Voice Agent Execution Loop V1

Compare the accepted plan with actual evidence, put waiting people and cost of delay first, then leave tomorrow with one number-one outcome and one protected first block. Matt still chooses the priority, writes the top three and their steps by hand, and performs the work.

flowchart LR
 A["Accepted plan"] --> B["Actual evidence"]
 B --> C["Cost-of-delay ranking"]
 C --> D["One protected first block"]
 D --> E["Truthful completion receipt"]

The August 14 reconstruction supplies the business intent. The planning audit supplies the reliability failures. Current local source supplies plan-versus-actual, evidence receipts, one durable browser and phone session, typed precedence, recovery, number one, schedule, and partial close. The local state still says ship false and requires separate staging authority.

flowchart TD
 I["August 14 intent"] --> R["Required behavior"]
 A["Planning audit"] --> R
 P["Planning V2 local source"] --> U["Reusable mechanics"]
 R --> C["Concept"]
 U --> C

Manual is immediate but fragile. Reuse preserves Planning V2 unchanged but misses the new focus contract. Integration adds only the compact execution state and is recommended. Custom builds the full Voice Platform and is too indirect for the immediate goal.

flowchart LR
 M["Manual"] -->|"same day"| H["Habit rehearsal"]
 R["Reuse"] -->|"existing flow"| G["Good baseline"]
 I["Integration"] -->|"recommended"| V["Execution Loop V1"]
 C["Custom"] -->|"defer"| P["Full platform"]

Show one top outcome, up to three waiting obligations, plan versus actual, a maximum of five active tasks, manual time-block and handwriting confirmation, and truthful save states. Keep text fallback, exact resume, and partial close. Do not launch Codex or change production in version one.

flowchart TD
 O["Orient"] --> N["Narrate naturally"]
 N --> P["Plan versus actual"]
 P --> Q["Choose one number one"]
 Q --> T["Confirm blocks and handwriting"]
 T --> S["Truthful close or partial close"]

The main choice is full durable transcript versus a confirmed minimal action packet. The plan recommends minimal-summary retention for the first sensitive pilot. It also recommends one local alarm before any reminder swarm. The exact evening time, daily language, and whether the health reset is scored remain open.

flowchart LR
 D["Matt decides"] --> T["Transcript boundary"]
 D --> R["Reminder contract"]
 D --> L["Language"]
 D --> H["Health scope"]

Outcome and options are complete. Review readiness requires playable audio, one shared voice agent, the append-only feedback ledger, a commentable live page, and verification. Concept approval remains pending. Only after exact approval may the techspec prompt be created.

flowchart LR
 O["Outcome locked"] --> P["Options ready"]
 P --> R["Review ready"]
 R --> A["Origin-task approval"]
 A --> H["Techspec handoff"]