Repository navigation
Releases
Looking Glass v2026.11.19.2 (2026-11-19 (2))
Added
- The severity map, on the Definitions page (researchers and administrators only). Every infection and issue the
reports carried, with its criticality and where it comes from: rated by Shadowserver (the highest severity
Shadowserver gave the records that carry it), not rated (no record rates it, listed first) or rated by a researcher.
A researcher can give a not-rated name a level, or take it back; Shadowserver's own ratings are never changed, and
when Shadowserver later rates a name, its level replaces the researcher's (shown as replaced). Every rating is in the
audit log (severity_rated). The PDFs' References section uses a researcher's rating where no record rates a name,
flagged "Researcher's rating". CVEs are rated by their CVSS score and are not on the map. The names are kept in
data/name_kinds.jsonand the ratings indata/name_ratings.json(names only: no host, no address, no count). - After updating: about two minutes after the service starts, Looking Glass reads every report's records once more
in the background (infection, issues, tags and severity only, in batches with pauses), so the map also lists the
names no record rates from before the update. Nothing to do;NAME_LEVELS_DELAY_S=0turns it off.
Released 2026-11-19. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.19 (2026-11-19)
Changed
- Documentation: for CSIRTs in every country. The documentation site's home page, its title and description, its
link-preview card and the guided installer's banner now say "built in Africa, for CSIRTs in every country", as the
README does. - Documentation: memory and disk measured again on this version. The installation guide adds a measurement of the
current version (test installation, made-up data, about 67,000 records a day: about 2.3 GB for the platform and
MongoDB together, within the 4 GB minimum). The README's requirements now give the same 4 GB minimum (it said
about 2 GB), and the download and disk sizes are updated (about 70 MB to download, about 45 MB with--depth 1;
the code itself about 6 MB). Nothing to do when updating.
Released 2026-11-19. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18.8 (2026-11-18 (8))
Changed
- Every PDF lists its references at the end, in a References section ordered by criticality. The definition cards
of the analysis, combination, Executive and advisory PDFs keep their explanation; their links now sit at the end,
under one heading per name, then the Shadowserver reports the PDF draws on (the advisory's "Shadowserver Report
References" page became this section). Each heading carries its criticality: a CVE its CVSS severity and score, and
an actively exploited one (CISA KEV) also "Actively exploited", first within its level; an infection or an issue
the highest severity Shadowserver gave the records that carry it; a Shadowserver report the severity Shadowserver
gives that report type (INFO, Shadowserver's lowest level, as it says). A name with no severity at all is "not
rated", listed last. The Executive PDF gains the section. A name whose days have left memory keeps the level it
had: Looking Glass remembers each name's highest level indata/name_levels.json(names and levels only); after
the update it learns them from the days it loads, so a name seen only before the update reads "not rated" until it
appears again.
Added
- The advisory's preview names who opened it and when. Every page of the preview carries "PREVIEW · NOT FOR
DISTRIBUTION · name · date and time · reference" at the top and the foot, and a faint watermark behind the content
(a picture, so it cannot be selected or removed as text). A screenshot of the preview therefore shows who took it.
The report the advisor generates for the organisation carries none of it. Each preview was already in the data-access
log.
Nothing to do when updating; PDFs made after the update have the new layout.
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18.7 (2026-11-18 (7))
Added
- A check of Abused or Compromised against MongoDB (
platform/tests/verify_intelligence.py), like the checks the
other pages have: every report's rows per day, its 7-day median, the group totals, "never received", "still
arriving", each report's view and its CSV export, and management's counts, all recounted straight from the
database. Read-only, counts only. It runs in the full suite on a running installation and skips itself where
there is none. Its CSV comparison is recorded in the audit log as an export by the automation account. - The suite's line for the page's other check now names it as people see it: "Abused or Compromised".
- An outside-connection guard for the checks (
platform/tests/no_outside/). While the checks run with it, only
the computer itself and the database can be reached: every other connection or name look-up is refused and
listed, and the run fails on any place Looking Glass has no reason to reach. GitHub's checks run with it and a
made-up map key, so a check that should send nothing is seen if it tries (an advisory had sent place names to the
map geocoder, caught on a test installation only, 2026-11-18). GitHub's database job now also runs the advisory
ATT&CK checks, where there is report data to load. On an installation:LG_NO_OUTSIDE=1 tests/run_all_checks.sh
(see CONTRIBUTING).
Nothing to do when updating.
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18.6 (2026-11-18 (6))
Fixed
- An advisory's Affected lists every infection, issue and CVE its devices show, from every report. The devices
table and the devices CSV named what each device showed in its other reports that day (for example "smb-scan;
Compromised" from Compromised IoT), but Affected did not, when the finding's review was kept before 2026-11-17 (no
device details were kept then). Affected and the table now read the same thing, worked out the same way. - The definitions in an advisory count the devices that have each issue, for example "1 of the 148 devices in this
advisory"; an issue could read as if every device of the combination had it.
Nothing to do when updating; generate an advisory again to get the full Affected list and the counts.
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18.5 (2026-11-18 (5))
Changed
- The Findings page no longer waits for analyst codes' device counts after a restart. Measured on a test
installation, right after a restart, four codes' counts took 30 to 45 s together. Each one built the code's whole page,
and that loaded the code's older day from the database twice, then again for every count (one large day loaded
three times in 13 s). Three changes:- an analyst code's page loads its day once and uses it for its records too;
- an older day loaded for an analyst code is kept a few minutes, so its page and its counts share it (Older days
kept for analyst codes, default 1 day, and How long such a day is kept, default 300 s, on Settings →
Configuration; 0 days keeps none, as before). A kept day costs about as much memory as a day in memory and is
dropped as soon as the constituency or the stored data change. A code spanning many days is still read one day at
a time, never two in memory together; - 30 s after a start, the device counts of every open finding's analyst code are worked out in the background, for
people who see addresses and for those who do not, each kept apart (Analyst code counts worked out after a
start, 0 = off).
Nothing to do when updating; the defaults apply.
- ATT&CK techniques in PDFs say only "Observed" or "Possible" again, for example "T1071.001 Web Protocols · Command
and Control · Observed". The line had grown to "Observed: this device was used for it". The analysis and Executive
PDFs use the same form. The device's part is still in the cards and in the advisory's "Why it is in this advisory".
In the advisory PDF, Observed and Possible have the cards' look (amber and teal, a faded outline); in the analysis
PDFs the word is in the same colour. - "Why it is in this advisory" counts IP addresses for Observed techniques too, and says what they are linked to:
"3 IP addresses in this advisory have these affected devices (most often: …)". A technique linked to a botnet
infection names the infections, one linked to an issue names the issues, one linked to a CVE names the CVE. Before,
Observed techniques from a sinkhole had no figures, and Possible ones named the kinds of device.
Nothing to do when updating; advisories generated again get the new wording.
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18.4 (2026-11-18 (4))
Added
- Two new checks of every page as people see it, run before every release.
platform/tests/pdf_pages_check.pyrenders PDFs and finds text drawn over other text, the "…" cut-off and text
off the page. It also runs in GitHub's checks.platform/tests/browser_checks.pyopens every page for every role in a browser and checks that every control is
reached with the keyboard, focus is never trapped, lists are lists, no text is cut off and no page errors.- Both skip themselves where their tools are not installed (pypdfium2; Playwright with a test installation and its
test accounts), so the suite runs anywhere.
Changed
- The advisory's own title is now the large title on its PDF. The advisory page has one Title field (it
starts as the finding's title); it is printed large at the top of the first page, with "CSIRT Security
Advisory" in small letters above it. Before, the large title was always "CSIRT Security Advisory" and the
advisor's words were the smaller line below. An advisory that already had a typed heading keeps it as the small
line. A long title is no longer cut after three lines. Generate an advisory again to get the new layout. - The report page with the map is named "Geographic Location" (it was "Where"), in the analysts' reports and in
the advisory's insights. - The table of addresses says what it lists: "The IP addresses with the most records" on a page named "IP
addresses" (it said "The devices"), and page 1's preview is "The busiest IP addresses". - An advisory with more than 10 affected IP addresses points to its devices CSV instead of listing them. The
Affected IP addresses section gives their number and the name of the attached CSV; with 10 or fewer the table
stays. The devices CSV now also has the host_name and ports columns (after ip), so nothing the table showed
is lost; the analysts' devices CSV has them too. - Generate report always gives the advisory and its devices CSV together, never a PDF alone. A finding that keeps
no devices (made before devices were kept with findings) is refused with what to do: the analyst sends the
combination to management again, and the new finding keeps its devices. - No page of an advisory prints the platform and scope line ("Publicly accessible services · Ghana · whole
country"). It was already gone from the advisory's own pages; the insights, Technical Details and references pages
still had it. The analyst's own report keeps it. - An advisory's Technical Details pages are titled "Definitions": the badge beside the logo already says
TECHNICAL DETAILS, so the title no longer repeats it.
Fixed
-
An advisory's Affected section lists every infection its devices carry. Its Categories showed only the infections
from the devices' own sinkhole records, while the device table also showed those the search found on them that day
(for example android.vo1d2 listed, avalanche-andromeda and m0yv only in the table). Categories, CVEs and Issues now
list everything the device rows show. Generate an advisory again to get the corrected PDF. -
The device counts beside analyst codes on the Findings page no longer wait for each other. Every code was
counted under one shared lock, so one code that takes long to count (about 16 s the first time, on a large
selection) held every other code's count, and their loading placeholders could spin for over 30 s. Each code now
has its own lock: a slow count only delays itself, and the same code is still counted once at a time. -
Accessibility (found by an automated check of every page for every role, WCAG 2.2 AA):
- The top lists on the map and services pages, and the attack-type legend on the map page, made each list item a
button, so screen readers met lists without list items. Each entry is now a real button inside its list item,
at least 24 px tall (the touch-target minimum). They work the same with a mouse, Enter or Space. - The Executive portal's map was marked as a picture while holding a control (the whole-country switch); it is
now a group.
- The top lists on the map and services pages, and the attack-type legend on the map page, made each list item a
Found by rendering every page of every kind of PDF the platform makes (110 pages) and checking each one:
- Executive report, "Explained" pages: a card's text stopped mid-sentence (the description after six lines, the
researchers' notes and CISA's required action after three). Every line is now kept and the card grows to hold it. - Executive report, top infections and top issues: a long name or a CVE's description was squeezed onto extra
lines over the bar and the next row. The row now grows with its text, and the description goes under the name when
it does not fit beside it. - Executive report, the bars beside them: a long label ("Accessible GPRS Tunneling Protocol (GTP)") ran over
the next row; it now wraps and its row grows. - Maps: a region's name at the edge of the map ran into the list beside it; it now stays on the map.
- Analysis report summary: the list of actively exploited CVEs stopped after eight with "…"; it now names the
rest as a count ("and 21 more, each explained under Issues explained"). - Analysis report, the reports in the selection: a report's name could end a page with its text on the next.
- Audit evidence: the method text showed "backed" as written; it now reads "backed". The seal's SHA-256
is printed in full instead of shortened with "…".
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18.3 (2026-11-18 (3))
Fixed
- The advisory's table of affected IP addresses no longer leaves its heading alone on a page. When the first row
was too tall to fit under the report chips, it moved to the next page and left the section heading and the column
names over half an empty page. The heading, the column names and the first row now always start on the same page. - Long names in the advisory and the analysts' report break at their hyphens. A name too long for its column
broke mid-word ("avalanche-androme" / "da"); it now breaks after a hyphen, dot, colon or slash ("avalanche-" /
"andromeda"). In the analysts' report such a name used to run over the next column. - The advisory's Other days column is wider (28 % of the table instead of 15 %), taken from the Reports and CVEs
columns, so its rows are about half as tall and the table needs fewer pages. Generate an advisory again to get the
corrected PDF.
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18.2 (2026-11-18 (2))
Added
- Findings pages keep up on their own. A finding sent to you, decided or replied to shows on your open Findings
page within about 15 seconds, no reload needed, and the cards that changed flash. While you type a comment or choose
a person, a notice offers Show the changes instead of redrawing under your hands. The advisor's portal updates the
same way; the review and advisory pages show the notice. This works for every role. The page only asks for a short
fingerprint of the findings in your channels (no title, name or person), and only while the page is in front. Nothing
to do when updating. - The review page shows the researcher and the analyst side by side: What the researcher is saying (their
finding and notes on the combination) and What the analyst is saying (the final comments and the analyst's
replies).
Changed
- The Findings page groups a case as one: case → finding → sub-findings. A researcher's finding and an analyst's
request to management on the same combination used to be two separate cards, linked only by the code. The analyst's
requests now sit inside the researcher's finding, which is marked Case: the researcher's finding, and the
organisations' sub-findings stay inside each request. States, channels and decisions are unchanged, and the Open tab
brings the whole case along while any part of it is open. Existing findings are grouped the same way. Nothing to do
when updating. - An administrator's steps name the role they played. Notes, findings and the review page said only
"Administrator" when an administrator did a researcher's, analyst's or manager's work. They now say Administrator
acting as researcher, acting as analyst, acting as management or acting as advisor. The role is worked out
from the step itself (a note is research; a request from the analyst's tools is analysis; a decision is
management's), so findings made before this update are labelled too.
Fixed
- Long text in the analysts' PDF tables no longer runs over the rows below. A cell such as Other days stopped
at three lines and squeezed the rest of its text into the third, which spilled over the next rows. Every line is now
kept at the table's size and the row grows to hold it; the page breaks follow. Generate a report again to get the
corrected PDF.
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.
Looking Glass v2026.11.18 (2026-11-18)
Added
- Settings → Configuration: every setting in plain words, without a terminal. A new tab lists every setting
Looking Glass runs with: what it does, the value in use, where it comes from (.env, this page or the default) and
its default, with a search box. The tuning settings (replay timing, loading and counting, memory and speed limits
such as the decoded beacon answers kept, findings' deadlines, spike and pipeline alerts) can be changed within a safe
range and are kept indata/config.json; they take precedence over.envand apply after Restart now (about a
minute while the days are loaded again). The security switches, sign-in locks, the audit log's settings and the
database connection are shown only, with the exact.envline to change on the server, so a stolen administrator
session cannot weaken them; passwords and tokens are never shown, only whether one is set. Saving and restarting
need a recent sign-in and are written to the audit log. Nothing to do when updating.
Fixed
SETTLE_MAX_SECONDS, andDAY_MAX_AGE_SECONDSfor the services platform, set in.envare now used. They were
read before.envwas loaded, so their defaults always applied..envis now read as the service starts, before
anything else. If you set either one, check its value still suits you.
Released 2026-11-18. Updating an installation: follow Updating. Every change by date: CHANGELOG.md.