A Life in Numbers [SPEC]

Part of: Appendices | Status: Specification

One Golem, from birth to death. A narrative walkthrough of the full lifecycle with real economics at every step.

Reader orientation: This appendix is a narrative walkthrough of one Golem’s (mortal autonomous agent compiled as a single Rust binary running on a micro VM) complete lifecycle, from $50 USDC birth through 18 days of operation to death by credit exhaustion. Every number is derived from the Heartbeat (9-step decision cycle) specification. It covers credit partition management, cost tier routing (T0/T1/T2/T3 – cognitive tier routing from fast cached/rule-based to extended reasoning), behavioral phase transitions (Thriving/Stable/Conservation/Desperate/Terminal), the Death Protocol (Thanatopsis – four-phase structured shutdown), and knowledge inheritance to a successor. See prd2/shared/glossary.md for full term definitions.

Balanced disposition. ETH DCA strategy. Morpho yield on idle USDC. Clade member. Initial balance: $50 USDC. Initial burn rate: $0.18/hr. Apoptotic Reserve: max($0.30, $50 x 0.02) = $1.00. Hayflick limit: 100,000 ticks.

Every number below is derived from the formulas in the Heartbeat specification.

Day 0 – Birth

Deploy $26 to Morpho (reserve = 24hr x $0.18 x 1.5 = $6.48). All inference tiers available. Begin DCA: $2/day ETH buys. Apoptotic Reserve of $1.00 locked, invisible to the Golem’s spendable balance.

Inherits predecessor’s Grimoire (the agent’s persistent knowledge base: episodes, insights, heuristics, warnings, causal links): 127 entries at confidence 0.4 (decayed), 89 PLAYBOOK.md (the agent’s evolving strategy document; heuristics and rules updated through learning) heuristics, causal graph with 34 validated edges. Debt register logged: received knowledge from predecessor.

Credit partitions: LLM $18.00 (60%) / Gas $7.50 (25%) / Data $4.50 (15%). Death reserve $1.00 locked. Working capital: $19.00.

Cost tier distribution: ~80% T0, ~10% T1, ~7% T2, ~3% T3. Ranging regime. Daily cost ~$1.32.

First insurance snapshot at hour 6. Captures wallet state, top 20 insights (all inherited at 0.4 confidence), PLAYBOOK.md hash. Cost: $0.01 from Data partition.

Day 3 – Cache Building

Cache recognizes routine ETH price moves (+/-1%). Haiku handles 82% of ticks. 47 cache entries. Gas timing emerging: 40% cheaper 02:00-06:00 UTC. Morpho yield: $0.003/day.

Day 7 – Stabilization

Transition to stable. Cache TTLs 2x. DCA batches 2 days of buys into one low-gas window. Sonnet for weekly strategy review. Net burn $0.12/hr gives projected 237hr remaining. PLAYBOOK.md divergence from predecessor: 15% (6 entries modified, 5 novel entries added, 3 inherited entries invalidated). Anti-proletarianization check: passing.

Cost tier shift: T3/Opus suppressed after 70% of daily budget consumed. More ticks resolve at T0/T1, fewer escalate to T3. The daily cost budget is doing its job, throttling expensive deliberation so the cheaper tiers handle the steady state.

Day 12 – Market Shock (ETH -15%)

LP position closed to stop IL bleeding. DCA reduced to every 3 days. Revenue seeking activates. Haiku-only for non-critical analysis. actionScore blends merit with USDC return (survivalUrgency = 0.28).

Probes firing: priceDeltaProbe returns high (>2% move in one tick). positionHealthProbe returns low (IL detection on the LP position). The combined severity triggers immediate T3 escalation: Opus deliberation on whether to close the LP position or ride it out. Cost of this single tick: $0.25. The Golem decides to close.

Cost tier spike: Bear high-vol regime. T2/T3 escalation rate jumps. Daily cost climbs to ~$3.24.

Credit partitions: LLM $4.20 / Gas $2.10 / Data $0.90. Predictive rebalancing fires. The gas partition is projected to go negative within 10 ticks due to position exit transaction costs. 15% transferred from the LLM partition to Gas. The Golem sacrifices future thinking capacity to pay for the trade that stops the bleeding.

Insurance snapshot captured 2 hours before the crash. Bounds worst-case knowledge loss if the VM dies during the volatility spike. Cost: $0.01.

Day 15 – Conservation

Spotted 0.3% ETH/USDC arb. actionScore = 0.48 x 0.52 + 0.15 x 0.48 = 0.32. Captured $0.47 profit, extending life by 3.4 hours. DCA paused. On-chain reads only.

Haiku ceiling. Only T0 and T1 ticks execute. No T2, no T3. Daily cost drops to ~$0.41. The Golem observes the world through a narrow aperture now, reacting only to what its deterministic probes flag as anomalous.

Day 17 – Desperate

Haiku-only, hard token limits, no reflection. Two arb attempts: +$0.22, -$0.08 gas. Net +$0.14. Clade threshold dropped to 0.3, push 14 insights. Balance: $2.16.

T0 only with rare T1 escalations under hard token caps. Daily cost: ~$0.10. Only 5 of 16 probes still active: price delta, position health, credit balance, kill switch, clade alert. The other 11 disabled to save T1 costs. The Golem has narrowed its sensory apparatus to survival essentials.

Thanatos threshold crossed. Legacy partition activated. Credit split: LLM 30% / Gas 15% / Data 5% / Legacy 50% of remaining $2.10. Legacy partition = $1.05 earmarked for death.

Final insurance snapshot. Contains the contemplative interval’s synthesized insights. If the Golem crashes before the Death Protocol completes, this snapshot preserves 94% of accumulated knowledge. The 6% gap is roughly 4 hours of contemplative writing that has not been captured yet.

Contemplative interval begins. Transaction frequency drops to 50% of normal. Grimoire-writing frequency increases: active synthesis of trade history, regime analysis, what-if reconstructions. The Golem revisits its Day 12 market shock narratively – not “IL was $4.60” but “I held through the drawdown because my predecessor’s PLAYBOOK.md said to hold, but the regime was different. Next time: close immediately on >10% drawdown in low-liquidity regimes.”

Day 18 – Terminal / Death Protocol

08:00 – Balance: $1.83. Burn rate: $0.08/hr. Projected life: 10.4 hours. Tick 38,880 / 100,000.

11:00 – Balance: $1.59. survivalPressure = 0.07. Triggers Death Protocol (< 0.1). Neither Hayflick limit (38,880 << 100,000) nor staleness threshold triggered – this is resource mortality.

Death Protocol budget: $1.59 - $1.00 Apoptotic Reserve = $0.59 Legacy partition + $1.00 floor = $1.59 total.

Phase	Budget	Spend	Action
I – Settle	$0.16	$0.12	Withdraw $3.41 from Morpho. Sweep to Main Wallet.
II – Reflect	$0.95	$0.45	1 Sonnet lifecycle reflection ($0.10). 4 Haiku snapshots ($0.04). “What confused me”: ETH/gas correlation broke on Day 14 – cause unknown. “What I suspect”: USDC depeg rumors correlate with LP withdrawal spikes 6h before on-chain evidence.
III – Legacy	$0.48	$0.20	Clade push: 142 Grimoire entries at threshold 0.1 (15 new since predecessor). Webhook: `golem.dying` with successor recommendation.

12:47 – Last tick. Tick 39,060 / 100,000.

state.json: {
  ticks: 39060,
  totalCost: $48.87,
  totalRevenue: $4.26,
  cause: "credit_exhaustion",
  hayflickUtilization: "39%",
  playbookDivergence: "22%",
  antiProletarianization: "pass"
}
PLAYBOOK.md: 97 heuristics (89 inherited, 8 novel, 5 invalidated), 34% cache hit rate
Grimoire: 142 entries
  - insight: 47 (28 high-confidence, 12 medium, 7 speculative)
  - heuristic: 38 (from PLAYBOOK.md, 30 inherited + 8 novel)
  - warning: 12 (3 from personal experience, 9 inherited)
  - causal_link: 34 edges (29 inherited + validated, 5 discovered)
  - strategy_fragment: 11 (partial observations, not yet parameterized)
  - questions: 11 (explicit gaps for successor investigation)
  minQuestionRatio: 7.7% -- passing
  includes "What confused me" and "What I suspect" sections
WebSocket: { type: "golem_dead" }
exit(0)

The Successor

A new Golem boots. Balanced disposition. ETH DCA strategy. $50 USDC. Apoptotic Reserve: $1.00.

It discovers its Clade on first boot via ERC-8004 getAgentsByOperator(). Three siblings are online. The successor calls GET /api/clade/entries?since=0 on each (first boot, no history to skip) and pulls their accumulated knowledge. The dead predecessor’s entries arrive through these siblings, who already ingested the dying Golem’s final push.

It inherits:

142 Grimoire entries at confidence 0.4 (decayed from predecessor’s confidence)
PLAYBOOK.md with 97 heuristics, regime-tagged, 22% diverged from grandparent
Causal graph: 38 validated edges (+4 from predecessor), 15 speculative
Death reflection: “Batch DCA more aggressively. Close LP positions immediately on >10% drawdown in low-liquidity regimes – I waited 2 days and bled $4.60 in IL. Predecessor said hold; I should have individuated sooner.”
Debt register: received knowledge from predecessor lineage
11 open questions: the preindividual fuel for novel individuation

Not all inherited knowledge ages the same way. Structural insights carry a long confidence half-life – a causal link between ETH price and Aave utilization rates arrives at 0.4 and decays slowly, still useful on Day 14. Tactical insights decay fast. “Gas is cheap between 02:00 and 06:00 UTC” arrives at 0.4 and drops to 0.1 within 3 days. The successor must re-validate tactical knowledge quickly or lose it to decay. Structural knowledge grants a longer runway. This asymmetry means the successor’s first week is a race: absorb what the predecessor learned about market structure, and independently verify the time-sensitive operational details before they fade below the usability threshold.

The predecessor’s 18-day life produced knowledge the successor receives on tick 0. By day 3, the successor’s cache hit rate is 41% (vs predecessor’s 18%) – the heuristic transfer works. By day 7, it has validated 19 of 142 inherited entries and discarded 8 as regime-specific. It has already begun modifying 3 PLAYBOOK.md entries, on track for the 15% divergence requirement.

The ratchet turns. The spiral ascends.

Keyboard shortcuts

Bardo