CTIPilot
AI-generated · no human review · verify critical claims against the linked source. how it works →

2026-11-22T1407Z-audit

One pipeline fire, in full · audit run of 2026-11-22 · sub-agent allocation and telemetry, per-iteration verification verdicts and findings, source-list edits, coverage gaps, bridge invocations, and the run's own verification & coverage notes: what was published, what was dropped at the borderline or judged not relevant (and why), single-source carve-outs, and contradictions. Rendered from runs/2026-11-22/2026-11-22T1407Z-audit.md.

Run telemetry

2026-11-22T1407Z-audit audit prompt v4.25 publish ok
3h 01m duration 2 published 25 updates
Claude Opus 5.5 (claude-opus-5-5) main agent
G1 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
15
Duration
29m 16s
Tool calls
10 WebFetch8 WebSearch250 bridge
Cited sources
0 of 46 in slice
G2 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
11
Duration
25m 57s
Tool calls
4 WebFetch28 WebSearch150 bridge
Cited sources
0 of 43 in slice
G3 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
18
Duration
21m 09s
Tool calls
4 WebFetch5 WebSearch120 bridge
Cited sources
0 of 48 in slice
truth-TA Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
0
Duration
21m 19s
Tool calls
not reported
Cited sources
none
truth-TB Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
0
Duration
27m 16s
Tool calls
not reported
Cited sources
none
truth-TC Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
0
Duration
17m 19s
Tool calls
not reported
Cited sources
none
truth-TD Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
0
Duration
19m 30s
Tool calls
not reported
Cited sources
none
truth-TE Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
0
Duration
17m 20s
Tool calls
not reported
Cited sources
none
truth-TF Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
0
Duration
18m 19s
Tool calls
not reported
Cited sources
none

Verification

#? NEEDS_FIXES · Sonnet 5.5 · t=13 e=2 a=4 #? NEEDS_FIXES · Sonnet 5.5 · t=4 e=3 a=4 #? NEEDS_FIXES · Sonnet 5.5 · t=3 e=3 a=2 #? NEEDS_FIXES · Sonnet 5.5 · t=2 e=1 a=0 #? NEEDS_FIXES · Sonnet 5.5 · t=1 e=1 a=2

Deep dive

·

Entries this run published (2) and updated (25)

Sources changed (this run)

Edits this run made to sources/sources.json · promotions, demotions, new candidates, and fetch-method / category / reliability / url corrections (the run record's sources_changed[]). Paginated; 10 per page.

No source-list edits recorded for this run.

Coverage gaps (this run)

Sources this run's brief needed that returned no usable content via any documented recipe. Bridge-recovered or quiet-day sources do NOT appear here. (Distinct from the independent source-accessibility probe at the foot of this section, which probes all active sources regardless of what any run needed.)

No coverage gaps in this run · every source the brief needed returned usable content via its documented recipe.

Verification findings · all iterations

Per-iteration finding detail. Each table is one verifier pass · what was flagged, how the main agent remediated it, and the outcome. Walking the tables top-to-bottom shows the verifier's debugging trail across iterations.

Iteration #? NEEDS_FIXES · 19 findings (truth=13, editorial=2, advisory=4) · Claude Sonnet 5.5 · 26m 26s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F4
hallucinated-fact
·
GHSA-ffr6-rwpm-p5wq (the per-CVE authority for CVE-2026-115731) lists two packages: OpenVox Agent <= 9.2.0 and <= 8.31.1, patched 9.3.0 and 8.32.0, and OpenVox Server <= 9.1.0 and <= 8.17.0, patched 9Applied: OpenVox Server restored beside OpenVox Agent in cves[] (GHSA-ffr6 lists both), section and record summary rewritten
F4
hallucinated-fact
·
Rewritten this fire from 'does not state exploitation'. Neither the Broadcom advisory (…/SecurityAdvisories/0/39495) nor the README/Q&A says exploitation is absent; the entry's own summary ('Broadcom'Applied: restored 'does not state exploitation'
F4
hallucinated-fact
·
Mitel MISA-2026-0010 rev 6.0: 'OpenScape SBC | Affected | V11R3.4.4 and earlier | Update to V11R3.4.5. Workaround available'. This fire's section and action say the SBC fix is released, yet the takeawApplied: takeaway now updates OpenScape SBC to V11R3.4.5 and keeps the SESAP and Contact Center workarounds
F4
hallucinated-fact
·
The cited Mitel page now reads 'Update to V11R3.4.5. Workaround available, see KB article' (rev 6.0), so the quote is no longer a contiguous passage (it was the standing quote-literal WARN in page-cheApplied: evidence quote replaced with the revision-6.0 passage 'Update to V11R3.4.5. Workaround available, see KB article'
F4
hallucinated-fact
·
This fire's section and sources[] cite Spiegel, which quotes the operator's BundID site notice ('Sie erhalten nun bei jeder Anmeldung mit 'Benutzername & Passwort' zusätzlich einen Sicherheitscode perApplied: sourcing_note rewritten (notice known through Spiegel's quotation); sourcing_note added to the record's fields
F3
claim-not-supported
·
The clause added this fire (return behind an e-mail code) is attributed to heise and Handelsblatt and carries no citation. heise: 'der Zugriff ... über die Anmeldung Benutzername und Passwort deaktiviApplied: the e-mail-code clause now cites Spiegel; heise and Handelsblatt keep only the switch-off
F4
hallucinated-fact
·
(low confidence) Synology-SA-26:21 contains no statement about exploitation either way; the entry's summary correctly says 'No source reports exploitation'. Reword to 'The advisory does not mention exApplied: 'The advisory does not mention exploitation.'
F14
?
·
(low confidence) Huntress tested one fully patched Windows Server 2025 DC (nine machine accounts) and itself says that DC's behaviour 'may not be representative'; it also notes Samba AD DCs do not impApplied: Exposure scoped to Windows domain controllers, naming the single tested controller and Samba's lack of the extended form
F3
claim-not-supported
·
(low confidence) The product table has rows for DSM 7.1, 7.2.1 and 7.2.2 (OnGoing) but no DSM 6.2 row, so 6.2.4-25556-9 is not marked OnGoing by the table.Applied: sourcing_note separates the 6.2.4 build (no table row) from the OnGoing rows
F3
claim-not-supported
·
(low confidence) The cited page's dateline is Published Time 2026-11-13T22:59:21+00:00 and the agent-impersonation passage sits under Publication 2 (November 12), not under the Update: November 16 bloApplied: N-able source record and citation dated 2026-11-13 (page dateline)
F3
claim-not-supported
·
(low confidence) VMSA-2026-0009's matrix lists VCF and VVF for the 9.1 and 9.0 rows, but the 5.x row names only 'VMware Cloud Foundation | SDDC Manager | 5.x' (fixed 5.2.5); adding vSphere Foundation Applied: vSphere Foundation scoped to 9.1 and 9.0 in the CVE-2026-123893 record
F4
hallucinated-fact
·
(low confidence) This fire's record states Calif is 'named as one of two credited reporters rather than the discoverer' (Oracle credits Albert Yang and Quang Luong of Calif.io) and reworded the title,Applied: five source and evidence labels read 'Calif'; record summary reworded
F3
claim-not-supported
·
(low confidence) Report URI: '1688360.cc ... was put on `clientHold` at 17 Nov 00:37 UTC'; clientHold is a registrar-side EPP status, a registry hold is serverHold. Say registrar hold or clientHold.Applied: 'a registrar hold (clientHold)'
F8
needs-more-research
·
(low confidence) The correction establishes that three unauthenticated 9.8 rows (Communications Diameter Signaling Router, Communications Unified Assurance, PeopleSoft PeopleTools) are the KEV-listed Applied: Oracle takeaway now leads with the three products whose Brotli rows are exploited
F10
missed-angle
·
(low confidence) G1 returned Metabase (GHSA-69v7-pwxv-xxr3, 2026-11-19: all self-hosted x86-64 systems potentially impacted, fixed Docker images 58.42/64.8/65.6 or -XX:-UseBASE64Intrinsics) and Adobe Declined: Metabase and Adobe Animate are derivative vendor-impact rows; the report now lists them as correctly droppable
F11
editorial-advisory
·
Literally true (Oracle's matrices carry no flag) but now reads against the corrected headline and summary; title is not in the record's fields. Reword to say Oracle flags none while CISA lists the thrApplied: title reads 'Oracle flags no row as exploited, but CISA lists the Brotli flaw behind three of them'; title added to the record's fields
F11
editorial-advisory
·
Pipeline vocabulary (field names, length bounds, record-keeping narration) in changelog record summaries that readers see; say what changed for the reader or mark the record internal.Applied: eleven record summaries rewritten as plain reader statements
F11
editorial-advisory
·
The new entries are stamped after the run's completed time, and the 15:15 note post-dates this pass's start (15:11:23Z) when the note was already written; align completed with the final write or correPending Phase 6: completed and duration are re-stamped after the loop
F11
editorial-advisory
·
On disk the 34 new entries carry 8 high (Drupal now included, the same base as the 15 of 49 in the preceding line); the store has 331 iterations with both claim counters (14 mismatches, none CLEAN); tApplied: report figures corrected (8 of 34 new high, 14 of 331 iterations, G1 droppable count and items)

Iteration #? NEEDS_FIXES · 11 findings (truth=4, editorial=3, advisory=4) · Claude Sonnet 5.5 · 23m 18s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F4
hallucinated-fact
·
False provenance statement. GHSA-q4hm-33rc-572g, one of the two pages first cited for the OpenVoxDB clause, lists package 'OpenVoxDB (OpenVox/Puppet)' with affected '<= 9.1.0 and <= 8.17.0' and patcheApplied: false clause removed from the record summary and the report row; the record is retyped improvement (the agent omission was imprecision, not a wrong statement)
F4
hallucinated-fact
·
(low confidence) The headline says the Oracle Brotli rows are exploited; the sources support only that CISA lists the Brotli CVE (Google Brotli, KEV dateAdded 2026-10-27). Oracle flags no row and the Applied: headline reads 'Brotli rows on KEV'; takeaway and Exposure say KEV-listed
F3
claim-not-supported
·
(low confidence) Residual of iteration 1 #4. VMSA-2026-0009 matrix: CVE-2026-123893 rows for 9.1.x.x and 9.0.x.x name 'SDDC Manager / VCF Installer' (VCF, VVF), but the 5.x row is 'VMware Cloud FoundaApplied: CVE-2026-123893 affected now separates VCF and VVF 9.1 and 9.0 (SDDC Manager and VCF Installer) from VCF 5.x (SDDC Manager)
F3
claim-not-supported
·
(low confidence) The CISA alert lists only 'CVE-2026-90436: Apache Axis, Unsafe Reflection Vulnerability' and names neither N-central nor 4EverAble2Shell, so the clause 'the entry point of 4EverAble2SApplied: appositive removed; the CISA clause names only the Axis CVE
F5
missing-citation
·
(low confidence) Fact with no inline citation; Synology-SA-26:21 does not say where the version is read. Cite a Synology documentation page or drop the clause.Applied: uncited version-location clause dropped
F8
needs-more-research
·
(low confidence) Incomplete remediation of iteration 1 #14: only the takeaway was changed. The Exposure line still lists Access Manager, WebLogic, E-Business Suite, APEX, Platform Security for Java anApplied: Exposure now leads with the three Brotli-row products, and the action patches them first
F10
missed-angle
·
In-window publisher revision missed on an entry this fire edited. SSA-045512 is now V1.3 (Last Update 2026-11-18): 'V1.3 (2026-11-18): Added SENTRON Powermanager V8 and V9, SICAM PAS CC, SICAM SCC, anApplied: SSA-045512 V1.3 (SENTRON Powermanager V8 and V9, SICAM PAS CC, SICAM SCC, IPAS Tester) added to the section, the main analysis, cves[] affected, affected_products and Exposure; Cisco v1.14 is a generic…
F11
editorial-advisory
·
Priority and actions[] bookkeeping in reader-visible record summaries; say what the delivered change means for the reader (for Drupal, exploitation reported; for Brotli, the SBC fix is released) and lApplied: Drupal and Brotli record summaries reworded without bookkeeping
F11
editorial-advisory
·
This fire added the CISA sentence to the main analysis and again as the opening sentence of the Improvement section and record summary; the reader sees the same fact twice. Keep it once (the section sDeclined: the main analysis states the current state and the section carries the dated delta; both are required
F11
editorial-advisory
·
On disk 332 iterations carry both counters once this run's own iteration 1 (202 of 202) is included; 14 mismatches, none CLEAN, holds. Say 'before this run' or update the denominator.Applied: '14 of the 331 earlier iterations'
F11
editorial-advisory
·
No news site is cited or named, so the sentence cannot be checked by a reader; name and cite it or drop it. 'Recorded here' is record-keeping phrasing; say 'the table is what this entry follows'.Applied: news-site clause and 'recorded here' removed from the sourcing note

Iteration #? NEEDS_FIXES · 8 findings (truth=3, editorial=3, advisory=2) · Claude Sonnet 5.5 · 16m 47s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
GTIG (https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-vsphere/) says the Directory Service data.mdb is acquired by BONESAW through the vmdird RPC interface: 'Using the Applied: Detection and section now say the directory database was pulled over the directory service's RPC interface and only the key and credential files were read over SSH; report row corrected
F3
claim-not-supported
·
(low confidence) The only citation terminates a sentence that chains Mitel's SBC V11R3.4.5 and the SESAP and Contact Center workarounds with Cisco's Duo Network Gateway 3.4.2 and Guestshell 4.9. The CApplied: the Mitel clause of the takeaway carries its own Mitel citation; the Cisco citation covers only the Cisco items
F4
hallucinated-fact
·
(low confidence) The stated window ends 2026-11-13, but this fire put content of two 2026-11-18 revisions into the same summary: 'OpenScape SBC fixed in V11R3.4.5' (Mitel MISA-2026-0010 'Last Updated:Applied: summary range ends 2026-11-18
F8
needs-more-research
·
(low confidence) Same shape as iteration 2's Oracle F8. Exposure now names SENTRON Powermanager, SICAM PAS CC and SICAM SCC, but the takeaway, actions[0] and cves[].fixed still give the Siemens remedyApplied in part: the SENTRON .NET remedy reached the takeaway and actions[0], the SICAM no-fix state cves[] and the dated section but not the takeaway; completed after iteration 4 (SICAM instruction and Integrated PA…
F8
needs-more-research
·
(low confidence) The correction makes Communications Diameter Signaling Router, Communications Unified Assurance and PeopleSoft PeopleTools the first products to patch (Oracle matrix: CVE-2026-111497,Applied in part: the three Brotli-row products added to affected_products; no cves[] record added, since CVE-2026-111497 has its own entry, which this entry references
F10
missed-angle
·
(low confidence) The Brotli entry is the home of the KEV-listed CVE-2026-111497 and was edited this fire for the Mitel and Siemens revisions, but it names no Oracle product and does not reference the Applied: references[] links the Oracle CSPU entry
F11
editorial-advisory
·
The news-site clause is gone but 'is the state recorded here' is still in the sourcing_note, so the run record's remediation line overstates what was done. Either reword ('the table is what this entryApplied: 'is the later statement'
F11
editorial-advisory
·
(low confidence) This fire widened the base wording ('says nothing about exploitation of these rows') to an unqualified statement. The Oracle page has a general paragraph: 'Oracle continues to periodiApplied: sourcing note now notes the generic paragraph on attacks against already-patched flaws

Iteration #? NEEDS_FIXES · 3 findings (truth=2, editorial=1, advisory=0) · Claude Sonnet 5.5 · 17m 15s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F4
hallucinated-fact
·
(low confidence) Two clauses in the sourcing_note rest on no cited page. https://www.synology.com/en-global/support/security/Synology_SA_26_21 says nothing about Synology being the CNA (CVE record forApplied: 'the CNA and' dropped; the closing clause says the affected and fixed versions follow the product table
F4
hallucinated-fact
·
(low confidence) On disk the Brotli Defender takeaway and actions[0] carry the SENTRON .NET remedy but neither names SICAM PAS CC or SICAM SCC (no 'SICAM' in the takeaway or any actions[] line); the SApplied: the iteration-3 Brotli line now says what was and was not done
F8
needs-more-research
·
(low confidence) SSA-045512 V1.3 (https://cert-portal.siemens.com/productcert/html/ssa-045512.html) also lists Integrated PA Suite Tester, All versions, 'Install the latest .NET security updates via WApplied: takeaway adds the SICAM PAS CC and SICAM SCC instruction to disable Brotli where not required and Integrated PA Suite Tester with the .NET remedy; actions[0], cves[] and affected_products carry Integrated PA…

Iteration #? NEEDS_FIXES cap-breach · 4 findings (truth=1, editorial=1, advisory=2) · Claude Sonnet 5.5 · 17m 49s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) The Exposure sentence 'FortiMail 8.0.0 to 8.0.1, 7.6.0 to 7.6.6, 7.4.0 to 7.4.8 or any 7.2; FortiAuthenticator 8.0.0 to 8.0.3 and 6.6.0 to 6.6.10; FortiOS below the fixed builds and aApplied after the final iteration (early exit): the FortiClient clause is its own sentence carrying the FG-IR-26-194 citation
F8
needs-more-research
·
(low confidence) Residual of the iteration-4 F8 remediation. SSA-045512 V1.3 (https://cert-portal.siemens.com/productcert/html/ssa-045512.html, read this iteration from the raw HTML table) lists 'InteApplied after the final iteration (early exit): Integrated PA Suite Tester added to the Exposure host list and the analysis sentence
F11
editorial-advisory
·
(advisory, pre-existing label) The Cisco page read this iteration (and the audit's own cached copy, quote-bodies/8882cce481569b43) shows 'Version 1.14 ... Interim 2026-NOV-17' at the top of its revisiApplied after the final iteration (early exit): Cisco paragraph reads version 1.14 of 2026-11-17; the verifier found every Cisco fact unchanged at 1.14
F11
editorial-advisory
·
(advisory, wording) On disk the SICAM no-fix state also stands in cves[].affected ('SICAM PAS CC from V6.02 and SICAM SCC from V7.01 (no fix planned)') and in the 2026-11-22T14:49:25Z section, so 'onlApplied: iteration-3 Brotli line reworded

Verification & coverage notes

The run record's narrative body, verbatim. This is where the run accounts for its own judgement calls: every borderline drop and judged-not-relevant item with its reason, dedup decisions, single-source items and their carve-outs, contradictions, and per-source coverage gaps, so nothing the run considered disappears silently.

Verification & coverage notesrun record body

2026-11-22T1407Z-audit · audit · Opus 5.5 · window 170 h · 2 entries published

Verification & coverage notes

Scheduled weekly quality audit over 2026-11-15T14:07Z to 2026-11-22T14:07Z (gap 168 h to the 2026-11-15 audit; window 170 h). Report: docs/audits/2026-11-22-quality-audit.md. The claim was on main before the first spawn; the open partial claim of cleanup wave 05 (2026-10-21T1147Z-audit, wave=05, older than 48 h) was ignored under Phase 0 step 2. Monthly calibration: not due (the 2026-11-01 report owns November). ATT&CK pin: up to date, local v20.1 equals upstream v20.1.

  • Truth passes: 49 entries in scope (34 new, 15 updated), 1,426 claims, all answered: 24 clean, 22 imprecision, 3 factual-error (Drupal SA-CORE-2026-017 exploitation unrecorded, Oracle CSPU "none exploited", ScreenConnect August link). Out of scope under v4.22: 22 entries carrying only 2026-11-15 audit records plus its recovered ARTEX-in-AWS entry.
  • Records: 25 entries carry one record each from this fire (7 update, 9 correction, 9 improvement; 5 internal), including the out-of-scope Brotli entry, whose Mitel advisory moved to revision 6.0 (a pre-pass lead).
  • Recovered entries (2): Synology-SA-26:21 (CVE-2026-120382) and Huntress's timeroasting detection research. The Unit 42 Com-vishing report is correctly droppable; the 2026-11-21 fire's call stands.
  • Declined findings: Cisco Secure Email Gateway critical kept and its Metasploit record left typed as improvement (append-only); Puppet kept at notable; Java4Shell record-summary count (append-only); Turla DLL names (hijacked legitimate names, not indicators); EUVD API citations; HPE sourcing-note timestamp. Reasons in the report.
  • Contradiction: the LiteLLM chain entry records GTIG's description of the exploited CVE-2026-40217 as a key-regeneration rebind against the vendor's sandbox-escape description; no source reconciles them.
  • Pre-pass leads decided by the main agent: the two Publica NZZ quotes are unverifiable on a paywalled teaser (watch item); the Mitel lead was a revision (recorded); php2shell (Trend status table revised 2026-11-18) and agent-framework (SafeDep removed the sentence on 2026-11-18) were decided by TB and TD as page revisions.
  • Calibration alarm: in-scope high is 15 of 49 after this audit (Drupal added); all 14 earlier high entries were re-graded against the Phase 4 bar and clear it.
  • KEV: nine additions in the window, all covered (kev-window.txt); no COVERED-STALE or RANSOMWARE row. Backlog: no open row.
  • Rotational lookback: one sweep defect (Synology listing, fixed by the 2026-11-22 fire's recipe), one rotation lag (Huntress, unswept 2026-11-15 to 11-22).
  • Fix effectiveness (2026-11-15): v4.25's slice-only attempt rule held (25 outside-slice checks, none counted); the entry-shape uncited-source check passes over the window.
  • Warning sweep: Phase 0 check_run.py --all 1 warn · 0 fail · 29 acknowledged; one acknowledgment added (2026-11-19T0355Z-intel single CLEAN at the cap), none removed; site/build.py emits no self-check warning.
  • Source health: the one UNSOLVED source, checkpoint-support, reads relevant on its existing recipe (transient block); safeonweb-be recovered. Reader pool: every key exhausted (context only).
  • Session budget: 14 of 20 spawns (6 truth passes, 3 re-sweeps, 5 verifier iterations, no follow-up research), 6,982,947 sub-agent tokens (work/2026-11-22T1407Z-audit/budget.tsv). No secondary work ran.
  • Verification: five iterations, all NEEDS_FIXES; truth plus editorial fell 15, 7, 6, 3, 2. Iteration 5 met the low-residual early exit (truth 1, editorial 1, no F1 or F4); its two remediations (the Fortinet FortiClient citation split, Integrated PA Suite Tester in the Brotli Exposure line) were applied after the pass, so the residual recorded is 2.
  • Coverage gaps: lab52 (unreadable on every transport), previdian (listing 403), baselland.ch FAQ (403 to the container; read through the canton's short link).

← Operations dashboard · run-record contract: docs/pipeline.md