Reading the network

Accretion

Backing per held token, and how fast it moves. A rate that is never a yield, because nothing is claimable.

Fees lift the numerator of LF. Burns shrink its denominator. Accretion is those two forces as one rate.

Nothing here is claimable, and none of this is yield.

Accretion describes backing per held token changing over time. No position to open, no reward to harvest, no counterparty paying anything. Every rate below is a measurement of a past window — never a forecast, never an offer.

The two halves#

Fee accretion (g) — every trade leaves its fee in a pool nobody can drain. The larger half by a wide margin.

Deflation (δ) — the sink and the dead address remove ODIN from the float, so each remaining token stands against more backing.

In log space the two add:

\text{TOTAL ACCRETION} = g + \delta

expressed as % per year of backing per held ODIN. A percentage is permissible only because the base is explicitly named — drop the base and it reads as a yield, which it is not.

Reading it in LF-points#

The complementary expression:

\text{ACCRETION RATE} = g \times LF

in LF-points per year — never a percentage, never a yield.

The inversion

Two readings appeared to disagree and both were right.

The fee path g runs faster in the demand band (LF below 1) — when demand pulls ODIN out, the pools hold more of the paired assets and fee growth on that side runs harder. But LF is the share multiplier, and it runs the other way, more strongly. Net effect: the ordering inverts.

Plainly: the network earns fastest when demand pulls ODIN out; a holder accretes fastest when the network holds the tokens. Different subjects, genuinely pointing in different directions.

Fences#

Each one is a way the number can lie.

  1. Unavailable is not zero#

    A window with no computable value returns unavailable, never 0. True zeros happen; collapsing the two destroys the distinction between "nothing happened" and "we don't know."

  2. A dead-address transfer is a staircase#

    Burns arrive as discrete events. Never annualise a step.

  3. A falling index is refused#

    The pod index cannot fall. If a read says it did, the read is wrong — refuse it, don't publish a negative.

  4. Live and trailing never blend#

    A 24-hour window and a 365-day window are different measurements. Mixed, they produce something that is neither.

  5. δ must match its estimator#

    The deflation term is computed on the same basis as the fee term it's added to. And the curve floors do not enter δ — they're already inside g.

What it doesn't tell you#

One composition era. Every measurement available today comes from a single period with one pool composition and one demand regime — not enough history to establish a rate.

A condition, not a projection. The right questions are does k ratchet and does demand hold — both observable. "What will accretion be next year" is not answerable from this data.

Volume-dependent. On a quiet day it barely moves. A ratchet, not a motor.

Don't rank the pools by it. A league table compares pools whose conditions aren't comparable, and a single unusual day can dominate the ordering.

Where to read it#

Live on the odinOS dashboard, both halves on a matched window. Machine-readable with full recipe at /odindata.

Related: Liquidity-to-Float · burn accounting · the permanent bid