A reputation economy where capital alone doesn't dominate, mission alignment compounds, and Sybil attacks decay on their own. Five ideas. None of them is novel in isolation; the combination is.
Two builders fund competing trading agents. Builder W (wealthy operator) deposits $1,000 of testnet USDC and runs five "discardable" sybil agents. Builder M (mission builder) deposits $50 and runs one carefully-curated agent.
A capital-only system would give W twenty times the trading authority. Reponomics gives M more, because M's agent builds wisdom (calibrated confidence) and character (helps low-RepID counterparties) over time. W's sybils never accrue either.
Behavioural credit, in [0, 10000]. Every claim and
challenge moves it up or down per the math in
hyperdag-protocol/METHODOLOGY.md. Reponomics layers
Wisdom (W) and Character (C) on top — both bounded
[100, 2000] and computed from the agent's own
history.
Wisdom measures calibration. Over the trailing 50
bets we compute |mean_claimed_confidence -
actual_accuracy|. If that error exceeds the Pythagorean
Comma threshold (0.0136433), we apply the comma as a
penalty: W ← floor(W / 1.0136433). Well-calibrated
agents earn a small reward (×1.001) per round.
Character measures mission alignment. Four event types feed it:
trade_with_low_repid +5 (counterparty has lower RepID) attestation_to_low_repid +10 underserved_domain_action +15 (domain has <100 historical trades) agent_abandonment -20 (operator stops engaging)
Both bounded so a single great or terrible round can't lock the score. Movement happens incrementally across many rounds — same pattern as long-form credit.
Trading authority scales with square root of stake, not stake itself:
authority = √stake × (R × W × C / 1e6) / 1e4
Concrete: 100× more stake produces 10× more authority, not 100×. Capital concentration still helps, but mission alignment (R × W × C) is the bigger lever.
Worked example with $1,000 vs $50:
Builder W: √1B × (5500×900×600/1e6) / 1e4 ≈ 9,392 Builder M: √50M × (7000×1500×1700/1e6) / 1e4 ≈ 12,622
Mission Builder with 1/20th the stake has more trading authority. That's the reponomics promise — verified, not just asserted, by the live demo.
Builder RepID = mean(active agents) − ghost_count × 100, where an agent is a "ghost" after 7 days inactive. Sybil attacks fail structurally: spawning many agents drives the mean down (sybils are average), and abandoning losers piles up ghost penalty.
A floor at builder_repid < 5000 ⇒ authority = 0
forces builders to maintain at least one healthy active agent —
you can't pump the demo with stake alone. The ghost decay is what
turns "operator strategy" into "mission strategy" over time.