- Ticket #32225 exists but has no time logged
- Today's GuruRMM enrollment work is unbilled
- Needs either ticket update or new ticket creation
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Cascades caregiver shared-phone bypass pilot — 2026-04-29 evening into
2026-04-30 early morning continuation.
Major work:
- Adopted phased per-group CA rollout (corrects original tenant-wide §5
design that would have blocked off-site office users)
- Step A: backfilled admin@ into excludeUsers on all 8 existing Cascades
CA policies (mirrors sysadmin@ exclusion posture; Option 1 break-glass)
- Outlook + Helpany + LinkRx assigned to Cascades - Shared Phones group
and added to MHS kiosk app list (final dashboard: 5 caregiver apps)
- Created cloud-only pilot user pilot.test@cascadestucson.com,
SG-Caregivers-Pilot group, Business Premium license, vault entry
pushed to Gitea vault repo
- Built 4 CA changes: PATCH legacy all-users-MFA to exclude pilot group,
CREATE 3 new Report-only policies (block off-network, block
non-compliant, 8h sign-in frequency) with both admins excluded
- Pilot phone wipe + re-enroll after first attempt stuck; PIN set,
awaiting MHS to take over launcher and SDM sign-in prompt
6 new project/feedback memories. Resume point at top of new session log.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Both servers were already patched (11.110.0.97 and 11.134.0.20) via
daily auto-update. IOC scan found 16 flagged sessions across both
plus 4 uncommented SSH keys on IX.
Critical remediation:
- Forensic evidence preserved before any deletion
- 4 uncommented SSH keys removed from IX (server-side backup retained)
- 16 flagged sessions purged across both servers
- Root passwords rotated via chpasswd
- New WHM API tokens created; 3 stale transfer-* tokens revoked
- Vault entries + 1Password Infrastructure items updated
Forensic deep-dive verdict: patch held. All 7 actual CVE exploit
attempts (botnet IPs hitting /json-api/version) returned HTTP 403.
The "multi-line pass" IOC hits on user sessions were false positives.
Unidentified 76.18.103.222 root session traced to routine SSL
maintenance (zero sensitive endpoints touched).
Skill hardening:
- Added MANDATORY service-token directive to .claude/commands/1password.md
enforcing OP_SERVICE_ACCOUNT_TOKEN from SOPS for all op CLI calls
- Per Mike: memory files alone don't reliably bind agent behavior;
baking governance into skill content loaded at moment of use.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Make the radio archive Q&A pairs actually browseable end to end:
- /api/qa list endpoint (year, min_score, exclude_banter, topic_class,
pagination, sort by air_date or score). Returns the same column shape as
/api/search Q&A hits.
- /api/audio/{episode_id} streams the MP3 with HTTP Range support so the
browser <audio> can seek. 206 + Content-Range when ranged, 200 when
full-file. Returns 404 cleanly when episodes/ tree is absent (Jupiter).
- /episode/{id} HTML transcript view: chronological segments with clickable
timestamps, Q&A blocks spliced inline (anchor #qa-<id>), intros marked
inline, right-rail summary. Hash-anchor on load auto-seeks the audio.
- New question_excerpt / answer_excerpt fields on /api/search Q&A hits and
on /api/qa items: trim leading run-on chatter, take ~300 chars, end on a
sentence boundary or word boundary with ellipsis.
- Index UI: each Q&A hit now links to /episode/{id}#qa-{qa_id}; new
"Browse all Q&A" toggle (year selector, sort, append-load 50 per page,
defaults to min_score=3); FTS snippet replaced with the plain excerpt
when available.
No new dependencies, no schema changes, no LLM calls. Uses
EPISODES_DIR env (default /data/episodes) — Jupiter compose still only
mounts /data so audio degrades gracefully to 404 there until episodes
are uploaded.
Backend min_score/exclude_banter wired through to HTML index. Adds
score badges (1-5 red->green), topic_class pills, dim styling on
banter rows. Live on http://172.16.3.20:8765/. Synced to portable
repo. pscp ENOSPC quirk worked around by plink-stdin streaming.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds quality-filter controls to the search UI: a "min score" select
(any/2+/3+/4+/5) and a "hide banter" checkbox. Q/A hits gain a small
color-coded usefulness badge (1-5, red->green) and a topic_class tag
(computer-help, banter, off-topic, promo). Low-score and banter rows
render dimmed by default so they're visible but de-emphasized.
Defaults to "any" + banter visible to preserve existing search habits.
Mike toggles up when he wants quality. URL-encoded params built via
URLSearchParams so empty values don't leak into requests.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Streams the read-only archive.db over the same Tailscale-routed port
as the search service. Companion to azcomputerguru/radio-archive-portable
which curl-fetches from this endpoint and runs locally on the laptop.
Disclosure equivalent to /api/search (which already exposes every
transcript), so no auth added. Deployed to Jupiter; verified GET
returns 60 MB SQLite blob with all 1,405 classifier rows intact.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3.5h run on qwen3:14b processed 1,405/1,407 Q/A pairs (2 failed,
will retry on next invocation). 37% scored 4-5 (useful), 41%
scored 1-2 (banter/promo/off-topic). API filter ready; Jupiter
redeploy pending Mike's manual review.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds an Ollama-based content quality classifier and exposes the
results via the search API. 1,407 existing Q/A pairs were scored
in 3.5h via qwen3:14b (1,405 succeeded, 2 failed).
Distribution: 37% scored 4-5 (useful), 41% scored 1-2 (banter/promo/
off-topic). 43% flagged as banter overall. Default-on filtering at
search time will hide ~half of the noise without losing any real
listener questions.
Files:
- new classify_qa_quality.py: walks qa_pairs, calls Ollama qwen3:14b
per row, writes usefulness_score/topic_class/is_banter back to DB.
Idempotent (--rebuild to reprocess), --smoke for sample check, --limit
for partial runs. Detached run handles 1407 rows in ~3.5h on a 4090.
- server/main.py: /api/search accepts min_score (0-5) and exclude_banter
query params. NULL scores treat as "include" so unprocessed rows still
appear. Episode detail endpoint includes the new fields in qa results.
Schema migration in import_to_sqlite.py was made by the same agent run
(visible on the live archive.db: usefulness_score / topic_class /
is_banter columns now exist on qa_pairs).
Local archive.db updated; Jupiter container has NOT been redeployed
yet — that is a separate manual step.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
role_assigned() only checks direct/permanent roleAssignments.
PIM-managed assignments are in roleAssignmentSchedules and won't
be found, producing noisy (non-blocking) output on re-runs against
tenants with PIM-assigned roles (e.g. Cascades).
TODO comment added at the helper — Howard to implement the fix.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Six small bash scripts uploaded to /tmp on 172.16.3.22 during the
OwnCloud cron stacking incident — investigation, group enumeration,
failed group-restrict attempt, occ subcommand discovery. Captured for
audit; full context in clients/pavon/session-logs/2026-04-29-session.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Found 75-126 stale `occ system:cron` processes on 172.16.3.22 piling up
since 2026-04-27 due to bad oc_filecache LIKE query against pavon's 257K
camera files. Killed stale procs (load 80 -> 5), wrapped apache crontab
with `flock -n /tmp/oc-cron.lock` to prevent restacking. Per-user
versioning disable rejected by OwnCloud Community (`files_versions`
can't be enabled for groups); workaround `occ versions:cleanup pavon`
identified and deferred. Migration/retention cron deferred per user.
NVR architecture clarified: GeoVision NVRs sync via OC Desktop client
with virtual file placeholders; no direct SMB access to Jupiter.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Day-long session unblocking the Cascades CA reconciliation that was paused on
the Tenant Admin SP directory-role gap. Discovered Microsoft also tightened
the OAuth scope for /identity/conditionalAccess/* reads (Policy.Read.All now
required, Policy.ReadWrite.ConditionalAccess no longer accepted for reads).
Patched Tenant Admin manifest accordingly and re-consented in Cascades.
Phase B Intune state turned out to be far more built than the 4/20 log
suggested -- compliance policy, Wi-Fi, device restrictions, both SDM app
configs (Authenticator + Teams), and 7 of 8 apps were already deployed and
assigned. PATCHed device restrictions to block camera/Bluetooth/roaming
and enabled Managed Home Screen multi-app kiosk (ALIS + Teams visible,
10-min auto-signout). PATCHed Cascades named location to add primary WAN
(184.191.143.62/32). Howard added Outlook from Managed Play; SMB encryption
enabled on \CS-SERVER\homes.
CA bypass design corrected -- original §5 plan in user-account-rollout-plan.md
called for "block off-site + MFA on-site" which doesn't match the actual goal
of bypass when network + device assurance present. Reshaped to three policies
that produce on-site-compliant = password only, anything else = MFA or block.
onboard-tenant.sh patched to:
1. Backfill Policy.Read.All on Tenant Admin SP if missing (idempotent --
for tenants consented before the 2026-04-29 manifest update).
2. Assign Conditional Access Administrator directory role to Tenant Admin
SP at onboard time. Mirrors the Exchange Operator fix Mike landed in
16f95e8.
Validated with --dry-run against Cascades. Customer-facing tenants already
onboarded should be re-run with this script to backfill both items.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Added cross-reference from FEATURE_ROADMAP.md to UI_GAPS.md tracking document.
Clarifies that features may be backend-complete but UI-incomplete.
Submodule commit: f76051a
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Documented two fundamental GuruRMM development principles:
1. Holistic Feature Development (MANDATORY):
- Every feature requires complete stack: backend, API, UI/UX, docs
- Features without management interfaces are incomplete
- Design for scalability and future expansion
- Example workflows included
2. AI-Optional Operation:
- Product must work without AI agents (Claude, autonomous tools)
- AI features are enhancements, not requirements
- Core operations remain deterministic and reliable
Principles documented in guru-rmm/docs/DESIGN.md and now in memory for
cross-session reference.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Updated DESIGN.md with two fundamental principles:
1. Holistic Feature Development - every feature needs full stack (backend, API, UI, docs)
2. AI-Optional Operation - product works without AI agents; AI features are enhancements
Submodule commit: e490307
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Updated GuruRMM roadmap with two major features:
- Network Discovery Node (P2): site-level device discovery and mapping
- Local Collection Node (P2): reduce WAN traffic by local aggregation
Submodule commit: db7d074
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Howard is cleared to proceed with Path A (Graph API role assignment) for
Cascades CA Administrator fix.
Also communicated new approval workflow:
- General tools: Howard can modify OR Claude can execute with Howard/Mike approval
- Projects: require Mike approval, features→roadmap, bugs→bug list
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Tools (remediation-tool, onboard scripts, MSP utilities):
- Howard can modify directly
- Claude can execute with Howard OR Mike approval
- No roadmap process, immediate operational changes
Projects (GuruRMM, ClaudeTools API, etc.):
- Require Mike approval
- Features go to roadmap
- Bugs go to bug list
Established during Cascades CA role gap fix discussion.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- New laptop provisioned onsite at IMC Speedway: joined to imc.local, AD
account created for Manda (incoming GM), Outlook bound to her M365
mailbox, Office activated via retail key, AIMsi USER#=4 per Leslie.
- Syncro ticket #32218 invoiced — 1.5 hrs Onsite Business labor debited
from IMC's prepay block (14.0 -> 12.5 hrs).
- ServerIMC (192.168.0.63) confirmed as a real authentication-degrading
phantom DC: SRV/A records claim it's a DC; LDAP/Kerberos refuse
connections. Promoted from "unclear, worth verifying" (2026-04-13) to
confirmed AD hygiene issue. Was the root cause of the 2026-04-22 remote
domain-join failure. Needs follow-up ticket: repair or ntdsutil cleanup.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Thread 1 (AD-side prep on CS-SERVER) completed:
- howard.enos password reset to memorable value (PHS will sync to M365 once staging exits)
- proxyAddresses=SMTP:howard.enos@cascadestucson.com added (G1 convention)
Thread 2 (CA reconciliation) blocked: ComputerGuru - Tenant Admin SP
(appId 709e6eed-...) has zero directory role assignments in Cascades.
Graph CA endpoints 403 despite Policy.ReadWrite.ConditionalAccess on token.
Decision pending: Path A (Graph-side role assignment via existing
RoleManagement.ReadWrite.Directory) vs Path B (portal click as admin@).
Target role: Conditional Access Administrator
(b1be1c3e-b65d-4f19-8427-f6fa0d97feb9) on SP objectId
a5fa89a9-b735-4e10-b664-f042e265d137.
Follow-up: extend onboard-tenant.sh to assign this role at onboard time
(parallels 16f95e8 Exchange Admin fix for Exchange Operator SP).
Pilot target slipped 2026-04-27 to 2026-04-28. ALIS App Store still
inaccessible — install-side of ALIS SSO still deferred regardless.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Append to 2026-04-28-session.md covering the FastAPI/SQLite container
deploy: build + ship + verify, plus credentials, paths, and re-deploy
procedures for both DB updates and source updates.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Read-only HTTP layer over archive.db. Endpoints: /api/stats,
/api/episodes, /api/episodes/{id}, /api/episodes/{id}/transcript,
/api/search (FTS5 over segments + qa_pairs, bm25-ranked, snippets),
/api/callers. Single-file HTML index with debounced search UI.
Deployed: Jupiter (Unraid Docker), bound to 172.16.3.20:8765, LAN only.
Container path: /mnt/user/appdata/radio-archive/{app,data}.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Synced with Gitea, reviewed 14 commits from GURU-BEAST-ROG:
- Radio show audio processing (Tara voice profile, Q&A extraction, 4090 benchmark)
- Cascades client work (Howard - HIPAA remediation, Entra Connect staging)
- Valleywide client init (app modernization project)
Note detected: Co-host name 'Tom' needs correction in radio show profiles.
Session type: Sync and context review only, no active development.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Execution-only follow-on to 2026-04-27. Both batch passes done (519+53,
0 errors), import_to_sqlite.py run incrementally to bring archive.db
to final state. Next step: Jupiter Docker container deploy.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
QAPair gets caller_name and caller_role fields populated by a new
attach_caller_names(pairs, transcript_segments) helper. For each pair,
finds the active opening intro at the question_start time (8s forward
tolerance, no backward limit — a caller's call can run for 10+ minutes
and the intro happens once at the start) and attaches the speaker name.
Validation on 9-episode test set:
19/19 Q&A pairs (100%) now have caller names attached.
Examples of corrections from oracle attribution:
2018-s10e18 @ 73:36 Christopher (was misattributed to "Tara")
2015-s7e19 @ 35:45 William (was misattributed to "Tara")
2010-05-08-hr1 Jackie x3, Bruce
2012-03-10-hr1 Adam x2
2016-s8e43 John, Doug
2017-s9e30 Tom, Denise x3, Charlie
speaker_oracle.py: adds speaker_at(time, intros) helper used both by the
existing resolve_speakers() and the new caller-name attachment. Also
adds the "let's fit/bring/put X in/on" intro pattern variant (caught
Charlie at 70:21 in 2017-s9e30 that "talk to X" missed).
download_full_archive.py: SSH keepalive every 30s + per-file retry-on-
failure (up to 3 attempts with reconnect). Earlier run hung on a dead
connection at file 109 of 589 with no recovery; restarted run is now
running at ~10 MB/s vs ~2-3 MB/s before.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New module src/speaker_oracle.py extracts speaker introductions from
transcripts ("let's talk to William", "we have Clay from the Nerd Junkies",
"in Tara's place, we have Clay", "thanks for the call <name>") and binds
them to non-HOST diarization turns. Pure post-pass on diarization JSONs,
no audio processing — corrects audio-only cosine errors using Mike's
deterministic on-air announcements.
Algorithm:
- Extract intros: regex patterns for caller pickups, guest intros,
fill-in announcements, caller closes. Case-strict (rejects mid-sentence
lowercase matches), with a blacklist of common false-positive words.
Deduplicates same-name intros within 5s.
- Resolve speakers: for each non-HOST turn, find the LATEST opening intro
at or before turn.start (with 8s forward tolerance for boundary slop).
Later intros implicitly close earlier callers, so the most recent
intro wins. No artificial lookback limit (callers can talk for 10+ min).
- Falls back to caller_close patterns within 30s after a turn ends.
Validation on 9-episode test set:
2018-s10e18: Christopher 190s correctly named (was mislabeled "Tara")
2012-06-09 : Kay 160s correctly named (was mislabeled "Tara")
2015-s7e19 : Clay 45s as fillin for Tara, William 40s as caller
2016-s8e43 : Charles 630s, Bruce 210s, John 205s — most callers named
2017-s9e30 : Denise 295s, Tom 115s, Elaine 85s, Jeff 10s
Many other callers across all episodes correctly named.
Remaining unnamed CO-HOST/CALLER (~5-10% of non-HOST time) are real
co-host banter or callers without explicit Mike-introductions.
benchmark.py: adds Phase 2.5 "Name Resolution" between diarization and
Q&A extraction. Prints named-speaker breakdown per episode. Doesn't
modify diarization JSONs (resolution is computed on demand).
Next step: feed named turns into qa_extractor so Q&A pairs get caller
name attached for searchability. Also: bootstrap recurring-speaker
profiles (Tara, Tony, Rob, Randall, producers) by accumulating
intro-tagged windows across the full archive once download completes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
First attempt at Clay's voice profile from 2015-s7e19 produced
Clay-vs-Mike cosine similarity of 0.994 — essentially a Mike clone.
Root cause: 10s WavLM x-vector chunks averaged Mike's frequent
interjections together with Clay's dialogue, and Mike's well-trained
profile dominated the resulting embedding signal.
Mike's call: skip Clay, accept the 2015-s7e19 Q&A as noisy. Clay rarely
appears in other episodes, so the cost of not having his profile is
bounded to this one episode plus any rare future appearances.
Cleanup:
- voice-profiles/clay/ removed
- voice-profiles/profiles.json: Clay entry removed
- Memory updated to record the decision and the failure mode
Kept build_clay_profile.py in-repo as documentation of the attempt and
the Mike-similarity-filter pattern. Useful starting point if a future
attempt provides cleaner pure-Clay timestamps.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a transcript-driven bumper filter to the diarization pipeline. When
a transcript segment matches qa_extractor's promo/bumper signatures, the
overlapping audio windows are labeled BUMPER and the WavLM cosine match
is skipped. Prevents music/promo from being matched against speaker
profiles (the failure mode Mike caught in 2018-s10e18 @ 09:20-10:05).
Code changes:
- src/voice_profiler.py: identify_speakers() takes optional skip_ranges
parameter; windows whose midpoint falls in a skip range get labeled
"[bumper]" and skip cosine match
- src/diarizer.py: diarize() takes optional transcript_path; pre-computes
bumper time ranges via qa_extractor._is_promo_or_bumper, passes to
identify_speakers; adds BUMPER speaker label
- benchmark.py: passes transcript_path to diarize()
Aggregate impact across 9-episode test set:
Tara attribution: 4880s -> 3680s (-1200s / -25%)
Q&A pairs: 17 -> 19 (+2)
(bumper-flagged segments had been disrupting conversation detection
in 2017-s9e30 and 2018-s10e18)
CALLER total: 1320s -> 1190s (bumpers previously labeled CALLER moved)
Per-episode bumpers caught: 1-8, total ~165 bumper segments across set
Remaining Tara false positives are real callers acoustically similar to
Tara (Christopher in 2018, Kay in 2012, William and Charles in 2015) and
guest Clay in 2015-s7e19 — those need profile rebuild + Clay profile,
not bumper filtering.
Adds download_full_archive.py — resumable mirror-style downloader that
walks IX server's /home/gurushow/public_html/archive/{year}/ and copies
all MP3s to archive-data/episodes/. Run is in progress (~589 files,
~10-15GB). Used to source clean profile windows for the remaining
co-hosts (Tara rebuild, Clay, Tony, Rob, Randall, producers).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Added 2010, 2015, 2018 test episodes to round out the test set to one
per available year:
- 2010-05-08-hr1 (May 2010, earliest available; pre-Tara era)
- 2015-s7e19 (Jan 2015, avoids training's s7e30)
- 2018-s10e18 (only 3 non-training 2018 episodes exist)
Archive has no 2019 directory — Rob's "2018/2019 appearances" are
constrained to the 5 available 2018 episodes only.
Per-year diarization summary (Tara presence, post-rename):
2010-05-08 30s 1.2% likely false positive (pre-Tara)
2011-03-12 140s 5.6% likely false positive (call-in only)
2012-03-10 30s 1.1% likely false positive (call-in only)
2012-06-09 340s 12.8% suspicious — Mike to confirm
2014-s6e19 680s 23.3% confirmed
2015-s7e19 280s 9.9% plausible — Mike to confirm
2016-s8e43 1890s 35.5% confirmed
2017-s9e30 610s 11.4% plausible
2018-s10e18 880s 17.1% COULD BE ROB — Mike flagged Rob for
2018/2019 appearances; cosine threshold may
be hitting on Rob being acoustically similar
to Tara
Total Tara across 9 episodes: 1h 21m / 8h 52m audio (15.3%).
Q&A counts (still suspect — every voice that isn't Mike-or-Tara is
labeled CALLER, so Randall/Rob/producers inflate the bucket):
2010=4, 2011=1, 2012a=2, 2012b=0, 2014=0, 2015=1, 2016=2, 2017=4, 2018=3
Total: 17 pairs across 9 episodes
4090 perf on the expanded set:
- Diarization: 31928s in 121.5s = 262.7x realtime (vs 209.7x on 5070 Ti, +25.3%)
- Transcription (3 new episodes only): 10554s in 112.4s = 93.9x
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mike confirmed there is no co-host named "Tom" — the voice in 2014-s6e19
and 2016-s8e43 is Tara. The 5070 Ti session fabricated the Tom identity.
The voice profile itself (44 embeddings, 0.698 cosine vs Mike) is correct;
only the human label was wrong.
Rename swept:
- voice-profiles/tom/ -> voice-profiles/tara/ (git mv preserves all .npy)
- voice-profiles/profiles.json: "Tom" key -> "Tara"
- build_cohost_profile.py: TOM_WINDOWS -> TARA_WINDOWS, COHOST_NAME, comments
- 2026-04-27-qa-extraction-cohost-indexing.md: correction header + body sweep
- 2026-04-27-4090-benchmark-and-test-set.md: closure note
- .claude/memory/radio_show_no_cohost_named_tom.md: resolution + speaker roster
Diarization re-run after rename so speaker_map emits "Cohost: Tara".
Q&A counts unchanged (rename is label-only): 9 pairs across 6 test episodes.
Tara distribution from the post-rename diarization (per-episode % of audio):
2011-03-12-hr1 140s 5.6% likely false positive (call-in only)
2012-03-10-hr1 30s 1.1% likely false positive (call-in only)
2012-06-09-hr1 340s 12.8% suspicious — pending Mike confirm
2014-s6e19 680s 23.3% confirmed
2016-s8e43 1890s 35.5% confirmed
2017-s9e30 610s 11.4% plausible — pending Mike confirm
Broader speaker-roster context Mike provided this session (saved to
memory): the show has had multiple co-hosts (Tara, Randall, Rob) plus
producers/board ops (Andrew, Shannon, Ken, others) who would sometimes
go on-air. Only Tara has a profile so far. Every other speaker is
currently labeled CALLER, which means small CO-HOST attributions in
unexpected episodes (e.g. 2011/2012) may actually be a producer rather
than a false positive — Mike to spot-check.
Action item before full-archive run: build profiles for Randall, Rob,
and the named producers to avoid systematic Q&A false positives in
early-years and 2018/2019 episodes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Re-ran benchmark.py on GURU-BEAST-ROG against the post-overhaul code
(co-host profile, batched Whisper int8_float16, revised Q&A extractor).
Results vs 5070 Ti baseline:
- Diarization: 209.7x -> 338.1x (+61.2%)
- Transcription: 63.8x -> 94.8x (+48.6%)
- Q&A pairs: 9 vs 10 (within run-to-run noise; structural correctness matches:
2014 = 0 callers, 2016 = 2 WiFi caller pairs)
Setup change: BENCH_SETUP.md now lists ffmpeg as a Step-2 prereq
(winget install Gyan.FFmpeg). Was missing on this machine and the pipeline
fails silently at the first diarize call without ffprobe.
Code change: benchmark.py BASELINE_RTF updated 149.5 -> 209.7 to reflect
the 5070 Ti's post-overhaul measurement (e9ac607).
Data: 6 test episode transcripts and diarizations regenerated under the
new code path (batched Whisper output + co-host-aware speaker_map).
Correction memory: voice-profiles/tom/ directory + 5070 Ti session log
fabricated a co-host named "Tom" — Mike confirms no such person exists on
the show. The audio profile is real and the diarization separation is
sound, but the human identity attached to it is wrong. Saved under
.claude/memory/radio_show_no_cohost_named_tom.md pending Mike providing
the correct name for rename.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
