Product changes, security hardening, public-site changes and corrections of our own copy, newest first. Each entry is written on the day it shipped from our own decision record, and a correction of something we said wrongly is listed like any other change.
Public site
A guide for every one of the 23 dashboard features, and the documentation as Markdown
Twelve new guides, each written from the shipped screen: Test Folders, Test Runs, Project Management, the Analytics Report, Enterprise TDM, the credentials vault, Data Shift, QA Generator, Page Analyzer, QA Testing beyond the canvas, Billing & Invoicing, and your plan with the trial, the free period and a paused account. The documentation index now lists every feature the dashboard shows with the guide that explains it, and a test fails the build when a feature has none. Thirteen existing guides were brought up to date (file requests and downloads during a recording, co-admins and ownership transfer, the November accessibility rounds, the Web Analyst tabs, the full export list). Every guide and the index now have a Markdown version at the same address with .md appended, listed in llms.txt, for readers that prefer text.
The quick start and llms.txt said the Free plan has no time limit; since 14 November it has one
Since 14 November 2026 the Free plan runs as a 14-day free period after the Solo Pro trial, and an account whose free period has ended is paused until a plan is chosen: everything is kept, nothing is charged or deleted, and only the profile and plan pages open. The quick-start guide answered "Does the free plan expire?" with "No", and llms.txt said "no time limit". Both now describe the free period, and a new guide explains what each stage allows. The quick start also still named sign-in with Google or GitHub, withdrawn on 21 October; that sentence is gone.
A step-by-step guide to SOC Analyst, written from the shipped screen
A new guide walks through SOC Analyst screen by screen: where to find it and which plans include it, cataloguing your critical assets first so priorities are not provisional, uploading a log or a raw Windows .evtx, reading the report in the order an analyst works it, exporting the report before you leave the page, working the incident register, and every error message the upload can show, each with what it means and what to do next.
SOC Analyst described wrongly in four places on 14 November, corrected the next day
Writing the guide from the shipped code showed that the module page, the decoder article, the home page, llms.txt and this changelog had described SOC Analyst wrongly. They said detections run as Sigma rules; the analysis runs 17 deterministic detections of its own, and Sigma appears only in experimental detection drafts. They said findings are labelled suspected or confirmed; the SOC report shows a severity, the arithmetic and a computed confidence instead. They said every tool carries a statement of what it can and cannot prove; the report grades what the corpus can support instead. And they described opening an incident from a finding as a step on the page; there is no button for that yet, and incidents are opened through the API. Every sentence was rewritten, and the article records the correction.
SOC Analyst is on the site: the thirteenth module page, and raw .evtx uploads
SOC Analyst has been in the product since 13 October and reachable at its own route since 13 November, and the public site named it nowhere — so a customer could not find out it existed. It now has a module page of its own, and the home page, the product hub, the Product menu, the entity data and llms.txt count thirteen modules and three jobs instead of twelve. The module takes one log corpus of up to 64 MB and returns a triaged assessment: 17 deterministic detections mapped to MITRE ATT&CK, the arithmetic and a computed confidence behind every finding, a grade of what the corpus can support, and priority weighted by the asset inventory a team owner maintains. From 14 November it accepts the raw binary Windows Event Log (.evtx) with no conversion step, alongside CSV, TSV, Excel, JSON, JSONL, syslog, CEF, LEEF, Apache and nginx access logs, plain text and an EvtxECmd CSV export. The corpus and the report are never stored; the one stored record is an incident, which keeps its priority and response clocks. It is on the security-scanning entitlement, Solo Pro and up, and not on Free.
Our .evtx decoder was wrong about the format twice; 1,961 real records found it
The decoder that reads a raw Windows event log passed every test we had written for it and was wrong about the format in two independent ways: an inline element name left the BinXML cursor twenty bytes short, and EventData arrives as a nested fragment rather than a scalar value. The effect was the worst shape a parser defect takes — every record’s envelope decoded perfectly while the entire payload, the account names and command lines a detection actually reads, came back empty, with nothing in the report saying so. Measured against 1,961 records from four real Windows channels, records decoding in full went from 27.4% to 86.9% and recovered EventData values from zero to 5,984, with 83% of records carrying at least one. Records we still cannot fully read are marked partial and counted on every run rather than quietly shortened. The article publishes the method, both numbers and the correction of a figure we had got wrong ourselves.
Enterprise test data: mask a production extract and keep its joins
Enterprise test-data management now masks (pseudonymises) a CSV, TSV or JSON extract you upload, from "Mask an extract" in the dataset catalogue. It proposes a rule for each column from a sample and a person reviews the plan; the run then masks the whole extract with a fresh key that is destroyed when the run ends, so we cannot recreate the mapping. The same value masks the same way in every file, so joins survive, and the run counts orphaned keys before and after and stops if the count changes. National identifiers are re-issued from ranges never issued to a person and e-mail addresses move to example.com. Every column the plan keeps is checked for e-mail addresses, US social security numbers, UK national insurance numbers, card numbers and IBANs, and the run stops until a person acknowledges a hit; the checks match patterns and do not find a name inside free text. The masked copy is saved as a dataset in your workspace, and a smaller subset that keeps its links can be cut through the API. It is still personal data under the GDPR and India’s DPDP Act, so keep treating it as such.
Masking claims withdrawn on 11 November and restored, with their limits, on 13 November
On 11 November a review found masking sold as ours on six surfaces (the pricing table, the Katalon comparison, the test data page, the GDPR-safe test data article, llms.txt and the signed-in dashboard) while no part of the product masked a dataset a customer supplies, and those sentences were withdrawn. Masking shipped on 13 November, so the same surfaces describe it again, each with its limits: the output is pseudonymised and still personal data, and the checks on the columns you keep match patterns.
Four more module pages: smart test selection, QA workflows, document intelligence and synthetic monitoring; the platform counts twelve modules on every page
Four services that every signed-in user already had, and that the site named only as capabilities around the modules, now have pages of their own under Product, each with its capabilities, how it works, an evidence ledger, the limits we state and the plans it is on: smart test selection (a requirements document or a Jira, Linear or GitHub issue in, the smallest scenario set within a time budget out), QA workflows (keyword nodes on a canvas, 81 keywords, RPA steps, connectors, Search & Collect, a test-file library per user, an exception desk), document intelligence (answers cited to your own files, a read-only database assistant, a local model) and synthetic monitoring (a recorded flow or an API check on a timetable, uptime, alert rules with hysteresis, alerts to Slack, Microsoft Teams, Opsgenie or a webhook). The home page, the product hub, the header menu, the entity data and llms.txt now count twelve modules and three jobs; the accessibility page describes the measured checks, site audits, finding lifecycle and exports shipped in November; and a line in llms.txt that said smart test selection maps a commit range to tests was corrected, because it works from requirements.
The k6 export page shows the script it writes, five glossary terms with examples, and the vendor documentation behind each integration
An outside review of the site’s search presence asked for artefacts a developer can read. The k6 export page now prints the script the load screen writes for an illustrative profile, generated by the product’s own exporter and held equal to it by a test, with what the export leaves out stated beside it. The glossary gained five terms (401 Unauthorized against 403 Forbidden, an idempotent request, a proof of concept, a per-virtual-user test data set, vendor lock-in), three with a short example. Every live integration page links the other product’s own documentation, read on the date shown, and names that product in its structured data. The trust centre answers how to run a vendor security assessment, the Postman alternatives page whether one platform can take over Postman and JMeter work, and the smart test selection guide whether it can choose the tests for a pull request. Documentation guides are typed as articles as well as technical articles in their structured data. Nothing was removed.
Load testing: different data for each virtual user, headers and a body per target, and an export
The load screen gained three things. A CSV file can be attached so that each virtual user sends different data: a column is written into the address, a header or the body as ${column_name}, each user starts on its own row, and the screen says so when there are fewer rows than users. Each target has its own headers and body. And a profile set up on the load screen leaves the platform: as a JSON file the screen reads back and as a k6 script, with the test data as a CSV file, and a summary line per stored run (average, P95, P99, error rate, requests per second) as CSV or JSON. Three settings the screen showed and the engine did not apply are corrected: every target after the first now runs (on runs of up to 500 virtual users), the wait between requests is applied, and Stop stops the run on the server; three more that are shown and not applied (profile type, ramp-down, pacing) say so on the screen. A plan built from a recording, with its steps, assertions and thresholds, has no export, and there is still no JMX import or export.
The AI provider is named wherever it applies, and other statements that had fallen behind are corrected
A reader’s review of the site found statements that had fallen behind the product. More than twenty sentences said nothing is sent to an AI provider; each now carries the exception that has applied since 1 November: a team that buys and switches on Work with AI sends redacted text to Anthropic, which is listed as a sub-processor with a thirty-day notice period. The Terms and the Cookie Policy carry the product name throughout, the vendor checklist states its own tally correctly, two claims with no source behind them were withdrawn, and the one-line description of Work with AI names the product areas it covers instead of two features.
Pricing re-cut for the global launch: Solo Pro $39, Team $249 for 16 seats, Enterprise from $799
The launch sheet raises Solo Pro from $29 to $39 and Team from $199 to $249 (still 16 seats, about $15.56 a seat when full), and starts Enterprise quotes at $799. Single tools were re-priced to read as a way to start rather than a way around a plan: Extra recordings is $9 (was $25), the Credentials vault $5, QA workflows, Projects and AI tools $15, Document intelligence and Work with AI $19, Security scanning stays $35. Annual billing stays 20% less and every allowance is unchanged. The page now opens with the three ways to buy, answers what happens at a cap and what is actually paid in which currency, and names Work with AI as the one paid add-on that sends redacted text to an outside model. The rupee sheet follows the same published rule as every tool: ₹3,399 and ₹21,899 a month.
Every comparison re-read against the vendors’ own pages, with sources, owners and a working correction link
An outside legal review and a re-audit of 586 statements against the vendors’ own pages changed every comparison and alternatives page. Each claim about another product now links the page it was read from and the date; capability rows carry one of four verdicts (built in, via add-on, not offered, not confirmed), and "not offered" is used only where the vendor’s own pages say so. Each page names every brand it mentions with its owner’s legal name (the full list is on the new trademarks page), logs its corrections, and links a correction form that we answer within five business days. The Postman, Selenium, Apache JMeter, Katalon and Playwright comparisons moved into their alternatives pages, where the same searches landed. ZAP is named as ZAP: it left OWASP in August 2023. Our own claims were corrected where the code disagreed: a recorded run has no heal review queue, the API studio times HTTP requests to the microsecond but not its gRPC and SOAP testers, there is no self-hosted option, the admin audit log has no integrity check, and account and recording records are deleted by hand.
A test evidence page, and country pages built around the rules buyers there must meet
A new page says what every run keeps for a reviewer (seven linked files per recorded session, a screenshot and the network calls per failed step, proof with every confirmed security finding) and what it does not, including no CI quality gate and no attestation of our own. Each of the thirteen country pages gained the local rules a testing buyer is asked to show, each with its primary source, the date read and where Dokimore fits or falls short. Sitemap dates now move only when a page’s content changes.
The UI/UX validation engine that measures every control of a recorded journey, live since 24 October, now has a public page: what it checks and which standard each check cites, the friction it counts, the gate verdict, the Definition of Done, the website evidence, and the limits (your testers, not recruited users; one lab sample for vitals). A head-to-head with Testsigma joins the comparisons, every row read on the vendor’s pages with its date, strengths first, no price figures. Each module page gained a "Works with" list of the integrations its work passes through and a list of related modules.
Country pages carry the words buyers use for regression, no-code and UX testing
The United States, United Kingdom, India, Canada, Germany, France and Netherlands pages map the phrases their buyers type for regression testing, testing without code and UX testing (tests de non-régression, regressietesten, Regressionstest, no code automation testing among them) to the page that answers each, in the market’s own words, with the limits stated: no CI trigger yet, a lab proxy rather than recruited users.
The review earlier today corrected most sentences that said a person approves each heal. A second sweep found 38 more, on the home page and its search description, the about, landing and solutions pages, eight country pages, the comparison and alternatives pages, the glossary, two guides and two blog posts. Each now says what happens: when a locator breaks, a fallback rescues the step during the replay, and the run detail shows the locator that failed, the replacement and the strategy. There is no customer review queue yet. The test that refuses the old claim now also refuses the paraphrases that slipped past it.
Answers for the searches people type in 2026, and two new glossary terms
Short answers where buyers search: whether the LLM tester tests AI agents, how WCAG testing works, whether the security scanner is a penetration testing tool (partly: an automated DAST scanner, not a manual pentest), ADA compliance testing for US teams, and Playwright MCP among Playwright’s strengths. The glossary now defines quality engineering and shift-left testing.
Every public promise checked against what a customer can use
An audit of the public pages against what a signed-up customer can reach found claims with nothing behind them, now corrected. The CI pages described a pipeline integration only operators can open; they now show how to run exported tests in each CI. Slack, Teams, Opsgenie and webhooks receive synthetic-monitor alerts only, and PagerDuty is not connected. Load profiles have no k6 export, and HAR files are shown, not imported. Google and GitHub sign-in are not switched on. Orders are paid by bank transfer, not card. Cloudflare, in front of the site since 17 October, is now listed as a sub-processor, and the Terms name the allowances we actually count and when they reset. A second pass corrected the legal and trust pages: the database has no transparent encryption, row-level security covers the billing, access and usage tables rather than every table, the audit log is not hash-chained, erasure and export do not yet reach account and recording records automatically, and backups stay on the same server for up to six months. A rescued locator is applied during the replay and shown in the run, rather than waiting for a person to approve it; accessibility audits run the A and AA rules, and the 66 national standards are a reference table.
Cypress and Katalon alternatives, two new guides, and answers to the questions people search
Alternatives pages for Cypress and Katalon that state when to stay on the incumbent before comparing seven other tools by job, and name where Dokimore falls short; every Cypress and Katalon statement was read from the vendor’s own pages that day and checked by a second reader. A guide to flaky tests that quotes the studies that measured their causes; a guide to the seven files one recorded run produces; and short answers on the product pages to questions buyers actually type, such as what a DAST tool tests and whether a load testing tool is a performance testing tool.
Katalon comparison corrected from Katalon’s own pages
Re-reading Katalon’s pages for the new alternatives page found three statements of ours with no source. We said Katalon had basic security checks: its pages describe no security scanning, and its blog points to OWASP ZAP and Burp Suite. We called its plugin ecosystem and community larger: it runs a plugin store, but no source measures the size. Our EU and UK compliance answer left out that Katalon states SOC 2 Type II and ISO/IEC 27001, 27017 and 42001 certification and offers hosting in Europe, which Dokimore cannot match today. The comparison now says so, and adds rows for application types, assurance and how tests leave each platform.
One more import claim corrected, on the recorder page
The correction below missed one line: the recorder page’s list of limits said Cypress tests could be imported. They cannot. No Selenium, Cypress, Playwright or JMeter file is imported, and a Cypress flow is recorded again in the browser. The same line also said Cypress export was unavailable everywhere; that is true of recordings only, so it now says so. For this correction the test that guards public copy was widened: it now recognises more ways of wording an import claim, and a negation excuses a claim only when it sits in the same clause.
Public copy corrected: no Selenium, Cypress, Playwright or JMeter importer
An audit of the live site against the code found two claims with nothing behind them, now corrected: a "Migration Setup" that imported Selenium, Cypress, Playwright and JMeter assets (the screen of that name is a database-migration wizard), and load profiles that exported to JMeter and Locust. The API studio imports Postman, Bruno, Insomnia and Hoppscotch collections, OpenAPI and cURL. (This entry also said HAR files are imported and load profiles export as k6 scripts; neither was true, corrected 2026-10-21.) Every page, FAQ and pricing row that said otherwise was rewritten, and a test now refuses the old claims.
Seventeen integration pages, each naming the file in our code that implements it and checked by a test that the file exists; a dated changelog; and meta descriptions rewritten to fit a search snippet on every blog post, guide and country page.
Twenty tools compared by job, alternatives pages for Postman, Selenium, Playwright and JMeter, and two new head-to-head comparisons
A hub that compares twenty tools by the job each is built for, with Dokimore in the same table and never first; one page per incumbent that states the incumbent’s case before ours and names where Dokimore falls short; and comparisons with testRigor and mabl whose rows carry a three-state verdict, the vendor page it came from and the date it was read. Comparison pages gained breadcrumbs and links to the product they compare.
LLM tester: prompt-engineering toggle and output grammars
The API-key tester can correct and engineer the text you send for the model it goes to, with the sent text kept as the stored message, and it validates structured output against a grammar (JSON schema, regex, GBNF and more); an assertion fails when a grammar cannot be verified.
Row-level security enforced at the database for billing, access and usage tables
Tenant isolation is now enforced by PostgreSQL row-level security on the native tables, with forced policies, a tenant context set per transaction, and an isolation suite that proves a cross-tenant read returns nothing. Corrected 2026-10-21: this entry first said "the tenant tables"; the policies cover the tables that hold billing, access-control, usage and security records (138 of 664 tables), while accounts, recordings and teams are kept apart by checks in the application.
Write-ahead logs are streamed continuously beside the daily encrypted backups, and a drill replays them into a scratch database every week and checks table, policy and row-count parity against production.
A sweep of the HTTP surface closed object-level and tenant-scoping gaps, standardised input validation at the boundary and added rate limits where they were missing.
The SOC analysis engine detects the format of uploaded evidence by content rather than by file name, reads fourteen formats, grades its own evidence before it reasons from it, and refuses a confidence it cannot support.
SOC console: five views, a ranked queue, ATT&CK v20 mapping
The security operations surface became a console rather than one long report: a ranked queue, an ATT&CK mapping updated to v20, and the full report kept one click away.
LLM tester as a chat app, with datasets, rubrics and a judge
Connect a key, chat, run an agent or compare up to four models side by side; keep datasets and session references; grade meaning with rubrics and a judge whose backend is selected by environment with a local fallback.
robots.txt stopped enumerating the product, the sitemap carries loc and lastmod only, client-only app routes answer noindex from the edge, and three independent launch audits were verified claim by claim: 24 real defects fixed in copy and markup.
LLM tester: multi-turn probe chains and three adversarial probe domains
Nine multi-turn chains with a turn-of-flip measure and single-turn controls, probes for manipulation, bias and sensitive data, and a two-level slash menu that groups the probe catalogue by category.
Templates for REST, GraphQL, gRPC, SOAP, SSE, WebSocket and MQTT with an honest executability field per template, so a pack says which protocols it can run rather than passing by absence.
One authorised account per browser with its stored session reused, visibility per user, team and enterprise, and a dedicated browser profile per identity.
API Studio: one import pipeline and an OpenAPI core
Postman, Bruno, Insomnia, Hoppscotch, OpenAPI, HAR and cURL imports share one pipeline with a per-item report of what was skipped or translated; OpenAPI 2.0 to 3.2 operations keep their contract snapshot for schema assertions.
Warm paper, one accent, one signal colour and a type pairing committed to across the public pages; the LLM tester and project management followed inside the app.
Billing under bank transfer, privacy rights requests, accessibility fixes
The billing lifecycle runs on bank transfer with enforced quotas and no overage; access, export and erasure requests can be made in the product; the pre-launch accessibility audit findings were fixed. Corrected 2026-10-21: this entry first said the requests "reach every table"; the automated part covers the registered data classes, and account and recording records are handled by hand until they are registered.
Evidence ledger, exposure revalidation, asset graph, remediation and retest
A tamper-evident hash-chained ledger for findings, continuous revalidation of each finding as a fact with a half-life, an asset graph with criticality and exposure, and a remediation, retest and closure lifecycle.
Every outbound request from the platform passes an egress guard; per-tenant egress and tenant-scoped realtime delivery were enforced; a passive detector for AI/LLM response data leakage and cross-scan finding correlation were added to the scan pipeline.
A benchmark of five small local models on 250 labelled vulnerability descriptions, and how the React site became readable by answer engines without a new dependency.