What we are standing behind

Newest first. A withdrawal that was itself withdrawn is shown as such rather than removed: letting the record be edited by deletion is the thing a register prevents.

• KCS

  movements · 2026-08-21 to nowstruck out 9 Sept 2026

  The KCC total-supply series these rows are derived from does not hold still, so a difference between two of its readings measures the endpoint rather than the chain. Every day from 2026-08-26 onward carries between 258 and 287 readings from a single endpoint holding 25 distinct totals — 26 on the 30th — and the day's opening and closing figures reverse from one day to the next: 168,937,756.49 falling to 168,937,299.93 on the 26th, rising to 168,937,911.64 on the 27th, falling to 168,936,754.24 on the 28th, rising to 168,938,222.11 on the 29th. KCS is capped at 200,000,000 and has no mint. A total that rises 1,467.86 one day and falls 1,157.40 the next is an explorer answering unsteadily, and nothing drawn from it is a measurement of anything created or destroyed. Both directions, because the swing has no preferred sign. The rule this replaces retired burns only, so the nights the subtraction came out positive were withheld and the nights it came out negative were published: seven derived mints reached the page as issuance, $32K in all, under the words "measured from supply". What the published KCS burn figure is made of is untouched by this. It is 43 collected transfers to the sink read from Etherscan's API, each with its transaction, 2021-02-25 through 2026-06-14, plus the single derived burn of 0.018376294 KCS on 2026-08-20 that a reading of the sink itself confirms to the last digit — which is why the bound opens on the 21st and not before.

• PUMP

  movements · 2026-08-26 to 2026-08-26struck out 31 Aug 2026

  A day the collector already read, derived a second time. Two transactions carry 2026-08-26 on the chain — 987.00 under `2Wx7HYKVqRbb` and 8,124,984,503.92 under `4nv7nZ2qyiBK`, both written on the 27th — and on the 28th the stand-in derived the same day again as one row of 8,124,985,490.92, their sum to the cent. Nothing moved twice; the day was measured twice. What is withheld is the derivation, because the transactions are the better evidence: they name hashes, addresses and blocks, and a reader can check them. The chart was never wrong — `preferDerivedSelect` splits derived from observed and the day takes one, and here the two agree to the cent — but the event ledger listed the day three times and the cadence strip published 16,249,970,981.83 for a day on which 8,124,985,490.92 moved. `drop-tiled-derivations` cannot reach this: its candidates are derived on both sides of the comparison, so a collected transaction can never be the narrower row covering a wider one. The stand-in has since been taught that a transaction hash is a source naming itself and answers for its day, so re-deriving no longer puts this back.

• RENDER

  movements · 2026-08-30 to 2026-08-30struck out 31 Aug 2026

  Not a movement: the component changed meaning under it. `solana:jupiter-lock` stopped being the cumulative total ever deposited and became the balance still locked, and the reading changed with it. Seven Solana tokens gained the component between 11:50:45 and 11:52:08 on 2026-08-30 and every one of them fell within twenty-eight seconds of the same minute — HNT at 14:36:52 by 78.3%, JTO 14:37:02 by 95.9%, JUP 14:37:04 by 66.7%, PENGU 14:37:16 by 97.5%, PUMP 14:37:18 by 60.0%, RAY 14:37:20 by 90.5%, RENDER 14:37:20 to zero. Seven independent escrow sets do not empty by seven different proportions in the same half minute; the reader changed, the chain did not. The nightly derivation at 00:18 the next morning differenced the two definitions and booked the result as a release for four of the seven. What is withheld here is that difference. The methodology guard that declines a pair whose readings were taken under different versions reached production an hour later, and the deriver now refuses this span; JUP's row was withdrawn the same day. An earlier justification on this record called it a registration, which is the JUP escrow-address case and not this one: the component was already being read here, and it was its meaning that moved.

• PUMP

  movements · 2026-08-30 to 2026-08-30struck out 31 Aug 2026

  Not a movement: the component changed meaning under it. `solana:jupiter-lock` stopped being the cumulative total ever deposited and became the balance still locked, and the reading changed with it. Seven Solana tokens gained the component between 11:50:45 and 11:52:08 on 2026-08-30 and every one of them fell within twenty-eight seconds of the same minute — HNT at 14:36:52 by 78.3%, JTO 14:37:02 by 95.9%, JUP 14:37:04 by 66.7%, PENGU 14:37:16 by 97.5%, PUMP 14:37:18 by 60.0%, RAY 14:37:20 by 90.5%, RENDER 14:37:20 to zero. Seven independent escrow sets do not empty by seven different proportions in the same half minute; the reader changed, the chain did not. The nightly derivation at 00:18 the next morning differenced the two definitions and booked the result as a release for four of the seven. What is withheld here is that difference. The methodology guard that declines a pair whose readings were taken under different versions reached production an hour later, and the deriver now refuses this span; JUP's row was withdrawn the same day. An earlier justification on this record called it a registration, which is the JUP escrow-address case and not this one: the component was already being read here, and it was its meaning that moved.

• PENGU

  movements · 2026-08-30 to 2026-08-30struck out 31 Aug 2026

  Not a movement: the component changed meaning under it. `solana:jupiter-lock` stopped being the cumulative total ever deposited and became the balance still locked, and the reading changed with it. Seven Solana tokens gained the component between 11:50:45 and 11:52:08 on 2026-08-30 and every one of them fell within twenty-eight seconds of the same minute — HNT at 14:36:52 by 78.3%, JTO 14:37:02 by 95.9%, JUP 14:37:04 by 66.7%, PENGU 14:37:16 by 97.5%, PUMP 14:37:18 by 60.0%, RAY 14:37:20 by 90.5%, RENDER 14:37:20 to zero. Seven independent escrow sets do not empty by seven different proportions in the same half minute; the reader changed, the chain did not. The nightly derivation at 00:18 the next morning differenced the two definitions and booked the result as a release for four of the seven. What is withheld here is that difference. The methodology guard that declines a pair whose readings were taken under different versions reached production an hour later, and the deriver now refuses this span; JUP's row was withdrawn the same day. An earlier justification on this record called it a registration, which is the JUP escrow-address case and not this one: the component was already being read here, and it was its meaning that moved.

• SUI

  movements · the whole seriesstruck out 31 Aug 2026

  Derived while module:stake_subsidy was classified a schedule, which it is not. The reserve pays staking rewards, and a staker being paid is not an unlock any more than a validator is — the same call the register already makes on Aster Ecosystem Distributor Safe, in the same words, for the same shape of balance. These eleven rows are that reserve draining by a flat 282,429.536481 an epoch, net of the storage fund rise, published as tokens entering circulation. The classification was cleared on 2026-08-31 on the argument that a balance which never rises is not two-way churn; that is true and it is not the question the flag answers.

• SUI

  movements · 2026-08-19 to 2026-08-29struck out 31 Aug 2026

  Derived from `module:stake_subsidy`, which the register calls a queue. The release rule sums only the parts the register puts on a schedule, and Sui has exactly one — `module:storage_fund` — which rose on every pair in this span. No unlock is derivable from these readings at all: each of these ten rows is the subsidy reserve draining by a flat 282,429.54 a day, net of the storage fund's rise, published as tokens entering circulation. A queue's drain booked as a release is the defect the keep-list exists to stop, and what it cost before was 55 billion LUNC of invented releases out of the oracle pool. Written by one re-derive on 30 August at 15:25 and left behind when the register settled; the nightly derivation has produced only the storage-fund lock ever since, which is why 30 August carries no unlock and the days before it each carry one.Withdrawn again 31 Aug 2026: Taken back within the hour. Retiring these ten left Sui with no ledger at all — its locks were already retired unbounded — and the classification the retirement rested on is itself the thing in question: `module:stake_subsidy` falls by a flat 282,429.54 every single day, which is the shape of an emission schedule and not of a queue. Whether that reserve is a queue or a schedule decides both these rows and every day after 30 August, and it is a register decision, not one to settle by withdrawing the evidence.

• JUP

  movements · 2026-08-30 to 2026-08-30struck out 31 Aug 2026

  Read during the deploy that replaced Jupiter Lock deposits with the balance still locked: for 34 revisions between 11:50 and 14:32 UTC the escrow component was captured as 1,258,791,583 JUP, the cumulative deposited figure, against 418,973,365 still locked. The 838,323,261 difference is the two readers disagreeing, not a release.

• SUI

  movements · the whole seriesstruck out 30 Aug 2026

  Derived under a classification the register has since corrected. Sui registers three lock components; module:stake_subsidy was marked a queue, whose fall is churn rather than a release, and the derivation therefore summed only module:storage_fund — 27,909 tokens across the year, drawn as movement into a lock. The subsidy is not churn: it pays out 282,429.54 SUI on every day we have read it, without deviation, which is the shape this register calls a schedule everywhere else. It is now one, and the same spans derive from the full schedule set as releases of about 280,000 a day. Both cannot stand: they are the same days measured on two bases, and the newer one is the one the rest of the ledger uses. What is still not established, and is not claimed here, is whether that daily figure holds across a step in Sui published subsidy schedule - eleven days is the whole basis for calling it constant.Withdrawn again 31 Aug 2026: Taken back with the classification it rested on. It withheld the storage-fund locks as derived under a classification the register has since corrected; that correction has itself been reverted, so the locks are what this token measures again.

• FIL

  movements · 2026-08-12 to 2026-08-12struck out 30 Aug 2026

  Not a movement. The lock component is named two ways across revisions of the same day — fil:locked on one and module:fil_locked on the next — and only the second is registered, so the reading that spoke for 12 August summed to zero while the one for the 13th summed to the whole balance. The difference, 65,400,095 FIL, was booked as a lock. The balance itself never moved: 65,407,510 on the 12th and 65,400,095 on the 13th, a fall of 7,415, in line with the 154 to 982,000 a day either side. Rebuilding this token's whole year from the readings produces 368 flows identical to these except for this row and two already withdrawn. The deriver no longer differences a pair whose set of summed components changed.

• FIL

  movements · 2025-08-25 to 2025-08-25struck out 30 Aug 2026

  Not a movement. The lock component was renamed between these two readings — fil:locked and module:fil_locked are the same balance under two spellings — and the deriver, reading a missing key as zero, booked the whole balance as a movement. Real change across the span is under 50,000 FIL. The deriver no longer differences a pair whose set of summed components changed.

• FIL

  movements · 2026-08-19 to 2026-08-19struck out 30 Aug 2026

  Not a movement. The lock component was renamed between these two readings — fil:locked and module:fil_locked are the same balance under two spellings — and the deriver, reading a missing key as zero, booked the whole balance as a movement. Real change across the span is under 50,000 FIL. The deriver no longer differences a pair whose set of summed components changed.