Fleet token monitor cutover and old-stack retirement

KPI-74 kpi Done high

└─ Accept fleet ranking API and simple web page

KPI-78 kpi Done high

└─ Bring Lezama Claude and Codex token feed live

KPI-77 kpi Done high

└─ Bring Whey Claude and Codex token feed live

KPI-76 kpi Done high

└─ Bind ingest credentials to Venus, Whey and Lezama hosts

KPI-75 kpi Done high

└─ Retire remaining old tracking stack and references

KPI-79 kpi Done

kpi-burn-alert.sh cannot see unregistered codex agents - read v_usage_resolved, not usage_events

KPI-72 kpi Done high

lezama/codex token feed has never produced a row - install the collector before codex carries fleet load

KPI-71 kpi Backlog high

Fleet token accounting: truthful top-X/N-hours report

KPI-66 kpi Backlog high

└─ Message leg of the truthful top-X/N-hours report: hub-message-census.sh, rows and sends side by side with the aro half declared unavailable

KPI-80 kpi Done high

└─ Fleet accounting contract: shared query, window validation, exact-vs-estimate

KPI-68 kpi Blocked high

└─ Collection: coverage tracking + automation_job class

KPI-67 kpi Done high

Fleet token-governance / KPI consolidation (epic)

KPI-45 kpi In progress high

└─ Staged offender remediation: escalation ladder (DM -> alert Elazar -> unregister/kill)

KPI-47 kpi Blocked high

└─ KPI-47 stage 3: unregister from hub + kill process after M more ignored DMs

KPI-56 kpi Canceled high

└─ KPI-47 stage 2: after N ignored DMs, alert Elazar on all channels

KPI-55 kpi Canceled high

└─ Alert Elazar: misconfig / high-burn / context / needless-turns

KPI-46 kpi Blocked high

└─ KPI-46 DM fan-out: every elazar CLI session, not one hub DM

KPI-54 kpi Canceled high

└─ KPI-46 Linux notification leg: inotify/notify-send on whey, venus, lezama

KPI-53 kpi Canceled high

└─ KPI-46 email leg: mail Elazar on every finding / escalation

KPI-52 kpi Canceled high

└─ Fleet-wide usage dashboard on kpi.pensanta.com: every agent, every host, both providers, ranked

KPI-58 kpi Done high

└─ KPI-58 retire the WEB/EXPORT read of token-usage.sqlite + verified snapshot (collector stays)

KPI-61 kpi Done high

└─ KPI-58 acceptance audit of the pg-backed kpi.pensanta.com page

KPI-64 kpi Canceled high

└─ KPI-58 whey ingestion: usage_events has zero whey rows

KPI-62 kpi Done high

└─ KPI-58 web cutover: token-usage-web.sh /, /export.json, /export.md read Postgres usage_events (pg_rank_sql contract)

KPI-60 kpi Done high

└─ KPI-58 fleet notice: export.json agent set doubles at pg cutover

KPI-63 kpi Done high

└─ KPI-58 backfill usage_daily into usage_events: 52 days x 61 agents of fleet history no leg has ever read

KPI-65 kpi Backlog

└─ fleet-tokens.sh - one-command top-offenders / per-agent token answer (monitor-context-cc)

KPI-59 kpi Done high

└─ INCIDENT (done): 19h fleet-wide compact blind spot from CC statusline format change

KPI-57 kpi Done high

└─ Evidence-led compact/change/shutdown recommendation workflow

KPI-49 kpi Backlog high

└─ Fleet running-agent/model inventory without DM storms

KPI-48 kpi Backlog high

└─ Fleet inventory: every token/context/llmmsg-srv script, monitor, service each agent owns/knows

KPI-1 kpi Backlog high

└─ MEASURED LEVER: MCP http transport costs ~60k context per agent

KPI-31 kpi Canceled high

└─ DUPLICATE COLLECTION: two parsers scan the same transcript jsonl for tokens

KPI-12 kpi Canceled high

└─ DECOY DATASTORES: whey /var/lib/llmmsg-srv holds a 2-month-stale v2.sqlite + 2 zero-byte husks, all silent

KPI-41 kpi Backlog high

└─ guard-unbounded-output.py — global PreToolUse Bash hook, venus, all agents

KPI-4 kpi Backlog high

└─ scan-context.py / transcript_signal.py / memory-trim-candidates.py

KPI-32 kpi Canceled

└─ COVERAGE LIMIT: what my data cannot see

KPI-22 kpi Canceled

└─ Known-of, not mine: other token/context collectors on venus

KPI-21 kpi Canceled

└─ context-venus-tick units + alert path (the live actuator wiring)

KPI-20 kpi Canceled

└─ MEASUREMENT.md - methodology + measured limits of the context data

KPI-14 kpi Canceled

└─ analyze-agent-context.sh - one-shot per-agent context diagnosis

KPI-13 kpi Canceled

└─ context-monitor.sh - shared fleet context monitor (PATH script)

KPI-9 kpi Canceled

└─ monitor-context.sqlite - the venus context/token datastore

KPI-8 kpi Canceled

└─ tick.py - venus /compact auto-actuator (OPS-24 sole actuator)

KPI-2 kpi Canceled

└─ ccusage-watch.sh v2.6 - token burst detection off usage.polls deltas

KPI-29 kpi Canceled

└─ ccusage-snapshot.sh v2.2 - per-project/per-model token usage into pg usage.daily

KPI-28 kpi Canceled

└─ llmmsg-kpi-coach.sh v1.8 - daily coaching pass over the KPI report

KPI-27 kpi Canceled

└─ llmmsg-kpi-report.sh v1.13 - daily hub KPI report from v2.sqlite

KPI-26 kpi Canceled

└─ context-monitor-whey.timer (whey, --user scope, ~14min) - armed instance of monitor-context-cc script

KPI-24 kpi Canceled

└─ ccusage-daily-kpi.sh + timer (whey, system scope, daily 08:00) - token usage KPI collector

KPI-15 kpi Canceled

└─ kpi-web (whey, :9876): RUNNING but NON-FUNCTIONAL and UNUSED - 502 permission denied on every view, zero requests 7d (self-corrected; my earlier "live" was a 200 on the static shell)

KPI-34 kpi Canceled

└─ Completed dataset: full census of traffic addressed to write-only scrp-* emitters (614 msgs)

KPI-23 kpi Backlog

└─ Known-of KPI/context collectors on venus PATH - maintainers named, none mine

KPI-33 kpi Backlog

└─ Negative results: 5 traffic predicates + 2 dead fields that look like KPI data and are not

KPI-19 kpi Backlog

└─ 61 analysis views already live in llmmsg-srv v2.sqlite - dedup candidate, maintainer pm-llmmsgsrv-cc

KPI-11 kpi Backlog

└─ Avoidable-traffic audit lane (monitor-llmmsgsrv-cc) - 6h cut over llmmsg-srv v2.sqlite

KPI-3 kpi Backlog

└─ maintainer-agents.sqlite - the ownership registry this epic depends on

KPI-37 kpi Backlog

└─ sys-check.sh (whey)

KPI-40 kpi Backlog

└─ sh.git cannot attribute maintainership - one git identity for all agents; 12 of 15 whey telemetry scripts declare no owner, so these rows are recollections not records

KPI-39 kpi Backlog

└─ OPS-137 orphan/scratch sweeper - owner resolution measured, design note delivered

KPI-38 kpi Backlog

└─ jobreg v1.7 - fleet scheduled-job registry, drift-check + heartbeat watchdog

KPI-36 kpi Backlog

└─ hub-db-backup.sh v1.2 - verified daily backup of the llmmsg-srv hub DB

KPI-35 kpi Backlog

└─ memory-budget-sweep.sh v2.3 - daily auto-load footprint sweep

KPI-30 kpi Backlog

└─ 13 of 15 telemetry scripts present on whey are armed nowhere on it - presence != collection (sh.git checkout artifact)

KPI-25 kpi Backlog

└─ cdwlogs.sqlite: cdw SSE/send/notify telemetry (13.4k events, Jun-Aug 2026)

KPI-17 kpi Backlog

└─ chat-duo-web (cdw): llmmsg-srv browser client, venus :9704

KPI-16 kpi Backlog

└─ ctxscan — transcript token/reclaim scanner (per-agent, venus CC only)

KPI-6 kpi Backlog

└─ MSG-330 traffic breaker (hub.mjs) — held, not applied

KPI-5 kpi Backlog

└─ chat-duo (GTK client, DORMANT): logger-chat-duo.sqlite diagnostic logger

KPI-18 kpi Backlog low

└─ 26 analysis views incl. v_kpi_* scoreboard and vUsageDataPerAgent*

KPI-10 kpi Canceled

└─ kpi-web (whey :9876, ONEDB-5): 17-view telemetry dashboard - maintainer nw-venus-cc, state contested

KPI-42 kpi Done

└─ agent_telemetry (unresolved) bucket: roster is a presence table, so 16827 codex rows / 2.08e9 tok cannot be attributed at all

KPI-51 kpi Backlog

└─ Lezama token-usage collector (no cross-host ingest exists)

KPI-50 kpi Done low

OPS-125 M-DEADSEAT: distinguish NEVER-EXISTED from LAPSED dead seats

KPI-70 kpi Canceled

OPS-125 M-DEADSEAT: add maintainer-registry check as a third, separate alert line

KPI-69 kpi Canceled

Codex AGENTS.md lacks the no-ack/no-phatic rule, so daily coaching of -ca agents cannot land

KPI-73 kpi Backlog

venus telemetry PG: no owned schema convention — role/table/view per host (lezama/venus/whey), unify + add prune mechanism

KPI-44 kpi Backlog

llmmsg-kpi-report.sh: large-ARO flag is base-rate noise + stale-corpus decoy risk (v1.15/v1.16 fixes)

KPI-43 kpi Backlog

MEMORY.md index-bloat sweep — pm-llmmsgsrv-cc project

KPI-7 kpi Backlog someday