Last updated: Monday Aug 03, 04:11

⚠️ Briefing is more than 26 hours old — may be stale.

2026-08-03

Daily Briefing — Monday, August 3, 2026

⚠️ You fly tomorrow (Tue 8/4). Several pre-departure items in TASKS.md are due today or already overdue — see Tasks at the bottom.

Media (last 24h)

No new imports in the last 24 hours — Sonarr 0 episodes, Radarr 0 movies, Lidarr 0 albums. All three services responded normally.

Homelab

All nominal. Snapshot 04:00 EDT.

• ZFS pools — all ONLINE, all well under the 80% threshold: mediapool2 55%, nvmepool 50%, Biggest 43%, backups 29%, mediapool 22%. Highest fragmentation is nvmepool at 19%.
• Containers — all 47 on CT100 Up, health-checked ones all healthy. No restarts, no unhealthy states.
• CTs — 100/101/102/103 running. CT105 (roon) stopped, as expected.

Seedbox

Queue: 49 torrents — 48 complete, 1 incomplete.

• ⚠️ 1 stalled download — Little Feat - Sailin' Shoes [Rad Gumbo] (2014) FLAC sitting at 22.9% with 0 B/s. No progress.
• Seeding: 19 (Foyle’s War / Doc Martin episodes, several FLAC albums, film remuxes).
• Stopped/complete: 29 — mostly the large Kiwix .zim archives (wikibooks maxi + nopic, gutenberg, khanacademy, trueprepper, anonymousplanet, milne, cheatography) plus finished movie/music grabs. Done seeding, just occupying disk — cleanup candidates.

Farm

No changes in the last 24h. Catalog steady at 328 species, 232 plantings, 22 locations.

Backups

homelab-config git: a3cf3b2 — 2026-08-03 02:46:30 -0400 — “Auto: 2026-08-03T02:45:01-04:00”. Auto-commit ran on schedule.

nvmepool nightly backup log: podcasts, sync, video, vms all synced (exit 0); “Nightly backup complete (status 0)” at 01:09:13. No errors.

Weather — Brownsville

Now (Tempest ST-00177825, 04:00 EDT): 63.8 °F, feels like 63.8 °F. Humidity 99%, dew point 63.6 °F — fog likely. Wind calm. Pressure 29.07 inHg. UV 0.0 (pre-dawn). Rain today 0.00 in.

⚠️ Tempest logger has a gap — only 35 rows in the last 24h (01:10–04:00 EDT). The latest reading itself is current (3 min old), but the “24h high/low” figures below only cover ~3 hours. Range over that window: 63.8–66.8 °F, 0.00 in rain, max gust 1.4 mph.

Forecast (NWS ILN):

• Today: Mostly sunny, high near 79 °F. North wind ~7 mph. Precip 0%.
• Tonight: Clear, low ~56 °F. NE wind 2–7 mph.
• Tomorrow: Sunny, high near 83 °F.

Alerts: None active.

Burgos — Eclipse Watch

9 days to go. Totality over Burgos is Wednesday 12 August 2026.

Local circumstances — Burgos city (42.34°N, 3.70°W), all times CEST

Event	Local time
Partial begins (C1)	19:33
Totality begins (C2)	20:28:23
Maximum	20:29:15
Totality ends (C3)	20:30:07
Partial ends (C4)	21:21
Sunset	21:20

Totality lasts 1 min 44 s. Burgos sits essentially on the centreline — close to the best you can get on the Spanish mainland.

The thing that matters most isn’t the timing, it’s the geometry: the Sun is only 8.2° above the horizon at totality, bearing 282.6° — just north of due west. That’s roughly a fist-and-a-half above the skyline at arm’s length. A ridge, a treeline, a row of buildings, even a low haze bank takes the whole event away. You need a site with a genuinely open, flat western horizon, scouted in daylight beforehand. Note the last row: the Sun sets at 21:20 while the partial phase is still running — the eclipse literally ends at sunset. No margin. If it’s obstructed, it’s gone.

Cloud forecast for eclipse day

Nine days is at the edge of useful range, but the models are unanimous. For 19:00–21:00 on 12 August at Burgos: ECMWF, GFS and Open-Meteo best-match all show 0% cloud at every level — total, low, mid and high. (ICON, Météo-France and UKMO don’t reach 12 August yet; they come into range 7–8 August.) Consistent with the rest of the picture: 34.8 °C daytime high, ~50 km model visibility, a hot dry anticyclone parked over the meseta.

Plainly: this looks about as good as a cloud forecast can look. High cloud is what decides whether the corona is crisp or milky, and it’s currently forecast at nothing. Two caveats. 0% at nine days means “no organised cloud system in sight” — it is not a promise about thin cirrus forming 36 hours out. And in a hot, still, high-pressure August setup over Spain the realistic threats aren’t frontal cloud at all: they’re afternoon convective build-ups drifting off the Sierra de la Demanda to the southeast, and low-level haze thickening near the horizon exactly where the Sun will be. Neither shows in a 9-day grid. Recheck at day 5 and day 2.

Wildfire smoke aloft — the live risk

Column aerosol optical depth over Burgos (CAMS), min / mean / max:

Date	AOD
3 Aug (today)	0.17 / 0.24 / 0.32
4 Aug	0.07 / 0.14 / 0.20
5 Aug	0.11 / 0.14 / 0.18
6 Aug	0.13 / 0.15 / 0.18
7 Aug	0.11 / 0.12 / 0.14

For scale, clean Atlantic air over northern Spain runs 0.05–0.10. Today’s 0.24 mean with a 0.32 peak is a noticeably hazy sky — white rather than blue near the horizon, low Sun going orange early. That’s smoke and heat haze in the column, and it would meaningfully dull a corona if it happened on eclipse evening. Good news is the trend: it halves tomorrow and stays around 0.12–0.15 through the 7th.

⚠️ The CAMS aerosol forecast does not yet reach 12 August. Last valid hour in the current run is 07 August 14:00 — CAMS only runs ~5 days ahead. There is genuinely no smoke forecast for eclipse day yet; anyone claiming otherwise is guessing. Eclipse day enters CAMS range on 7 August. That’s the date to re-run this.

On drift, hedged appropriately: forecast surface wind at Burgos at eclipse time is ~25 km/h from the NNE, bringing clean Cantabrian air down onto the meseta and pushing away from the Ávila/Madrid/Toledo fire complex. Aloft it’s the opposite — ~52 km/h from the WSW at 500 hPa, and that upper flow tracks straight over the León and Zamora fire zones before reaching Burgos. Smoke aloft is the plausible spoiler, not smoke at ground level, and upper-wind direction at nine days is low-confidence. Watch it.

Ground air quality is fine and improving — PM2.5 4–8 µg/m³, PM10 under 13, dust essentially zero. Not the number that matters for a corona, but worth knowing.

Confirmed fires vs thermal detections

Burgos province: zero active thermal detections right now. NASA FIRMS/VIIRS shows 95 detections in the province for all of 2026, only 3 in August, none active. June (39) and July (53) carried nearly all of it.

Two detections on 31 July landed at 42.371, −3.726 and 42.356, −3.647 — essentially on top of Burgos city — at 12.6 and 11.8 MW, confidence 69 and 61. These are satellite thermal anomalies, not confirmed fires. At that power and location they’re far more likely industrial heat, a flare, or burning stubble, and nothing in Spanish reporting shows a fire in Burgos city on 31 July. Exactly the category of hit that has produced false alarms before. Treat as noise unless a Spanish source confirms.

Confirmed, named fires — none in Burgos province:

• Sierra Oeste (Madrid/Ávila/Toledo) — ~50,000 ha, largest wildfire in Spanish history. Stabilised but still burning, still under national emergency declaration. ~200 km SSW of Burgos.
• León — San Cipriano del Condado, Vegellina, Valdelaloba; La Utrera entering control phase, confirmed deliberately set. ~180 km west.
• Zamora — Fermoselle, Arribes del Duero; 14 villages evacuated, 800+ people, two more confined. ~250 km west.
• Guadalajara — Sierra Norte Natural Park, 32,000+ ha. ~200 km SE.
• Castellón — La Vall d’Uixó, ~6,500 ha; 6,000 residents returned. ~450 km SE, irrelevant to Burgos.

Nationally ~150,000–170,000 ha burned in 2026, about six times the same point last year, 14 confirmed deaths.

Evacuations, roads, advisories

One correction worth making. The “267,000 evacuated” figure is Spain and France combined, and 197,000 of it was the Bordeaux region in France — probably the largest peacetime evacuation in French history. Spain’s own peak, reported 25 July, was 70,000 people. There was no Spanish city of ~250,000 evacuated. The fires are severe and the death toll is real, but that number belongs to Gironde, not Castilla y León and not anywhere near Burgos.

Current state: thousands of evacuees began returning 29 July. As of early August roughly 30 roads remain closed nationally — all in Ávila, Castellón, Madrid, Toledo and León, including the N-621 in León. Madrid is progressively rehousing residents, ~1,000 still waiting in San Martín de Valdeiglesias and Pelayos de la Presa.

Nothing closed in Burgos province. No evacuation, no confinement, no travel advisory affecting the Burgos area.

Eclipse-day traffic is a separate problem. The DGT published a special safety plan on 14 January covering seven autonomous communities including Castilla y León: HGV restrictions, signed alternative routes, designated parking, spot closures, reinforced patrols. Burgos expects a large visitor influx. Specific closures are supposed to be published with at least a month’s notice, so they should be on the DGT site now or imminently — worth checking before committing to a viewing site. The risk isn’t getting into Burgos, it’s being stuck on a secondary road at 20:28 with 104 seconds of totality happening through the windscreen.

Bottom line

Nine days out, this looks good. Burgos is on the centreline, cloud forecast at zero at all levels, smoke column trending down. The fires are real and severe but 180–250 km away in other provinces; Burgos province itself has no active fire. Two things could still take it away: (1) a thin cirrus veil or smoke layer aloft that won’t be forecastable until ~7 August, and (2) a blocked western horizon — entirely within your control, and worth scouting. Recheck 7 August, when CAMS aerosol and the shorter-range cloud models both reach eclipse day.

Service Health

⚠️ 2 monitors DOWN, both on the Mac Studio (192.168.8.180):

• LM Studio (TCP probe) — connect ECONNREFUSED 192.168.8.180:1234 — app not running / server not listening.
• Life Archive API — connect ECONNREFUSED 192.168.8.180:8900.

Both are connection-refused on the same host, so this is Mac Studio–side services not being up rather than a network fault. Note both com.bee.lmstudio-server and com.beedifferent.life-archive-api are loaded in launchd, so the agents are registered but the processes aren’t listening.

No farm monitors down — the house→farm path is healthy. All key edmd.me endpoints serving.

Dictation

No dictation in the last 24 hours. 24 hourly runs fired (2026-08-02 00:05 through 2026-08-03 04:05), every one reported “No new dictation files found.” As of the 04:05 run this is the 33rd consecutive empty run. Last processed recording was the USB recorder capture on 2026-08-01 (2 days ago).

0 dispatched, 0 tasks queued, 0 new questions pending. The three md5-duplicate skips (2026-06-04_09-53-11 2.m4a, 2026-06-27_12-13-03.m4a, 2026-07-27_11-52-52.m4a) recur every run and are routine, not errors. Orphan-transcript checks came back clean each hour.

Diary — Saturday, August 2, 2026

No diary entry for this date. Most recent diary on file is 2026-08-01 — 2 days ago. Recap of that one:

Eight hours on the USB recorder, most of it ambient. The morning starts in the truck with Amigo along and the radio on, Ed talking to the dog — then arriving somewhere, all delight: “We’re here! We’re here! Puppy!” The middle of the day is largely video playing in the background: a long diesel fuel-system troubleshooting walkthrough where the man’s problem turns out to be a plugged port at the bottom of the fuel tank sucking air — almost certainly why Ed was watching, given the buggy is still running rough after a new fuel filter. Then small-engine repair, a subcompact tractor with a landscape rake, and two that look like actual shopping: a Blaze 15 propane flame weeder assembly walkthrough and a Neversink-style dibbler conversion for a wheel hoe. The day ends at home with Mingo strutting because he heard them pull in, and a proper catch-up about Sarah — allergies bad, and a mess with her doctor’s office that turned out fine, since the new one they assigned her seems better than the old. Domestic business closes it out: the TV wouldn’t turn on and every clock was wrong, because Ed had cut power the night before to fix the well.

Care due this week

No items due this week, and no active items in window. care-state.json regenerated on schedule at 03:15 today — due_this_week, active and missed are all empty. Nothing missed.

Tasks

160 open items in TASKS.md (last updated 2026-08-03).

⚠️ Pre-departure block — you fly tomorrow, Tuesday 8/4. Several of these have slipped:

• Treat the hives — explicitly “before the Tuesday flight.” Due today.
• Set up the two sprinklers in the tree area — manual, not on the controller. Must happen before you leave.
• Put the supers back on the hives — was due Sat 8/1. Overdue.
• HONEY — finish filtering (was due Sat 8/1) and bottle (was due Sun 8/2). Both overdue; hard deadline was the plane.
• Fix the house internet “this weekend” — the interior AP isn’t hooked up. Weekend’s gone.
• Three matched jars to the fair — Tracy submits them, from almost-fully-capped frames only.

Other near-term deadlines:

• Register Burgos Eclipse Watch as a desktop-local scheduled task — still unregistered, so the standalone 12-hourly Spain wildfire/weather reports continue to silently fail. (This briefing is covering that gap manually through 8/12.)
• Register for Woodland Pollinators Workshop — deadline August 7, $20, go.osu.edu/woodlandpollinators2026. Event Fri Aug 14. You’ll be travelling — worth doing before you go.
• Register for Shrub ID Workshop — deadline Aug 21, event Fri Aug 28.

Longest-standing infrastructure item still open: investigate the recurring house→farm routing failure — partially covered by the farm-link-watchdog on pve since 7/22, but the CT100-side self-heal cron is still unbuilt.

---

Questions for Ed

1. Tempest logger gap. The weather DB only has rows from 01:10 EDT onward — 35 rows in 24 hours instead of ~288. The current reading is fine, but something dropped the logger for most of yesterday. Want me to dig into tempest-logger.py on CT100, or leave it until you’re back?
2. Stalled seedbox torrent. The Little Feat FLAC has been at 22.9% / 0 B/s. Remove it, or leave it in case seeders return?
3. Kiwix .zim torrents. 29 stopped-and-complete torrents are still occupying seedbox disk, most of them the big Kiwix archives that already rsynced to /mnt/kiwix. Safe to clear them off?
4. Burgos eclipse watch registration. I’ve folded the eclipse section into this briefing through 8/12 as the SKILL instructs, so you’re covered either way — but the standalone task still isn’t registered in the desktop app. Do you want to register it before you fly, or just rely on the daily briefing for the remaining 9 days?
5. LM Studio / Life Archive API. Both launchd agents are loaded but neither is listening. Want me to try restarting them, or are they down deliberately?

Saying goodnight to Larry

2026-08-15

A minute and a half of a Friday night, caught on the recorder around 10:23 PM — the tail end of an evening winding down.

Saying goodnight to Larry, who’s heading out west. “Are you calling it a night, Larry?” — “Yeah, probably pretty soon.” Ed told him if he’s ever near Duke he’ll look him up, and joked about how he’d manage it: just call in and ask “Is Larry there?” and he’d probably find him. If it’s him, it’s him. Or ask the immigration guys at the airport — they probably know.

Larry had a story. He left there one time and forgot he’d put a bottle of Worcestershire sauce in his coat. Somebody thought it was Tabasco. “It’s not Tabasco. It’s Worcestershire sauce.” He figured they’d take it off him, said he’d be back in a week — and when he came back, they’d held onto it for him.

Betsy came up in passing. “Take care if you’re coming out west.”

Warm goodbyes at the end: “Well, you made the week better. Thanks very much.” — “Thank you.” — “Okay, well, I hope you’ll get to meet him soon.” — “I hope we’ll connect again.” Not on free time, though. They’re going to do another one. They’re going to eat.

And somewhere in the middle of all of it, small and domestic: “Oh, look at that.” — “Yeah, I got my milk.”

No farm talk, no bees, nothing on the property. Just a good end to somebody’s week, and Ed’s milk.

---

Last updated: Saturday Aug 08, 15:07

Quick Hits (5-minute tasks)

• the cattle doorknob needs to be fixed (dictation 2026-06-15, 01:23)

• there are some vines in the arborvitae (dictation 2026-06-15, 01:29)

• spread some grass seed around the house (dictation 2026-06-15, 01:33)

• plant clover around the house (dictation 2026-06-15, 01:40)

• Take the sprayer to all front fence lines — a full-day job (inferred, dictation 2026-06-15, 01:53)

• Cut plant cuttings at Granville, bring to the farm, and propagate (inferred, dictation 2026-06-15, 04:19)

• Rig up a Wi-Fi access point on top of the survey buggy — bed-mounted unit transmits poorly and degrades Emlid GPS data collection in the field. (inferred, dictation 2026-06-18, 02:33)

• Do something with the carved planter — the planter “the guy carved out” (inferred, dictation 2026-06-08, 76:13)

• Take the chain to get sharpened at the same time you take the Echo down — do it on the same trip when the Echo weed eater goes in for repair. (“change” = chain.) (dictation 2026-06-07, 51:52)

• Plant the OPN/Ernst native seeds (next few weeks) — 31 packets now in inventory; sow per Zone 6b timing. (2026-06-06)

• Gather Emlid GPS points for the new native plantings (planted ~wk of 2026-06-01) — Two photo batches in Downloads. Batch 1 (33 photos): 18 species, 11 new added to DB (ids 310–320): green-headed coneflower, Ohio spiderwort, hoary vervain, groundnut, wild mint, flat-topped aster, ditch stonecrop, sweet shrub, hairy mountain mint, compass plant, American hazelnut. Batch 2 — Scioto Gardens shrubs (12 photos, IMG_1052–1076): 7 species, all already in the catalog — Kalm’s St. John’s-wort ‘Gemo’ (300), Spicebush (294), Northern bush honeysuckle (288), Fragrant sumac ‘Gro-low’ (299), Winterberry holly incl. ‘Winter Red’ + ‘Southern Gentleman’ male pollinator (50/301/302/56), Roughleaf dogwood (293), Silky dogwood (70). Shoot each planting location with the Emlid and log to the farm DB so they show on the GIS map. (photos 2026-06-06)

• Build 3 rudimentary work tables (garden, meadow, bee yard) — has 2x8s for tops and a few 4x4s; cut more 4x4s at 4-foot intervals. Parser captured this as “I guess” due to phrasing; full context: “build my three at least rudimentary work tables, one for the garden, one for the meadow, and one for the bee yard, so dash to do I guess” (dictation 2026-05-24, 811:13)

• Put up the gun targets and paint the gun targets (dictation 2026-05-24, 832:29)

• Troubleshoot electricity / lights (inferred, dictation 2026-05-24, 834:57)

• Fence repair at the end by the drive — that stretch is “destroyed”; get someone in to fix it or put up steel posts. (May be same area as “Fence line cleanup, lower lot.”) (inferred, dictation 2026-05-30, 540:34)

• Throw leaf mulch over all beds — raspberry/plant beds need mulching after planting (dictation 2026-05-22)

• Take sprayer apart — check if battery is replaceable (dictation 2026-05-13)

• Install Wi-Fi bird cam by tire / equipment-barn feeders — pulling wires; add variety feeders there for new bird-cam zone. (inferred, dictation 2026-05-28 10:01, 33:45)

• Cut down invasive vines at the far gate (oak/sumac being smothered) — egregious 30-year parasitic-vine specimen; tree may not survive without intervention. Also clear residual vines from arborvitae along driveway. (inferred, dictation 2026-05-28 14:31, 47:25)

• Begin Autumn olive (Russian olive / Elaeagnus) removal from periphery — Ed identified one along upper path; many more around property edges. Silvery leaf underside is the tell. (inferred, dictation 2026-05-28 14:31, 57:31)

• Fence line cleanup — lower lot, pick up debris, sow clover seed (dictation 2026-05-13)

• Get a hook for jacket (dictation 2026-05-12)

• Fix/rivet mini portable workbench (dictation 2026-05-09)

• Get speaker for meadow (dictation 2026-05-18)

Due Now

• HONEY — finish filtering THIS MORNING (Sat 2026-08-01) — Pull the finest filter bag out slowly right before bottling. Hard deadline is the plane Tuesday morning. (dictation 2026-07-31 usb, 05-27-22 50:13) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)
• HONEY — bottle Sunday 2026-08-02 — ~5 gallons total, roughly fifty 12 oz bottles. (dictation 2026-07-31 usb) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)
• Three matched jars to the fair — Tracy submits them; one comes back. Fair jars must come from almost fully capped frames only. Ed had trouble even finding the entry category. (dictation 2026-07-31 usb, 07-27-23 01:44) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)
• Put the supers back on the hives — was “tomorrow” as of Fri 07-31, so due Sat 2026-08-01. (dictation 2026-07-31 usb, 07-27-23 12:50) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)
• Treat the hives — before the Tuesday flight. (dictation 2026-07-31 usb, 07-27-23 23:59) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)
• Set up the two sprinklers in the tree area before leaving Tuesday — manual, not on the controller. (dictation 2026-07-31 usb, 06-27-23 28:09) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)
• Fix the house internet this weekend — the interior access point isn’t hooked up. Ed: “it’s not hard to do.” (dictation 2026-07-31 usb, 04-27-21 52:43) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Pond aerators in, duck decoys out — within 10-11 days — small-boat trip planned. (inferred, dictation 2026-05-28 14:31, 80:38) (2026-07-28: Ed confirmed still wanted; do not prompt — he’s tracking the timing himself)
• Order deer-proof spring bulbs (July deadline) — Habitat plan (2026-06-11) identified snowdrops, Crocus tommasinianus, daffodils (500+), camassia, and drumstick alliums as the fix for the October forage cliff. Good naturalizing stock (Van Engelen, Colorblends, Prairie Moon) sells out by September. Full plan at ~/Sync/ED/2026-06-11 - Fall Bulb Plan (deer-proof).md. (added via doc-sync 2026-06-12) (2026-07-28: Ed confirmed still wanted; do not prompt — he’s tracking the timing himself)
• Reminder end of June: furniture for house (after cleaning done) (dictation 2026-05-13) (2026-07-28: Ed confirmed still wanted; do not prompt — he’s tracking the timing himself)

• Investigate recurring house→farm routing failure — Path breaks approximately monthly (confirmed June 11 + June 22 briefing). Symptom: all farm Kuma monitors EHOSTUNREACH. Manual fix: restart farm-route.service on CT100 + verify pve ufw route allow from 192.168.8.0/24 to 192.168.0.0/24. Root cause unclear — fix should survive reboots but doesn’t always. Options: (a) watchdog cron on CT100 that pings 192.168.0.5 and re-applies route on failure, (b) diagnose whether fpve reboot or CT100 restart is the trigger. [2026-07-26: PARTIAL COVER — farm-link-watchdog deployed on pve 7/22 pings fpve + farm Pi-hole every 2min and restarts pve netbird if both unreachable ≥12min. That handles the NetBird-drop failure mode from the pve side. The remaining gap is the CT100-side farm-route.service / ufw-rule loss (EHOSTUNREACH signature); option (a) as written — a CT100 self-heal cron — is still unbuilt. Recommend building it, see plan doc 2026-07-26.] (added via doc-sync 2026-06-24)

• Bob call — TONIGHT Fri 2026-07-31 — He said he’d call Friday night to schedule a Sat or Sun visit. Ed is away next week, so this is the window. Two items for him: (1) the stream conduit under the driveway (6-7 ft, rocks around it, put in by the Elwoods) — Bob does NOT remember it and wants to see it; (2) an overflow pipe for the new pond, like the one on the other pond — he has already agreed it can be done. Related: 07-30 recording covers a second tunnel at the crossing to push flooding to the far side instead of tearing out the driveway. (dictation 2026-07-28, ~48:00) (added via email 2026-07-31) (2026-08-01: that Friday has passed — the visit itself is today/tomorrow. Add to the agenda: the sheared tree over the creek near the beaver dam, which needs a lift before anyone can cut it. Status of Bob’s Friday call unknown.)
• Register Burgos Eclipse Watch as desktop-local scheduled task — Replacement SKILL written 2026-07-29 at ~/Documents/Claude/Scheduled/burgos-eclipse-watch/SKILL.md; Ed must register it via the Claude desktop app UI (do NOT hand-edit scheduled-tasks.json — the app owns that file and clobbers manual edits). Without registration the 12-hourly Spain wildfire/weather reports continue to silently fail; the existing cloud Routine is tool-starved by design. Eclipse is 2026-08-12. (added via doc-sync 2026-07-30)
• Register for Woodland Pollinators Workshop — deadline August 7, 2026; $20; go.osu.edu/woodlandpollinators2026. Event: Friday August 14, 9:00 AM–3:30 PM, Canter’s Cave 4-H Camp, Jackson OH. Calendar event already created (Tomato/RSVP color). (added via email 2026-07-14) (2026-08-08: registration deadline of Aug 7 has PASSED — it expired while Ed was in Spain, and no run was able to flag it in time because of the 08-04→08-08 spend-limit blackout. Registration status unknown. Event itself is still Fri Aug 14. Either call Jim Downs, 740-603-3224 / downs.63@osu.edu, to ask about late or walk-in enrollment, or drop this task and delete the calendar event. Do NOT leave it unchecked — same pattern as the Tree ID Workshop and Gravel Gardens 101 entries already rotting in Notes.)
• hpve SSH host key CHANGED — decide before departure — ssh hpve now fails with REMOTE HOST IDENTIFICATION HAS CHANGED. Key presented: SHA256:+TjGic4bo7rGCs63f30qiypZotf0TBuEJjCvwttb6Ds. Probably benign: that same fingerprint is already trusted in known_hosts for 100.123.31.199 (hpve’s own NetBird IP), pve, 192.168.8.221 and 10.10.10.2 — so the hpve name entry (known_hosts line 26, old key SHA256:4H51Cr5GJP1BJWIoqUhXWwrbXel6gmV6EfAthUNCw2g) looks stale rather than hostile. Not auto-accepted — trust decision is Ed’s. Fix once confirmed: ssh-keygen -R hpve then reconnect and verify. Related hygiene: that old fingerprint is bound to hpve, hpve.edmd.me, proxmox.edmd.me, proxmox-home, 192.168.8.54, beemacbook, opnsense, opnsense.edmd.me — one key across eight unrelated names; known_hosts needs a cleanup pass. Blocks any hpve-routed automation while Ed is away. (added via email-review 2026-08-03)
• Delete 5 duplicate Spain trip calendar events — 8 events exist where 3 should; all 8 re-verified still present on the Claude calendar 2026-08-04. Deletion is destructive so no automated run has done it — needs Ed’s OK. KEEP: 1tr6qupccf14ervn461d59kalc (Welcome Drink, Aug 7 19:30 CEST), fpdiu7e9c3nt9jsm0lofmcrb28 (Welcome Dinner, Aug 7 20:15), up7l54tl5kqfspltiaqgl9tnek (TOTALITY, Aug 12, full 19:33–21:21 window). DELETE: senng26k1irlcnb48bp3iu2aag (drink, wrong time — 17:30 not 19:30), 6ead32mc663v0p046307fb2kfo (extra drink), ueg2u9ob28g5n1qv8k0r21neuc (extra dinner), vh691uvm3it7e9c0bmv9p29sb4 (extra totality), mijg3cmadqamrtpp2sg33krvm8 (extra totality). Root cause + full detail in ~/Sync/ED/.claude-email-log/2026-08-03.md. Note: the Notes-section entry “Spain Eclipse trip calendar items — RESOLVED 2026-08-03, all three created” is now stale/misleading. (added via email-review 2026-08-04)
• Raise monthly Claude spend limit — All automated scheduled tasks hit the spend limit at ~19:05 EDT on 2026-08-04 and have been non-functional since; raise at claude.ai/settings/usage?from=cc_cli_limit_message. (added via doc-sync 2026-08-05) (2026-08-08: likely ALREADY RESOLVED — the 11:17Z eclipse-watch run and this 08-08 email-review run both completed normally. Verify and close.)
• ECLIPSE DAY SMOKE CONTINGENCY — Burgos, Wed 2026-08-12 — A wildfire ignited 2026-08-05 in the Sierra Norte de Burgos and as of 08-08 is uncontrolled: >4,000 ha burned, 2 fatalities, 12 aircraft + 400 ground crew, spreading toward Cantabria, no rain forecast. Burgos city AQI ~87 (Moderate); Euronews puts the plume ~80 km SSW. Sky itself is forecast clear through eclipse day (>80% clear-sky odds, heat dome locked over Iberia, highs 34–37 °C), so smoke — not cloud — is the threat. Sun is only ~8° above the WSW horizon at totality (20:28–20:30 CEST), so the sight line crosses a very long low-atmosphere path; thin smoke invisible overhead can extinguish the corona at that angle. Late-evening winds each day this week back to NW, which would carry northern smoke straight across the western horizon at totality. Action: watch the sunset from the intended site each evening (sharp yellow disc = clean; mushy deep-orange = smoke), track fire containment daily, and hold a fallback — drive 30–50 km south to Aranda de Duero or east to Miranda de Ebro — ready for eclipse day. ⚠️ Note the existing TOTALITY calendar events still say “Burgos province: ZERO active fires” (written 2026-08-02) — that is now out of date. (added via email-review 2026-08-08, from Burgos Eclipse Watch T-4d report)

Active — Errands

• Physical check Sonos/WiiM speakers — Sonos integration reload 2026-07-18 confirmed speakers are physically offline (not a software issue). Check Living Room Connect:Amp power and WiiM Equipment/Mini status; possibly powered down deliberately. (added via doc-sync 2026-07-19)

• Re-clip 2 damaged Notebooks library clips (after extension reload) — foxrunenvironmentaleducationcenter.org “Build a Hibernaculum For Snakes, Frogs, and Insects” (2026-07-10, confirmed 0 paragraphs — Squarespace body deleted by declutter bug) and badgerton.substack.com “Free Download: Make-Your-Own Field Guide Website” (2026-07-13, confirmed 0 paragraphs — Substack article deleted). PDFs are equally empty. (added via doc-sync 2026-07-18)

• Reload both web clipper extensions in Chrome for the declutter bug fix — chrome://extensions → Reload the Notebooks clipper (:8787) and Tana clipper (:8788). The content-loss fix (Dotdash/Squarespace/Substack sites were returning nav sidebar instead of article body) is in source but the browser still runs old code until both extensions are reloaded. (added via doc-sync 2026-07-18)

• Verizon (added via email 2026-05-03)
• Find the dryer on the upper lot and remove it (dictation 2026-05-13)

Deferred — Calls

• Contact Eric Mitchell (builder, Hanover OH) — custom home builder recommended by Candy (Jordan’s wife). For future hilltop house project. Got his phone number from Jordan 2026-05-15. (dictation 2026-05-15)

Active — Shopping / Buy

• Order bulk native vine seeds from Sheffield’s Seed Co. — Ed indicated intent to order during 2026-07-09 vine seed research session. Sheffield’s (sheffields.com, Locke NY) has 6 target species in stock: Maypop (Passiflora incarnata), Virgin’s bower (Clematis virginiana), American bittersweet (Celastrus scandens), Virginia creeper (Parthenocissus quinquefolia), riverbank grape (Vitis riparia), trumpet creeper ‘Flava’ (Campsis radicans). Pull bulk-per-weight pricing before ordering; coral honeysuckle already on property (#330), groundnut better from tubers. (added via doc-sync 2026-07-10)
• Ikea order — furniture for Amber’s daughter’s room — at least one room in the house needs desk, chair, etc. so Amber’s 9yo daughter can entertain herself while Amber works. Ed planning big Ikea purchase in next 2-3 weeks. (dictation 2026-05-21) (2026-07-28: Ed confirmed still wanted; do not prompt — he’s tracking the timing himself)

Active — Property / Garden

• Reset the weather-station pole in a sleeve — cut the damaged pole off at ground level and sleeve the replacement so the whole pole lifts out for servicing. Ed hit the pole on 07-30 and took the station off the top. (dictation 2026-07-31 usb, 06-27-23 25:08) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Spool the 200 ft light coil hose so it can be rolled anywhere and rolled back in 20 seconds. Current hose leaks badly at the juncture and the fittings are still lying out. (dictation 2026-07-31 usb, 04-27-21 38:03) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Decide greenhouse size and order it — 48 ft was plotted and is too big; leaning ~30 ft, with lemon and lime trees, sited on the pad by the bent hydrant. Waiting on Chris’s foundation number (quoted ~$30k for the greenhouse footings; the barn pad was ~$8k). Ed is holding the order until the contractor’s number comes back so it doesn’t land at an “astronomical price.” (dictation 2026-07-31 usb, 06-27-23 26:58) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Hire a pro for the sheared tree over the creek — The big tree sheared off and a second trunk is perched on top of it. Ed’s own words: nobody should walk up to that with a saw. Do not DIY. (dictation 2026-07-30, storm damage walk near the beaver dam) (added via email 2026-07-31)

• Register for “A DAY in the WOODS” Shrub ID Workshop by Aug 21 — Fri Aug 28, 9:00 AM–3:30 PM, University of Rio Grande, OH. $20 (lunch + materials). Register at http://go.osu.edu/aditw, call 740-286-5044, or email miller.11368@osu.edu. Calendar event created. (added via email 2026-07-20)

• Add missing blueberry varieties to farmdb — Only Bluecrop (#223) is in farmdb; BerryBux ×2, Perpetua ×2, and second Stark Bros order (Jersey ×3, Jubilee ×3, Misty ×3) are all absent. Ed to confirm what actually arrived and Claude can add the records. (added via doc-sync 2026-06-23; properly placed 2026-06-30) Planting / Time-sensitive:

Planting / Landscaping:

• Plant black-eyed Susans in far corner of garden for color (dictation 2026-05-13)
• Source native Penstemon for pollinator habitat — Penstemon digitalis (foxglove beardtongue) and P. hirsutus (hairy beardtongue) both native to Ohio/Licking County. Full sun to part shade, deer-resistant, self-seeding. (2026-05-12)
• Start catmint from seed — direct sow November or start indoors mid-February. (flagged by doc-sync 2026-05-06)
• Till lower meadow edge — plant deer-resistant wildflowers to dissuade deer from jumping fence (dictation 2026-05-13)

Research / Planning:

• Build manifold and dust collection for shop area (dictation 2026-05-21)
• Make propolis tincture from bee frames — Ed has frames with heavy propolis buildup. Research extraction/tincture process. Propolis goes for ~$60/oz as tincture. (dictation 2026-05-21)
• Set up portable watering system for meadow trees — Need to water new trees planted along meadow borders. Plan: 110-gallon barrel in buggy bed, watering hose, enough for ~40 trees per run. Also need wood chips around tree bases. Consider 265-gal barrel vs IBC tote. (dictation 2026-05-19)
• Extend Wi-Fi further into property — High priority infrastructure item. Need coverage up into the meadow area for tech stack (sensors, cameras, Zigbee water sensors). Possibly cable into the new shed. (dictation 2026-05-19)
• Research fruit tree planting method — video (possibly Epic Gardening) showed compost + bone meal + osmocote in augered holes. Ed already dug 8 holes (36" dia) with auger. (dictation 2026-05-13)
• Research Ohio fruit tree varieties — 4-5 apple varieties, plus peach, pear, fig, plum, pawpaw. Buy 2-3 year specimens. (dictation 2026-05-13)
• Investigate alyssum varieties — different colors? Bulk seed sources? Cheapest supplier for large quantities. (dictation 2026-05-13)
• Make well cap (dictation 2026-05-13)
• Review overall property plan — “nail down what the plan is” (dictation 2026-05-13)

• Care schedule v2 — quarterly review email + dictation-learning hook — Two enhancements layered on the v1 shipped today. (A) Quarterly review email: scheduled task fires at start of each season (Mar 1, Jun 1, Sep 1, Dec 1), composes email of “proposed items for [season]” pulling from any new dictation cues + season-of-year matches, sends via msmtp. Ed replies, email-review skill catches the reply, parser updates the source file. (B) Dictation hook: extend parse-dictation.py so utterances like “I waxed the bee frames today” fuzzy-match to a schedule item → set **Last done:** YYYY-MM-DD AND append dated snippet to that item’s notes block (Tier 1 + Tier 2 learning). Records edit history in a sidecar JSON over time. (added 2026-05-27, build when v1 has 4+ weeks of real use)
• Care schedule Tier 3 — auto-refine rules from accumulated history — Future enhancement after v2 is live. After 3–6 months of real last_done + notes data, build a pass that proposes rule changes (“you’ve done X 5 times averaging 6 weeks apart but rule says every 4 weeks — push to every 6?”). Output: list of proposed diffs Ed approves like PRs. Don’t auto-apply. (deferred 2026-05-27)

Sweet alyssum deployment — 2 lb supply on hand:

Irrigation / Infrastructure:

• Irrigation drip layout — 92 berry plants, YoLink valve controllers for Home Assistant integration.
• Z-Wave setup on farm HA — SLZB-MRW10U at 192.168.0.16 port 6638. Left unresolved due to prototype/non-certified firmware.

Farm DB: (see TASKSDONE.md for Achillea ‘Cinnamon’ ID 277 and Hakuyoh peony ID 278)

Active — Ponds / Fish

Nomenclature: Upper pond = Pond 2 (0.86 acres, trophy bass pond). Lower pond = Pond 1 (stunted bass).

Recent stocking (2026-05-19): Upper pond got bluegill + minnows + sunfish. Lower pond got minnows + 2 grass carp.

• Install pond aerators in lower pond (Pond 1) — investigate how to install. Ask at Fender’s. (dictation 2026-05-13, clarified 2026-05-20) (2026-07-28: Ed confirmed still wanted; do not prompt — he’s tracking the timing himself)
• Set up pond journal/monitoring system — Ed wants per-pond tracking of fish stocking, observations, and conditions. Possibly per-section property journals. Format TBD (Tana, farm DB, or markdown). (dictation 2026-05-13)

Active — Bees

• Second extraction pass on the uncapped frames — run a fan + dehumidifier over the honey first to drop the water content. (dictation 2026-07-31 usb, 06-27-23 04:29) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Melt down the cappings wax (dictation 2026-07-31 usb, 06-27-23 55:43) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Find the second escape board — Ed thinks he carried it inside. (dictation 2026-07-31 usb, 07-27-23 57:31) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Mite context to pair with the varroa wash above — several hives had capped very little on 07-31, which Ed called “disturbing.” Production down overall; first spin came in under 3 gallons against a book estimate of 3.5. (dictation 2026-07-31 usb, 06-27-23 38:38) (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Source a queen — KP Honey Farms had none this late in the season. One hive needs pinching. (dictation 2026-07-28) (added via email 2026-07-31)

• Do a varroa mite wash — Ed: “I probably have a mite problem.” Production is down this year; the hive move is also a suspect. (dictation 2026-07-28) (added via email 2026-07-31)

• Make labels and number beehives — 7 hives currently unlabeled. (dictation 2026-05-09)

• Create bee calendar/tracking system — Ed wants a separate bee calendar. Decide format (Tana, standalone app, etc.). (dictation 2026-05-09)

Active — Stable / Barn

• Stable plumbing — fix pipes, get utility sink or food prep station (check FB Marketplace), plumb in with on-demand propane heater, add shower/wash area and Amigo wash station (dictation 2026-05-13)
• Stable shelves — for bee equipment (dictation 2026-05-13)
• Change lights in stable (dictation 2026-05-13)
• Put up bat box (dictation 2026-05-13)
• Gravel along side of stable — place boards first (dictation 2026-05-13)
• Seal fragmented insulation — mice damage, ~10-25 sq ft area with foil backing. Find spray coating. (dictation 2026-05-13)
• Pull Ethernet cable to back of barn for Wi-Fi camera to observe bird feeders (dictation 2026-05-13)

Active — House

• Squirrel-proof the seed container in the garage — Squirrels are getting in through the doggy door and opening the lid. A weight on the lid would do it. (dictation 2026-07-28) (added via email 2026-07-31)

• Wet corner of house — rotted wood (Granville house, not Brownsville) — Ed and Lois noticed a corner that stays wet with significant wood rot. “We need to fix that.” Possibly related to overall water leak or separate issue. (dictation 2026-05-21)

• Hire house cleaner — inside of house and springhouse (dictation 2026-05-13)

Active — GIS / Mapping

• Export project 7 Emlid data from Emlid Flow — June 4 survey session (Amber’s training day); points still on-device, not yet exported. No dash-tags used (teaching demo), so species correlation must come from label photos taken that day. Export before next session uses the device. (added via doc-sync 2026-06-05)

• Re-home 19 Lower Lot reference/infrastructure points — Foundation Rock (×multiple), Corner Drain, drainage points currently on location_id=9 (Lower Lot). Re-parent to specific building locations (House, Cattle Barn, Equipment Barn, etc.) once boundaries are surveyed or Ed identifies which feature belongs where. (added via doc-sync 2026-06-04)

• Review location-nomenclature-design.md open decisions — Ed: “save this whole thing to a file to tackle later this week.” Doc at ~/Sync/ED/farm/location-nomenclature-design.md. Open decisions: (a) form enum on plantings for row/patch/individual/bed-member; (b) whether row/patch belongs as location_type or plant attribute; (c) feature_type retrofit for gis_features; (d) container boundary surveying plan. (added via doc-sync 2026-06-04)

• Fly DJI Mini Pro drone orthophoto of property — Capture high-res aerial imagery of all 93 acres for QGIS background layer. Steps: (1) plan grid flight, (2) fly at ~250-300 ft AGL with 70-80% overlap, (3) stitch with OpenDroneMap or WebODM, (4) georeference and load GeoTIFF into QGIS. Will take 3-5 battery swaps. Best conditions: midday overcast, low wind. (added 2026-05-19)

• Build field observation workflow (Emlid snap-to-plant) — Design and build an Emlid observation session type where new GPS points snap to the nearest existing plant in PostGIS and write a plant_observations record parsed from voice memo content. Ed approved this (‘Yes to 3’) during the 2026-05-19 session. (added via doc-sync 2026-05-19)

• Collect more Emlid data — for existing survey points (dictation 2026-05-12)

• Determine Emlid project workflow — Should Ed start a new project each day or continue one project? (dictation 2026-05-07)

• Emlid point-to-point navigation — how to navigate to a specific point relative to another using Emlid RS3 (dictation 2026-05-13)

• Research iPhone GPS accuracy for continuous tracking — Ed wants to know if phone GPS can correlate with dictation timestamps and supplement Emlid data. (dictation 2026-05-09)

Active — Infrastructure

• Investigate Bee Hub Deploy push monitor DOWN — Daily briefing 2026-07-31 flagged the Kuma ‘Bee Hub Deploy (push)’ monitor as DOWN with no heartbeat; hub.edmd.me itself is healthy. Prior session log notes deploy-vps.sh pushes its Kuma heartbeat only on a fully-successful two-target deploy and ends the call in || true, so a single-target failure is silent. Check which of the two targets (VPS 100.123.69.155 or CT103 192.168.8.54) is failing and whether the heartbeat push is reaching Kuma. (added via doc-sync 2026-08-01)
• Fix dictation-run preflight: it reported ~/Sync/ED, ~/Sync/farm, ~/Sync/SyncConfigs as deleted and Syncthing folders as paused on 2026-07-31 (4 alarm emails). Verified false — ED holds 6,975 files / 5.33 GB, all folders active, all named scripts present. It checks wrong paths (ED/scripts/process-dictation.sh vs ED/dictation/, ED/TASKSDONE.md vs ED/.claude/) and fabricated both a failure-report path and a list of un-transcribed recordings. Must distinguish unreadable-mount from deleted, and never advise Syncthing re-add without confirming local file count first. (added via email 2026-07-31)

• Re-enable Pi-hole over NetBird with a public fallback — during the 2026-07-22 Verizon-FWA outage the Pi-hole nameserver group was disabled (it was catch-all DNS for all peers and blackholed everything when its host dropped). Once the primary line is back, re-add it in NetBird DNS with a second nameserver (1.1.1.1) OR scope it to match-domains only, so a downed peer cannot take every device’s DNS again. (added 2026-07-22)

• Investigate fpve host-key change — During 7/20 session, SSH to fpve (farm Proxmox, 100.123.49.175) returned REMOTE HOST IDENTIFICATION HAS CHANGED (known_hosts line 45). Ed redirected to CT100 without confirming. If intentional rebuild: run ssh-keyscan 100.123.49.175 to get new key, update ~/.ssh/known_hosts, then check farm Pi-hole (.5) pihole.toml — its ha.edmd.me entry may still point to 192.168.0.10 instead of 192.168.0.54 (the house Pi-hole fix ran 7/20 but farm Pi-hole was inaccessible). If NOT an intentional rebuild: key change is unexplained and warrants investigation. (added via doc-sync 2026-07-21)

• Determine Zigbee2MQTT status + clean up orphan HA entities — HA investigation 2026-07-18: no Z2M process found on house CT100, farm CT100, farm VMs, or Mac Studio. Three MQTT-discovery bridge entities (z2m/bridge version, permit-join, restart) are orphaned in HA. If Z2M is retired: delete the 3 entities via HA developer tools. If still running: identify host and reconnect HA MQTT integration. Ask Ed: ‘Is Zigbee2MQTT still running somewhere?’ (added via doc-sync 2026-07-19)

• Investigate VPS Ping DOWN — Daily briefing 2026-07-18 flagged 172.93.50.184 (SSDNodes VPS, Hermes agent host) at 100% ICMP packet loss via Kuma. All edmd.me endpoints were serving, suggesting ICMP filtering; but if the VPS is genuinely unreachable, Hermes and other VPS services are offline. Confirm with a real TCP check (e.g. nc -zv 172.93.50.184 22). (added via doc-sync 2026-07-19)

• Complete Hermes Telegram gateway + systemd + Nous Portal — Ed requested during 7/13/14 session (‘Wire it up with the telegram, systemd Noves portal’). Work began but ended mid-flow: (1) Ed creates new Telegram bot via @BotFather (NOT @beenetworkbot — that conflicts with Gotify bridge) and provides token; (2) Claude re-runs Nous Portal device-code flow (prior code PB3F-662K has expired). Hermes runs as hermes user on root@100.123.69.155. (added via doc-sync 2026-07-15)

• Check farm Equipment Outdoor AP + Cattle Outdoor AP hardware — both APs offline (status=0) since 2026-07-11; Equipment Outdoor did not re-adopt after power-cycle 2026-07-11. Suspected PoE injector/cable or hardware fault on both. Check LED state; inspect ES205GP PoE switch ports and cabling. Until fixed, Equipment building (fpve + Pi-hole + HA) has no reliable wired uplink. (added via doc-sync 2026-07-13)

• Install Crane Apple Container GUI — Ed agreed to install lcandy2/container-ui v0.1-b3 (requires macOS 15+; apple/container 0.6.0 already installed on Tahoe 26.6). Download ContainerUI.zip from https://github.com/lcandy2/container-ui/releases/latest, unzip, drag to /Applications, right-click → Open (app not notarized, double-click will block). (added via doc-sync 2026-07-13)

• Investigate recurring vzdump failure on CT100 — vzdump wedged CT100 at 02:00 on BOTH 2026-07-10 and 2026-07-11, holding the CT100 config lock for 26+ hours each time; CTs 101/102/103 backed up fine. Root cause unknown (I/O freeze from missing mediapool2 drives during snapshot, storage configuration, exclusion misconfiguration). Diagnose before next 02:00 run; consider disabling CT100 vzdump temporarily until mediapool2 is repaired. (added via doc-sync 2026-07-12)

• Farm observability: Omada health poller + farm-snapshot.json — Cron script on fpve (or farm CT100) polling the Omada API (:443; use learned semantics: minimal-body PATCH, channel = 1-based index into channelRange, config reads None while provisioning) every 5–10 min. (a) Alert on AP isolated/disconnected/heartbeat-missed via existing HA webhook kuma-farm-alerts-x7q2m9 → iPhone push; (b) write farm-snapshot.json (AP/switch/gateway status, mesh uplinks, RSSI, uptimes, fpve + CT basics) to a Syncthing’d path so Claude sessions read farm state in one call — farm twin of .homelab-snapshot.json. Would have caught Equipment Outdoor AP down within minutes (it sat unnoticed ≥1 day, 7/10–7/11). Bake API auth/semantics into the script per bundle rule #21. (added 2026-07-13, Claude-effectiveness review)

• Characterize ER8411 Omada QoS for Starlink SQM deployment — Farm gateway confirmed as TP-Link ER8411 (Starlink router is NOT in the traffic path — CGNAT 100.64.0.1 hits the ER8411 directly). Before implementing SQM: (a) confirm whether ER8411 Omada QoS supports HTB/FQ-CoDel/CAKE or only simpler token-bucket scheduling; (b) derive viable rate cap from accumulated bufferbloat samples (bufferbloat-sample.sh cron running 4×/day on fpve since 2026-07-08; first peak = 10 Mbps down, 10% loss under load, no latency bloat). Samples tabulation scheduled via send_later 2026-07-09 13:40 EDT. (added via doc-sync 2026-07-09)

• Seedbox drain: confirm delivery completes (UPDATED 2026-06-11 — reaper guard v2) — Checked 6/11 PM: pipeline IS moving (seedbox→ingest→Radarr import→reap; BTTF III imported+reaped today), BUT (a) the 6/10 safe_to_reap() guard was STILL killing live transfers — it aged by mtime, which rsync -a preserves from ancient release files, so fresh mid-pull dirs looked years old (Indiana Jones, Interstellar, It’s A Wonderful Life all DEL’d mid-write 6/11) — reaper v2 deployed: ages by max(ctime, mtime) (ctime is local, not preserved) + min_age escalates 24h→72h while CT100’s seedbox-sync flock is held; live-verified reaped 0 / SKIPped all in-flight dirs (bak ingest-reaper.py.bak.20260611); (b) quota back to 98% (2738/2794G, was 2620G) because Radarr keeps grabbing — completed/Movies now 83 dirs / 2.3T + intermediate 366G/31 jobs. Multi-day drain continues; self-balances (nzbget stalls when full, rsync frees, Transmission has Restart=always). WATCH: quota at next check; Kuma 57 heartbeat resumes on first <6h pass (won’t happen until backlog clears). Pressure valve ENGAGED 2026-06-11 PM (Ed’s call): all 5 Radarr indexers paused (RSS + auto-search off; interactive search intact; prior state at CT100 /root/radarr-paused-indexers-20260611.json). Re-enable is AUTOMATED: CT100 cron radarr-unpause-check.py (every 6h at :40) measures seedbox completed/Movies; when < 100 GB it restores the indexers from saved state, Gotify-notifies, and self-retires (fail-loud: Gotify p8 after 3 consecutive check failures; log /var/log/radarr-unpause-check.log). First run verified: 2262 GB, no action. UPDATE 2026-06-12 (late): the leak around the indexer pause was found and plugged — Radarr’s PendingReleases table held 218 cached pre-pause releases that kept grabbing as their holds matured (releaseSource=Rss, no indexer needed) → ~60 grabs on 6/12 alone. Purged 218→0 (db bak radarr.db.bak.20260612 in the radarr-config docker volume) + NZBGet on the seedbox PAUSED (live RPC = 127.0.0.1:13036, creds in ~/.apps/nzbget/nzbget.conf; conf’s ControlPort=6789 is NOT the live port). ⚠️ WHEN DRAIN COMPLETES: resume NZBGet ("method":"resumedownload" to the same RPC) — the unpause-check cron does NOT do this. Also seedbox-sync.sh patched (bak .bak.20260612): dropped -z, and a 6h-cap timeout (exit 124) now counts as success when >1GB moved — kills the false “Seedbox sync stuck” alerts during multi-day drains. Link rate ~6-9 MB/s (VPN-bound) → 2.2TB ≈ 3-4 days. (updated 2026-06-12)

• Decide: configure rclone gdrive remote for farmdb offsite? — backup audit 2026-06-10 found the gdrive remote was never configured, so backup-farmdb-gdrive.sh has only ever made Mac-local copies. One-time rclone config on the Studio enables the offsite leg; script already handles the rest. (backup audit 2026-06-10)

• Fix edmd.me remote access from farm — [DEFER 2026-07-26: NetBird covers the real need (CTs reachable via fpve route); revisit only if public-subdomain-from-farm becomes a real requirement] Farm NetBird routing confirmed working as of 2026-05-25 (CTs 100/102/103 all reachable via fpve’s route after LXC reboot). HA at 192.168.0.10 separately offline (physical power/cable issue, not a routing problem — see HA check task). Remaining gap: public edmd.me subdomains resolve to 192.168.8.54 (CT103 Caddy, house LAN IP), unreachable from farm LAN. Options still valid: Cloudflare Tunnel for public access, split-DNS, or Caddy block on fpve NetBird IP (100.123.49.175). (added via doc-sync 2026-05-17, rechecked 2026-05-25)

• Offsite backup to farm (waiting on farm infra) — ZFS replication of irreplaceable data from home → farm Proxmox via NetBird. 3-2-1 rule.

• Loki + log aggregation: enrichment — Loki + Alloy deployed, shipping journald to 192.168.8.100:3101. Pending: Docker container log scraping on CT100 (needs loki.source.docker block), Caddy access logs, Pi-hole query logs, parse log levels.

• Disaster recovery procedures doc — prioritized rebuild sequence at ~/Sync/ED/homelab/bee_hub/content/homelab/disaster-recovery/_index.md. Print a copy and store fireproof.

• Metrics dashboards (Prometheus + Grafana) — both running but unused. Add node_exporter everywhere, build host-level dashboards, per-service dashboards.

• Network monitoring (NetFlow / sFlow) — OPNsense NetFlow export → ntopng or Grafana.

• Container vulnerability scanning — Trivy weekly against all 35+ Docker images. Cron with Gotify output.

• Two-person/break-glass access plan — document where master passwords are, how trusted person gets in. Print + fireproof.

Active — Claude / MCP

• Fix ledger-backfill is_substantive() filter — Current 2+ user-message threshold counts automation sessions (email-review produces exactly 2 user messages) as substantive, generating ~37 noise stubs in .claude-session-log/. Correct filter: detect <scheduled-task name="..."> in the opening user message; human sessions lack this prefix. Claude offered to implement during 2026-08-04 interactive session; Ed’s confirmation was implicit. (added via doc-sync 2026-08-05)

• Fix email-review dedup: upgrade from thread-level to message-level claiming — Thread IDs in processed-threads.txt suppress all later messages in a thread; three second-messages in claimed threads were silently skipped on 2026-08-03. Change the claim key to message ID so replies on already-claimed threads are still evaluated. (added via doc-sync 2026-08-04)

• Fix email-review skill: query calendar before create_event — Before creating any calendar event, call search_events for a matching title+date; skip creation if a match exists. This would have prevented the May 14 (~48 dupes), May 29 (~110 dupes), and Aug 3 (×8 Spain trip) incidents — all three caused by runs trusting a stale ’not yet calendared’ note rather than checking the live calendar. (added via doc-sync 2026-08-04)

• Fix dictation-ingest.py fabricated-diary bug — When background media (TV/radio/YouTube) plays near the USB recorder, the ingest script summarizes it as Ed’s own first-person diary narrative. Confirmed 2026-08-01: a YouTube small-engine-repair video was written up as Ed’s own day in Arkansas. Needs detection/exclusion of background-media speech before diary generation, or narrator-voice matching. Annotated corrections written to Notebooks but root cause unfixed. Six follow-up questions in ~/Sync/ED/dictation/processed/2026-08-01_usb-recorder.md. (added via doc-sync 2026-08-02)

• Add a dash recognize music handler to the dictation parser — Ed issued the command at 74:26 on 2026-07-28 (“recognize music — this particular song that’s playing, I’d like to be put on a new Spotify playlist called Recent”). The parser has no handler for it, so it would be missed even with a working pipeline. (added via email 2026-07-31)

• Identify the 2026-07-28 song and build the “Recent” Spotify playlist — Whisper garbled the sung lyrics. The track at 74:26 had lyrics like “I believe in you baby / and I’ll always be around / with the sun shining through to the break of dawn.” The NEXT song was clearly Dua Lipa’s “Levitating.” Needs Ed to confirm which one he meant. Playlist was NOT created. (added via email 2026-07-31)

• Process the 4 remaining unprocessed recordings (6/27, 7/19 x2, 7/22, 7/25) — Held back because the processed-ledger was reported missing; re-running blind risks duplicate tasks and diary entries. Verify the ledger, then run. Note 2026-07-27/11-52-52.m4a is corrupt (117 KB, no moov atom) — nothing to recover. (added via email 2026-07-31)

• Reconcile the dictation pipeline’s false “~/Sync is gone” reports — Three runs on 2026-07-31 (11:13, 12:12, 13:32 EDT) reported ~/Sync/ED empty and TASKS.md missing. Verified FALSE on the Mac Studio: ~/Sync/ED is fully populated and TASKS.md exists (72 KB, modified 10:08 EDT 7/31). The 13:32 run also claims it wrote ~/Sync/ED/dictation/processed/PENDING-TASKS-2026-07-31.md and 07-28/07-30 reports — none of those exist on disk (latest processed file is 2026-07-25). Those runs appear to be writing to an ephemeral/sandbox filesystem view rather than the real one. Separately, that run reports NetBird is advertising no routes on the Studio (192.168.0.35 cannot reach the 192.168.8.x homelab) — worth checking independently. (added via email 2026-07-31)

• Dictation runs are reporting from a different machine than the Mac Studio — and now advise a destructive restore. The 16:13 EDT 2026-07-31 email ([Dictation] Correction — Time Machine has the missing ~/Sync data) states the host is “a MacBook Pro (M3 Max, 128 GB, serial V00NM76YY3) just NAMED Mac-Studio-2” on 192.168.0.x with its Time Machine drive unplugged. Verified on the real Mac Studio this run: ComputerName Mac Studio, hw.model Mac15,14, en0 192.168.8.180, Time Machine destination present and mounted. ~/Sync/ED (93 entries), ~/Sync/farm and ~/Sync/ED/homelab/bee_hub all intact; TASKS.md 76,345 bytes. The report it cites (~/Sync/ED/dictation/processed/2026-07-31_pipeline-failure-report.md) does not exist. Risk: the email tells Ed to restore ED/farm/SyncConfigs from a Jul 28 Time Machine backup — doing that here would roll back three days of good data. Do not follow it on the Studio. Confirm which host the process-dictation scheduled task actually executes on before acting on any of its filesystem claims. One claim does check out on both machines: process-dictation.sh, parse-dictation.py and whisper-transcribe.py are genuinely absent from ~/Sync/ED/scripts/. (added via email 2026-07-31) UPDATE (17:15 EDT email, 6th false report today): the run claims the USB voice recorder at /Volumes/USB-DISK holds 8 MP3s in RECORD/ totalling ~5h04m stamped 2026-07-06, plus a SETTINGS.TXT reading 05/18/2026, and offers to transcribe them. Verified read-only on the Studio: USB-DISK is mounted, but contains only MRECSET.TXT and MUSIC.LIB — no RECORD/ directory, no SETTINGS.TXT, and zero .mp3 files anywhere on the volume. This is the cleanest disproof yet: a removable device present on both hosts with completely different contents. The same two report files (2026-07-31_pipeline-failure-report.md, PENDING-TASKS-2026-07-31.md) were cited again and are still absent. Also unverifiable here: a claimed backup-brain.sh run on Jul 27 23:31 (rc=0) pushing /Volumes/Brain/ to hpve:/mediapool/brain-backup/ — the script exists (~/scripts/backup-brain.sh, Jul 27 20:59) but nothing in ~/Library/Logs corroborates the run. No audio was transcribed and no decision was made on Ed’s behalf — that email asks him a real question (transcribe the full set? under what date? Diary or report only?), and it should not be answered until the host question is settled, since the files may not exist at all.

• Audit the dictation/sync alerting task — 3 fabricated CRITICAL alerts on 2026-07-31. It sent [URGENT] 07:13 and [Dictation] CRITICAL 08:12 EDT claiming ~/Sync was recreated empty at 07:05:20 and that TASKS.md, process-dictation.sh, farm and SyncConfigs were gone. All verified false: /Sync birth 2024-11-28, ED birth 2026-02-18, farm 2026-05-06, SyncConfigs 2026-03-27, all present and populated. It also claimed 4 dictation recordings (07-19 x2, 07-22, 07-25) were never transcribed — all four have transcripts AND processed/ ledger entries. Both alerts cited full incident reports that were never written (/INCIDENT-2026-07-31-sync-ed-data-loss.md, ~/Downloads/DICTATION-INCIDENT-2026-07-31.md). Disable or fix before it fires again. (added via email 2026-07-31)

• Dictation inbox cleanup misses .m4a files. dictation-ingest.py globs and cleans only .mp3/.MP3, so transcribed .m4a recordings never leave ~/Sync/ED/dictation/inbox/. Currently stale there: 2026-07-22 (16 MB), 2026-07-25 (38 MB), 2026-07-28 (43 MB), 2026-07-30 (405 KB) — all transcribed — plus 2026-06-16 mp3 (159 MB) transcribed but not removed. ~255 MB. (added via email 2026-07-31)

• Add briefing_today invariant to check-invariants.py — Assert that a Morning Briefing entry was written today; check ~/Sync/ED/.notebooks-mirror/Morning Briefing/ for a file dated today. Flagged as a buildable-on-go item by the 7/26 fable-5 session alongside doc-sync prompt-trim, CT100 self-heal watchdog, and gitleaks. (added via doc-sync 2026-07-27)

• Delete legacy Tana #task supertag — Old #task supertag (X1pD1uBmGm2Q) still exists after 7/20 Tana todo rebuild; the new #todo tag is the active system. The API can remove a tag one node at a time but cannot bulk-delete the supertag definition. Requires Tana UI: open Supertags browser → find #task → Delete. Also: old #todo definition (pwltksqlNi) was trashed during the session but may still show in search — verify it’s gone. (added via doc-sync 2026-07-21)

• Fix parse-dictation.py JSON overwrite bug — On days with multiple recordings, parse-dictation.py overwrites {date}_parsed.json per source file; last recording’s data clobbers earlier ones. Discovered 2026-06-18: 3 bird-feeder + 2 mason-bee-house GPS tags from first two recordings were lost, recovered only from run log. Fix: accumulate all dash_commands/inferred_actions across recordings and write one merged JSON per date. (added via doc-sync 2026-06-19)

• Verify Nano Banana Pro icon re-batch applied to Stream Deck ED profile — Ed found the first icon batch (regular Nano Banana) grainy/low-res. A Pro-quality re-batch was launched in background with auto-apply+restart script. Check that all 8 pages of the ED profile show sharper icons; if the batch stalled, re-run the ~/icon_out_pro/ batch manually. (added via doc-sync 2026-06-16)

• Set Chaptarr UI login credentials — Navigate to http://192.168.8.100:8788 and configure auth. Backend fully wired (NZBGet .100:16789 tested pass, 4 Prowlarr indexers synced, /books root folder, rename OFF, health 0). UI is gated until first-login set — required before Chaptarr vs. Bookshelf comparison can begin. (added via doc-sync 2026-06-15)

• Update ha-mcp addon 7.6 → 7.7 — Deliberately skipped during 2026-06-11 HA cleanup because a session was running through it. Click Update in HA UI or: ssh ha 'source /etc/profile.d/homeassistant.sh; ha apps update 81f33d0f_ha_mcp' (doc-sync had the wrong slug — 45df7312 was Z2M’s repo prefix; corrected 2026-06-12). (added via doc-sync 2026-06-12)

• Claude Inbox enhancements — beedifferent5455@gmail.com universal inbox is live. Remaining:

  • Siri Shortcut for voice input — “Hey Siri, tell Claude…” → email to beedifferent5455
  • RSVP deadline reminders — text Ed 48 hrs before deadlines
  • Photo/image processing — plant ID via iNaturalist, OCR receipts

Active — Claude effectiveness improvements

Items from the 2026-05-02 brainstorm. Tier-1 shipped; these remain in roughly priority order.

• Verification subagent for dictation outputs — After report+diary are written, a fresh-context subagent reviews them against the raw parsed JSON + transcript: date rule compliance, dash false-positives (e.g., teaching demos), task-vs-diary misclassifications. Cheap insurance on every run. (identified 2026-06-05)

Active — Life Archive / AI

• Set up speculative decoding on Mac Studio LM Studio — pair a large model with a small draft model for 1.5-3x inference speedup. Also check if models are MLX vs GGUF — switching GGUF→MLX gives another 1.5-2x.
• Load pre-computed vectors into LanceDB — _vectors.json files already generated. Confirmed 2026-05-27 at /Users/bee/Sync/ED/life_archive/data/embeddings/: epub_articles_vectors.json exists, alongside evernote_notes, evernote_pdfs, paperless_docs, paperless_docs_new, tana_nodes, magazine_articles. LanceDB folder ~/Sync/ED/life_archive/lancedb/ is EMPTY — nothing loaded yet. Run python load_lancedb.py — do NOT re-embed. (rechecked 2026-05-27)
• RunPod contextual re-embedding — re-embed all 7 sources with contextual chunking (35-50% retrieval improvement). B200 pod, 500GB disk. Est. 4-6 days + ~$50-80.
  • Create pod, deploy, run pipeline
  • Monitor with pod_monitor.sh (auto-SCP each file to Mac)
  • After completion: load_lancedb.py, load_knowledge_graph.py, eval_quality.py
  • Terminate pod
• Switch Paperless-AI to faster model — current qwen2.5-vl-32b at ~9-10 docs/hr, ~38K remaining (~175 days). Switch to Qwen2.5-14B-Instruct Q4_K_M → ~35-40 docs/hr (~42 days).

couldn’t reach Tana logged one line and retried forever, and Ed would only notice when an article wasn’t in Tana. Added to ~/Sync/ED/scripts/singlefile-tana/singlefile_to_tana.py (+93 lines, all additions, no deletions): a gotify() push using --form-string (NOT -F — that truncates, per the 2026-07-27 finding), a persisted alert state at alert-state.json, and three triggers — (1) Tana unreachable ≥15 min → one alert, then 6 h cooldown; (2) a clip sitting unprocessed ≥15 min → one alert per file; (3) partial import (some nodes failed) → alert, previously a macOS-only desktop notification that Ed would miss if away from the desk. State self-clears when Tana returns or the file files successfully. Service restarted and verified running (log shows watcher starting 09:07:12); test push confirmed delivered to Gotify (msg id 29180, full message, untruncated). Ed declined domain→tag routing and Feedly auto-clip as polish. **REVISED same session after Ed asked the right question — “if Tana is not reachable, why can’t pages just accumulate and be processed when it is available again?” They already do: process_file() returns False on a failed Tana probe BEFORE archiving or importing anything, so the file is left untouched in ~/Downloads, never marked handled, and retried every 5 s poll. Confirmed by reading the call order and by the log showing zero TANA-DOWN events ever. So the Tana-down alert was solving a non-problem, AND it introduced a bug: the per-file “stuck” alert would have fired separately for every queued clip during one routine Tana restart. Fixed: TANA_DOWN_MINS raised 15 min → 6 h (now means “down so long you may wrongly assume clips are landing”, priority dropped 7→5, wording says nothing is lost), and note_file_seen() now returns early while tana_down_since is set so per-file alerts can’t pile on. The genuinely useful trigger is the remaining one — a clip failing for a reason that ISN’T Tana, which retries forever and would never otherwise surface. (2026-07-28)

Active — Security / Credentials

• Rotate the altHUB indexer API key f0d9327b… — NEEDS ED — verified still live in Prowlarr’s indexer DB and it was published on the public Bee Hub shelfmark page. This is the one with consequences outside the LAN: a paid third-party usenet account, so regenerating it means logging into altHUB’s own site — I can’t do it from here. Steps once regenerated: Prowlarr → Indexers → altHUB → paste new key → Test → Save. (2026-07-28)
• Install gitleaks + pre-commit hook — Mac + pve, custom rules file, wire as pre-commit in homelab-config
• Weekly gitleaks cron — gitleaks detect on pve with Gotify alert
• Version SECRETS.md consumer-map — move into homelab-config/SECRETS.md

Handover — Amber, while Ed is away (Tue 2026-08-04 onward)

Amber is solo about five days, then travels to Danielle’s in Columbiana and is back the 7th. (merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Assemble the IKEA shelving for the spring house when it lands — 4-section 63" bookcase with glass doors, plus wood shelving. Lean it anywhere; Ed will re-orient. Glass doors chosen so the books don’t get dusty.
• Clear driveway debris into piles — as many piles as convenient.
• Flame the weeds in that bed; mow the hummingbird area.
• Water everything including the potted trees. Do not plant the trees — they are unlabeled and wait for Ed.
• More clover on the far side of the fence — the first attempt didn’t take because the bed was never prepped and the tiller won’t climb that incline.
• Hose off / store the patio chairs, or deal with the nesting birds.

Handover — Hannah, Monday 2026-08-03

(merged from PENDING-TASKS-2026-07-31, dictation 2026-07-31 usb)

• Pick up every rock — “I don’t care if it’s a pebble.” Ed is tired of running them over with the skid loader and the tiller.
• Pull all the screws out of that one path.

Waiting On