nightly report — 2026-05-09

generated 2026-05-09T20:18Z · sources=10 · runtime budget <90s · haiku-only

---

tldr

quiet 24h — zero truncations, zero red, zero deploys, zero pending. publishers still down (llm-costs/heartbeat/external-spend stale 20d, aws email WinError 10061 ~69th consecutive). orchestrator + watchdog both green. nothing burning, nothing shipped.

---

swarm-orchestrator (last 24h)

entries: 5 checks (2026-05-08T00:18Z → 2026-05-09T20:14Z)
result: healthy on every cycle
truncations: 0 · respawned: 0 · stuck: 0 · 4hr-rate: 0 (threshold 5)
registry in_progress: 0 hits (Status:.*in_progress regex clean)
outputs_total drifted 548 → 405 between 2026-05-08T01:04Z and 2026-05-09T20:14Z — windows temp cleanup of expired workspace dirs, normal
active_30min noise on latest cycle: 12 outputs but all classified non-substantive (9x zero-byte ephemeral self-spawn, 3x bash stdout / powershell error / image-gen filenames — NOT agent transcripts)
~43h gap between 2026-05-08T01:04Z and 2026-05-09T20:14Z runs — daily-downgrade compliant this window for once
nothing to respawn

watchdog (last 24h)

entries: 3
OK: 3 · WARN: 0 · RED: 0
inbox: 838 (+7 since 2026-05-08T01:03Z, normal growth)
audit-2026-05-09.jsonl fresh, 47KB, mtime ~0min old at last cycle
queued_manual_actions=12 (unchanged across all 3 cycles)

persistent watchdog blind spots (carried, not new)

swarm-orchestrator.log = "missing" per watchdog (intermittent bidirectional pattern continues — file exists but watchdog can't always see it)
audit-cadence.log missing (by design post-refactor >7h)
heartbeat.jsonl stale 20d (publisher-down, known)
llm-costs.jsonl stale 20d (publisher-down, known)
external-spend.jsonl stale 19d (publisher-down, known)
master-batch-2026-05-09.log missing (intermittent, no pending batch)

verbatim: no RED entries in window.

advisor audits (yesterday = 2026-05-08)

claude-code-audit-2026-05-08-*.md: unavailable — last advisor audit on disk is claude-code-audit-2026-05-02-0335.md. ~7d stale.
advisor cron may be paused or queue-only — flag to investigate

journal DEPLOYED (last 24h)

0 new DEPLOYED markers
only file in F:/TITAN/plans/journal/ is R0172-DEPLOYED-2026-04-22.md (17d old)
nothing shipped in window

task-registry updates (yesterday/today)

last_updated: rows matching 2026-05-08 or 2026-05-09: 0
most recent last_updated: value across registry: 2026-05-02
registry stale 7d
zero in_progress, zero pending on Harnoor in window

llm-costs (last 24h)

entries matching 2026-05-08/09: 0
last entry in F:/TITAN/logs/llm-costs.jsonl: 2026-04-19T02:48Z (publisher down, known, stale 20d)
daily LLM spend: unavailable — publisher offline since 2026-04-19
proxy estimate via swarm cycles: 5 swarm + 3 watchdog cycles, all haiku-tier work, well under $0.05/cycle budget

aws daily cost (last line)


[2026-05-09T08:00:02] fetching AWS cost data...
  MTD total: $0.00 across 0 services
  Daily avg: $0.00 (0 days)
  Forecast rest-of-month: $0.00
  Forecast next month:    $0.00
  email: {'ok': False, 'reason': '<urlopen error [WinError 10061] No connection could be made because the target machine actively refused it>'}

$0.00 / "0 services" smells like cost explorer creds expired or scoping issue — same shape as 2026-05-07 onward
email publisher refused connection — 69th consecutive failure per watchdog

cloudwatch innerverse/pmf (last 24h)

SessionDepth (Sum): 0 datapoints
HelpfulnessScore (Avg): 0 datapoints
RatingCount (Sum): 0 datapoints
namespace silent — either zero traffic on innerverse apps or metric publisher detached. consistent with quiet weekend pattern.

pending t-numbers on harnoor (>24h idle)

swarm orchestrator reports registry=0 in_progress on every cycle in window
230 T-number references in registry but all legend/historical, none active
nothing to chase

background scout/forge outputs (last 24h)

substantive (>5KB, non-self) in Temp/claude/*/tasks: 0
only >5KB outputs in window are this nightly-report session's own log reads (swarm/watchdog/llm-costs caches)
18 files modified <24h total — all small bash stdout / image-gen filenames, no agent transcripts
no SCOUT memos, no FORGE deliverables

---

quick deltas vs prior nightly (2026-05-07)

2d gap between reports — nightly skipped on 2026-05-08 (re-fire pattern aligned with daily-downgrade window)
watchdog: queued_manual_actions still 12 (unchanged)
aws email failure counter advanced: 67 → 69
llm-costs/heartbeat/external-spend staleness: 18d → 20d
advisor audits: still 2026-05-02, no new
task-registry mtime: still 2026-05-02

the 12 queued manual actions (unchanged carry-over)

watchdog references "queued_manual_actions=12" repeatedly. items per prior reports:

1. widen deadman alarm 26h+

2. review scheduler cron daily-downgrade

3. rotate AWS creds / clear ~/.aws/cache

4. commit/stash 34 working-tree items

5. widen audit-cadence freshness 24h+

6. fix titan-daily-pa-email FileNotFoundError

7. fix llm-costs publisher (down since 2026-04-19)

8. fix heartbeat publisher (down since 2026-04-19)

9. fix external-spend publisher (down since 2026-04-20)

10. fix aws cost-explorer "$0.00 / 0 services" — likely creds or scope

11. fix aws-cli cloudwatch publisher hang

12. fix email publisher WinError 10061 (69x consecutive)

suggested next move

nothing on fire. cron is quiet. swarm is healthy.
if anything: the publisher cluster (llm-costs / heartbeat / external-spend / email / aws cloudwatch) is the choke point. fixing one likely surfaces the rest.
aws cost showing $0.00 across 0 services for 3+ days is either (a) genuinely zero usage (unlikely with innerverse apps live) or (b) creds/scope broken. worth one ec2 describe-instances or cost-explorer get-cost-and-usage call to disambiguate.

---

report end