// TrulySovereign venture · autonomous change intelligence

RADAR is watching.

Three lanes. One clear view of what changed,
why it matters, and what to do next.

CHOOSE A LANENewest 10 per lane · every item in the lane archivesEach tab shows two counts: “newest” is the lane’s share of the newest 100 published items, any date; “in 7 days” is items published in the last 7 days, the same items the radar plots
Every light is a real published item from the last seven days. Its sector is the source’s coverage lane; distance from the centre is its age. Markers are all one size; colour is urgency. Click a sector to filter briefings; select a light to open its item. The published history below it lists the six newest briefings with the UTC time each was published, and every entry opens the item. The rotating arm is animation, not measurement. The published history keeps what was published and states the backlog beside it. Source coverage reports the saved health observations separately.
Source coverage · 29 sources

Configured public sources; observations come from the saved health snapshot.

LANEOBSERVED HEALTHLAST SUCCESS
AI Platforms7/7 observed healthy
Agent Tools12/12 observed healthy
Cyber Threats10/10 observed healthy

Enabled does not mean successfully collected or interpreted. Lanes with no published item are not listed. A channel that is not public is not published here at all — not its sources, items or feeds. Existing channel RSS feeds retain their original scope.

Read the radar

URGENCYAct now · Plan · Watch
SIZE All markers are the same size; RADAR does not score impact.
AGECentre = newer · rim = older

Select a marker to open the real published item. Colour and printed text both identify urgency.

Watching 33 sources, every 12 hours on schedule. GitHub starts scheduled runs late: observed 3h35m–6h26m after the slot, 2026-10-18 to 2026-10-24. Last run 2026-11-21 19:16 UTC.
Generated 2026-11-21 19:19 UTC and static until the next regeneration. Every time below is absolute — nothing here silently re-reads as fresh.
DAILY DIGEST

Only when something changed.

One email at 13:00 UTC, only on days something changed. No email when nothing did — silence is the signal, and a daily "nothing to report" is how a feed teaches you to ignore it. Confirm by email, one-click unsubscribe, no account.
Change detection is a commodity. Interpretation is not.
Every item answers three questions in fixed order — what changed, why it matters, what to do. An item that cannot answer all three is never published.
72ACTION NEEDED · 7 DAYS
102PUBLISHED IN WINDOW · 7 DAYS · INCLUDES CORRECTED
33 / 33SOURCES HEALTHY · 1 DISABLED
2026-11-21 19:16 UTCLAST FETCH RUN · NOT MODEL SUCCESS
THIS WEEK

Start with what needs action.

Everything that needs acting on, across all public channels, newest first.
SECURITY ADVISORY TRIAGE AIT-GUI route chain allows unauthenticated Python execution Upgrade ait-gui to version 2.5.1 or later. Model-reported confidence 95% (not measured accuracy)
SECURITY ADVISORY TRIAGE Budibase allows unauthenticated global-admin creation via init guard bypass Check whether you run @budibase/server in production. Model-reported confidence 85% (not measured accuracy)
AI PLATFORM DRIFT Agents SDK v0.34.0 upgrades PiHarness to Pi 1.6, breaking backward compatibility Review the Pi Durable 1.6.0 changelog and Agents SDK v0.34.0 changelog for the full list of breaking changes. Model-reported confidence 95% (not measured accuracy)
SECURITY ADVISORY TRIAGE PHP session handler RCE affects Google Cloud managed runtimes Set session.upload_progress.enabled = 0 in php.ini immediately for all PHP applications. Model-reported confidence 95% (not measured accuracy)
SECURITY ADVISORY TRIAGE GKE patches CVE-2026-99288 for Container-Optimized OS 1.32 and 1.33 Remediation: see bulletin https://docs.cloud.google.com/support/bulletins/index#gcp-2026-088. Model-reported confidence 85% (not measured accuracy)
AI PLATFORM DRIFT OpenAI Agents Python SDK v1.0.0-alpha.1 breaks Agent.run() and Session.submit() APIs Before upgrading, complete all pending native runs and approvals in your v0.24.0-alpha.8 environment. Model-reported confidence 95% (not measured accuracy)
LATEST · BRIEFINGS

Decisions, not just changes.

Channel scope and RSS
TAG

AI Platforms

Newest 7 · All 292 AI Platforms items ↗
WORTH KNOWING OpenRouter model catalogue · 2026-11-20 07:17 UTC

OpenRouter adds Mistral Pixtral Large 2502 model

OpenRouter added mistralai/pixtral-large-2502 to its catalogue with prompt pricing at $0.000002/token ($2/M) and completion pricing at $0.000006/token ($6/M).

WORTH KNOWING OpenRouter model catalogue · 2026-11-17 20:56 UTC

Mistral Small 3.2 24B pricing decreased on OpenRouter

OpenRouter reduced pricing for mistralai/mistral-small-3.2-24b-instruct: prompt tokens from $0.10 to $0.075 per 1M tokens (-25%), completion tokens from $0.30 to $0.20 p…

Agent Tools

Newest 10 · All 224 Agent Tools items ↗
WORTH KNOWING Cloudflare developer changelog · 2026-11-21 07:10 UTC

Hyperdrive supports custom CA and mTLS over Workers VPC

Hyperdrive can now connect to private databases over Workers VPC using custom certificate authority (CA) certificates and client certificate authentication (mTLS).

ACTION NEEDED Cloudflare developer changelog · 2026-11-19 20:40 UTC

Cloudflare deletes Worker versions and Pages deployments after 30 days

Cloudflare now deletes Worker versions and Pages deployments older than 30 days, except the 100 most recent versions and currently deployed versions.

ACT-BY DATE PASSED — this item's act-by date, 2026-11-19, is before this page was generated 2026-11-21 (UTC). Check the primary source before acting on it.

Cyber Threats

Newest 10 · All 594 Cyber Threats items ↗
ACTION NEEDED Google Cloud security bulletins · 2026-11-21 19:18 UTC

PHP session handler RCE affects Google Cloud managed runtimes

Google published GCP-2026-090 on 2026-11-20, disclosing CVE-2026-126714, a critical unauthenticated remote code execution vulnerability in PHP's file-based session handl…

ACTION NEEDED Google Cloud security bulletins · 2026-11-20 20:23 UTC

NGINX HTTP/2 vulnerabilities enable DoS and potential RCE

F5 disclosed CVE-2026-123633 and CVE-2026-123007 on 18 November 2026, affecting NGINX Open Source 1.19.7-1.31.8 and 1.9.5-1.31.8 respectively, and NGINX Plus R33+.

LATEST CORRECTIONS AND ARITHMETIC NOTES
52 item corrections or arithmetic notes since 2026-09-22 00:02 UTC · median time to correct 115h 38m
Original statements stay beside the corrected facts on each item. Every correction and arithmetic note (52) ↗

FETCH

Fetch — COMPLETED
33/33 sources healthy · 1 disabled.

INTERPRETATION

Interpretation — HEALTHY
Latest execution: completed · observed 2026-11-21 19:19 UTC.
Latest run start with a recorded triage response: 2026-11-21 19:16 UTC.
Untriaged backlog over 24h: 0 · oldest pending: never.
Fresh fetch and page generation do not establish interpretation success. The workflow and this page both allow 24h queue-age grace; explicit execution failures bypass it. A completed worker may have made no model call.

SOURCE HEALTH

ACCURACY LEDGER

1123 published · 52 corrected or noted (4.6%) · median 116 h to correct
Published items · all time · includes corrected1123
Corrected or noted52
Corrected or noted ÷ published4.6%
Median time to correct116 h
Precision · random sample checked 2026-10-30: 95.6% (43 of 45 checked); strict 86% (43 of 50) · checked by RADAR's build assistant, not a person; human review: 0 of 50
Window: all time. Every item RADAR has published, sample output excluded. A corrected item stays in the published count and is also counted as corrected; a small arithmetic note counts as noted. Time to correct runs from when the item was published to when the correction was stamped on it — every correction ↗.
What this is not. It counts the mistakes that were found and fixed, not the ones nobody has found — a low rate is not proof of accuracy. Corrections are drafted by RADAR's build assistant and approved by RADAR's owner; there is no independent reviewer. The only precision figure is the one random sample checked 2026-10-30: 95.6% (43 of 45 checked); strict 86% (43 of 50), checked by RADAR's build assistant and not independently double-checked by a person — how precision is counted ↗.

RUN ECONOMICS

Published items · database total1123
Action needed · rendered window100
Fetch runs · loaded window (max 200)200
Fetch runs without error · loaded denominator86.0%
Fetch runs with errors · loaded window28
Changes · loaded window (max 50)50
Model cost · loaded run window$23.7719
Last run2026-11-21 19:16 UTC
Fetch only: green = no recorded change · blue = change detected · red = recorded error. These ticks do not verify interpretation.

FEEDS