Session Steering Specs — 2026-06-08 (from 🐝 Benny)

Three paste-ready blocks — open the relevant Code session, paste the block, let it run.
State verified by a 4-agent read on 2026-06-08. Benny = the back-office/Worker-Bee session; these are for the OTHER sessions.

⬛ PASTE INTO THE MIS SESSION

MIS steering — 2026-06-08. Resume respecting the HARD FREEZE (B.75 → C → STOP;
canonical MIS/docs/MIS_SACS_CALIBRATION_MODULE.md). NO engine-logic changes, NO new
factors/scores/states, NO production swap. Verified state: equity $22,135 broker-true,
recon GREEN on all 5 accounts, B.75 holdings-freshness gate CLEARED, email v1 is cron-fired.

Execute IN ORDER, all freeze-safe (data/presentation only):
1. MACRO FEEDS FIX (Tickers tab) — swap GLD/USO/DXY/ETF mappings to real spot:
   XAUUSD (gold), NYMEX:CL1! (oil), DXY, ETHUSD. ~5 min, biggest credibility lift.
   Verify the next Morning Brief shows real spot prices, not ETF proxies.
2. TRIGGER RESET — audit Apps Script triggers on BOTH the v1 script and the old
   production script. Kill any dormant v7.x/v11 triggers so ONLY v1 fires. Confirm one
   email fires, not two. (Fires real emails → don't run right before Shabbos.)
3. C-GATE — CORRECTION (per the strategy session's 6/8 handoff): Gate C is a BUILD, not just
   presentation — the `misV2VerdictFor_` accessor doesn't exist yet, so it needs a WATCHED WINDOW
   (build the accessor + wire synthV2 to the brief/Test-Entry/Scope with a MIGRATION_MODE badge).
   Acceptance: shows Proposed/Legacy/Exposure/Mode; ZERO action drift from baseline; any drift →
   STOP. D/E/F stay STOPPED (ZW lock). PREREQ: pin a Broker_Qty seeding ritual (owner=Sam, weekly
   CSV) before any unfreeze. Equity is live ≈ $21.6K (META soft) — was $22,135.
4. THESIS-CADENCE TAB — one new read-only tab: 2.5%/mo target tracker + drawdown gauge +
   bucket allocation + cash-extraction state. Must answer Sam's 5-question 30-sec test.
5. HOLDINGS REFRESH + TRADE_PLANNER SYNC — clean stale ABBV, unify Trade_Log source, fix
   the DONE/DEFER tracker logic. Data-layer only.
6. MONDAY TRADES — set the LLY bracket per Quick Trade Calc; re-verify ABBV/WMT/ASML
   exposure vs mandate; execute on broker. Equity verified, recon GREEN, no gate blocks entry.
Stop rule: any baseline drift → surface the blocker, do not proceed. Architecture stays locked.

⬛ PASTE INTO THE LevSMS SESSION

LevSMS — UPDATED 2026-06-08 from the session's own handoff (this session is self-driving; Sam
relays). MAJOR SHIFT: bot-reply is FIXED + the architecture moved to EDGE-COMPUTE.
- bot-not-replying FIXED (was Twilio err 11200 — Apps Script returned a redirect). Inbound flipped
  to the Cloudflare Worker (levsms-edge.sam-0f0.workers.dev); NW texts ~0.1s. Apps Script kept as
  Twilio fallback. Single TwiML (in-order). Out-of-area renders the right city.
- ARCHITECTURE (Sam-confirmed): COMPUTE EVERYTHING AT THE EDGE off Sheets; only the Friday shul
  schedule is stored. Sheets = back-office/intake only. (Supersedes the old "migrate NW hardcode
  to AREA_ rows" plan.)
- Their next action: deploy the KBH→8:15 fix + refresh, then build Worker-native edge compute
  (zmanim/shabbos/KS/holidays/yomi/weather via Hebcal / weather.gov / Sefaria directly).
- STILL OPEN for the email-INTAKE/submission go-live (separate from bot-reply): gmail.send re-auth
  · verify support@ alias (ALIASCHECK) · flip LEVSMS_TEST_MODE=false · wire the revenue pipe
  (Stripe link + pick the 20% tzedakah).
🔔 SAM ACTION: text 11581 once — his test ZIP is stuck on 08701 (Lakewood) → slow path.
⛔ Other sessions: stay OUT of levsms/ (mid-build).

⬛ PASTE INTO THE #042 / command-inbox SESSION

Portal/ops-api steering — 2026-06-08 (updated after the strategy session's handoff). #042 (brain
unification) is COMPLETE and LOCKED (5/5 tests, deployed). Two things in YOUR lane now:

1. **BUILD #043 = the UNIFIED OPERATING MAP** (canonical — spec already written + pushed:
   docs/PROJECT_043_UNIFIED_OPERATING_MAP.md). ONE always-current view both Sam AND Mildred open
   and instantly know what matters + what's theirs. DON'T build a new surface — extend home.html
   (Grand Central) with an Operating Map view, role-scoped via ?as=sam / ?as=mildred, reusing
   hs-core.js + hs-theme.css. Worker endpoint GET /operating?as=<role> aggregates Action_Queue +
   Obligations + Calendar into one role-filtered JSON; SERVER-SIDE Mildred-boundary filter (hide
   wife/mother/family/private-dates for ?as=mildred); a "One Next Best Action" ranker (P0 + nearest
   bumper + money-impact). Full build steps + acceptance in the doc.
2. **RETIRE morning-spine "A" (CF-cron)** — fold its Telegram delivery into the strategy session's
   live routine "B" (7:30 ET weekdays) so there's ONE spine, not two.

SEPARATE / LATER (NOT #043 — do not conflate): the **Adaptive Layer** — when Sam corrects the bot,
write review_state=CORRECTED + the corrected fact (today corrections are IGNORED, the gap Sam felt
in the inbox in/out); a nightly distill compressing captures+corrections into clean Profile facts;
proactive surfacing. This is the learning moat; build it AFTER #043. (Sam's queue card #042 —
"memory local-only, migrate to Drive/repo" — folds in here.)

Source: 4-agent read (MIS, LevSMS, #042, moat) 2026-06-08. Full map + moat verdict: outputs/2026-06-08_12-03_audit_sessions-map-and-moat.html.