What changed

This week brought a simpler experience for companies preparing their own privacy documentation, with a unified Questions screen for all users, clearer document states, and fixes to how facts flow into what gets drafted and published.

What changed

Documents and what goes into them

The DPA's Article 28(3) particulars (subject matter, nature, purpose, types of data, special categories, data subjects, duration), its breach-notification term, its sub-processor terms and signature block are now built by code — facts where confirmed, gaps where not, the same on every run. The sub-processor list attaches to the DPA as "Annex 1 — Sub-processors", the same code-built table as the standalone sub-processor list document. Documents are prepared by deliverable and regenerated only where something changed, with no required-answer gate. An answer on a document now marks only the documents resting on it out of date, instead of regenerating everything. The Documents page shows "Available" with Prepare and "Update documents (N)".

Documents are the company's own and say only what it confirmed — no "consultant" in a document or a gap question, controller or processor only from a confirmed answer (asked on the document with a suggestion and its reason), and "I don't know" is a gap, never printed. A document a running job is still drafting is not queued again — the server refuses a second prepare/regenerate with 409 "… is being prepared already". A document's Failed state follows its own last attempt, not the client's latest job.

A document with open gaps can now be published to the trust page. The editor shows "Publish with gaps"; it stays marked as needing review until filled. The trust page, its tables and the PDF a visitor downloads read each gap as "Not stated". The Trust tab says how many gaps each published document shows.

Privacy policy, DPA and transfers

The privacy policy's "Who we share it with" is now built by code from the linked vendors — what each does, its country and each enabled law's transfer basis, with the providers' instructions-and-contract rule and authorities only where the law requires. The privacy policy's and the DPA's transfer sections now list only the vendors whose transfer needs a safeguard beyond adequacy (for EEA and UK alike), and point to the recipients table or Annex 1 for the rest. The UK section no longer repeats all nine rows. A vendor with no country is asked once per row, and its basis cells read "Depends on the country". Annex 1 shows the UK basis when UK GDPR is enabled.

The DPA's and the ROPA's data subjects are the people the product facts name (employees, end users, customers' customers), else

This week we improved how we handle certain scenarios when gathering information from external sources, and refined how draft letters are created.

What changed

• When a source refuses to answer (with certain error types), we now treat that as a final answer rather than retrying indefinitely. Sources that fail are checked at most three times a day.
• Draft letters are now written only when there's a genuine reason to do so, based on your decisions about which cases need them.

This week brought improvements to data matching and integrity across the register, reporting, and discovery features. We also fixed issues with scheduling and text handling.

What changed

Registers and vendor data

• The DPF register now matches covered entities to their parent's certification, including vendors on contract platforms and trust centres.
• The queue searches companies by their trading names (d/b/a and brands), catches banks named without the word "bank", and recognises group companies on registry vendor sites.
• A list's country now confirms a candidate when its legal form and domain match. Research you've done is never replaced or refilled by an automated run — complete candidates are added but stay unverified.
• The nightly registry job now tracks which lists it has read in chunks. A fetch failure on one of 417 lists meant nothing was collected or added between 14 and 19 October.

Law pages

• Page text now drops NUL characters before it is stored. A law page's PDF text carried them, which broke the nightly law walk from 16 October.

Outreach scheduling

• New York channel hours now run to 22:00, so the Pacific coast keeps 7:00–19:00.
• A company's sending day runs 7:00 to 19:00 its time on weekdays, and the daily cap counts the provider's UTC day. (Letters held past midnight UTC went out at 20:00 New York.)
• A company's sending day now opens at 7:00 its time, not 8:00.
• The daily cap now counts what the provider sent today (UTC) and what it still holds, not what we handed over today.

Next-wave testing

• One follow-up was read as the whole sequence in testing.
• The test now sends ten first letters and five follow-ups a day (one follow-up to role mailboxes, none to others). Versions are judged by replies and clicks — opens within two minutes are classed as machines. The subject line names the customer whose list names the recipients.

Scoring

• One signal row per kind and page is now written in a single write. A scoring run failed on three job adverts pointing at the same board.

Discovery and sub-processors

• A business site is now read from what such sites say: docs, API, for-teams pages, enterprise, SSO, integrations, trusted-by lists.
• A company that another company names as its sub-processor is always marked B2B.
• Companies from other companies' lists are now sized by Hunter. Over 500 people were excluded as too large, and unchecked ones are never auto-chosen. A group naming its own

This week brought improvements to vendor research, letter writing and sending, and company discovery for outreach.

What changed

Vendor research and registry

• Services now include a category based on what the company's research says it does.
• Registry pages let you copy vendor names and select all vendors at once. You can do the same on the New vendors page, for each filter (All, Verified, To verify, Missing a link).
• Vendor search now finds documents nested inside other documents, checks parent companies, and handles sites that have moved. It also respects legal form restrictions by country.
• Registry calls are cached by step and read at 10k per page.

Finding contact details

• Hunter now searches for named people first, and falls back to a general mailbox only if no person is found.
• Hunter looks for leadership roles only: executive, legal, management, IT and operations. It returns ten results per search.
• Marketing, product and support roles count as eligible contacts, after the four main roles.
• Technology officers and directors of engineering are treated as CTO-level. Hunter confidence is set to 60.
• Addresses come only from public pages, not from search engines or LinkedIn. We don't track the recipient's own pages.

Letters and replies

• Letter 1 now asks for a reply. Follow-up 1 brings the report, and shows the company's domain.
• Letters name the services found, and say who operates clausebench. They leave any changes the vendor made to their own documents for the report.
• Letter 1 includes a fix. There's a new version B for the buying moment, and a new follow-up 2. The recipient's working day is respected, and results are grouped by version.
• Named people are contacted first; a general mailbox is used only if no person is found.
• US companies whose Hunter zone is outside America are sent to during New York's working day.

Sending letters

• Letters are sent through Smartlead. Telegram alerts you when they are actually sent, not when handed over.
• No more than 15 letters are handed over in a day (the limit the mailbox can send), counting letters the provider still holds.
• A first letter written and sent in one run is announced in Telegram too.
• Each letter sent and each reply is reported to you in Telegram, without naming recipients.

Company discovery

• Discovery now finds companies for outreach automatically, scored by signals and kept to a three-day stock.
• Discovered companies are found by web search (US, 10–500 people). Discovery leaves out competitors and companies past Series B or with over 500 people, and requires the company to be named on its own site.
• Discovery records

This week brought fixes across search and the vendor registry, alongside improvements to how the workspace handles vendor data and site scans. We also resolved several issues with notifications, empty states, and the site's navigation.

What changed

Vendor registry and research

The registry now consolidates vendor information from multiple sources: legal pages, SEC filings where applicable, and domain whois data all feed into country and legal entity fields. When you add vendors in bulk, records publish immediately. The registry processes up to 50 candidates and 50 additions daily, with Claude finishing incomplete entries each day and code review confirming each response.

When you mark vendors as verified in bulk, you see the cost upfront. You can now copy vendor names in group actions on the candidates page and mark items "to verify". Sub-processor lists appear in the vendor's trust-centre tab, pulled from the DPA itself rather than from trust-centre naming conventions.

The registry no longer silently drops scan findings when you merge vendor records. When a browser retries a request after the registry declines it, the /scanner endpoint now reports this. A vendor with no documents of its own is marked as a duplicate rather than queued for review.

Banks are rejected as candidates before any search begins. Vendors closing their service are rejected by their own stated closure language. The registry reads legal pages in EU languages, and recognises trust-centres on their own domain even if they don't use the word "trust".

Site search and scanning

The site now has search (powered by Google Search Console hints) and a proper 404 page. Ten site searches no longer return the same records repeatedly. The scan progress bar now appears on the scan page, with a menu toggle showing how much work awaits you.

When a site blocks every page up to /trust, you still get enough renders to map it. A site that won't let us in is now named on the report, with an explanation of who can open it.

Free checks

Free checks now report against both your company and the vendor. Results for free checks now show timelines as a visual bar, not just listed. The retention period and domain for the sandbox—both promised to customers—now work as described.

Vendor pages

Vendor pages appear as a search result on the site. The subprocessor list within a vendor's own trust-centre is now visible.

Admin menu and task tracking

Admin pages had generic "Admin" titles; each now has its own name. The menu now shows how much work is waiting for you in each queue. Records awaiting review count towards your workload in the menu too. When you view candidates or records, icons in the menu show both queues combined.

Notifications

Notifications have moved from a menu row to a bell icon in the top right.

Empty

This week brought improvements to interviews, pricing, internal linking across guides, and a new public API for the registry. We also launched US state privacy packages and made the product speak to companies rather than individuals throughout.

What changed

Interviews and questionnaires

Interviews now display as a dialogue rather than a list of exchanges. Old interviews play in order, and the hint in the field refers to the current question only.

Pricing and payments

Stripe is now live. Payment options are card, Apple Pay and Google Pay. The pricing page no longer reads as though payments are unavailable—and it no longer promised a tax that does not exist.

Guides and internal linking

Guide text now links itself: a name in a guide or article becomes a clickable link to that page. These internal links are worked out automatically, not maintained by hand. Links are also filled in for records already in the reference. Ten of the eleven guides had no internal links at all; guides are now 1,500 to 2,700 words each and link to one another where the text already names the thing.

A hub for the DPA was added, and an alternate name for that document now works.

Registry API

The registry now answers other systems. Seven endpoints on /api/v1 serve: the list and a single record; your own domains and names matched against the registry; a vendor's history and a feed of all changes since a moment; a sub-processor summary by jurisdiction; and reference values. Only verified records leave the building. Keys are shown once and stored as a hash; revoking a key stops it on the next request. Rate limits and monthly quotas are counted from the log; requests refuse with 429, a reason and Retry-After rather than an empty answer. Unchanged answers return 304. Each API consumer carries three real test records. The admin issues and revokes keys and sees monthly usage. /registry-api documents the API with curl, TypeScript and Python examples, and /api/v1/openapi.json describes it in OpenAPI format.

Addresses and company details

The API page quotes prices from the price table and an address someone reads. Company pages use the name from the footer, and industry pages speak to companies in that sector.

Campaigns and visitor tracking

The campaign report now reads the ad account, and answers from a check can be read back. Offline conversions are switched on. Three campaigns are running, and a daily check knows their labels. Adverts now buy brand searches rather than category names.

Visitors are counted as visits, not people—a visit is each time someone opens and reads.

Landing pages and site structure

All twenty-five landing pages now follow GDPR-page structure. Email fields no longer collapse on phone. The admin screen

This week we shipped firm lookups, email through Resend with a custom template, and the UK transfer mechanism. We built out the moderation queue for registry changes, restructured how interviews and questionnaires work, and made large changes to how the workspace looks and behaves.

What changed

Client overview and setup

A client's overview now opens showing what it holds — vendors, questionnaires, open breaches and data requests — each a link to its tab. Your company is marked "Your company" after the account becomes a practice.

The setup no longer has navigation and does not ask again what the sign-up quiz answered. A new workspace is set up before its dashboard, reading your website in the background to add vendors and proposed interview answers.

Interviews and questionnaires

Questions are stored before answers, so history reads in order. A free-text question is answered by the reply itself when the model is unavailable, shown as a proposal.

Answered questionnaires now return to you by a link with answers on a page and the original file to download, with everything unconfirmed left blank. What was answered once is proposed where it is asked again.

Optional questions in the interview panel are answerable but never asked or counted. Inviting comes first on the team page, with roles shown as Colleague, Admin colleague and Viewer, each with what they may do.

Questionnaires

The questionnaires page now shows what it is for and how to get there — interview, regimes, documents — instead of a bare prompt. Structured questionnaires show reply, control owner and cleared foreign answers; CAIQ is recognised by its headings.

A Yes/No column beside the questions is the reply column, with answer text going to the next free one. Our own questionnaire may name us under the same exemption as generation.

Documents and gaps

Every gap comes with its question and is answered where it can be closed. An answered gap is not asked again under another name — the question is kept with the answer.

A gap inside a table cell no longer breaks the table in the editor. A gap without its question no longer costs the whole document.

Fields completed on signature — the customer's details in a client's DPA — are blanks to fill in, not gaps.

Document import

You can now import existing documents — DOCX, PDF, TXT or pasted text — read in the background. Facts are kept only with a passage that is in the document. A review screen shows fact and quote, with confirm, fix or skip. Conflicts with confirmed values show side by side and are never overwritten in bulk. Vendors are matched through the registry. ZIP import is available with an emailed report.

Trust pages and documents

The trust page shows the version marked final,