v0.8.0

Net application changes since public 0.7.3.
The documentation/landing site is a separate repository and is not shipped here.

Traffic analysis and administration

• A bearer-authenticated MCP endpoint exposes nine read-only tools:
  list_campaigns, get_campaign, traffic_summary, traffic_timeline,
  traffic_breakdown, list_clicks, visitor_journey, conversions_summary
  and pixel_summary. Report calls use database-enforced read-only transactions
  and a 15-second statement timeout. Settings → MCP provides key generation,
  regeneration/revocation and client setup snippets; keys are shown once and
  stored as hashes. IPs are masked to IPv4 /24 and IPv6 /48 by default.
  The bundled traffic-analysis skill is available at /mcp/skill.
• Date-range filters on clicks, pixel events, sessions and conversions, rolling
  period presets, a country filter on pixel events, and click KPI cards/top
  lists. Pixel charts now use the same filters as their lists. Filter URLs no
  longer carry CSRF tokens, and nested filter/action forms are fixed.
• Signed campaign and flow test links use _t; _geo overrides the test
  country and _dbg=1 returns a routing trace. Campaign tests follow normal
  routing; flow tests pin the selected flow. Pixel lander buttons retain the
  test key. Reserved route names cannot be campaign aliases.

Sessions and removal of Clickunder

Session lists and replay pages now share the admin design and light/dark theme.
Replay uses a scaled viewport without enlarging mobile recordings, localized
controls and guarded load/player errors. Browser regression tests exercise real
rrweb recording, persisted events and replay controls; a dedicated Docker
browser-test runner supplies Chromium and its supported dependencies.

Sessions now display an interaction badge and filter. Recordings without mouse
movement, button presses, clicks or touch are marked possible bots; this is
a heuristic, not a confirmed bot verdict. Scrolling or typing alone do not qualify. The default hides these recordings
and includes all durations. Unclassified recordings remain visible. Historical
recordings can be processed in bounded batches with bin/console rrweb:classify --all; missing or damaged chunks stay unclassified and replay data is preserved.

Clickunder is fully removed from the UI, routes (including /p/play), pixel,
active application code and campaign/flow schema. Overlay/redirect/random modes
and _eng support are gone; ordinary signed test links remain. Historical
migration files are retained for forward upgrades.

Runtime and reproducible builds

PHP 8.5 is now the minimum. The pinned runtime uses PHP 8.5.11,
FrankenPHP 1.12.7, Slim 4.15.3 and PostgreSQL 18.6. Dependencies move to Symfony
8.1, Pest 5, PHPUnit 13 and PHPStan 2, with compatibility updates for those APIs.
Root Bun dependencies and locks are refreshed (including rrweb 2.1.7, ECharts
6.1.0, FingerprintJS 5.2.0, Alpine 3.17.4 and Tailwind 4.3.3). Docker images
and CI actions are pinned by SHA; builds consume frozen locks.

This is not a claim that every advisory is fixed: the recorded root Bun audit
still reports a high-severity braces advisory in the build/watch dependency
chain. Hosted CI execution and runner support for the updated action runtimes
remain unverified locally.

Saved engine benchmark

Median throughput from the saved 2026-10-04 comparison:

Response	Connections	Before req/s	After req/s	Change
Redirect 302	16	1581.9	1698.6	+7.4%
Redirect 302	50	1574.4	1685.9	+7.1%
No matching flow 403	16	1589.0	1774.4	+11.7%
No matching flow 403	50	1679.8	1776.8	+5.8%

Before is archived source a367ffc on PHP 8.4.24 / FrankenPHP 1.12.6;
after is the updated application/runtime stack. This is not a direct 0.7.3
comparison. Both databases use PostgreSQL 18.6. There were 24 trials,
400100 expected responses and zero unexpected responses/transport errors.
Each cell uses three 10-second trials on a shared host. Several ranges overlap;
these synthetic engine results do not guarantee a production or universal
speedup and do not isolate the effect of any single dependency.

Methodology and limitations
and per-trial CSV
are pinned to the release tag. Full raw
artifacts are retained outside the public repository.

Upgrade and recorded validation

Take a full backup before upgrading, including the database and deployment
configuration. Follow the update procedure
and run all pending migrations before serving the new application. From 0.7.3,
these include 20260919000001_engagement_mode (historical addition),
20260921000001_auth_events_mcp_key (MCP key audit events), and
20261004000001_remove_engagement_mode (drops campaign/flow engagement columns), and
20261004000002_rrweb_session_interaction (nullable activity flag and indexes), and
20261005000001_rrweb_pointer_activity (invalidates earlier, broader activity labels).
After upgrading, run bin/console rrweb:classify --all to rebuild labels from preserved recordings.
Do not skip the historical addition. Rolling back the removal recreates disabled
defaults; it cannot recover removed settings. Older code needs those columns
restored before it can serve traffic; recovering their values requires the backup.

Prior application validation, recorded before this release preparation: 646
Unit/Integration/Arch tests, 12 Chromium browser tests, and PHPStan level 6 clean.
The removal and session-interaction migrations were applied, rolled back and reapplied on an isolated
restored database copy. These are recorded results, not application tests rerun
for this documentation-only preparation.