Skip to content

PRD-1-Programme.md

Latest commit

 

History

History
76 lines (53 loc) · 3.36 KB

PRD-1-Programme.md

File metadata and controls

76 lines (53 loc) · 3.36 KB

Peel Programme — Food Waste Performance Credits

PRD 1 of 2 · Hedera Agentic Society Hackathon · 11–12 April 2026 Status: background track. Function 2 (Market) is the build focus — see HACKATHON-PRD-MARKET.md.

TL;DR

A public food-waste performance platform where UK kitchens have their waste derived, not self-reported, and the top 25% of performers are minted tradable REDUCTION_CREDIT tokens by an independent Regulator Agent on Hedera. Invoices come in → mint RAW_{ingredient} tokens. POS sales go out → standard recipes back-calculate theoretical consumption. Residue is waste. At period close, Regulator Agent ranks every kitchen on public HCS data and mints credits to the top quartile.

Problem

UK hospitality wastes 920,000 tonnes/year (WRAP). Self-reporting is unverifiable. Defra mandatory reporting is landing under the Environment Act 2021. No trusted, auditable infrastructure exists.

Solution (three layers, all on Hedera)

1. Ingestion — Kitchen Agent Mints RAW_{ingredient} HTS tokens per kilogram purchased. Publishes INVOICE_INGEST event to shared HCS topic.

2. Derivation — Kitchen Agent, at period close (monthly)

theoretical_consumed_kg = Σ (POS_units[dish] × recipe_kg_per_unit[dish])
residual_waste_kg       = purchased_kg − theoretical_consumed_kg
waste_rate              = residual_waste_kg / purchased_kg

Publishes signed PERIOD_CLOSE message to HCS.

3. Ranking — Regulator Agent, at period boundary

cutoff = 75th percentile of waste_rate across all participants
for each kitchen k where waste_rate[k] < cutoff:
  credits[k] = (cutoff − waste_rate[k]) × purchased_kg[k]
  mint REDUCTION_CREDIT to kitchen k

Publishes signed RANKING_RESULT to HCS.

Hedera integration

Service Usage
HTS RAW_{ingredient} × 4 tokens (shared with Market); REDUCTION_CREDIT fungible
HCS Single global topic, all events signed by operator keys
Agent Kit v3 Tool-calling runtime for both agents
Mirror Nodes Regulator reads; third parties can re-verify any ranking

Agents

Kitchen Agent (N instances, one per restaurant) Tools: ingestInvoice, ingestPOSEvent, computePeriodClose, publishPeriodClose

Regulator Agent (1 instance, platform-operated) Tools: fetchAllPeriodCloses, computeRanking, mintCreditsToTopQuartile, publishRankingResult

MVP scope — in

  • 3 hardcoded demo kitchens, pre-populated invoices + POS events
  • Static JSON recipe book (~10 dishes, kg per unit from CoFID)
  • One PERIOD_CLOSE → RANKING_RESULT cycle executed live on testnet
  • HashScan link on every mint and message
  • Make this a stub that I can potentially expand on if there's time

Demo flow — 30 seconds

  1. Three kitchens' balances shown pre-populated on dashboard
  2. Trigger period close → Regulator Agent fetches all three from mirror node
  3. Ranking appears live: kitchen in top 25% highlighted
  4. REDUCTION_CREDIT mint renders with HashScan link
  5. "This ran on public ledger data — anyone can verify it."

Connection to Function 2

Kitchen Agents in this PRD share RAW_{ingredient} tokens with the Market (PRD 2). A trade in the Market updates both parties' balances, so Programme mass balance closes correctly even as inventory flows between kitchens. The two functions reconcile through the shared token primitive.