650,000 requests a day on Professional against 1 million on Enterprise, both at the same 190 requests per 10 seconds, so an upgrade raises the daily ceiling and not the burst rate.
AffectsHubSpotZapier
Baseline
HubSpot Import is on by default, and it points outward
From ZoomInfo's HubSpot integration guide, read in full for the setup and error sections. HubSpot Import is enabled by default and allows ZoomInfo to import and analyse information about your team and your prospects and customers from HubSpot. That is the only setting on this pair that moves your data outward rather than ZoomInfo's inward, and the default decides it unless somebody looks. Also recorded: ZoomInfo lists Free, Basic, Professional and Enterprise as supported HubSpot plans, so HubSpot gates nothing and every gate here is ZoomInfo's. The connection is either a service account, which ZoomInfo calls the default and preferred option because every export and filtering job runs through it, or individual user accounts. The connecting account needs View, Edit and Delete on contacts and companies, where ZoomInfo gives a minimum of records the user owns and recommends all records; the gap between those two is silent and shows up as exports reaching fewer records than expected. Duplicate handling is an admin default with per-export options for reps, including creating the duplicate anyway.
AffectsHubSpotZoomInfo
Baseline
Outreach publishes a FAQ for this pair, and most of it is behaviour nobody expects
Recorded from Outreach's own support articles. Reading a Sales Navigator tile inside Outreach does not notify the prospect; clicking through to the profile does. The Save as a lead button saves a lead in Sales Navigator and creates nothing in Outreach. An error shown inside the tile, such as CRM not connected, is usually LinkedIn's rather than Outreach's, because the tile is an embedded iframe of LinkedIn's own code. There are no scrollbars in it and the arrow keys are how you move. A wrong match is undone with Not the right person, and a no-match has a back button to search by hand. One operational note that costs a month if missed: the Outreach Salesforce plugin has to be current for LinkedIn tasks to sync to Salesforce at all.
AffectsLinkedInOutreach
Baseline
The connection type decides whether private channels work at all
Read from ClickUp's own integration article. A personal connection can sync ClickUp activity to public and private Slack channels; a Workspace connection can sync to public channels only. That is the opposite of what a team expects, because the Workspace connection is the one that looks like the governed choice. Recorded with four more conditions from the same page: a ClickUp Workspace connects to one Slack workspace at a time while a Slack workspace can serve several ClickUp Workspaces; initial Connected Search indexing can take 24 hours and a new message up to ten minutes to become searchable; a newly created or joined channel takes five minutes to appear in the picker; and the integration is available on every ClickUp plan, so price gates nothing here.
AffectsSlackClickUp
Baseline
Apollo publishes what each reveal costs, and a phone number is eight emails
Apollo's pricing page carries a credit price per action and it was not recorded here before: 1 credit for a verified email, 8 for a verified phone number, 1 to 8 for contact and company enrichment, 1 per AI research run and 1 per 30 seconds on the US dialer. The eight-to-one ratio between a phone number and an email is the number that explains an allowance disappearing without a bulk job, because the spend is per click by a rep rather than per campaign by an admin. This pair is also the first to carry no plan grid, deliberately. The constraint on it is not a tier on either side, it is LinkedIn's User Agreement, and inventing an editions table here would imply a gate that does not exist.
AffectsApolloLinkedIn
Baseline
The two meters on ZoomInfo Enrich, and the settings that cap them
Read from ZoomInfo's 44-page Enrich for Salesforce guide while writing the setup and error sections. Credit Usage Limits defaults to No limit, so an unbounded job enriches and charges for every record it matches; a per-run cap is available and is the single most useful setting on the page. Batch size is selectable at 10K, 1K, 500 or 200 records per Bulk API call, and a smaller batch increases Salesforce API consumption for the same work. An accuracy threshold can be set per job, and records below it are marked No Match with the reason "Below selected Accuracy Score" rather than being enriched on a guess. Contact creation can be hard-capped at up to 10 per account. Field mapping is per field with an overwrite or complete-if-missing option on each. Some features need the ZoomInfo Sales native application at 1.33 or higher.
AffectsSalesforceZoomInfo
Our correction
A ZoomInfo citation had gone to a Page Not Found
The first source on this record, university.zoominfo.com/hubspot-integration-setup, now returns ZoomInfo's own Page Not Found. It has been repointed at the ZoomInfo University index that lists the current HubSpot guides, and the PDF integration guide cited alongside it is unaffected and still live. Found by fetching every URL the dataset cites and reading what came back, which is now a step rather than a hope: `python research/docs.py --audit`. Across 160 cited documents it found two genuinely dead pages, this one and the Salesforce and Slack one recorded separately today. No claim on this page rested on the dead URL alone, so nothing stated here changes.
AffectsHubSpotZoomInfo
Baseline
Two automation engines, and only one of them maps a field
ClickUp's own integration article was read for the first time for this record and it settles what the previous "custom fields are the ceiling" limit only gestured at. Custom field mapping is not supported in HubSpot workflows at all: a deal-based workflow can create a Space, Folder, Subfolder, List or task and can carry deal name, type and amount as tokens in the name and description, and that is the whole of it. Field mapping exists only inside ClickUp Automations and only between equivalent types, so a HubSpot email field maps to a ClickUp Email custom field and to nothing else. One further rule that breaks chains quietly: an object created by a HubSpot Automation does not trigger another HubSpot Automation. Authenticating also needs a ClickUp Workspace owner or admin. Recorded alongside a published figure rather than a claim of ours: the app sits at 2.2 out of 5 from 72 ratings on HubSpot's own marketplace, 42 per cent of them one star.
AffectsHubSpotClickUp
Baseline
Three rules that decide where a HubSpot message can land in Slack
Recorded from HubSpot's own Slack workflow actions article while writing the setup and error sections. A record can be associated only with a public channel, or a private channel the HubSpot app has already been added to, which is why a private deal room is missing from the picker. When two records are merged, the Slack channel association comes from the primary record, so the wrong merge order silently drops the link. And the Slack workflow action sends to a statically chosen channel with no option to follow the record being processed, which HubSpot confirms in its own community thread and answers with branch-per-channel workflows. The first two were not recorded here before; the third was, and now has the workaround attached.
AffectsHubSpotSlack
Baseline
Six new conditions on the cheapest integration in the directory
This pair was re-read across five HubSpot articles for the setup, field and error sections, and the conditions that came out of them were not previously recorded here. A Gmail account enrolled in Google's Advanced Protection Program cannot be connected as an individual work email at all, and neither can a HubSpot View-Only seat. One email account cannot be both an individual work email and a team email, which is the rule behind every shared support@ mess. Email tracking and logging do not work with customised Gmail layouts. A message only logs and creates a contact if the sending address is a user, a connected address or a registered alias, and nothing errors when it is not. With data privacy settings on, opens are tracked only for contacts carrying a legal basis for processing. None of these changes what the integration costs, which is still nothing: they change who it silently does not work for.
AffectsHubSpotGoogle Workspace
Our correction
Embedded Profiles on Salesforce needs Advanced Plus, not Advanced
This record said the embedded profile widget runs on Sales Navigator Advanced and that only writeback, record creation and matching need Advanced Plus. That is true of the HubSpot pair and false here. LinkedIn's admin setup guide for Salesforce requires Advanced Plus for CRM Sync, Embedded Profiles and data validation alike, so on Salesforce there is no cheaper tier that buys you the card. The error came from assuming one vendor behaves the same way against two CRMs, which is the assumption this directory exists to check, and it was caught by reading LinkedIn's Salesforce-specific guide rather than its shared CRM page. The same guide settles the Salesforce side too: Enterprise, Performance, Unlimited and Developer are supported, and Professional only with API Access and Chatter enabled, which LinkedIn says may carry an extra Salesforce fee.
AffectsSalesforceLinkedIn
Our correction
Custom object sync never had a limit of ten, and we said it did
This record claimed custom objects synced "up to ten on Enterprise". That conflated two different numbers. HubSpot's product catalogue caps custom object *definitions* at ten on Sales Hub Enterprise, which is a limit on how many custom objects you may create in HubSpot at all; it is not a limit on how many of them the Salesforce integration will carry, and HubSpot's own custom object sync article states no such cap. Two further facts from that article replace the guesswork around it: the feature is Enterprise-only across the HubSpot hubs, and the HubSpot object must have at least one required field or it cannot sync. Found while writing the field reference for this pair, by reading the page the claim should have come from.
AffectsHubSpotSalesforce
Our correction
The Salesforce and Slack page cited a page that no longer exists
The single source behind this pair, salesforce.com/sales/slack-integration, now returns Salesforce's "the page you're trying to view isn't here". It was found by fetching every cited document in the dataset rather than by anybody clicking it, which is the point: a citation nobody re-reads is a citation that quietly stops being one. It has been replaced by the four documents this pair was re-read from, two on each side. The rewrite that followed corrects a real vagueness as well. The old record said feature availability "varies by cloud and edition" and told you to check your own licences, which is true of everything and useful for nothing. What the documentation actually says: the connection is available on every Slack plan and caps at one Salesforce org, with up to 20 more on Pro, Business+ and Enterprise Grid; Slack Connect channels are unsupported; a Government Cloud org cannot be connected at all; and the connection is a three-step handshake whose first two steps report success while nothing works. One thing is deliberately left unsettled rather than guessed. Salesforce does not name Professional on the permission-scopes page for the Salesforce for Slack app, and the older Slack-built app documents an API Access add-on on Professional, so the page says to price that add-on and confirm it against your own contract instead of asserting a requirement from a page about a different app.
AffectsSalesforceSlack
Vendor change
LinkedIn moved its CRM Sync guide, and the field list moved with it
Both documents behind this pair were re-read on this date. The tier facts are unchanged: CRM Sync still needs Sales Navigator Advanced Plus and a Sales Hub Professional licence or higher, and the embedded profile card still runs one edition down on Advanced. What changed is where LinkedIn keeps the answer. The enablement-guide URL we cited now redirects to a restructured CRM help hub, and the guide itself is published as a PDF linked from it. The anchor that used to list the writeback fields is gone. That PDF is now cited directly, because it is what was read: it carries the fixed object mapping, the format each Sales Navigator action takes once it lands in HubSpot, the per-user authentication requirement and the Business Manager conflict, none of which the HubSpot article states. Those are recorded on the page as a field reference, a connect sequence and an error list rather than as four more lines of prose.
AffectsHubSpotLinkedIn
Baseline
Twelve Reverse Contact guides, and five new tools behind them
The Reverse Contact hub goes from 7 pairs to 15. Four records that existed with no page were written up in full, Salesforce, Attio, Zapier and Make, and eight are new: Pipedrive, Zoho CRM, monday CRM, Outreach, Salesloft, Smartlead, lemlist and Instantly. Five of those needed a tool record first, read from each vendor on this date: Pipedrive, Zoho CRM, monday CRM, Salesloft and Instantly. The dataset goes from 52 integrations to 60 and from 18 tools to 23. The fact worth recording rather than the count: these pages were written to answer one question per tool, which is whether it can make an outbound HTTP call itself and read the answer, and the answers are further apart than any feature list suggests. Attio hands a workflow the response status and body with a 20-second ceiling. Zoho runs Deluge with invokeUrl and 30 seconds, on Enterprise and above. Pipedrive can fire a webhook from an automation and has no documented way to use what comes back, so the same enrichment needs a round trip. monday has no HTTP step at all and a daily ceiling of 1,000 API calls until Pro. Salesforce can do it from Flow or Apex and caps a transaction at 100 callouts and 120 cumulative seconds. On the sequencer side the spread is in the metering: 10,000 requests an hour per user on Outreach, 600 cost a minute for a whole team on Salesloft, 1,000 an hour on Smartlead Standard, 20 per 2 seconds on lemlist and 6,000 a minute on Instantly.
Three sequencer auth and rate-limit records were vague, and two were misleading
Was
lemlist auth was recorded as "API key" and its rate limits as "published in the lemlist API documentation". Smartlead auth was "API key" and its limits "published per plan". Outreach limits were "published in the Outreach API documentation".
Now
lemlist uses HTTP Basic rather than Bearer, with an empty username and the key as the password, so the string to base64-encode is :YourApiKey with a leading colon, and the limit is 20 requests per 2 seconds per key. Smartlead takes the key as a query parameter rather than a header, and allows 60 requests a minute and 1,000 an hour on Standard, 120 and 3,000 on Pro. Outreach allows 10,000 requests an hour, per user rather than per org.
Recorded as a correction rather than a vendor change because none of these moved; they were read properly for the first time. Two of them were actively misleading. "API key" for lemlist implies a bearer token, which fails against an API that wants Basic auth, and lemlist's own documentation puts that warning in red because of how often it catches people. "API key" for Smartlead hides that the key travels in the query string, where URLs get logged, which is a decision about where the integration may run rather than a detail. A directory that exists to save a reader an afternoon should not have shipped either as a shrug.
Unlimited plans replace pay as you go, and credits now expire in 3 months
Was
Pay as you go from $100, credits valid for 12 months, with custom annual plans above it and no published tiers in between.
Now
Credit packs from 5K to 1M credits, valid 3 months, still from $100. Above them three Unlimited plans: Go at $249 a month for unlimited cached lookups, Plus at $449 adding unlimited search, Max at $699 adding unlimited reverse lookups from an email, a name or a domain. Platform from $2,500 a month on an annual contract is the only option that keeps the 12-month rollover, and the only one licensed for showing the data to your own customers.
The shortened credit life is the part to design around rather than the new plans. A pack bought to cover a quarterly enrichment sweep used to last a year, and the same pack now expires between sweeps, which turns unused credits into a write-off for exactly the lumpy usage pattern a credit pack suits. Their own pricing page draws the line at roughly 15,000 lookups a month, above which an Unlimited plan costs less. Every cost table in this directory was rewritten against this.
Credit costs per call are published now, so this directory prints them
Was
No public per-lookup rate was published, and no page here printed a cost per enriched record for that reason.
Now
A published cost in credits for every endpoint: free for a freshness check and a search count, 1 credit for a cached people or company lookup, 1 per profile delivered by search, 2 for a reverse lookup from an email, a name or a domain, 2 for a live refresh, 3 for a business email and 20 for a direct phone number, with the last two charged only when one is found.
This changes a load-bearing rule in this directory rather than one field. The old rule was "print no cost per enriched record", which was right while nothing was published and became a refusal to repeat their own numbers the moment something was. The rule now is to print the published credit cost and never a dollar rate, because a dollar rate still is not published: pack prices exist, and dividing one by a credit count is our arithmetic wearing their authority. The full credit matrix sits in the dashboard rather than on the pricing page, so what is quoted here is what the public page and the MCP documentation state. The check in build/verify.py moved with the rule: it used to fail the build on the phrase "per lookup" and now fails it on a currency figure against a lookup, a record, a row or a credit.
Three native apps now exist, and none of them is in a CRM
Was
No marketplace app anywhere. Clay was recorded as an HTTP API column, Zapier as the generic Webhooks action, and Make as an HTTP module.
Now
An app on the Clay marketplace, listed there since May 2025, with four actions: find a person behind a work or personal email, enrich a person from a profile URL, find a company from a domain, enrich a company from its profile URL. A Zapier app with seven actions and two searches, including enrich and check for both people and companies and a workspace usage read. A Make app marked Verified and Official Vendor, with four modules: a LinkedIn profile, a LinkedIn company, a reverse email lookup, and an arbitrary authorised API call.
The pattern matters more than the count. All three sit on the orchestration layer, and there is still no app inside any CRM and none inside any sequencer, which is the argument the Reverse Contact pages here are built on and which survives intact. What changed is the shortest honest route to a tool Reverse Contact has never heard of: it is a step in a Zap or a Clay table rather than a day of plumbing. Their own integrations page invites requests for tools they have not built an app for, which is a fact worth knowing before a team pays for a build.
The Clay pair said you pay two meters, and on the native app you do not
Was
A row that resolves costs a Clay action and a Reverse Contact credit, and the pair required a Reverse Contact API key and a Clay plan allowing HTTP enrichment.
Now
The native app runs on Clay credits by default and needs no Reverse Contact account at all. Connecting your own key moves the lookups onto your Reverse Contact workspace instead, which is one meter either way rather than two. Clay still bills the attempt rather than the result, so a miss still costs a Clay action.
Written as a correction rather than a vendor change because the old wording was describing the HTTP column, which was the only route in September, as though it were the pairing itself. The practical consequence runs the other way from most corrections here: this pair is now the only route in the directory you can test before opening an account with the vendor, which is the right way to find your own match rate on your own list before committing to anything.
The MCP server went from 8 tools to 13, and the monthly refresh figure fell
Was
An MCP server with 8 tools. 300M requests a month. A 7-day evaluation with limited API access.
Now
13 tools: four enrichment, five search including a free count and two filter catalogues, and four free utility tools for checking a profile, a company, your usage and a running task. Streamable HTTP, JSON-RPC 2.0, OAuth 2.1 where the client supports it and a bearer key everywhere else. 200M refreshes a month. A 7-day free trial with no card, and 100 credits on a new account.
The headline number falling from 300M requests to 200M refreshes is a change of measure as much as of magnitude, and both wordings are theirs, so it is recorded rather than reconciled. The tool count is the part that affects a build: company lookup by domain and contact email or phone are deliberately not MCP tools, so an agent going from a domain to a company profile has to search first and enrich the profile URL it gets back. An agent skill that assumed otherwise would fail on its first call.
LinkedIn Sales Navigator becomes a hub, because five pair keywords worth around 1,700 searches a month had nowhere to file without one, and because the honest answer to "what integrates with LinkedIn" is a page the market does not write. Ten more pairs were read in full on this date: LinkedIn with Outreach, Zapier and n8n; Clay with Outreach, lemlist and Smartlead; HubSpot with Make, Attio and Smartlead; and Slack with Outreach. Five are new records, taking the dataset from 47 integrations to 52. Three pairs with better search volume were again left unwritten: Clay with Sales Navigator, because Clay publishes no document we could read for the import limits that page would rest on, and Apollo with ZoomInfo and Slack with LinkedIn for the reasons recorded below.
LinkedIn's API access is closed, not merely undocumented
Was
Auth was recorded as OAuth through the CRM Sync app, and the rate limits as not published for CRM Sync.
Now
OAuth 2.0, with sales permissions requiring approval as a Sales Navigator Application Platform partner, which LinkedIn documents it is not currently granting to new applicants. Rate limits are set per endpoint, reset at midnight UTC, carry both application and per-member ceilings, and the standard figures are not published anywhere; your own are visible only in the Developer Portal.
The old wording described one integration rather than the platform, and it made a closed door sound like a documentation gap. The distinction matters commercially: a vendor who says they will build you a Sales Navigator integration is describing a partnership they almost certainly cannot obtain, and a buyer who thinks the limits are merely unpublished will plan to discover them rather than to design around not knowing. This is the fact the new hub is built on.
HubSpot and Attio is a migration, and was recorded as a two-way sync
Was
Two-way, across contacts, companies and custom objects.
Now
One way, out of HubSpot. Attio's Import2 reads HubSpot once and writes into Attio, matching people by email and companies by domain, and it does not update attributes on records Attio already holds. There is no ongoing sync in either direction.
The record already said in its own summary that this was a migration rather than a sync, and then carried a sync direction that said the opposite, which is the kind of contradiction a reader resolves in favour of the field rather than the sentence. Anybody planning to run both CRMs in step on the strength of that field would have lost a week finding out. Corrected to one way, out.
The second batch read in full from vendor documentation on this date: Salesforce with ClickUp and with n8n, Slack with n8n and with Google Workspace, HubSpot with n8n, Gong and lemlist, Apollo with Salesforce and with Outreach, and Slack with Gong. Four of those are new records, which takes the dataset from 43 integrations to 47. Three pairs that rank well on search volume were deliberately not written up: Apollo with ZoomInfo, because the two compete and do not integrate, and Slack with LinkedIn and Clay with Sales Navigator, because neither vendor publishes documentation we could read for the claims those pages would have to make. An empty cell in the matrix is a better answer than a page built on forum posts.
Both Gong pairs had their sync direction backwards
Was
One way, out. On the Salesforce page, five of the five rows in the what-syncs table also pointed the wrong way, showing CRM records as arriving in Salesforce and Gong's call activity as leaving it.
Now
Two-way on both pairs. Gong reads accounts, contacts and opportunities out of the CRM and writes calls, meetings and Engage activity back into it, and the table now says so.
Direction on this site is written from the hub tool's point of view, so on a Salesforce page "out" means out of Salesforce. The Gong rows were authored the other way round and the headline flow was inherited from the shorter record that preceded the guide. The Salesforce page was live with the inverted table for part of a day. A sync direction is the field this directory exists to get right, which makes this worth an entry rather than a quiet edit, and the fix was found by checking the convention against four pages that already shipped rather than by reading the vendor documentation again.
A paid lemlist plan. lemlist publishes a 14 day trial and no free tier, and lists CRM integrations including HubSpot on every paid plan from Email upwards.
The record claimed a free tier that lemlist does not offer, which made the pair look cheaper to reach than it is. The integration itself genuinely needs no upgrade once you hold lemlist, which is why the cost band stays at included, but a reader deciding whether to try it should know the entry price is a subscription rather than nothing.
Ten records that carried only a direction, a cost band and a one-line summary were read in full from vendor documentation on this date, and now carry what syncs object by object, the documented limits and the pages every claim rests on: HubSpot with Google Workspace, Zapier and Outreach; Salesforce with ZoomInfo, Outreach, Gong and Zapier; Slack with ClickUp and Zapier; and Apollo with Sales Navigator. Salesforce with Zapier and Slack with Zapier are new records rather than deepened ones, which takes the dataset from 41 integrations to 43. Reading them closely moved four claims we had already published, and each of those has its own entry below.
HubSpot and Outreach connect through HubSpot data sync, not an Outreach plugin
Was
Per-object, per-field mapping in both directions across contacts, companies, deals and activities, as a project measured in days.
Now
HubSpot built the integration on its own data sync engine and runs it. Contacts sync one way or two way, email syncs from Outreach into HubSpot only, and Outreach's own documentation sends sync failures to HubSpot support.
The first reading described the plugin Outreach ships for Salesforce and assumed the HubSpot pairing was the same product pointed at a different CRM. It is not, and the difference is material: there is no per-field outbound control, no task or event mapping, and no way to sync a single record on demand. The record now lists two objects rather than four and hours rather than days. This is the most consequential thing we got wrong in this batch, because a team could have chosen HubSpot over Salesforce on the strength of a capability that only exists on the other CRM.
ZoomInfo Enrich writes into Salesforce and nothing comes back
Was
Two-way sync between Salesforce and ZoomInfo.
Now
One way, into Salesforce. Instant Enrich appends data as records are created and Scheduled Enrich runs jobs over existing ones, and the integration has no return path.
ZoomInfo does sell a two-way sync, on its HubSpot integration at the Elite edition, and the Salesforce record inherited that shape without being checked against ZoomInfo's own Salesforce guide. That guide describes Enrich as two applications that append ZoomInfo data to Salesforce leads, contacts and accounts, with no export back. A sync direction is the single most misread field in this category, which is the reason this directory records one per pair, and it is why getting it wrong in our own dataset is worth an entry rather than a quiet edit.
Saving a LinkedIn profile into Apollo is metered, not free
Was
Included on both sides, with no upgrade and no metering.
Now
Metered by usage. Apollo documents that saving a contact, or revealing an email or a phone number, consumes credits, and that credits are pooled across the team, granted upfront each billing cycle and expire rather than carrying over.
The cost band said this pair costs nothing extra on either side, which is true of installing the extension and false of using it. Every save is a credit, and Apollo's free plan does not include the CRM integrations that would get a saved contact anywhere useful, so the cheapest version of this workflow ends at Apollo. Corrected to the usage band, which also takes the number of integrations here that genuinely cost nothing extra from five to four.
Two marketplace install counts removed as unverifiable
Was
Gmail at 532,000 installs and Google Calendar at 289,000 as the top two apps on the HubSpot marketplace, and 173,000 installs for Zapier.
Now
No install counts on either page. The HubSpot marketplace renders its listings in the browser and returns an empty document to a fetch, so these figures could not be read again and nothing here now rests on them.
The figures were published in the first reading and we can no longer show the page they came from. The rule on this site is that a claim carries the document it was read from, and a number that cannot be re-read fails that test however plausible it is. Both pages now lead with a gate we did verify instead, which is a more useful sentence than an install count was.
AffectsHubSpotGoogle WorkspaceZapier
Our correction
HubSpot workflow webhooks need Data Hub, not Sales Hub
Was
Sales Hub Professional carries the webhook and custom code workflow actions.
Now
Both actions require Data Hub Professional or Enterprise, which HubSpot sells separately from Sales Hub. Custom code is also capped at 20 seconds and 128 MB.
This was wrong in the first draft of the HubSpot and Reverse Contact guide and it was the most consequential error in the build. A team on Sales Hub Professional has workflows and still cannot call an external API from one, so the recommendation flips from "call it from a workflow" to "call it from a workflow if you own Data Hub, otherwise Clay or n8n". Corrected before launch, recorded here because a change log that only shows other people being wrong is marketing.
Every tier requirement, sync direction, object list and documented limit in the dataset was read from a vendor knowledge base, developer documentation or pricing page on this date, and each record carries the pages it was read from. This is the line every later entry is measured against.
Freshness
Every record, and when it was last read
All 68 integrations with the date the facts behind them were last read from a vendor page, and how many vendor documents each one rests on. Most recently checked first.
68 of the 68 records carry a citation, and they are exactly the 68 with a full guide written. The other 0 were read on the same date but their source list has not been written up, which is why this column says zero rather than hiding the row. Writing those 0 guides is the next thing on the list.
Facts move. So does the data in your CRM.
A tier requirement changes a few times a year. A contact record goes stale continuously: people change jobs, companies get acquired, domains move. Reverse Contact resolves an email, a domain or a profile URL into a verified people and company record through one API, with a confidence score your logic can branch on.
Free trial, no card required, and 500 credits to start.
enrich.persons200 OK
FormResolveWriteAlert
Fit92VP Sales, 220 staffRouted to an owner✓
P95500msRecord completeTitle, seniority, company✓
FAQ
The change log, asked and answered
What this tracks, what it does not, and why that distinction matters.
Why are there so few entries?
Because the directory was first checked on 28 October 2026 and this log starts there. A log that opened with twenty entries would mean twenty changes had been invented. The baseline is the line everything later is measured against, and it will fill up on its own.
What counts as a change?
A claim a reader could act on: a tier requirement, a sync direction, what objects move, the cost band, the setup effort, a documented limit, or the vendor pages a record rests on. Rewording how a fact is explained is not a change and is deliberately not tracked, or the log would be noise.
Does this watch the vendors automatically?
No, and no page here says it does. The fingerprint catches a change to our own records, which is what stops a fact moving without being recorded. Re-reading a vendor page is a person opening it, and the last-read date on every record in the table above is when that happened.
Why publish your own mistakes?
Because the version of a page you read last month is the one you may have budgeted against. The HubSpot Data Hub correction changed the recommendation on several pages from "call it from a workflow" to "call it from a workflow if you own Data Hub", which is the difference between a build that works and a plan that does not.
How do I get told when something changes?
There is no subscription yet. The machine-readable surface is the way to watch it for now: data.json carries every record with its verified date, and the MCP server exposes the same log as a tool an agent can poll.
We sell B2B people and company data, and we publish this directory. Every number on this page is derived from those records rather than typed, and each record was read from a vendor's own knowledge base, developer documentation or pricing page on the date below. Nothing on it is restated from a press release or a third-party listing.
Where a vendor publishes no public figure, this page says so rather than estimating, which is why no page here prints a cost per enriched record. Reverse Contact appears in this directory as a data provider and publishes it. No other vendor pays for placement, and every link out of a listing or a citation carries nofollow, including the ones that point at Reverse Contact. The only followed links on this page are the two calls to action and the reviewer.
Tier requirements, sync behaviour and documented limits change without notice. Confirm against your own contract before you budget. What has changed since · How we checked, and what we will not claim
Researched and wrote every page in this directory, and checked every tier requirement, sync direction and documented limit against the vendor's own knowledge base, developer documentation or pricing page.
Tier requirements, sync behaviour and limits read from each vendor's own documentation in September, October and November 2026. Vendors change these without notice.
Reverse Contact appears in this directory as a data provider and publishes it. Every other tool is here on merit and nobody pays for placement.