not signed in[ sign in ]
vibecodestats.devpersonal insights station

What's getting finished, which models actually work, what keeps breaking, and how the systems are running — across Claude Code, Codex, Grok & Kimi. Usage trends underneath.

RAI content & authority system: 23 red checks today
Interactive completion rate 69% this wk vs 43% last
Opus 4.7 completes 70% of interactive sessions (30d)

Today

2026-07-27 · America/New_York
295.1M
tokens today
13h 8m
active
0/0
plans done
0
blocked
0/0
automations
$72.86
cost
codex220.0Mclaude code75.0M

Plans & sessions

interactive outcomes over time
69%
completion · this wk (39)
43%
last wk (7)
▲ +26%
donepartialblockedabandoned
No interactive sessions recorded today yet.

Problems I keep fixing

recurring failures, ranked by recent recurrence
Api socket closed mid session
Headless run terminated by 'Connection closed mid-response' API error before the briefing could be posted.
api-socket-closed-mid-session · 07-18→07-26 · 5 days active
13
hits
Cc dashboard classifier permission blocks
Claude Code auto mode classifier blocked an action, requiring a workaround.
cc-dashboard-classifier-permission-blocks · 07-22→07-26 · 4 days active
6
hits
Rai backlog drain scheduler miss
Missed scheduled runs (RAI measurement backfill, video factory) required 23 historical backfills and catch-up rather than self-recovering.
rai-backlog-drain-scheduler-miss · 07-23→07-26 · 2 days active
2
hits
Mcp granola auth unavailable
granola MCP server required OAuth that cannot be completed in a headless run.
mcp-granola-auth-unavailable · 07-25→07-26 · 2 days active
2
hits
Macos tcc reprompt per cc version
macOS Full Disk Access grants are keyed to the Claude Code version, so each update triggers a fresh round of per-directory prompts.
macos-tcc-reprompt-per-cc-version · 07-25→07-26 · 2 days active
2
hits
Dirty worktree breaks scheduled runners
Uncommitted finished work (rai-ops v3 follow-up feature, 458 tests passing) sits in the worktree and keeps triggering dirty-worktree warnings that block/annoy scheduled runs
dirty-worktree-breaks-scheduled-runners · 07-26 · 1 day active
2
hits
Auto mode model unavailable blocks bash
Auto mode could not determine Bash safety because the sonnet-5[1m] safety model was temporarily unavailable.
auto-mode-model-unavailable-blocks-bash · 07-26 · 1 day active
2
hits
Report findings schema validation error
ReportFindings tool call rejected because the findings payload was not an array, forcing a retry.
report-findings-schema-validation-error · 07-26 · 1 day active
1
hit
Rai producer message length limit
RAI producer adapter rejects oversized Slack posts without surfacing the actual char cap, forcing repeated trim-and-retry cycles.
rai-producer-message-length-limit · 07-16→07-25 · 9 days active
9
hits
Cross repo relative path lookup failure
Read tool rejected the diff's repo-relative file paths, requiring absolute-path retries
cross-repo-relative-path-lookup-failure · 07-20→07-25 · 4 days active
6
hits
Ai updates stamp check race false retry
Scheduled ai-updates retry fired again despite a successful 10:30 post because the already-posted stamp check failed.
ai-updates-stamp-check-race-false-retry · 07-24→07-25 · 2 days active
2
hits
Firecrawl tool not found
The firecrawl CLI invoked by the firecrawl-scrape skill is not installed/on PATH, failing with 'command not found'.
firecrawl-tool-not-found · 07-16→07-24 · 2 days active
2
hits

Systems board

mission-control health · 30d uptime
16 ok · 2 amber · 7 red today
RAI content & authority system
13%126 red
Brain · directory checker
25%7 red
RAI relationship agent
66%22 red
RAI Slack collection
86%21 red
RAI operations core
95%8 red
Vibecodestats collector
98%1 red
Pipeline morning brief
99%1 red
HoldenGR market reports
0%no reds
Local automation host
88%no reds
Holden repo nightly stats
90%no reds
FUB autopilot
91%3 red
HoldenGR email & calendar
95%8 red
RAI email & calendar
97%5 red
Granola
97%4 red
Mission Control
98%no reds
AI Updates
100%no reds
Brain System Review
100%no reds
HoldenGR market snapshots
100%no reds
HoldenGR videos
100%no reds
FUB content regeneration
100%no reds
Outcomes miner
100%no reds
RAI Notion project sync
100%no reds
Save work
100%no reds
System health history
100%no reds
Usage watchdog
100%no reds
RAI Upwork opportunity monitor
32%73 red

Model effectiveness

interactive task outcomes per model — which models work for me
modelsessionsdonepartialblockedabandonedcompletionfrictiontokenscost
Opus 4.72014——670%0.782.2M$135.04
Opus 4.8149—5—64%1.11.44B$1,598
Opus 5743——57%1.3312.4M$300.19
Fable 5532——60%1.02.24B$4,275

Interactive sessions only — automations are tracked separately and never mixed in here. Chat-only sessions are excluded from completion + friction. Friction is 0–3 (higher = more stuck).

Model mix

daily tokens, stacked by model
Opus 4.8Opus 4.7Fable 5GPT 5.6 SolHaiku 4.5 20251001Sonnet 4.6Opus 5Sonnet 5other

Productive hours

tokens by hour × weekday (30d)
hover a cell for detail
12a
1a
2a
3a
4a
5a
6a
7a
8a
9a
10a
11a
12p
1p
2p
3p
4p
5p
6p
7p
8p
9p
10p
11p
Mon
Tue
Wed
Thu
Fri
Sat
Sun
lessmore

Projects

top projects by tokens · model mix · ships
Documents/Claude/Projects/RealSavvy Pipeline System606.3M
6,028 turnsFable 5Opus 4.8Sonnet 5
holden-alt/watchreference569.9M
6,817 turnsOpus 4.8Fable 5Sonnet 5
holden-alt/holdengr-video480.7M
2,325 turnsFable 5Opus 4.8Sonnet 4.6
holden-alt/holdengr-updated377.3M
1,955 turnsOpus 4.8Fable 5Opus 4.7
holden-alt/richardsonapplied-brain255.5M
1,031 turnsFable 5Opus 4.8other
~146.2M
688 turns1 doneOpus 4.8Fable 5other
holden-alt/fub-autopilot99.0M
443 turnsFable 5Opus 4.8Opus 4.7
realsavvy-alt-org/john-tait82.2M
205 turnsOpus 4.8Fable 5other
realsavvy-alt-org/michael-oday-real-estate-help-utah80.3M
296 turnsFable 5other
holden-alt/richardsonappliedai66.3M
142 turnsFable 5Opus 4.8other