בס״ד

Context Update — 2026-06-01 (Mon ~4:45 PM ET) · Sam's "don't repeat mistakes" check

docs/CONTEXT_UPDATE_2026-06-01.md · last changed (pre-VM history) · rendered from GitHub master

Context Update — 2026-06-01 (Mon ~4:45 PM ET) · Sam's "don't repeat mistakes" check

Sam asked for a full honest context update so we stop repeating the scattered/bloated/slow pattern. Answers verified by exploration, not memory.

Straight answers to Sam's questions

  1. AI Rethink — built but PARTIAL. Reads CARDS, finds duplicates (found 7 real). Does NOT yet: read groceries, re-number, re-order whole queue. Dedupe-only today.
  2. Obligations vs Plaid — TWO SEPARATE SYSTEMS, no sync. Obligations = Monthly Obligations v1 Google Sheet (1L_rxCSOnc…, manual). Plaid = live cash balances only (cached in Worker KV, never written to sheet). NOT on Plaid: mortgages, utilities, tuition, camp, loans (all manual). The "cadence that chases Sam for what Plaid can't see" (STR statements he just uploaded, mortgages, utilities) = NOT BUILT. This is the money-chase layer, specced + parked.
  3. decide.html tiles: Cash/Home→home.html (real Plaid cash). Family→chanie.html (real 10KB but NOT confirmed with wife). Command Inbox→start-here.html (ANCIENT, May 19 — should point to start.html or retire). MIS/Stocks→peek.html ("Scope", pulls MIS v2; underlying MIS v7.1 known-broken per FSE arch doc).
  4. OpenAI voice in Telegram — LIVE (shipped 5/31, commits "voice LIVE (OpenAI)"). Per-request; bot default may still be free MeloTTS unless flipped. NOT lost.
  5. Swipe — NOT built. Tap + dropdowns only. Still owed.
  6. Master Plan board — strategic MAP (Protect/Advance/Waiting + Engine/Plane). Review, not daily-doer. Overlaps decide.html (consolidation owed).
  7. Per-person obligations source (Sam=Plaid live, wife=sheet) — right design, doable via ?as= scoping (engine exists in pwa-deploy), not wired into new pages.
  8. MIS review — best done as a SEPARATE ChatGPT/Codex adversarial session (MIS is big + broken). Claude Code preps the context pack.

THE META-PROBLEM (Sam nailed it)

Improvements get built into ONE page and DON'T propagate (dropdowns→only decide.html; bounce-fix→only start.html; crest fix→only decide.html). This IS the scattered-pages disease + same root cause as today's stale-facts miss: no propagation discipline. "Getting smarter not more bloated" requires: build a lesson ONCE in a SHARED place, all pages inherit. Right now every page is a hand-copied silo.

Active pages sprawl (the consolidation debt)

start.html · home.html · home-cockpit.html · decide.html · start-here.html (ancient) · obligations.html · pwa-deploy · peek.html · chanie.html (+8 chanie variants) · family-kiosk · the board · 142 briefings. TOO MANY. Sam wants ONE, fast, that all the lessons live in.

What Sam wants next (his words)

Recommendation (the honest one)

STOP adding features to individual pages. The next real move = Grand Central consolidation done right: pick ONE home (decide.html is the strongest base), make ALL shared behavior (dropdowns, crest, colors, actions, scoping) live in ONE shared layer so a fix propagates everywhere, make it snap (parallel+cache), retire the duplicates. THEN layer swipe + full-rethink + money-chase. Doing more one-off page edits = feeding the disease Sam just named.

Parked for dedicated sessions

Source trail · docs/CONTEXT_UPDATE_2026-06-01.md @ master · rendered 2026-07-02 7:23 PM EDT by scripts/build-docs.py · the .md in the repo is the truth; this page is the phone-readable view