claudetools/clients/cascades-tucson/reports/2026-06-16-2.4ghz-remediation-runbook.md

Cascades of Tucson — 2.4 GHz Remediation Runbook (execute 2026-06-16 night)

Goal: cut 2.4 GHz airtime/co-channel contention (the "bad for some users" cause) by (1) powering down over-strong 2.4 cells and (2) disabling redundant 2.4 radios where a nearby active-2.4 neighbor still covers the area — without breaking mesh backhaul, orphaning SSIDs, or opening coverage holes.

Reference data: SNR matrix .claude/tmp/cascades-nbr.json (fresh tonight), audit reports/2026-06-16-unifi-full-audit.md. Regenerate any list with the commands in each phase.

---

Safety facts established before touching anything

• Mesh topology (live): 2nd Floor Atrium is the wireless-mesh parent for CC Bridge + salon (backhaul ch36 / 5 GHz); 206 U7 Pro carries 108. MESH-PROTECTED — do NOT disable: 2nd Floor Atrium, CC Bridge, salon, 206 U7 Pro, 108. (coverage-thin now auto-excludes these.)
• SSID pinning: none — CSC ENT / CSCNet / Guest broadcast on all APs (broadcasting_aps off). No client is locked to one AP; disabling a radio just makes clients roam to a neighbor. No orphaning risk.
• Reversibility: every change is apply-radio with a rollback JSON saved to .claude/tmp/. Re-enable = apply-radio <site> ng enable --ap "<name>" --apply; restore power = ... ng power auto --ap "<name>".
• AP-hang recovery: if an AP goes/stays offline after a change, do NOT force-provision (that took 445 offline on 06-16). Recover with device-control.sh cascades poe-cycle "<AP name>" --apply.

---

Phase 0 — Pre-flight (do first, ~5 min)

1. Site VPN up + stable; controller + APs reachable:
   • bash .claude/skills/unifi-wifi/scripts/live-stats.sh cascades | head -3 (expect 77 APs reporting)
2. Baseline snapshot (to compare after):
   • bash .claude/skills/unifi-wifi/scripts/live-stats.sh cascades > .claude/tmp/baseline-pre.txt
   • bash .claude/skills/unifi-wifi/scripts/radio-usage.sh cascades ng 77 > .claude/tmp/usage-pre.txt
3. Pick a watch AP per floor (a KEEP neighbor) to eyeball during changes, e.g. bash .claude/skills/unifi-wifi/scripts/watch-ap.sh <ap-ip> in a second terminal.

---

Phase 1 — Power-down 2.4 to Low (biggest safe win, fully reversible)

Smaller 2.4 cells = less mutual interference, no coverage loss, no radios turned off. Do per floor; DRY-RUN first (omit --apply), eyeball, then --apply.

# SCOPE (Howard 2026-06-16): Floors 5 & 6 EXCLUDED. Floor 4 already powered-down (pilot).
# Floor 3 has no disabled radios + no mesh parent -> clean by-zone apply (17 radios):
bash .claude/skills/unifi-wifi/scripts/apply-radio.sh cascades ng power low --zone "Floor 3" --apply
# Floor 1 PER-AP (skip 128 = keep it disabled; skip 108 = offline mesh child) -> no pointless re-enable of 128:
for ap in "132 (Rec Room)" 133 127 140 145 121 122 109 116 115 139 146 103 102; do \
  bash .claude/skills/unifi-wifi/scripts/apply-radio.sh cascades ng power low --ap "$ap" --apply; done
# (Floor 4 = DONE; Floors 5 & 6 = SKIP per scope; 128 stays disabled; 108 left alone.)

For Floor 2 and misc (contain mesh APs), power-down per-AP to skip the mesh parents — do the non-mesh APs only, and WATCH CC Bridge/salon stay online:

# Floor 2 (skip 2nd Floor Atrium, 206 U7 Pro):
for ap in 203 204 209 210 217 221 222 229 237 240 241 248 236; do \
  bash .claude/skills/unifi-wifi/scripts/apply-radio.sh cascades ng power low --ap "$ap" --apply; done
# misc (skip CC Bridge, salon): Kitchen, Memcare*, memcare piano, Dining Room
for ap in "Kitchen" "Dining Room" "Memcare Nurse Station" "memcare piano" "Memcare TV room"; do \
  bash .claude/skills/unifi-wifi/scripts/apply-radio.sh cascades ng power low --ap "$ap" --apply; done

GATE: after each floor, confirm client counts hold and no dead spot (live-stats, watch-ap). If CC Bridge/salon blip during Floor 2/misc, STOP and reassess.

---

Phase 2 — Settle + re-measure (~15 min)

Let it converge, then:

bash .claude/skills/unifi-wifi/scripts/live-stats.sh cascades > .claude/tmp/baseline-postA.txt

Compare retry% / cu_total vs baseline-pre.txt. Expect 2.4 retry + cu_interf to drop. If Phase 1 alone looks good, it is a legitimate stopping point for tonight — disables can be a separate night.

---

Phase 3 — Disable redundant 2.4 radios (per-AP, floor by floor, with gates)

Each AP below is low-client (<2 avg), non-mesh, non-memcare, and stays covered by ≥2 active-2.4 neighbors. Do ONE FLOOR, validate 15 min, then the next. Recommend Floor 1 or Floor 4 as the pilot.

Floor 1 (7): 139 115 102 145 127 116 109 Floor 2 (8): 240 221 209 229 241 248 203 236 Floor 3 (5): 342 317 304 336 347 Floor 4 (5): 445 441 403 407 450

# one floor at a time, e.g. Floor 4 pilot:
for ap in 445 441 403 407 450; do \
  bash .claude/skills/unifi-wifi/scripts/apply-radio.sh cascades ng disable --ap "$ap" --apply; \
  sleep 5; done
# then VALIDATE before the next floor:
bash .claude/skills/unifi-wifi/scripts/live-stats.sh cascades --clients | head -40
bash .claude/skills/unifi-wifi/scripts/coverage-thin.sh cascades 14   # re-check remaining coverage (NEIGHBOR_JSON=...)

GATE (per floor): clients on disabled APs moved to neighbors; no AP shows a sudden client-drop to ~0 in a covered area; no user complaints. If a dead spot appears → re-enable that AP: apply-radio.sh cascades ng enable --ap "<name>" --apply.

---

Phase 5 — 5 GHz tuning (NEXT phase, after the 2.4 power-down settles)

Both changes are reversible and coverage-safe (width/channel don't change coverage). Per zone, dry-run (omit --apply) then apply; Floors 5 & 6 excluded per scope.

5a. Width 80 -> 40 MHz (currently 80 MHz on 76/77). 40 MHz doubles the non-overlapping channels -> better spatial reuse in this AP density. (Halves peak per-client rate, but the density win dominates.)

for z in "Floor 3" "Floor 1" "Floor 2" "Floor 4"; do \
  bash .claude/skills/unifi-wifi/scripts/apply-radio.sh cascades na width 40 --zone "$z" --apply; done
# rollback: na width 80

5b. Channel plan — DECISION NEEDED (non-DFS vs include clean-DFS):

• DFS reality here: dfs-check.sh found ZERO radar events fleet-wide -> DFS channels are empirically CLEAN.
• Tradeoff at 40 MHz: non-DFS-only = ~5 usable channels (36-48 + 149-165) for 77 APs -> heavy co-channel. Including the clean DFS channels (52-144) gives ~20 channels -> far better reuse, tiny resilience risk (a radar hit would force a vacate, but none seen near Davis-Monthan in the logs).
• RECOMMENDATION: use a plan that INCLUDES clean DFS for channel diversity. Choose non-DFS-only only if you want maximum resilience and will accept more co-channel overlap. (55/77 are on DFS today.)

SURVEY=.claude/tmp/cascades-survey.json; SURVEY_JSON=$SURVEY bash .../survey-collect.sh cascades
SURVEY_JSON=$SURVEY bash .../channel-plan.sh cascades na     # dry-run; review, then apply per zone w/ validation

Phase 6 — 6 GHz adoption + steering (the untapped clean band)

Root cause 6 GHz is empty (1 client): the production SSID CSCNet is NOT broadcasting on 6 GHz (wlan_bands=["2g","5g"]); only Guest is. So step 1 is enabling the band, then assisting steering.

# 6a. add 6 GHz to CSCNet (keeps 2.4/5 for legacy):
bash .claude/skills/unifi-wifi/scripts/apply-wlan.sh cascades bands all --wlan CSCNet --apply   # -> [2g,5g,6g]
# 6b. enable 802.11v BSS-transition on CSCNet (Guest already has it; assists up-band steering + roaming):
bash .claude/skills/unifi-wifi/scripts/apply-wlan.sh cascades bsstm on --wlan CSCNet --apply

Band-steering (no2ghz_oui) is already ON for CSCNet. After 6a/6b, 6E/7-capable clients gravitate to the clean 6 GHz, offloading 5 GHz. Validate: live-stats client mix shifts toward 6e; confirm no SSID-visibility loss for legacy devices (none expected — 2.4/5 stay on). 6 GHz has shorter range, fine with 77 APs. (CSC ENT is legacy/being deprecated — leave as-is or add 6g too.)

NOTE: this controller's 6 GHz token is 6g (not 6e); apply-wlan.sh fixed accordingly 2026-06-16.

---

NOT doing tonight (explicit — so nothing is ambiguous)

• Power-DOWN instead of disable (redundant but high-client, keep serving): 3rd Floor Atrium, 318, 303, 4th Floor Atrium, 406-408, Dining Room — already set Low in Phase 1; leave ON. (505 was here but Floor 5 is now EXCLUDED — Howard 2026-06-16 — so 505/517 and Floor 6's 615/608/622 are left fully alone.)
• HOLD (leave fully on): Memcare TV room (memory-care area); all MESH-PROTECTED APs.
• Remaining 2.4 levers deferred (isolate cause/effect; do after disables validate): min-data-rate raise (1→12 Mbps) and 2.4 min-RSSI on the OFF APs that aren't on excluded floors. (Band-steering no2ghz_oui is already ON for CSCNet/CSC ENT/Guest.)
• 5 GHz tuning + 6 GHz adoption are now Phase 5 & Phase 6 above (folded in per Howard 2026-06-16) — the next phase after the 2.4 power-down settles.

---

Phase 4 — Wrap

bash .claude/skills/unifi-wifi/scripts/live-stats.sh cascades > .claude/tmp/baseline-post.txt
# re-snapshot 2.4 usage + audit; then /save the session log.

Success = 2.4 retry% and cu_interf down on kept radios, client counts steady, zero coverage complaints. Abort/rollback at any point: re-enable/raise-power the affected APs (rollback JSONs in .claude/tmp/), or device-control poe-cycle a hung AP. Power-down is always safe to leave; only disables can hole coverage.

---

Tunable to regenerate the disable plan live

NBR=.claude/tmp/cascades-nbr.json; NBR_JSON=$NBR bash .claude/skills/unifi-wifi/scripts/neighbor-collect.sh cascades
MINCOV=2 CLIENT_CAP=8 NEIGHBOR_JSON=$NBR bash .claude/skills/unifi-wifi/scripts/coverage-thin.sh cascades 14