# Design Spec — Futures Gate-Tree Forecasting Project

**Date**: 2026-05-13
**Author**: Tamir + Claude Operator
**Status**: Approved → research phase

## Context

Tamir wants a structured, research-backed forecasting artifact covering the impact of AI and robotics on cost of living and quality of life over a **near-to-medium horizon (1–10 years)**. Outputs serve five overlapping uses, in priority order: (1) personal decisions, (2) long-term family planning, (3) public-facing artifact, (4) living monitor calibrated over time, (5) basis for colleagual discussion and business navigation.

The unit of forecasting is a **gate** — a specific, measurable trigger event with a P10/P50/P90 timeline and a downstream essay on what changes for Tamir's life when it passes.

This document is the design contract. It does not contain research findings; those land in `src/gates/<slug>.md` per gate.

## Scope (locked)

- **11 gates**, listed below. Each is a peer-level trigger event (not a sub-step of another), measurable, near-to-medium horizon.
- **Approach A — scenario-driven**: each gate's research includes a "downstream essay" mapping its impact across Tamir's dimensions.
- **Dimensions covered**: food, housing, utilities, education, childcare, global metals supply, food availability, travel.
- **Depth per gate**: ~2–4 hours of synthesis (Deep tier, not Sketch).
- **Cross-gate dependencies**: captured as structured frontmatter (`sub_gates`, `cross_gate` with relation types `enables` / `correlates` / `substitutes`) plus a post-pass `cross-gate-synthesis.md`.

## The 11 gates

| # | Slug | Trigger threshold |
|---|---|---|
| 1 | `humanoid-retail-20k` | Mass-market general-purpose humanoid robot (Figure / Optimus / 1X SKU) reaches **< $20,000 retail** in an OECD market |
| 2 | `ai-agent-30pct-knowledge-work` | An AI agent autonomously completes ≥ **30%** of tasks in a typical knowledge-work role (legal / dev / content / support) without human review on each step |
| 3 | `residential-solar-storage-0.04` | Residential solar + storage system delivers **< $0.04/kWh** over 25-yr amortized full-system cost in 3+ US states |
| 4 | `robotaxi-unit-economics-5-cities` | Robotaxi (Waymo / Tesla / Zoox) reaches **profitable unit economics** in 5+ US cities concurrently |
| 5 | `ai-tutor-k8-parity-20mo` | Conversational AI tutor matches a human teacher on standardized K-8 metrics at **< $20/month** retail, used by ≥ 1M families |
| 6 | `cell-meat-beef-parity` | Cell-cultured beef achieves **retail price parity** with conventional ground beef in a major grocery chain |
| 7 | `construction-robot-40pct-labor` | Construction-automation robot stack cuts **new-home labor cost by ≥ 40%** on representative single-family builds |
| 8 | `metals-bom-30pct` | EV/robot bill-of-materials cost drops **30%** via Li + REE supply diversification (China dependency on key minerals < 60%) |
| 9 | `evtol-1k-trips-major-city` | eVTOL / air-taxi commercial service runs **> 1,000 trips/day** in a major OECD city |
| 10 | `smr-first-oecd-deployment` | First commercial **SMR (sub-300 MW)** deployment in an OECD country providing baseload to the grid |
| 11 | `autonomous-freight-delivery` | Autonomous trucking + last-mile delivery cuts **labor share of US freight + last-mile cost by ≥ 30%** on national average (compound trigger across long-haul and last-mile) |

## File structure

```
~/projects/futures/
├── BRAIN.md
├── README.md                                       # later
├── docs/
│   └── superpowers/
│       └── specs/
│           └── 2026-05-13-futures-gate-tree-design.md   # this file
├── src/
│   └── gates/
│       ├── humanoid-retail-20k.md
│       ├── ai-agent-30pct-knowledge-work.md
│       ├── ... (11 total)
│       └── cross-gate-synthesis.md                 # post-pass
└── (V2) astro project files
```

## Per-gate `.md` template

### Frontmatter (YAML)

```yaml
---
title: <human-readable label>
slug: <kebab-case-slug>
status: draft | in-research | reviewed | published
dimensions: [food, housing, utilities, education, childcare, metals, food-availability, travel]   # subset
horizon: near | medium
trigger: "<exact measurable threshold sentence>"
timeline:
  p10: <year>           # 10% chance the gate passes by this year
  p50: <year>           # median estimate
  p90: <year>           # 90% chance by this year (or "never" if multi-modal)
confidence: low | medium | high     # how solid the timeline is
sub_gates:                          # upstream causal dependencies (not in our 11)
  - slug: <sub-gate slug>
    p50: <year>
    why: "<one line>"
cross_gate:                         # links to OTHER gates in our 11
  - other: <slug>
    relation: enables | correlates | substitutes
    strength: weak | medium | strong
    note: "<one line>"
external_calibration:
  metaculus: <URL or null>
  manifold: <URL or null>
  expert_consensus: "<one-line summary or null>"
last_updated: YYYY-MM-DD
sources_count: <int>
---
```

### Body sections (markdown)

1. **TL;DR** — one paragraph synthesizing the timeline + thesis.
2. **Current state** — where the threshold sits as of `last_updated`, with 2–3 hard numbers (price, capability, market share, deployment count).
3. **Key uncertainties** — 3–5 things that would shift the P50 by ≥ 2 years if resolved.
4. **Evidence synthesis** — bulk of the 2–4h research. Sub-sections per stream:
   - Academic (arXiv + Semantic Scholar)
   - Industry / market (Exa + Firecrawl)
   - Public sentiment (Reddit)
   - Prediction markets (Metaculus / Manifold via WebFetch)
   - Policy / regulation (Exa + WebSearch + claude-in-chrome)
5. **Sub-gates** — list upstream causal dependencies (frontmatter mirror).
6. **Cross-gate dependencies** — explicit prose about how this gate interacts with the other 10 (enables/correlates/substitutes).
7. **Downstream impact essay** — one paragraph per affected outcome dimension. What changes for food cost, housing cost, etc., if/when this gate passes.
8. **Decision implications** — 1–2 paragraphs: what Tamir should consider doing differently at P10 vs P50 vs P90 outcomes (career, family, investments, real estate, kids' education).
9. **Sources** — numbered, annotated list of every source used.

## HTML rendering (V2)

Astro static site. Each `.md` → one route:

- **Top half**: rendered prose (sections 1–8 above), styled, with inline source citations linking to the Sources section.
- **Bottom half**: a `<details>` block titled "Full markdown" containing the entire `.md` file verbatim (frontmatter + body), with a copy-to-clipboard button. The `.md` is also served as a static asset at `/gates/<slug>.md` so the HTML page can link to it directly ("View raw" link).
- **Index page**: grid of all gates with filter chips for dimension + status. Shows P50 timeline distribution as a single overlay chart.
- **Per-dimension index pages**: aggregate all gates affecting `food` / `housing` / etc., compute "earliest meaningful change" date = P50 of the earliest gate delivering ≥ 20% of impact.

Visual widgets — timeline distribution charts, dependency-graph SVG, outcome-impact bars — are V2 add-ons. V1 is "markdown rendered to clean HTML + raw-markdown section + raw-file download link".

## Research workflow per gate

1. **Dispatch a research agent** (`general-purpose`) with all 5 evidence-stream MCPs available and a brief specifying the gate's trigger sentence, dimensions, and required output sections. Each agent runs ~30–90 min.
2. **Agent returns** structured findings + draft `.md` content.
3. **I synthesize and finalize** the `.md`: confirm the P10/P50/P90 against the evidence, write the downstream impact essay (Tamir-specific), commit, push.
4. **Repeat** for all 11 gates.
5. **Post-pass**: write `cross-gate-synthesis.md` after all individual gates are done.

For all 11 gates running in parallel: agent dispatch can be batched 3–4 at a time to keep context manageable.

## Cross-gate synthesis pass

After every gate is reviewed/published, write `src/gates/cross-gate-synthesis.md`:

- **Year clusters**: which gates' P50s land within ±1 year of each other? Compound effects.
- **Enables chains**: full graph of which gates accelerate which others (#8 metals → #1 humanoid, #4 robotaxi, #11 trucks; #3 solar → #1, #11; #2 AI agent → meta-accelerator).
- **Substitution loops**: which gates compete (#3 vs #10).
- **Outcome bottlenecks**: per dimension, which gate is the rate-limiting step.
- **Calibration summary**: aggregate Metaculus/Manifold deltas vs. our P50s.

## Verification

End-to-end test of the workflow on **one gate first** to validate template + agent brief + synthesis quality before parallelizing across 11:

1. Run research workflow on gate #2 (`ai-agent-30pct-knowledge-work`) — highest decision-relevance for Tamir.
2. Inspect the resulting `.md`: frontmatter complete, all 9 body sections filled, source count ≥ 8, P10/P50/P90 grounded in cited evidence.
3. If quality is good, fan out to remaining 10 gates in 3 batches of 3–4.
4. After all 11 are reviewed, write the cross-gate synthesis.
5. (V2) Stand up Astro scaffold, render the .md files, deploy to Vercel.

## Out of scope (V1)

- Astro / Next.js / any web app scaffold (deferred to V2 after research is solid)
- Vercel deployment + domain (deferred)
- Visual widgets — timeline charts, dependency graphs, impact bars (deferred)
- Monte Carlo propagation over inter-gate dependencies (V2 stretch)
- Notion mirror of working notes (out unless explicitly added)
- Gates outside the 11 listed (e.g. downside / climate / geopolitical gates) — can be added in a future scope expansion

## Open questions

None remaining at design time. Move to research.