[jjroelofs]

Summary

• Add MkDocs Material documentation site structured from README content, split into logical sections (getting started, API reference, guides, development)
• Add .gitlab-ci.yml with standard DrupalCI templates to enable automatic publishing to project.pages.drupalcode.org/rl/
• Add mkdocs.yml configuration for Material theme

Documentation pages

• Home: overview, features, use cases
• Getting Started: installation, configuration
• API Reference: PHP API, JavaScript API (Drupal.rl), HTTP API (rl.php)
• Guides: Drush commands, experiment decorators, cache management, AI assistant integration
• Development: contributing guidelines

Test plan

• Verify no em-dashes in docs content
• Verify GitHub CI linters pass
• After merge to 1.x on drupal.org, verify docs publish at project.pages.drupalcode.org/rl/

https://claude.ai/code/session_01Cbtzf2mpa3xQKCSkFzJuiR

[jjroelofs]

The MkDocs site builds successfully with CI=true mkdocs build --strict, and all GitHub checks pass. I found eight actionable documentation accuracy issues below. Several runnable examples currently call unavailable APIs or point to nonexistent routes, so these should be addressed before merge. GitHub does not allow an author to request changes on their own PR, so I am submitting this as a comment review.

[jjroelofs]

[P1] Point users at the registered settings route. rl.routing.yml defines this form at /admin/config/services/reinforcement-learning under Configuration > Web services, so the documented URL returns a 404. The same stale URL appears in docs/getting-started/installation.md:10; please update both occurrences.

[jjroelofs]

[P1] Use a setting that the command accepts. RlConfigCommands::SETTINGS_SCHEMA only contains debug_mode, enable_event_log, event_log_max_rows, and chart_line_threshold. Both exploration_rate commands return Unknown setting (and 0.1 is not a valid value for any current numeric setting), so readers cannot run this quick-start example.

[jjroelofs]

[P1] Use the public registry/manager methods. ExperimentRegistryInterface exposes isRegistered(), not exists(), and it has no delete() method. Running this sample therefore fails on both calls. The current cleanup API is rl.experiment_manager->purgeExperiment(); please update the example accordingly.

[jjroelofs]

[P1] Remove the obsolete external Plotly installation. The rl/plotly library loads js/vendor/plotly.min.js, which is already shipped by this module; no code reads web/libraries/plotly.js-dist-min/. These steps unnecessarily add a Composer plugin and alter the consuming project's installer paths even though the installed asset is unused.

[jjroelofs]

[P1] Do not offer a route fallback that the module does not provide. rl.routing.yml contains only admin report/settings routes and no tracking API route, so a site that cannot expose rl.php has nothing named “Drupal Routes API” to switch to. Either document that consumers must implement their own route or provide/link to a real module endpoint.

[jjroelofs]

[P1] Document batch rejection responses. Batch failures are not silent: handle_batch_request() returns an errors array, and rl.php responds with HTTP 422 and ok: false when every requested entry is rejected; Drupal.rl also surfaces those errors in the console. Please describe the errors response and add 422 to the error table so API clients do not assume only 400/500 are possible.

[jjroelofs]

[P1] Make the example service instantiable. MyExperimentDecorator above has no constructor and accesses the entity type manager statically, but this definition passes @entity_type.manager to its constructor. Symfony will fail to instantiate the copied example. Remove the arguments line, or add constructor injection and use the injected service in decorateArm().

[jjroelofs]

[P2] Do not claim Thompson Sampling stops loser traffic. Each decision draws again from every arm's Beta posterior, and the implementation explicitly preserves exploration, so a losing arm retains a nonzero chance of selection. It generally receives less traffic; it only stops once the experiment owner removes or locks the variant.

[jjroelofs]

Nonexistent config key: exploration_rate does not exist anywhere in the module. config/install/rl.settings.yml and the allow-list in RlConfigCommands define only debug_mode, enable_event_log, event_log_max_rows, and chart_line_threshold. Both documented commands (rl:config:get exploration_rate here and rl:config:set exploration_rate 0.1 on line 20) fail key validation, and the doc implies a tunable exploration knob the algorithm does not have. Use one of the real keys in the examples.

[jjroelofs]

Wrong settings path: the real route (rl.settings in rl.routing.yml) is /admin/config/services/reinforcement-learning, titled "RL: A/B Testing", and rl.links.menu.yml places it under Web services, not System. /admin/config/system/rl returns a 404. The same wrong path appears in docs/getting-started/installation.md line 10.

Suggested change

	After enabling the module, visit **Administration > Configuration > System >
	RL Settings** (`/admin/config/system/rl`) to configure the module.
	After enabling the module, visit **Administration > Configuration > Web services >
	RL: A/B Testing** (`/admin/config/services/reinforcement-learning`) to configure the module.

[jjroelofs]

Wrong settings path (same issue as in configuration.md): the real route is /admin/config/services/reinforcement-learning; /admin/config/system/rl returns a 404.

Suggested change

	After enabling, visit `/admin/config/system/rl` to review the default settings,
	After enabling, visit `/admin/config/services/reinforcement-learning` to review the default settings,

[jjroelofs]

Undefined methods: src/Registry/ExperimentRegistryInterface.php declares only register(), isRegistered(), getOwner(), getAll(), and getExperimentName(). There is no exists() and no delete(); code copied from this example fatals with "Call to undefined method". The existence check is isRegistered(), and experiment deletion has no registry API (the drush rl:experiment:delete command deletes the rl_arm_data/rl_arm_snapshots rows directly).

Suggested change

	// Check if an experiment exists.
	$exists = $registry->exists('my-experiment');
	// Remove an experiment and all its data.
	$registry->delete('my-experiment');
	// Check if an experiment exists.
	$exists = $registry->isRegistered('my-experiment');
	// Removing an experiment and all its data has no registry API;
	// use `drush rl:experiment:delete my-experiment` instead.

[jjroelofs]

Wrong error contract: rl.php does not drop invalid batch entries silently. Rejected entries are reported in an errors array with machine-readable reasons (unknown_experiment, invalid_arm_id, missing_arms, ...), and when every entry of a non-empty batch is rejected the endpoint returns HTTP 422 with {"ok": false} (rl.php lines 111-122). The error table below (400/500 only) omits the 422. A client written to this doc, expecting 200 plus ok: true in all cases, will misclassify the 422 as a transport/server failure and may retry endlessly; js/rl.js itself special-cases exactly this 422 (line 175).

[jjroelofs]

Section is dead weight: nothing in the module loads from web/libraries/plotly.js-dist-min/. rl.libraries.yml defines the plotly library as the bundled js/vendor/plotly.min.js (committed to the repo, v2.35.2) and ReportsController attaches rl/plotly; there is no hook_library_info_alter and no reference to the Asset Packagist path anywhere. Charts work with zero extra installation, so these four composer steps send every installer through pointless setup, and "required for charts" is false. Either drop the section or make the module actually consume the external library. (README.md carries the same stale instructions.)

[jjroelofs]

Opts into the full DrupalCI matrix, not just Pages: the pages/mkdocs job does live inside include.drupalci.main.yml (and gitlab_templates offers no pages-only include), but SKIP_COMPOSER_LINT, SKIP_PHPCS, SKIP_PHPSTAN, SKIP_STYLELINT, SKIP_ESLINT, and SKIP_CSPELL all default to '0', so every drupalcode push/MR now also runs those jobs with the templates' own configs. This repo has no cspell wordlist and no phpstan config, and standardizes lint on docker compose plus GitHub Actions, so the first pipeline will show unrelated red/orange jobs and any pages-job failure gets lost in the noise. (PHPUnit is safe: that job is gated by a tests-exist rule and this repo has only scripts/e2e/.) Consider setting the documented SKIP_* or allow-failure variables in this same commit, or note explicitly that adopting full DrupalCI is intentional and was verified against a test pipeline. Also worth knowing: the pages job deploys only on pushes to drupalcode's default branch that touch docs/** or mkdocs.yml; nothing deploys from GitHub, MRs, or feature branches.

[jjroelofs]

Broken edit links (currently inert): with a leading slash, MkDocs resolves edit_uri against the host root and drops the project path: urljoin('https://git.drupalcode.org/project/rl/', '/-/edit/1.x/docs/index.md') yields https://git.drupalcode.org/-/edit/1.x/docs/index.md, a 404 (verified against the mkdocs 1.6 _set_edit_url source). Nothing surfaces it today because Material only renders the edit button when the content.action.edit theme feature is enabled, which it is not; the moment someone enables it, every edit link on the site 404s. Drop the leading slash, and add content.action.edit to theme.features if the button is wanted (or delete edit_uri instead of shipping dead-but-broken config).

Suggested change

	edit_uri: /-/edit/1.x/docs/
	edit_uri: '-/edit/1.x/docs/'

[jjroelofs]

Stray file will be published (anchored here because the file predates this PR): docs/rl_project_desc.html is a tracked drupal.org project-description HTML snippet containing a dxpr.com/?ref=50701 referral link. MkDocs copies non-markdown files in the docs dir verbatim into the built site, and no exclude_docs is configured, so the Pages build this PR enables will publish it at https://project.pages.drupalcode.org/rl/rl_project_desc.html as an unstyled orphan page. Add exclude_docs: rl_project_desc.html or move the file out of docs/.

[jjroelofs]

Example contradicts itself: this registration passes arguments: ['@entity_type.manager'] to a class that has no constructor and instead calls \Drupal::entityTypeManager() statically (line 38). PHP silently ignores constructor args on a constructor-less class, so the injected service is dead weight, and the canonical example teaches the static-call pattern the module's own decorators avoid (modules/rl_menu_link injects via constructor). Separately, ReportsController calls decorateArm() once per arm, so the per-arm load($arm_id) in the example is an N+1 entity-load pattern on the reports page; the bundled VariantExperimentDecoratorBase shows the cheaper shape (resolve once, per-request cache). Suggest adding a constructor that stores EntityTypeManagerInterface and memoizing loads.

[jjroelofs]

Duplication without a canonical source: docs/** copies most of the 600-line README (install steps, Plotly section, Nginx block, API references, drush tables) and the README gains no pointer to the new docs site. Drift is already visible inside this PR: README's rl.php smoke test uses action=turns against an unregistered experiment (silently dropped, so it proves nothing) while installation.md correctly uses action=ping; README's PHP example calls getThompsonScores('my-experiment') single-arg while docs/api/php.md shows the three-argument form. Suggest trimming README's deep-reference sections to links into this site, or at minimum adding a prominent docs-site link, and naming which copy is canonical.

[jjroelofs]

Not LGTM yet. All eight findings from the first review are substantively fixed, the current head passes every GitHub check, and CI=true mkdocs build --strict succeeds. However, the integrations guide added after that review introduces two new copy/paste or factual blockers below; I also noted one remaining UI-label mismatch. The previously posted threads about the full DrupalCI matrix, inert broken edit_uri, and the published orphan rl_project_desc.html also remain applicable.

[jjroelofs]

[P1] Make the menu hook example executable and recursive. This snippet never initializes $selector, so copied code fails with an undefined-variable error. It also loops only over the top-level items even though the text says it walks the menu tree; nested links are skipped. The actual module obtains rl_menu_link.variant_selector and passes it through the recursive _rl_menu_link_process_items() helper. Please include the service lookup and recursion (or clearly mark the omitted pieces) so the worked example demonstrates the behavior it claims.

[jjroelofs]

[P1] Describe rl_sorting's actual transport. The current rl_sorting 1.x implementation does not call Drupal.rl.turn(): js/rl-sorting-tracking.js collects visible arms for 100 ms, then sends a legacy action=turns FormData payload with navigator.sendBeacon(); rewards are also direct legacy form posts. This sentence contradicts this PR's own JavaScript API page, which correctly documents that behavior. Please keep the guide consistent with the production consumer.

[jjroelofs]

[P2] Use the menu item's actual label. The route is fixed, but rl.links.menu.yml titles this entry RL: A/B Testing, not Reinforcement Learning. Users navigating through the UI will not find the breadcrumb as written; change the final segment to the registered menu title.

[jjroelofs]

Follow-up review at head 60e5808

The content fixes in af16ecc land well: the config keys, settings path, registry example (isRegistered() / purgeExperiment()), batch error contract, Plotly section, and decorator DI example are all verified fixed against the source, and the new integrations guide checked out clean claim-by-claim. Nice work.

Still blocking merge (unchanged since the first pass):

1. .gitlab-ci.yml: the include still opts every drupalcode pipeline into the full DrupalCI matrix (phpcs, phpstan, cspell, stylelint, eslint with the templates' own configs) with no SKIP_* curation; the repo has no cspell wordlist or phpstan config, so the first push to 1.x will produce unrelated red jobs around the pages build.
2. mkdocs.yml line 5: edit_uri: /-/edit/1.x/docs/ still has the leading slash, which resolves edit links to git.drupalcode.org/-/edit/... (project path dropped, 404). Suggested one-line fix is in the earlier inline comment.
3. docs/rl_project_desc.html: still inside docs/, so the Pages build will publish it verbatim (including the dxpr.com/?ref=50701 link) at /rl/rl_project_desc.html. 60e5808 grows the file further in that location. exclude_docs: rl_project_desc.html in mkdocs.yml or moving it out of docs/ fixes it.
4. README drift: README is untouched and now actively contradicts the docs; it still carries the Plotly/Asset Packagist install that installation.md correctly dropped, plus the action=turns smoke test and single-arg getThompsonScores() example. A docs-site link plus trimming those sections would close it.
5. Two wording nits: (a) "Traffic only truly stops when the experiment owner removes or locks the variant" appears in docs/guides/faq.md and now also in rl_project_desc.html; no lock feature exists anywhere in the module or submodules, so "or locks" should go. (b) docs/api/http.md's errors example pairs "ok": false with "when some entries are rejected", but rl.php sets ok: false only when every entry was rejected (partial success is 200 with ok: true plus errors, and the 422 body still includes "decisions": {}).

All five are small, mechanical changes; with them in, this looks mergeable.

[jjroelofs]

Still not LGTM. CI and a strict MkDocs build pass, but the latest commit only updates the project-description FAQ/headings and leaves the previously reported menu-hook example, rl_sorting transport claim, and settings label unchanged. The modified project description also repeats the unsupported Drupal Routes API fallback; see the new inline comment. The earlier MkDocs/GitLab configuration concerns remain unresolved as well.

[jjroelofs]

[P1] Remove the nonexistent route fallback. The module defines no tracking/API route in rl.routing.yml; its routes are admin settings and reports only. The installation page was corrected after the first review, but this project description still tells blocked clients to use an API that the module does not provide. State that direct access is required, or document a concrete custom proxy implementation instead.

[jjroelofs]

Tiny carry-over: this section got the pre-fix wording, so the example pairs "ok": false with "when some entries are rejected", while rl.php sets ok: false only when every entry was rejected (a partial success is 200 with ok: true plus decisions and errors). docs/api/http.md now has the corrected version; this brings the README in line with it.

Suggested change

	When some entries are rejected (invalid IDs, unregistered experiments,
	malformed entries), the response includes an `errors` array:
	```json
	{
	"ok": false,
	"errors": [
	{"kind": "decide", "id": "unknown_exp", "reason": "unknown_experiment"},
	{"kind": "turn", "id": "bad!", "reason": "invalid_id"}
	]
	}
	When some entries are rejected (invalid IDs, unregistered experiments,
	malformed entries), the response includes an `errors` array alongside
	any successful results. A partial success returns HTTP 200 with
	`ok: true`:
	```json
	{
	"ok": true,
	"decisions": {
	"hero_cta": {"armId": "v1"}
	},
	"errors": [
	{"kind": "decide", "id": "unknown_exp", "reason": "unknown_experiment"},
	{"kind": "turn", "id": "bad!", "reason": "invalid_id"}
	]
	}
	```

[jjroelofs]

Re-review at head 79be026: all five blockers verified fixed.

1. .gitlab-ci.yml: the six SKIP_* variables curate the DrupalCI matrix as suggested (PHPUnit stays gated by its tests-exist rule; the pages job is unaffected).
2. mkdocs.yml: edit_uri leading slash removed; verified the urljoin now resolves to git.drupalcode.org/project/rl/-/edit/1.x/docs/....
3. exclude_docs: rl_project_desc.html keeps the project-description snippet (and its referral link) out of the published site.
4. README: docs-site link added, Plotly section removed, ping smoke test, three-arg getThompsonScores(), decorator example now uses constructor injection; the known drift points are closed.
5. Wording: "or locks" removed in both files; docs/api/http.md's errors example corrected to ok: true with accurate 200/422 prose.

One trivial carry-over, noted inline: the README's new batch-errors section received the pre-fix example ("ok": false paired with "when some entries are rejected"); a suggestion bringing it in line with the corrected http.md text is attached. With that one-liner applied, this is good to merge from my side.

[jjroelofs]

Re-reviewed at 69946ec (two commits since the last "still not LGTM").

Fixed since the last round, threads can be resolved

• .gitlab-ci.yml: the SKIP_* variables now match the drupal-module-docs template, so only the Pages job runs.
• mkdocs.yml: edit_uri without the leading slash, rl_project_desc.html excluded.
• docs/guides/decorators.md: example now has the constructor that its arguments: line needs (both decorator threads).
• README: obsolete Plotly install section removed, route-fallback sentence corrected, batch error contract documented.
• docs/api/http.md: partial-success semantics now match rl.php (200 with ok: true, 422 only when everything is rejected).
• docs/guides/faq.md: "locks the variant" claim removed.

CI=true mkdocs build --strict passes on the head, no em-dashes anywhere in docs/, README.md or mkdocs.yml, and the docs FAQ uses ## headings as the skill requires.

Still open, all carried over from the previous review

1. docs/guides/integrations.md:196: menu hook example uses an undefined $selector and does not recurse.
2. docs/guides/integrations.md:302: rl_sorting transport claim is wrong; it beacons directly to rl.php, not via Drupal.rl.turn().
3. README.md:421: the partial-rejection example still shows "ok": false; the README and http.md now disagree.
4. docs/getting-started/configuration.md:6 (and installation.md): menu label is "RL: A/B Testing".

New
5. docs/rl_project_desc.html is stale: it equals 1.x commit 44bad48, but 1.x is already at 45b9c21. Rebase onto 1.x so the file drops out of the diff.
6. No GitHub check run exists for 69946ec; the last green run is on 79be026. Worth a re-push or manual trigger so the merge commit has a status.

The README-duplication thread from the first round is still applicable but not blocking; the README now links to the docs site, which is the minimum.

Not LGTM yet because of items 1 to 3; they are all small edits in two files.

[jjroelofs]

Still broken (carried over from the previous review). $selector is never defined in this snippet, and the loop only touches the top level of $variables['items'], so a copy of this fails with an undefined-variable error and ignores every child link. The shipped rl_menu_link_preprocess_menu() in modules/rl_menu_link/rl_menu_link.module delegates to a recursive _rl_menu_link_process_items() helper and attaches cache tags plus the rl_menu_link/tracking library. Either show that shape (get the selector with \Drupal::service('rl_menu_link.variant_selector'), recurse into $item['below'], add the cache tags) or replace the code with a prose description and a link to the real file.

[jjroelofs]

Still wrong (carried over). rl_sorting does not call Drupal.rl.turn(). Its js/rl-sorting-tracking.js uses an IntersectionObserver to collect visible entity IDs, then posts action=turns with a comma-separated arm_ids list via navigator.sendBeacon() straight to the rl.php URL that rl_sorting.module publishes in drupalSettings (rlEndpointUrl). Rewards are a separate action=reward beacon on link click, de-duplicated per page load with sessionStorage. Describe that transport, or drop the sentence; "fully server-side" is also inaccurate given the client-side tracking.

[jjroelofs]

Still open (P2 from the previous review). The path is right, but rl.links.menu.yml titles this entry RL: A/B Testing, not "Reinforcement Learning". A reader looking for the label in the admin menu will not find it. Same wording appears in installation.md.

[jjroelofs]

Example contradicts the sentence under it and the code. rl.php line 118 sets ok to $status === 200, and a partial success is 200, so this "when some entries are rejected" example must show "ok": true with a decisions object next to errors, exactly as docs/api/http.md now does after the last commit. "ok": false only ever appears with the 422 all-rejected case. The getThompsonScores(..., NULL, ['variant-a', 'variant-b']) call above matches the real signature, no issue there.

[jjroelofs]

This file is stale relative to 1.x and should leave the PR. The head version is byte-identical to 1.x commit 44bad48, but 1.x has since moved to 45b9c21, which replaced this blockquote with a bold intro paragraph and dropped the documentation note-tip box. Since the last commit's intent was "match 1.x", the clean way is to rebase onto current 1.x so docs/rl_project_desc.html disappears from the diff entirely; the docs-site links in the box only resolve after this PR publishes anyway. Line 81 ("use the Drupal Routes API instead") is a pre-existing 1.x problem now, so I would fix it in a follow-up rather than here.

[jjroelofs]

Pushed 4f93643 addressing the three blocking items and two carry-overs:

Fixed:

1. docs/guides/integrations.md:196: menu hook example now initialises $selector via \Drupal::service() and recurses through _rl_menu_link_process_items(), matching the shipped implementation
2. docs/guides/integrations.md:302: rl_sorting transport corrected: uses IntersectionObserver + sendBeacon() directly to rl.php, not Drupal.rl.turn()
3. README.md:421: batch error example now shows "ok": true with decisions alongside errors for partial success
4. docs/getting-started/configuration.md:6: menu label corrected to "RL: A/B Testing"
5. docs/rl_project_desc.html: updated to latest 1.x (45b9c21), re-added only the docs callout box as the single change, and removed the nonexistent "Drupal Routes API" fallback claim

Re: desc.html docs links: The note-tip callout links resolve once this PR merges and the GitLab CI pages job publishes the MkDocs site. The links are added in the same PR so both ship together.

[jjroelofs]

Re-reviewed at 4f93643.

Fixed, threads can be resolved

• Menu hook example is now recursive and resolves the selector service.
• rl_sorting transport is described accurately (IntersectionObserver plus sendBeacon() to rl.php).
• Configuration label matches rl.links.menu.yml.
• README partial-rejection example now says ok: true with decisions and errors.
• Project description: intro paragraph and the "Routes API" sentence now align with 1.x.

Still blocking

1. docs/rl_project_desc.html conflicts with 1.x: git merge-tree reports a content conflict in the note-tip region near line 134, and the branch is two commits behind. Needs a rebase before this can merge (inline).
2. README.md:423: the new example uses "hero_cta": "v1"; the real shape is {"armId": "v1"} (inline).

Minor
3. docs/guides/integrations.md:313: "fully server-side" now contradicts the corrected tracking description (inline).
4. No GitHub Actions run exists for 69946ec or 4f93643; the workflow is on: [pull_request] and the last run is on 79be026. A rebase push will trigger it, which also resolves item 1.

Strict MkDocs build still passes on this head. Not LGTM until the rebase lands and the README shape is fixed; both are small.

[jjroelofs]

Wrong decision shape. rl.php returns {"armId": "<arm>"} objects (line 284, and the success example twenty lines above this one shows "hero_cta": {"armId": "v1"}). This partial-success example has "hero_cta": "v1" as a bare string. Match it to docs/api/http.md line 73.

[jjroelofs]

Sentence now contradicts the paragraph. The new text correctly describes client-side IntersectionObserver plus sendBeacon() tracking, so "fully server-side" is no longer true. Suggest "The sort is server-side: ..." and keep the tracking description as is.

[jjroelofs]

Merge conflict with 1.x. git merge-tree 1.x feat/mkdocs-pages conflicts in this file around this note-tip block (1.x 45b9c21 reworked the same region). The branch is still two commits behind 1.x, so GitHub will refuse the merge. Rebase onto current 1.x and resolve; if the docs-link box should survive, re-add it on top of 1.x's version.

[jjroelofs]

Addressed all remaining items:

1. Rebased onto 1.x: desc.html now matches 1.x exactly (only the "Drupal Routes API" wording fix and the docs callout box differ). The three-way merge conflict is resolved.

2. "Fully server-side" wording fixed: Changed to "The decision is server-side ... Tracking is client-side" in integrations.md to avoid contradicting the sendBeacon description.

3. Docs callout box re-added: Lost during rebase, now restored after the "AI coding assistant integration" section.

4. README decision shape: Already fixed in the previous push (commit 3da9234, now rebased as 0d3fcb1).

The force push was necessary due to the rebase.

[jjroelofs]

LGTM at aa60486.

Verified on this head:

• Rebased onto current 1.x (zero commits behind), git merge-tree is clean, GitHub reports mergeable.
• All GitHub checks green (drupal-lint, drupal-check, drush-e2e, changes; eslint skipped by design).
• CI=true mkdocs build --strict passes; no em-dashes in docs/, README.md or mkdocs.yml.
• README partial-success example now uses {"armId": "v1"}; rl_sorting paragraph separates the server-side decision from the client-side tracking.
• The four docs/rl-experiment*.{jpg,png} files in the diff against my last review come from 1.x commits 44bad48/45b9c21, not from this branch; relative to 1.x the PR adds only the three referenced docs/images/*.jpg.
• The four docs-site links in the project description (/, /api/php/, /getting-started/installation/, /guides/integrations/) all map to pages in the mkdocs.yml nav.

Two non-blocking notes for after merge:

• Push 1.x to the do remote first and confirm project.pages.drupalcode.org/rl/ serves before deploying the project description to drupal.org, otherwise the callout links 404 for a while.
• The README/docs duplication thread from the first round stays open as a follow-up; the README now links to the site, which is the minimum.

GitHub blocks approving your own PR, so this is a comment review.