Narabi — daily

1. Fix the daily UTC window and read its block range from the chain, from_block to to_block.
2. Sum USDe burns and mints inside the range from the token's Transfer events; read the closing supply.
3. Form the velocity v: burns divided by the opening supply (reconstructed exactly as close plus burns minus mints), per hour of window. Burns only in the numerator; a true fraction of the stock, never the odds.
4. Classify the window: calm when burns stay under one percent of the opening stock and the stock is above the committed floor; then evaluable, non evaluable or clipped for the pair.
5. Score the pair against the persistence forecast (tomorrow's v is today's v), step the tracker on it, and append the line with its hash, the sentinel's version and its RPC endpoint pool (this page shows only its size).

task class

stable-run-velocity-24h

the task class of the served gate description for Narabi's redemption-flow velocity

committed key

narabi:persistence-v2@eip155:1/erc20:0x4c9edd5852cd905f086c759e8383e09bff1e68b3

the one (task class, predictor) key the committed region is locked to

calibration pairs

613

consecutive calm pairs behind q₁, counted at build from the committed calibration scores

calibration digest

c9793b28…59d9e86c

calibDigest over those scores, recomputed at build: the digest the gate's verdict carries, not the tracker digest below

Pull both files. Recompute any window from its block range with eth_getLogs and totalSupply. Rederive q_t from the s column with the tracker replay printed below. Recompute each line's hash from its fields, as this page does in your browser. The AttestedFlow documents are not served: rebuild one from its window's facts with the public sentinel code and compare its hash with the published one. If anything differs, that is the finding, and it is yours to publish.

Narabi runs an adaptive quantile tracker (Angelopoulos–Barber–Bates 2024, decaying step) on the attested daily USDe redemption flow: its state moves each 24h window from the realized outcome, and the full timeline is published so anyone can replay it. What it carries is a deterministic long-run bound that tightens as windows accumulate, printed daily with T, not a per-window coverage, not a probability; the gate's committed calibration does not depend on the tracker state. Until the pre-registered drift criterion fires and an ADR says otherwise, the gate's region is still the committed static calibration: the tracker adapts, the gate does not yet.

q_t

1.372794e-2

current threshold of the adaptive quantile tracker

q₁

1.311923e-4

the committed static calibration

T

6

pairs stepped so far, evaluable or clipped

bound_thm1

0.9767

deterministic long-run bound, tightens as T grows; a — means no stepped pair yet

bound ≤ target, projected

2031-08-11

assumes one stepped pair per calendar day from the first tracker step (2026-09-18), under the published parameters; a change of c or ε would open a new parameter segment and move the printed bound to the arbitrary-step theorem of the same paper.

tracker parameter segments

one: every published bound equals the long-run bound under the published parameters

params

alpha = 0.1 · c = 0.041666666666666664 · eps = 0.1 · t0 = 0 · B = 0.041666666666666664

digest

26f94023…ae3d8dbe

the tracker digest: folds q₁, params and every stepped score

criterion

rolling calm-miss ≥ 0.40 over the last 90 calm pairs

status

6 of 90 calm pairs — evaluable after the window fills

fired

no

# facts — window 2026-09-23
eth_getLogs address=USDe topics=[Transfer] fromBlock=26036464 toBlock=26043610
  burns = sum(Transfer to the zero address)   -> 11321263619380000000000000
  mints = sum(Transfer from the zero address) -> 5429629789870680000000000
totalSupply at the close block                 -> 4890564085271987753523503861
# attestation
sha256(AttestedFlow bytes)                     -> 67555aa0afca4ca694434f46aed6deec6392ccd2078cc34e1645abd885e6c10f
# tracker replay (recomputed in this page from q1, params and the s column)
trackerReplay(q1, params, timeline.s) -> q = 1.372794e-2
state.json replay_q (published)       -> q = 1.372794e-2
# hash chain
prev_line_hash -> 842c96328367857dcbf29582fa5a4d9bbd3fdc51d107f56f8d771088a0ac775c
line_hash      -> e69ed7b085c9a7cb9a521f0ee95391abc81e05ba5b85c59ba7d302e64dac1d31

Recomputed when this page was built, on its committed capture; a browser that runs the page recomputes all of it again, first on that capture, then on the files as served now.

hash chain

7 links intact

state.json digest equals the last line's digest_T

yes

tracker digest recomputed from q₁, params and the s column

equals state.json

tracker replay recomputed here

1.372794e-2, the same as the published replay_q to the printed digits, over T = 6 steps

supply identity (open = close + burns − mints)

holds on all 7 windows (recomputed here in exact integers)

the timeline served now extends, byte for byte, the one served at the capture of 2026-09-24

not applicable: this is the capture

state.json · timeline.jsonl · published files, read-only

sentinel_sha

e73866a8…5748f286

node_version

v24.21.0

endpoints

7 RPC endpoints in the read pool; each burns, mints and supply read needs two distinct providers that agree

lag

unknown until the files as served now are read

published at

—

the Last-Modified header of each served file

code versions (sentinel · node)

2026-09-17 → 2026-09-17 · sentinel b5a0ed…f81092 · node v24.21.0 2026-09-18 → 2026-09-19 · sentinel a87e86…5353f5 · node v24.21.0 2026-09-20 → 2026-09-20 · sentinel 2c58b8…d92295 · node v24.21.0 2026-09-21 → 2026-09-22 · sentinel a3ed49…64582d · node v24.21.0 2026-09-23 → 2026-09-23 · sentinel e73866…48f286 · node v24.21.0

a run of lines written by an unchanged sentinel build and Node version; a new build does not change the tracker parameters

source

committed capture of the published files, taken 2026-09-24; the files as served now are read once this page runs in a browser

read at

—

• ×Not a per window coverage.
• ×Not a probability of being right.
• ×Not a price, a peg gauge or a run alarm.
• ×Not a trend: seven steps is a week, nothing stronger.
• ×Not a switch: a drift opens an ADR, never an automatic change.

T

Number of pairs the tracker stepped on so far, evaluable or clipped. Read from state.json, never typed on this page.

q₁

The committed static calibration threshold, the one the gate reads.

q_t

The tracker's current threshold after T steps. Published, not consumed by the gate.

bound_thm1

The long-run bound for the current T. A dash means no stepped pair yet.

regime

calm, stress or below floor: calm when the opening stock is above the committed floor and burns stay under one percent of it.

pair

evaluable, non evaluable or clipped: whether the window contributes a step.

static miss

A stepped pair whose score exceeded q₁, the committed region. Counted, never a coverage.

tracker miss

A stepped pair whose score exceeded the tracker's own threshold before the step.

calm-miss

Share of calm pairs whose residual exceeded q₁, over the rolling window. Metadata for the criterion, never a tracker filter.

supply identity

What each window must satisfy: opening supply equals closing supply plus burns minus mints, checked on every line.

code version

A run of lines written by an unchanged sentinel build and Node version. A new build does not change the tracker parameters.

parameter segment

A run of unchanged tracker parameters. A change of c or ε would open a new one and move the printed bound to the arbitrary-step theorem.

under_calib

Too few calibration points for a population: the gate abstains for it, serving no region while its state is published. A named state, never a number.