Research library · checked 31 Oct 2026

Research reports and project dossiers.

Start with the comparison, then open the dossier for the layer that matches the job. Each page separates documented facts, project claims, operating requirements, and unresolved questions.

ScopeThis library contains dated source audits and test-gate reports through the 31 October Plausible CE handoff check, with selected primary pages checked through 31 October 2026. The 28 September handoff is a synthesis. The 29 September small-list comparison, 30 September Keila follow-up and 1 October Listmonk tagged-code audit read source and documentation only. The 2 October private Listmonk runtime test could not run for lack of a pinned artifact, isolated runtime and sink; no install, send, image inspection or restore was performed. The 3 October phpList audit reads the exact-tag confirmation send path and current documentation; signup/editor files and runtime behavior remain unverified. The 4 October decision contract synthesizes the earlier one-message findings into a task-first route without adding runtime evidence. The 5 October Ghost audit reads selected v6.63.0 source and living official help, not the installed send, provider or restore path. The 6 October Marketing Agent OS one-skill install attempt stopped before execution because the pinned checkout and network-denied boundary were unavailable; no file effects were observed. The 7 October Postiz audit checks a tagged release note and living first-party docs: preview comments are not an enforced approval gate; no channel, draft, preview, send, provider or security test occurred. The 8 October Umami v3.4.0 audit traces tagged page views, custom signup signals, goal and attribution rules from source and current documentation only. No instance, message or browser fixture was used. The 9 October Plausible CE v3.2.1 audit traces a tagged pageview, UTM ingestion, a separate signup event and goal rate from source and living docs only. No image was pulled, instance started, dashboard inspected or account verified; the v3.2.1 patch and /storybook exposure condition are explicit. The 10 October SiteOne v2.5.1 source audit compares manual five-route review with a conditional, owner-authorized and externally guarded crawler inventory. Its --url-list bypass and five-page noindex-alert limit require explicit human checks; no crawler or fixture was run. The 11 October YOURLS 1.10.6 tagged-source audit uses the direct approved URL as the off-ramp; private redirecting needs separate approval and rollback records, and 301 caching can limit retargeting. No redirect, cache or destination was tested. The 12 October promptfoo 0.124.1 source audit puts a fact card and human checklist before a script or eval matrix; no eval or copy approval occurred. The 13 October ResourceSpace decision is based on living vendor documentation and limited master licence notices; no fixed 11.0 package, install, expiry or replacement was tested. The 14 October localized-offer audit uses the Weblate 2026.10.1 release view and versioned documentation only; no tagged tree, translator review, channel export or live action was tested. The 15 October calendar-to-receipt audit reads Vikunja v2.7.0 release, licence and selected tagged models plus living help; no artifact or instance was tested, and its three-channel reconciliation is fictional. The 16 October two-newsletter migration audit reads Listmonk v6.2.0 tagged import and campaign paths plus official sender documentation; no app, send, migration or restore was tested. Its nine-contact mapping is fictional. The 17 October Mautic form-to-contact audit uses official release metadata and living 7.2 documentation, not a pinned installed build or CRM test. Mautic submission/contact IDs are documented, but a durable CRM receipt is not; the releases page conflicts on 7.2 support dates. The 18 October pre-send audit reads Mailpit v1.31.5 tagged configuration and current official documentation; no SMTP message was sent or captured and no sender application was inspected. The existing sender preview is the first conditional option; a local sink addresses only the narrower application-to-SMTP question. The 19 October SerpBear v3.1.0 decision reads tagged code and current first-party docs, not a running tracker, provider, Search Console account or restore. Its fictional disagreement starts with matched Search Console clicks and impressions; a first-page provider result and a page-versus-domain match are different observations. The 20 October page-refresh decision uses current Google documentation and earlier SiteOne/SerpBear tagged-source audits; its three-page card is fictional, with no property, crawl, content edit or SEO effect observed. The 21 October page-publication receipt is a documentation synthesis with a wholly fictional variant example; no customer page, deploy, CDN, index status or SEO effect was observed. The 22 October new-share trace stopped at DNS before HTTP; current metadata, image response and native platform card are unknown, and a saved assembled page is not a live response. The 23 October executed offline fixture compared 17 fictional planned links with one fictional approval contract; 17/17 expected outcomes matched, but no live URL, sent link, analytics event or signup was observed. The 24 October executed Python-standard-library fixture mapped nine invented identities across two newsletters. Its baseline matched 3 conditional and 15 no-send slots, and four planted unsafe proposals failed; no Listmonk path, real permission record, migration or send was exercised. The 26 October one-click header fixture parsed eleven invented messages: two valid A/B cases passed, nine planted unsafe headers failed closed, and an unknown case failed. It checks only advertised pre-send header strings against an independent mapping; no DKIM, POST, Listmonk, mailbox control, endpoint state, consent or delivery was tested. The 28 October Formbricks decision reads release metadata and current first-party documentation; exact 6.1.2 tag files, edition behavior and runtime were unavailable. No existing form or survey response was tested. The 27 October campaign-page decision checks Google primary guidance only; no authorized URL, Search Console property, indexed result or removal action was observed. The 29 October inquiry-to-conversation report adds an operator-defined ledger and an executed, invented nine-inquiry reconciliation: 6 ever-linked booking chains, 4 attended, 3 human-qualified and 2 ambiguous holds. It did not exercise a real form, CRM, calendar, Cal.com/Cal.diy build, webhook or identity match. The 7/14/28-day policy is not a general fixture windowing engine; Cal.diy post-split exact artifact and licence remain unverified. The 30 October MJML report reads the v5.4.1 release, tagged licence and package manifests, CLI documentation, and a first-party example. Exact-version compilation was unrun; no validated template, generated HTML, diff, send or client result was observed. Tagged MIT evidence has a narrower scope than dependencies or hosted services. The 31 October Plausible CE handoff uses selective crawler-rendered text and clicked link destinations, not a normal browser or full page archive. Its local assertion and controlled saved-link mutation test only the chosen CE and Cloud paths; no redirect trace, variant, form, account or conversion was observed. None of these reports is a hands-on benchmark, security audit, legal opinion, or approval for unattended outreach or publishing.

Research collection

Switch between the same permanent report collection in List, Table, and Calendar views.

Plausible CE announcement: which next action?

The 31 October handoff check finds a comparison link from the CE announcement, a CE repository path on the destination and a separate Cloud trial button. The saved-link assertion distinguished those paths in a controlled mutation. No normal logged-out browser, redirect, variant or conversion result was observed; CTA prominence needs human review against the intended offer.

Changed an email CTA or preheader?

The 30 October three-gate report puts the approved sender/editor first, a local MJML v5.4.1 draft-to-HTML compile second only if source is owned, and final-message plus authorized client checks third. The exact compiler test did not run: no generated HTML, diff, send or client result was observed. Tagged MIT scope covers the repository and named packages, not all dependencies or hosted services.

Are submissions becoming qualified conversations?

The 29 October ledger and fictional reconciliation distinguish accepted inquiries, CRM receipts, linked booking status, host-confirmed attendance and a human rubric. Keep the current form, CRM and calendar when their IDs and history can be reconciled; hold ambiguous joins. The fixture is not a live conversion rate, and Cal.diy has a post-split artifact/licence gap.

One optional post-campaign feedback question?

The 28 October decision checks an existing approved form or survey link first. A plain Formbricks Community link is conditional if that path lacks needed review or export. The 6.1.2 release is a current maintenance signal, not proof of exact-tag licence or behavior; privacy, edition and coordinated recovery remain open. This was documentation review, not a submitted survey or visitor measurement.

Public, retired, or private campaign page?

The 27 October decision table chooses crawlable noindex, a real 404/410, or access control by intended visitor state. Its exact-URL receipt keeps the current public response apart from a later Google indexed result. Documentation review only; no site or Search Console state was observed.

Does one-click unsubscribe point to the right list?

The 26 October offline header check accepted two valid A/B messages and rejected all nine planted unsafe headers in a fictional parser fixture. Compare a captured final message with an independently approved list mapping before sending. The test did not validate DKIM, run a POST, exercise Listmonk or prove endpoint, mailbox, consent or delivery behavior.

Reconcile two newsletters before a migration?

The 24 October local no-send fixture matched all 18 invented newsletter slots and rejected four planted unsafe target proposals. An evidence token is only a fictional locator, not permission proof. This did not exercise Listmonk or a real import; the 16 October source audit supplies the separate tagged-code boundary.

Check a planned campaign link before sending

The 23 October offline fixture compared parsed destination and UTM values with one fictional approval contract. All 17 expected outcomes matched; the checker caught absent, empty, duplicated and conflicting tags, wrong destinations, embedded credentials and fragments. This is pre-send string QA, not a live-link, sender or analytics test.

Will a new share show corrected metadata?

The 22 October share-card trace did not reach an ordinary HTTP response. The saved assembled page has metadata fields but does not establish current title, description, image or native platform display. Check the public page, selected image and an authorized native preview in that order; handle existing posts separately.

Did approved copy reach the public URL?

21 October page-publication receipt separates approved, deployed, public and indexed states. Start with a manual normal-URL response: final URL, status, exact approved marker, canonical/robots, cache headers, UTC time and variant. The three receipts are fictional; no customer page was checked.

Which page should be refreshed?

Read the 20 October one-page diagnosis. A fictional three-page card separates refresh, technical/provider diagnosis and wait/unknown. It uses exact Search Console page/query filters and two complete 14-day PT windows, then manual URL and content checks. SiteOne is conditional; no live site or SEO effect was measured.

Search Console and a rank tracker disagree?

Read the 19 October Search Console–first decision. Match the query, canonical page, country, device, Web search type and complete dates before reading average position. SerpBear v3.1.0 can add a fixed provider snapshot only with the matched result URL, checked depth and failure status. A fictional triage card shows when to investigate indexing, inspect content, wait or retain an unknown. No live rank was measured.

One email before a real send?

The 18 October pre-send decision compares a verified existing preview/test path with an isolated Mailpit v1.31.5 SMTP capture. No message was sent or captured. A local receipt would establish only what reached that sink, not provider acceptance, recipient display or permission to send.

One welcome and an occasional newsletter?

Start with the 4 October small-list welcome decision contract. It separates DOI request, one immediate message, ongoing campaigns and delayed sequences, then tests the existing-sender option against consent, approval, suppression, attempts, provider outcomes and recovery. For a new one-message system, phpList is a conditional private-evaluation route, Keila is a form-centered alternative, and Listmonk needs a reviewed external/manual trigger and ledger. None is send-ready or added to the four-layer shortlist. If you already operate a Ghost publication site, its native one-message welcome is a conditional off-ramp documented on 5 October, not a send-vetted choice. Ghost's opt-in timed-sequence beta requires a separate audit. Earlier dated audits remain below.

Client sign-off before a social post?

The 7 October Postiz v2.25.0 report distinguishes a draft preview/comment conversation from a required approval gate. Current docs do not show an enforced two-person sign-off, and a no-channel in-app draft is not established. Start with offline fictional copy; a later Postiz trial needs a permissioned test channel and private deployment. Postiz is not a vetted shortlist choice.

Newsletter campaign measurement?

The 8 October Umami v3.4.0 report separates sender delivery, tagged site views, a site-emitted signup signal and verified account records. It documents the page-view UTM query and model-based attribution but no end-to-end runtime result. A private synthetic pilot is the next step; Umami alone cannot verify an account or campaign lift. This is an investigation, not a shortlist addition.

Another self-hosted measurement route

The 9 October Plausible CE v3.2.1 audit finds tagged-source support for UTM-bearing pageviews and a separately instrumented signup event. First-party docs describe Campaigns and goal rates, but no CE dashboard was observed. CE is a conditional synthetic-pilot choice if the team can operate the app, PostgreSQL and ClickHouse, and verify the v3.2.1 security patch and deployment exposure. Unlike a verified account record, its event remains a browser-side receipt. This is not a four-layer shortlist addition.

Five-page site promotion?

The 10 October SiteOne Crawler v2.5.1 report puts manual review first. A crawl is conditional on owner permission and an external exact host/path guard; the tool's URL-list and noindex-alert limits make a crawler score unsafe as the decision itself. This was a source-only audit, not a runtime test or shortlist addition.

One campaign link?

Use a direct approved HTTPS URL with UTMs unless shortness or controlled retargeting justifies another service. The 11 October YOURLS 1.10.6 source audit explains core create, edit, 301 redirect and click-count behavior, with an external approval/rollback ledger. Clicks are not arrivals or signups; no runtime behavior was observed.

AI-written campaign copy before sign-off?

Start with an approved-facts card and a human checklist. Repeated stable URL, phrase and length rules may justify a tiny script; a matrix across many drafts, prompts or models may justify promptfoo. The 12 October promptfoo 0.124.1 source audit keeps claim review and exact-version publication approval with people. No eval or campaign approval was observed.

Does one campaign image need a DAM?

Use a controlled folder, versioned register and a three-placement ledger first. The 13 October ResourceSpace decision compares a small script and a self-hosted DAM, and distinguishes the expiry-field warning from plugin/cron state changes. No immutable 11.0 artifact or install was inspected; external placements require separate action.

One offer, two languages, three placements?

Start with the 14 October localized-offer release-control report. Its bilingual register gives each landing, email and social variant an exact version, language and campaign-fact review, destination owner and queue/live/sent receipt. The fictional v1 email stop is unconfirmed and remains held. Weblate 2026.10.1 is a conditional repository-string route, not a channel approval or release system; this was release/documentation reading only.

Planned versus observed across three placements?

The 15 October campaign calendar and receipt-ledger audit starts with a shared calendar plus versioned destination receipts. Its fictional landing delay, unknown email cancellation and already-public social post show why a planned slot or completed task cannot prove release. Vikunja v2.7.0 can coordinate owners and reminders if handoffs repeat; it does not replace channel checks. This is a source audit, not a live campaign or installed-board test.

Moving two newsletters without reviving opt-outs?

Read the 16 October source audit before exporting contacts. Check whether the current service can retain A/B-specific preferences, global suppression and permission history. Listmonk v6.2.0 tagged paths make disabled and single-opt-in unconfirmed unsafe generic holding labels. Keep unknowns outside sendable lists, investigate overwrite collisions and read back every A/B and global state after late-event replay. The nine-contact example is fictional; no migration or send was tested.

Can a form inquiry be reconciled to a CRM contact?

Read the 17 October form-to-contact audit. Start with the existing form and CRM: for every accepted submission, preserve an event ID, source and stated contact preference, destination CRM ID or owned exception. Mautic 7.2.1 documents form submissions and associated contact IDs, but its generic POST and plugin actions do not establish a durable, idempotent CRM handoff. No exact-build, connector or CRM test was run; current 7.2 support status needs maintainer confirmation.

Comparison and dossiers

ComparisonFour-layer comparisonScope, setup, license, and operating boundaries side by side.PlatformMautic dossier7.2.1 campaign and form-to-contact boundaries: CRM receipts remain unproven.OrchestratorActivepieces CE dossier0.92.0 execution, credentials, retention and recovery boundaries.Source-availablen8n dossierWorkflow and AI orchestration, Sustainable Use License boundary, and a direct comparison with Activepieces CE.Harness libraryMarketing Agent OS dossierPortable skills, provenance limits, and host-harness responsibilities.Exploratory local drafting CLILoopmark dossierEvidence gaps, a draft-only test, and the narrow retained label.

Use the evidence carefully

For terminology and full source links, read Evidence & limits. For a draft-only first experiment, read Safe first test. Test the exact connector, credentials, data path, retry route, and human approval before a production decision.