Weekly note CW15 — 2026-04-08

Fleeting Notes/Weekly/2026_15.md

@@ -0,0 +1,124 @@
+---
+id: 2026-15
+title: Weekly — 2026-04-08
+type: Weekly
+created: 2026-04-08T11:00:00Z
+modified: 2026-04-08T11:00:00Z
+tags: [weekly]
+---
+
+# Weekly - 2026 CW 15 (Wed 2026-04-08)
+
+## Accomplishments
+
+### 🎯 MAJOR: FRET Specifications & Reactive Synthesis Progress (April 3)
+**Significant formal verification work — specifications written, realizability conflicts identified and resolved, test traces analyzed.**
+
+- Wrote FRET specifications for the reactor automaton
+- Worked through heatup mode requirements: bounded temperature rate (dT_dt constraints), maximum heatup duration before SCRAM
+- Developed watchdog pattern for bounded time constraints — key insight: "within T seconds satisfy SCRAM" would force SCRAM always (no happy path to load following); watchdog only fires if STILL in heatup when timer expires
+- Typed input/output variables: Inputs (STARTUP, t_dot_exceeded, t_max_exceeded, t_power_min), Outputs (SCRAM, OP_MODE, Cold_Shutdown, Heatup, Load_Follow)
+- Hit and resolved realizability conflict (PWR-0000 vs PWR-2001 vs PWR-3002) — added "While !SCRAM" scope to PWR-2001 so startup only happens outside SCRAM
+- Analyzed 67 test traces against 8 requirements:
+  - Only 4/67 pass all requirements
+  - PWR-2001 fails in 49 traces (adversarial: STARTUP + t_max_exceeded simultaneously)
+  - 75 physical consistency warnings (Heatup=1 with OP_MODE=0)
+  - Traces appear to be counterexamples, not valid behaviors
+- Identified need for explicit both-sides specification of inequalities/implications — potential FRETless design improvement (implicit iff support)
+
+**Status:** DR-72 (FRET specs) marked Done. Understanding of FRET's limitations deepening — feeding into FRETless design ideas.
+
+### DICE 2026 Travel Booked (April 4)
+- Expedia booking: $999.77 (hotel + car for SLC)
+- DR-78 marked Done
+- Conference: May 12-13, Salt Lake City
+
+### Firefly III — March 2026 Fully Reconciled (April 3)
+- Imported 18 Apple Card transactions (March 28 - April 2)
+- Cross-referenced 136 statement transactions against 148 in Firefly
+- Cleaned up migration garbage: phantom $5,755.53 payment, duplicate entries, wrong-sign Daily Cash adjustments
+- Added 3 missing Daily Cash Adjustment clawbacks
+- **March Apple Card statement now reconciled clean** ($2,898.11 → $9,976.38 payment → $8,728.99 charges → $1,650.72 ending)
+
+### 🔧 Anthropic Billing Change — Infrastructure Pivot (April 4)
+**Major cost optimization after Anthropic announced third-party tool billing changes.**
+
+- Anthropic subscriptions no longer cover OpenClaw usage (every token billed separately)
+- Old config exposure: estimated $600-900/month at Opus 4.6 default
+- Reconfigured same day:
+  - Default model: Sonnet 4.5 (was Opus 4.6) — ~5-10x cheaper
+  - Heartbeats: Haiku 4.5, 60-minute intervals (was 30min on Opus)
+  - Prompt caching enabled: "long" retention (90% savings on repeated context)
+  - Opus available on demand via override
+- Target: ~$100/month total
+- Created PER project in Plane for personal tasks (non-thesis/car/website)
+
+### iMessage Echo Loop Bug — Diagnosed & Fixed (April 6)
+- Outbound iMessage replies were echoing back as inbound messages, creating infinite loops
+- Root cause: DM chat classified as "self-chat" + binary corruption in `imsg watch` text causing echo cache misses
+- Applied surgical patch to OpenClaw monitor — forced drop of all `is_from_me` messages in self-chat context
+- ⚠️ Patch gets overwritten on OpenClaw updates — needs re-application
+
+### Thesis Draft
+- Preliminary Results section skeleton complete with equations (from late March)
+- PKE, thermal-hydraulic, and reactivity feedback equations in place with LaTeX labels
+- Awaiting Dan's V1 feedback to unlock next steps
+
+## Pushed or Rescheduled
+
+**Dan Cole V1 Feedback:**
+- Proposal draft sent March 17 — now Day 22+ with no response
+- Expected — Dan busy with Luis finishing up
+- All downstream formal verification work blocked on his review
+
+**FRET Discussion:**
+- Punted from Easter Sunday (April 6) to this week
+- Topic: extracting actual automaton/AIGER circuit from FRET reactive synthesis output
+- Dane wants visualization of the automaton itself for easier trace analysis
+
+**Email Infrastructure:**
+- himalaya still broken on all accounts (since March 24)
+- Only default (zoho) accessible
+
+## To Do
+
+**This Week (CW15):**
+- Discuss FRET automaton extraction with Dan at group meeting
+- Present FRETless as publication roadmap (FRET gaps → enum/FSM extensions → continuous dynamics → nuclear case study)
+- Continue barrier certificate work (DR-74, DR-75 — load following + cold shutdown)
+- Continue reachability analysis (DR-73, DR-76 — heatup + SCRAM modes)
+- DICE presentation prep (May 12-13 approaching — ~5 weeks out)
+- Martha's Run 5K — April 11 (Friday)
+
+**Personal:**
+- PER-3: Find +1 for May 9 wedding (deadline passed ~April 6 — status?)
+- PER-1: Evaluate OpenAI Codex Plus as potential cost optimization
+- PER-4/5: Cancel YouTube TV / Subscribe Pittsburgh Sportsnet
+
+**Upcoming:**
+- May 9: Wedding (guest, not in party)
+- May 10: DICE departure to SLC
+- May 12-13: DICE 2026 conference
+
+## Context
+
+**Weather:**
+- April 1: 58-63°F, rainy morning
+- April 3: 85°F high, rain chance midday
+- April 4: 87°F high, sunny
+- April 6: Heavy rain, 54°F
+
+**Fitness Battery (April 6):**
+- Broad jump: 65 inches
+- Dead hang: 32 seconds
+- Bench press: 135 × 5
+- Lat pulldown est. 1RM: 172
+- Leg extension est. 1RM: 187
+- Dane felt good about it — saw Jackie and Andreas at gym
+
+**Notes:**
+- FRET specification work is building real expertise — realizability conflicts, trace analysis, variable typing are all thesis-level skills
+- The watchdog pattern insight is worth writing up — common control pattern that FRET handles awkwardly
+- FRETless iff-style specification is a concrete design improvement to pitch
+- Cost optimization was handled well — same-day response, minimal disruption
+- CW14 cycle was productive: 9 items created, most FRET/analysis work progressed significantly