Haven (Learning Garden System)

Overview

Haven is a documentation-first garden planning system for a backyard raised-bed garden in southern Denmark (USDA Zone 7b). The project is designed as a developmental agency-building tool for a neurodivergent 7-year-old learner, where the primary success metric is engagement and growing independence — not crop yield.

Core purpose: A young learner who independently checks on, cares for, and harvests from plants he chose to grow — and sees the direct connection between his effort and the food on his plate.

Target user: A neurodivergent 7-year-old, with the parent as facilitator.

Key Features

• 27 crop JSON database with growth stages, child-facing actions in EN/DA, difficulty tiers, companion plants, and Home Assistant alert triggers
• 30 weekly schedules (W15-W44) with themed tasks, bilingual prompts, prioritized activities, and vacation flags
• Neurodivergence-adapted session scripts with 5-10 minute attention windows, sensory transitions, timed steps, self-regulation exit ramps, and pick-2-of-4 task model
• Hyperspecific planting guides with centimeter-level grid positions for every plant across 2 raised beds
• Self-watering reservoir system with DIY PVC wicking build guide and 5-day dry-run test protocol
• Vacation handoff documentation — Neighbor guide with photos, pre-departure checklist, countdown session script
• Home Assistant integration — 10 automation rules (drought, frost, heat stress), sensor definitions, Lovelace dashboard YAML (copy-paste ready)
• Bilingual throughout (EN/DA) with child-facing prompts spoken by father
• Succession calendar for continuous harvest (W22-W43) ensuring something is always ready to pick
• Evidence-based slug defense — Ferramol pellets (jernfosfat) per Aarhus University research; copper tape, beer traps documented as less effective
• Daily routine card and weekly garden walks — Printed visual checklists for child independence

Architecture

Structure

This is a pure documentation + structured data project — zero application code.

Layer	Technology
Documentation	31 markdown guides (143 total MD files including planning)
Data	72 JSON files with JSON Schema validation
Automation	Home Assistant YAML configs (sensors, alerts, plants, dashboard)
Planning	GSD methodology (.planning/ directory)
Version Control	Git (125 commits)

Physical Layout

• Bed A "Family Bed" (230x120x40cm, corten steel by byJEMA) — Tomato, cucumber, pepper, lettuce, herbs, nasturtium, 3 zones
• Bed B "His Bed" (150x60x40cm, corten steel) — Strawberries, radishes, sensory plants (lamb's ear, thyme), flowers, sunflower

Data Architecture

data/
  crops/          # 27 crop JSON files (growth stages, EN/DA actions, difficulty tiers)
  beds/           # bed-a.json, bed-b.json (cm-position coordinates per plant)
  schedules/      # W15-W44 (30 weekly files with themed tasks, vacation flags)
  ha/             # Home Assistant configs (sensors, alerts, plants, customize)
  schemas/        # JSON Schema (crop, bed, week, checklist, season-review)
  succession-calendar.json

Documentation Architecture

docs/
  property-map.md              # ASCII site overview with sun zones
  bed-layout.md                # Physical dimensions, materials
  planting-grid-bed-a.md       # Per-cm grid positions (36KB)
  planting-grid-bed-b.md       # Per-cm grid positions
  setup-guide.md               # 12-step construction sequence
  soil-layers.md               # 3-layer fill recipe
  reservoir-build.md           # DIY self-watering PVC wicking system
  trellis-build.md             # Lean-to cucumber trellis
  slug-defense.md              # Evidence-based Ferramol strategy
  planting-day-script.md       # W16 session (neurodivergence-adapted, sensory transitions)
  warm-season-session-scripts.md  # W21-W22 transplant scripts
  troubleshooting-guide.md     # Symptom-based pest/disease diagnosis
  daily-routine-card.md        # Child-facing printed routine (visual checklist)
  weekly-garden-walk.md        # Guided observation structure
  neighbor-vacation-guide.md   # Plain-language handoff with photos
  end-of-season-guide.md       # Cleanup, perennial care, soil refresh
  [+15 more guides]

Development History

100% complete — 12 phases across v1.0 and v1.1, 25 plans executed, built March 26-30, 2026:

Milestone	Phases	Focus
v1.0	1-6	Bed infrastructure, spring/warm planting plans, data system (27 crops, 30 schedules), documentation, vacation prep
v1.1	7-12	Data consistency fix (Bed 1-5 -> A-B), child engagement (routine card, garden walk, treasure hunts), operational readiness, HA setup, season 2 prep, artifact maintenance

125 commits over 4 calendar days. Model mix: ~80% Sonnet, ~15% Haiku, ~5% Opus.

As of April 13 (W15): zero commits since March 30. The project is now fully planning-complete and waiting for the seasonal start of physical execution. The engagement assumptions, neurodivergence-adapted session scripts, and pick-2-of-4 agency model remain theoretical. This is the only project in the portfolio where "shipping faster" is meaningless — plants, seasons, and a young learner's attention operate on their own clocks. The real validation happens outdoors in W16, not on the keyboard.

Architectural Decisions

Decision	Rationale
Raised beds over in-ground	Defined ownership ("his bed"), better soil control, accessible height for child
Corten steel beds (byJEMA)	20+ year lifespan, food-safe, aesthetic match to villa
Manual watering is intentional	Watering IS the routine that builds agency — do NOT automate away
Staggered planting	Continuous harvest keeps engagement high (always something to pick, W22-W43)
JSON data from day one	Enables HA integration, alerts, structured schedules; no rework later
2-bed layout (not original 5)	Scoped down from initial plan — simpler, focused, less overwhelming for first season
Sensors deferred to v2	Manual observation teaches the child more in season 1; data format proven first
Difficulty tiers (1-3)	60%+ Tier 1 "can't fail" crops; stretch crops alongside guarantees
Ferramol over copper tape	Aarhus University research: pellets + manual patrols more effective for Danish Spanish slugs
Bilingual EN/DA prompts	Father speaks Danish to child; prompts in both languages; not child-readable text

Strengths

• Neurodivergent-first design — Every session script accounts for short attention windows, sensory needs, and self-regulation resilience; not retrofitted
• Extraordinary specificity — Centimeter-level planting positions, timed steps, visual cues ("See the V between stem and leaf? Pinch it off!")
• Continuous harvest engineering — Succession calendar ensures something is always ready to pick May through October
• Evidence-based decisions — Slug defense from Aarhus University research, companion planting from validated sources, HA sensor recommendations from community testing
• Absence-proofed — Reservoir design, neighbor guide, pre-departure checklist, and countdown session script cover the longest summer absence window
• Complete HA integration readiness — JSON crop files include temperature thresholds and moisture alerts; YAML configs are copy-paste ready
• Data-first, schema-driven — JSON schemas enable reusability and validation; no surprises when HA integration activates
• Decision documentation — Every choice recorded with rationale in PROJECT.md

Weaknesses & Risks

• Longer absences are a single point of failure — Entirely dependent on self-watering reservoir + neighbor cooperation; no redundancy
• Budget exceeded initial ceiling — 8,500-10,500 DKK vs. initial 2,000-5,000 DKK ceiling; accepted as 20-year infrastructure investment
• Danish climate risk for warm-season crops — Tomato, pepper, cucumber all marginal in zone 7b; cool/wet August could mean poor yields
• Sensor deployment deferred to v2 — HA configs written but no physical sensors; manual observation only for season 1
• Engagement assumptions not yet validated — Neurodivergent design derived from literature and parent knowledge; real sensory triggers only emerge in practice
• Single-season data — All planning is theoretical until first growing season validates assumptions
• Roof terrace structural assessment deferred — Load capacity for wet soil on older carport roof not yet verified
• Crop variety availability unconfirmed — Specific varieties (Sungold, Mini Munch, Korona) not verified at local nurseries
• No fallback if child disengages — The entire project premise depends on sustained child interest

Connection to Other Projects

• TuringPi — Home Assistant runs on the homelab cluster; HA configs in Haven target this infrastructure
• 2024.garden — Digital garden could document the physical garden journey
• CNC — Could potentially monitor HA sensor health
• Evolver — Shares neurodivergent-first design philosophy (session scripts, timed steps, sensory engagement, quick wins)
• No direct connections to Roughneck, Etyde, GoVejle, or Ollama (no async processing, no LLM, no web serving)

What Makes This Project Different

Haven stands apart from the rest of the portfolio:

• Zero application code — Pure documentation + structured data
• A non-technical primary user — A neurodivergent 7-year-old learner
• Physical-world dependencies — Weather, soil, plant biology, seasons, Danish climate
• Developmental goals — Agency, routine, responsibility as success metrics; not technical metrics
• Time-locked phases — Can't be compressed with faster coding; plants grow on their own schedule
• Evidence-based domain research — University slug research, therapeutic garden literature, HA community testing
• Bilingual, locale-specific — Danish suppliers, climate data, plant names, HA prompts throughout

This is the only project where "shipping faster" is explicitly counterproductive — the child's pace and engagement, not development velocity, determine success.