Status: Landed
Type: Standards Track
Category: Core
Created: 2026-08-26
Requires: CIP-3
Related: CIP-30, CIP-31, CIP-39
Type: Standards Track
Category: Core
Created: 2026-08-26
Requires: CIP-3
Related: CIP-30, CIP-31, CIP-39
Cowboy Cloud and Cowboy Chain. This CIP specifies the Cowboy protocol. Cowboy Cloud runs on this protocol today. Cowboy Chain, a public network on the same protocol, is under consideration for a future release. Where the two differ, this CIP says so.
Background
Cold state reads can delay block proposals.Goal
This CIP specifies store-I/O metering and its fee market.Status / landed state
The current auth-v3 values are: envelope 9, AccountAuth-aware signer reservations, inspection 14 and the current lane/derived-bound table in §8. Historical changelog and appendix measurements retain their named table versions. The current table version is 43 (2026-11-02), with fingerprints164002283194194565011527033313008254991 (Access),
73972974678857136944856577970255514568 (Gate-1 prefix) and
39651854689389225152307557612509439205 (session extension).
The per-transaction lane-tail floor owns the signed floor rule; this section only updates its numerical inputs.
This CIP describes the access meter as it landed. The CIP-40 fresh-genesis
amendment updates §7 to Transaction v3; it preserves the Access field widths
and relative positions and does not claim that the historical v2 pins below
already implement v3. The table-21 amendment below uses the complete table-21
schedule measured on 2026-10-05.
That measurement date does not declare a deployment. Earlier version and fingerprint
records remain historical; their appendix descriptions refer to those table versions.
- Landed as the metering-audit line, including the CBQS-v2 access rows (baseline table 4). Reflected here on 2026-09-07.
- Under the regenesis MRU amendment (2026-10-08), successful result parents add two structural MRU writes. Runner receives eight charged units from Timer (1,152/893); complete callback suggestions are 426/576/576 at identity/launch/maximum. The committed rows and real estimate/sign proof pin the common parent cost. That historical route formula also included the bounded-layout selection fixed term 43, so complete fixed work is 69.
- In the payer-lane rebalance (2026-09-08, decision 2d), Access payer lanes were rebalanced:
LANE_USER_ACCESSchanged from 2,294 to 2,051 andLANE_RUNNER_ACCESSchanged from 901 to 1,144 (withRUNNER_TX_ACCESS_LIMIT), and two runner-callback reservation constants were added,RUNNER_CALLBACK_HANDLER_ACCESS_ALLOWANCE = 152andRUNNER_CALLBACK_ACCESSES_LIMIT = 314. Timer (901) and System (4,096) are unchanged and the three payer lanes still sum to 4,096. These are historical values from that rebalance; the current auth-v3 Status amendment and §8 supply the current callback allowances and capacity table. See “Two-results / two-users lane derivation and the runner-callback tier” below for the derivation and the aggregate-policy qualification. - The CBSS amendment (2026-09-22) specifies genesis version-allocation and deletion rules in table 11 (
ACCESS_WEIGHT_TABLE_VERSION = 11). Appendix A.2 imports that table’sSetSecret,DeleteSecretVersionandDeleteSecretrows. The durable version high-water mark, delete cascade and retained evidence are specified in CIP-24 §3.2 and §3.5. Deletion reserves both fixed work and committed-state-dependent evidence operations; the displayed rows include the common pause read. Other appendix rows retain the baseline provenance below. - Baseline
ACCESS_WEIGHT_TABLE_VERSION = 4(2026-09-07). The two fingerprints in §5 identify that baseline. Appendix A retains historical CBSS and Public Store substitutions, uses table 21 for the amended signed CBQS/RAS rows, and includes the entitlement and OpenAuction substitutions pinned in §5; it is a mixed-provenance reference, not a complete copy of the baseline or table 21. - For the Public Store amendment (2026-09-23, CIP-33 §2.1.2), the Public Store table is access-weight table version 13. Version 12 added the Public destination-commitment read; version 13 retains that read and adds the active Public-volume pin.
TradingPostRegisterArtifactisRk(12)+Wk(5)andTradingPostRetireArtifactisRk(6)+Wk(4); direct and staged manifest commits and volume transfer also gain a pin read. Version 13 requires a new coordinated regenesis. The version-4 fingerprints below retain their historical provenance; the version-13 fingerprints are stated in §5. - For the hire-lifetime pin correction (2026-09-23, CIP-33 §2.1.2), the hire-lifetime table is access-weight table version 14. It keeps a Public volume pinned while a non-
RevokedStore hire uses it.TradingPostHirebecomes phase-oneRk(12)+Wk(6)plus the existing split term;TradingPostRevokeHirebecomesRk(9)+Wk(3). Hire already holds the artifact, so it adds one pin-count read and one write. A hire record does not store its volume, so revocation adds the version and artifact reads, the pin-count read, and one write. This measured table requires a coordinated regenesis. It has not been deployed. - For pinned-volume deletion (2026-09-24, CIP-33 §2.1.2), the pinned-volume deletion table is access-weight table version 15.
DeleteVolumegains one Public Store pin read, becomingRk(10)+Wk(8). The earlier artifact, hire, commit, and transfer rows remain. The existing occupied-ID check and permanent garbage-collection tombstone prevent reuse without a new pin-specific creation row. The three version-15 fingerprints are in §5. This proposed table requires regenesis. It has not been deployed. - For permanent volume registration (2026-09-25, CIP-33 §2.2), the permanent registration table is proposed table version 16. It adds the permanent volume reverse index for artifact 0 and Store artifacts.
TradingPostDeclareisRk(3)+Wk(4),TradingPostDeclareVersionisRk(5)+Wk(4), andTradingPostRegisterArtifactisRk(13)+Wk(6). Artifact 0 creation inDeclareandDeclareVersionadds the creatorStorageCommitmentownership read, the reverse-index uniqueness read, and the reverse-index write. Store registration already read the destination commitment, so it adds only the reverse-index read and write. The three version-16 fingerprints in §5 identify that registration table only. This proposed table requires regenesis. It has not been deployed. - For devnet-merge repricing (2026-09-25), the merged table 16, which merges devnet into the registration-table line, reprices the per-shard terms of direct
CommitManifestandCommitManifestFinalizeand changes the table-16 fingerprints; §5.5 gives the new terms and §5 the new fingerprints. Table numbers after 13 on that line collide with devnet numbering; this CIP identifies those tables by date and descriptor (§5). - Under the table-21 amendment (2026-10-05), table 21 adds the unconditional sender delta-merge re-read to the table-20 envelope (
7 = 5R + 2W), makingE = 8 = 6R + 2W; the wallet identity read/write remain included. That historical envelope, its derived capacities, CBQS signed totals and RAS rows used the table-21 schedule; §5 records its fingerprints. Clients use the matching shared protocol parameters. - Under the entitlement amendment (2026-10-06), table 22 permanently retires the three unauthenticated role instruction rows (wire tags 33–35). All surviving reservations remain unchanged by that amendment. §5 records its three fingerprints.
- Under the payment/settlement amendment (2026-10-07), table 23 changes
OpenAuctiontoRk(16)+Wk(8), including consumed-grant indexing and all writer pause reads. The role retirement remains at table 22. §5 records the table-23 fingerprints. - Under the Bank writer-guard amendment (2026-10-08), rules-v2 table 25 measures all lazy-expiry callers with production pause and vesting-floor guards. Deposit/Withdraw, CloseCard and TransferOwnership gain fixed reads; the full Bank audit also adds issuance/consent and fiat mint/settle writer reads. Writes and bounded index scans retain their reservations. The current Bank rows in §5.6 supersede the historical Bank rows in Appendix A.2.
- The meter is enforced from block 0; there is no observe-only mode (§11).
- Live scalar
access_weight_scalar_ppm = 1_350_000at genesis (COW3659_ACTIVATION_ACCESS_WEIGHT_SCALAR_PPM); the identity scalar1_000_000is the compatibility default (DEFAULT_ACCESS_WEIGHT_SCALAR_PPM).
Historical changelog: 2026-08-27 draft to version 4
This table is historical: later amendments, including the table-20 and table-21 envelopes, changed several of these values; §8 gives the current values.Abstract
Cowboy prices execution in three consensus dimensions: compute incycles, store I/O in accesses, and persisted bytes in cells. One access is one cold StateGet; every other store operation has a deterministic weight relative to that unit. memory_cells remains an unpriced liveness bound, while recurring storage rent remains a separate ledger. This CIP adds the access meter, its transaction and receipt fields, an independent EIP-1559 fee track, the enforced store gate, and the versioned weight schedule. Native work is metered: user-signed System instructions charge structural reservations before dispatch, the per-transaction envelope charges a fixed row, and each guest host-call crossing charges one unit. (Sources: CIP-3 §§2.1, 2.4; Technical Whitepaper §§2.3, 17.1.)
1. Why this is not the third meter §8.5 forbids
The Storage Whitepaper §8.5 states:RAS MUST NOT introduce a third on‑chain meter. On‑chain operations related to storage MUST consume Cycles and Cells per Technical Whitepaper §4; storage rent MUST be maintained in the Storage Manager’s separate billing ledger.That prohibition scopes the off-chain RAS/CBFS rent mechanism. Recurring rent for retained bytes MUST remain in the Storage Manager billing ledger described by CIP-31; it MUST NOT be folded into transaction gas. (Sources: Storage Whitepaper §8.5; CIP-31 §§1–5.) The
accesses dimension prices transient QMDB reads that can block proposal execution and writes durably flushed at apply_finalized. The state-transition function charges and settles these operations using EIP-1559 rules, as it does for cycles and cells. Accesses do not price bytes held over time. (Sources: Technical Whitepaper §§State Rent, 17.1, 17.5.)
CIP-30 §3.3 already requires CIP-3 or a revision to fix a deterministic, bounded cost for trie-update work on top of per-byte Cell cost. This CIP supplies the missing store-I/O dimension and gate; it does not alter CIP-31’s off-chain billing ledger. (Source: CIP-30 §3.3.)
2. Motivation
- Cold state reads during block proposal and writes flushed at finalization dominate validator wall-time. The flat cycle charge for this work let a read-heavy actor consume seconds of I/O at little cost and push a block past its proposal deadline.
- The
accessmeter prices that store I/O in its own unit (one cold read = 1, other store operations weighted), with its own EIP-1559 basefee, per-transaction and per-lane limits, and a per-block aggregate ceiling, charged before each store operation through a single enforced gate that every host-visible read and write must pass. - Separate compute, store I/O and persisted-byte meters bound every transaction’s validator time using protocol constants calibrated on the slowest supported validator, so validators do not depend on a wall-clock reap, which different machines trigger at different points.
accesses. Read-side bytes are priced separately in cycles by measured bandwidth (STORAGE_READ_CYCLES_PER_KIB = 7); they do not enter cells and do not enter the access set (§6). (Sources: CIP-3 §§2.1, 2.2.1; Technical Whitepaper §§17.3–17.4.)
The meters price the following resources:
(Sources: Technical Whitepaper §§3.2, State Rent, 17.1, 17.5; Storage Whitepaper §8.5.)
How host operations are priced
The meter MUST NOT charge by operation name. Each host call MUST be decomposed into the store operations it performs, and the transaction pays for the committed state it touches:- The store-operation weights are
R = 1per cold point read (W_READ),W = 1per cold staged write (W_WRITE),Scan(n) = 5 + ceil(n/8)per bounded scan (W_SCAN_BASE = 5, items-per-unit8),Copy(n) = ceil(n/8)per storage copy,Fanout(k) = ceil(k/4)per derived-key fan-out,NestedFrame = 75per nested actor frame (W_NESTED_FRAME) with a one-time40-unit first-nested setup per frame chain (W_FIRST_NESTED_SETUP), and a top execution frameW_TOP_FRAME = 99. Cold/warm accounting is transaction-scoped: the first read of a key and the first write of a key charge, while repeats in the same direction are free. - Every guest host-call crossing charges
W_HOST_CALL = 1through an invocation-scoped credit: a successful positive row already includes that unit, so it is not an additional surcharge; warm, no-op, and caught early-reject calls still retain exactly one crossing unit, and unused credit is never refunded across invocations. - Every host call MUST have a weight-table row in §5.1 or §5.1.1; execution-envelope and deferred-work rows are in §5.4. CI SHOULD reject any host method without a row. A nested frame costs
40 + 75on first use and75thereafter, a scan uses theceil(n/8)slope, and each crossing carries its1-unit credit. Scan,Copy, andFanoutMUST charge from a committed or input-visible bound before dispatch, never from what the operation happens to find. A charge MUST NOT depend on cache state, hardware, latency, or timeouts. The fixedMAX_LIBS_PER_ACTORscan bound applies because no O(1) pin count provides a smaller preflight bound.- Only trusted runtime sidecars are zero-access:
gas_left,charge_gas, and the instruction-gas flush/refresh charge zero accesses. Every other guest crossing still charges its one-unitW_HOST_CALL(§2, A2), includingcontextand the cryptographic host calls (context/keccak256/ed25519_verifyeach charge theW_HOST_CALLcrossing first). What stays outside the access meter is the underlying work of those calls: bytecode and crypto compute (charged in cycles), persisted byte volume (cells), and transient RAM (the memory bound). Native STF reads are charged as specified in §3.
3. The access dimension
The unit is one cold StateGet = one access. Other operations consume the relative weights in §5.
Cold/warm accounting MUST use one transaction-scoped access set with EIP-2929 semantics. The first logical in-access and the first logical out-access for a key are charged; repeats of the same direction are warm and free. Read-then-write of one key charges once in each direction. The access set MUST be shared by nested frames and MUST NOT roll back when a nested frame reverts.
The transaction warm set and the committed-value cache MUST survive a nested revert. Only cache mutations caused by staged writes roll back, in reverse order, to their pre-write values. A committed read learned inside a reverted frame therefore remains cached and cannot cause a later uncharged physical read.
An AccessKey is the tuple (committed store namespace/prefix, key bytes as stored). Within the in-access or out-access direction, two operations are warm-equivalent if and only if they resolve to the same physical committed key. Code(hash) and Library(hash) use different prefixes and are distinct keys; same-address account reads from sender, transfer-target, or balance-flush paths resolve to one key; system-record wrappers alias the underlying record key they read.
Native work is metered
Native state-transition work is metered wherever a signer or the envelope owns it. Every signed transaction has a non-zero intrinsic access cost, the signed envelopeSIGNED_TX_ENVELOPE_STRUCTURAL_ACCESSES = 9.
- Under the per-transaction envelope (CIP-49), every signed transaction reserves
E = 9 = 7R + 2W, including the primary account’s policy read. Each additional account reserves3R + 1W = 4structural accesses. The complete row is reserved before identity I/O, even if an identity marker is already committed. These reservations stand outside the cold/warm set; deferred work hasE=0. The pure identity floor is a separate subset of this envelope:3 + 4k, rounded once at maximum governed scalar, producing primary-only floor 5. - For user-signed System instructions (A1), native System-instruction handlers charge in two phases. A fixed, input-visible structural row is reserved before any admission gate, actor-pause/bank check, or handler dispatch. A handler that then decodes a committed live-set index reads that cardinality and reserves the exact cardinality-dependent row before the first per-item walk or mutation (the write side reserves the full committed
Neven when fewer rows are ultimately selected). Each reservation, once actually made, is charged in full and never refunded, on success or rejection. Committed state therefore cannot create a refund oracle, and exhaustion preserves state. A rejection before the second phase never incurs the phase-two reservation. Both phases stand outside the transaction-scoped cold/warm access set. - For host-call crossings (A2), each physical non-sidecar guest crossing charges
W_HOST_CALL = 1, credited into the row it belongs to as described in §2. Instruction-gas flush/refresh are the only trusted zero-access runtime sidecars. - Under the native rows for CBQS-v2, CBQS dispatch is priced in both cycles and access. Its five decoded v2 rows reserve the complete structural row before the first admission check and never refund it (§5, §12).
R and each reserved write one structural W. These accumulate in the same structural meter as every other §5 operation and convert to charged units once at the boundary (§5.0). No reservation is scaled individually. Remaining native STF reads that no signer or envelope owns stay out of accesses. They cover pure admission, nonce, and basefee bookkeeping charged elsewhere; only pure in-memory work is free. Read-side value bytes are priced in cycles (§2, §6). Accesses do not price them. (Source: CIP-39 §15.1.)
Reads MUST charge before the latency-bearing store operation on the proposal path. Writes MUST charge at stage time, before their later durable flush at apply_finalized. Read weights protect the proposal deadline; write weights protect fairness and finalization cadence.
4. The enforced gate
Every actor-reachable store operation MUST route through one before-dispatch charge point. A conforming client MUST make the raw state reader/writer inaccessible to host handlers, nested frames, system-route helpers, and post-effect code except through that gate. (Sources: CIP-3 §2.2.1.1 deterministic-ordering rule.) The reference sealed-capability design has these portable requirements:- A
MeteredHostStore-equivalent capability owns the only raw store handle. Raw state construction and the typed access trait MUST be sealed outside the gate module. - Sync and async entry points MUST share one charging core, and the blocking adapter MUST exist only behind the gate.
- Every scan, copy, or fanout MUST precharge from a maintained deterministic bound before dispatch: committed/input-visible where specified, and the syscall-time read-your-writes snapshot for CIP-27 fork copy (§5.1).
Scan(n)MUST charge from the validated bound and insert its returned keys, at mostn, into the in-access warm set; a later point read of a returned key is warm, while a key outside the returned set MUST chargeR.Copy(n)MUST insert thencopied child keys into the out-access warm set, so those keys are not chargedWseparately.Fanout(k)MUST insert thekderived per-item record keys into the in-access warm set; every per-item nested dispatch MUST still chargeNestedFrame. Each aggregate MUST charge from its validated or derived bound irrespective of prior warmth of any individual key; implementations MUST NOT reduce the aggregate ton - 1ork - 1. A key outside the exact precharged or returned set that is read anyway MUST pass through the gate and charge cold, so a bound mismatch cannot create free I/O. - CI SHOULD reject raw-store imports outside the gate and any HostApi method without a weight-table row. Equivalent static enforcement is conforming if it provides the same non-bypass property.
5. Weight table
The dated amendment records below are historical. The current table-43 schedule and its fingerprints are pinned in the auth-v3 amendment above. The entitlement amendment (2026-10-06) retires the three role instruction rows (wire tags 33–35) in Appendix A.2 and recordsACCESS_WEIGHT_TABLE_VERSION = 22. That entitlement change leaves surviving instruction reservations unchanged. The payment/settlement amendment (2026-10-07) records ACCESS_WEIGHT_TABLE_VERSION = 23 and sets OpenAuction to Rk(16)+Wk(8): the consumed-grant GC index and its pause checks add 3 reads and 2 writes, while 6 further reads account for previously omitted writer pause checks.
The prior amendment has ACCESS_WEIGHT_TABLE_VERSION = 21 (2026-10-05). Its envelope and amended rows supersede the historical values below. The baseline table has ACCESS_WEIGHT_TABLE_VERSION = 4 (2026-09-07). Appendix A combines baseline host/System rows with the historical CBSS and Public Store substitutions named there, plus current E=9 CBQS signed totals in A.4 and historical table-25 RAS rows in A.5, role-row removal from table 22, and the OpenAuction row from table 23; it is not a complete inventory of any one source schedule. The historical per-stripe CBFS amendment in §5.5 retained ACCESS_WEIGHT_TABLE_VERSION = 25 and RULESET_REVISION = 1, with 32 MiB object chunks at widths up to 6 (K+M ≤ 3 moved from 16 MiB). The table version is a protocol constant. Changing relative weights requires a coordinated release under CIP-3 §4; governance may scale the whole table through access_weight_scalar_ppm within the bound in §8 but may not rewrite individual rows.
Each object must match one complete allowed tier with uniform visibility overhead: zero for public data or one 16-byte AEAD tag per private chunk. Ranged reads, including FUSE cache reads, map byte offsets to chunk indexes using the persisted descriptor’s validated plaintext tier, excluding any private AEAD tag.
The historical baseline table is identified by two FNV-128 fingerprints
computed over the closed row set:
- Executable-access fingerprint
145510929673014535888950164850774767257. - Composed tri-cost fingerprint
202721689210127662310138855754496845765.
87050058421315405739685876342088125776, the Gate-1 prefix tri-cost fingerprint is 21174235201325959043110468468392591213, and the session-extension tri-cost fingerprint is 118638424846733099013270632665704235802. The first is computed from the closed executable access rows; the second extends that snapshot with System wire opcodes, three cost dimensions, and phase-two classifications for the Gate-1 prefix, then extends it with session-work values for the third fingerprint. Appendix A retains the baseline except for three CBSS rows from table 11 (version 11), three Trading Post registration rows from the permanent registration table (proposed version 16), the retirement row from the Public Store table (version 13), and two hire-lifetime rows from the hire-lifetime table (proposed version 14). The earlier appendix used version-13 RAS stage/finalize/transfer rows, version-16 direct CommitManifest and version-15 DeleteVolume; §5.5 and A.5 now supersede those rows with the rules-v2 rent amendment. This mixed appendix is not a complete copy of any one node table. It also incorporates the role-row removal in table 22 and OpenAuction substitution in table 23. The node’s full table version and row set enter the fingerprint; any row change requires explicit provenance.
In the hire-lifetime table (table version 14, 2026-09-23), the access fingerprint is 178321879494598796522930407062259511195, the Gate-1 prefix tri-cost fingerprint is 173192047303472327226438143062800880918, and the session-extension tri-cost fingerprint is 274138959066111445865989563955173012941. Version 14 inherits version 13’s rows and adds the Public Store hire and revoke rows below. The version-13 values above remain historical; they do not identify the version-14 table.
In the pinned-volume deletion table (table version 15, 2026-09-24), the access fingerprint is 254168922177092646589884203421089811033, the Gate-1 prefix tri-cost fingerprint is 13691609927054203654716208221041767188, and the session-extension tri-cost fingerprint is 27621886539640747869149653721981724283. Version 15 adds the DeleteVolume pin read below; the version-14 fingerprints remain historical.
In the permanent registration table (table version 16, 2026-09-25), the access fingerprint is 186358957115666140290847872422173554630, the Gate-1 prefix tri-cost fingerprint is 161809815397502607899677594500796353387, and the session-extension tri-cost fingerprint is 257838973940684647473105831716727650200. Version 16 adds the permanent artifact-volume index rows below; the version-15 fingerprints remain historical. These three fingerprints identify that registration table only. After the devnet merge (2026-09-25), the merged table still declares ACCESS_WEIGHT_TABLE_VERSION = 16, but pins access fingerprint 91843525370429136499481528771868231467, Gate-1 prefix tri-cost fingerprint 244711882150756174567760762831538975818, and session-extension tri-cost fingerprint 93920369198531761669027632991107518713.
Historical table-14/15/16 labels are scoped to the tables described above. The numbers 14, 15, and 16 in this CIP are the labels those tables used. Devnet uses the same numbers for different tables: devnet table 15, the native inventory revision table, prices the native inventory revision and does not include the pinned-volume deletion table’s DeleteVolume pin read, and devnet table 14 is not the hire-lifetime table. In the merged table 16, the version comment combines both meanings under 15. Each table in this history is identified by its date and descriptor as well as its number. Those historical labels do not identify the table-21 amendment below.
For table 21 (2026-10-05), the executable-access fingerprint is 135988636987504002179005556505266809626, the Gate-1 prefix tri-cost fingerprint is 228787906373042312038736524520585212954, and the session-extension tri-cost fingerprint is 301635596982272969729431529574676081897. These identify the complete schedule, including the eight-access envelope; the earlier fingerprint records identify only their named historical tables.
For table 22, the entitlement role-retirement schedule (2026-10-06), the executable-access fingerprint is 232553860638032868139545853097188225621, the Gate-1 prefix tri-cost fingerprint is 273469661632577949209879909545110220673, and the session-extension tri-cost fingerprint is 287809953087940110961618299974873066278. These identify the complete role-retirement schedule in that table, before the OpenAuction amendment.
For table 23, the payment/settlement schedule (2026-10-07), the executable-access fingerprint is 91300246119976964880136556331643606013, the Gate-1 prefix tri-cost fingerprint is 236875437954293307667852064751605696197, and the session-extension tri-cost fingerprint is 224447412884237802700138120771193997954. These identify the complete schedule with the OpenAuction amendment; the table-22 fingerprints remain historical.
For table 25, the Bank writer-guard and capture-discovery schedule (2026-10-08), the executable-access fingerprint is 202309279313112522397761736180121213458, the Gate-1 prefix tri-cost fingerprint is 333531851423930634533423003910038535572, and the session-extension tri-cost fingerprint is 169499624693929243432403378510294644297. These identify the sampled access/tri-cost schedule of that table, including the Bank writer-guard amendments in §5.6 and the four capture-discovery rows in §5.5; committed-phase MRU/TEE reservations have separate numeric pins and are not sampled by these three fingerprints. This record remains historical.
For table 25, the TEE binding schedule (2026-10-08), the executable-access fingerprint is 187294827414688968366620474990429781293, the Gate-1 prefix tri-cost fingerprint is 1869818338373724114580100038620239273, and the session-extension tri-cost fingerprint is 242698085518949219720755821437819546588. These identify the sampled access/tri-cost schedule of that table, including the Bank writer-guard amendments in §5.6, the four capture-discovery rows in §5.5, and the TEE binding rotation/deprecation rows in Appendix A.2; committed-phase MRU/TEE reservations have separate numeric pins and are not sampled by these three fingerprints. This record remains historical.
For table 25, the settlement non-delivery schedule (2026-10-08), the executable-access fingerprint is 280397173733679593457446784025773085433, the Gate-1 prefix tri-cost fingerprint is 86146913954463348964707149284935936397, and the session-extension tri-cost fingerprint is 249950869783308284539181363345473913832. These identify the combined schedule, including the Bank writer-guard amendments in §5.6, the four capture-discovery rows in §5.5, the TEE binding rotation/deprecation rows in Appendix A.2, and settlement’s operational non-delivery supplement. They supersede the TEE binding schedule’s table-25 values 187294827414688968366620474990429781293 / 1869818338373724114580100038620239273 / 242698085518949219720755821437819546588. Committed-phase MRU/TEE reservations have separate numeric pins and are not sampled by these three fingerprints. The earlier records remain historical.
The operational non-delivery supplement is 3 reads + 2 writes per affected runner, including an assigned never-committing member after its current commit window closes. The fingerprints explicitly mix this NREV committed-phase row. A maximum nine-runner, six-reveal, three-non-revealer container settlement measures 135R/60W with 264 structural accesses reserved without TEE, or 137R/64W with 270 reserved with TEE; these exclude the eight-access signer envelope. Each snapshot-active non-revealer fits the reserved 3R/2W. Ordinary result client floors do not change; RPC estimates include the additional 5F structural accesses.
The pre-SC reap-A draft schedule (table 25, 2026-10-08) recorded executable 78718387371152216953451162405054294493, Gate-1 prefix 197750054510346681902323161737670978504, and session extension 10636182524167953481576798998912084357. Those draft values predate composition with SC and remain historical.
For table 25, the reap-A token-index schedule (2026-10-08), the combined executable-access fingerprint is 181764417234658707493998335983437962361, the Gate-1 prefix tri-cost fingerprint is 761344322944648516915419303163561564, and the session-extension tri-cost fingerprint is 232930896964031914744951243466209908657. These extend the merged SC record above with the native token-index write and byte tariff. Both NREV and TIDX are sampled; TIDX covers zero, one and 64 tokens. This record supersedes the earlier table-25 records for this schedule and remains historical.
The paged delegated-slash amendment (2026-10-09) keeps table 25. Its executable-access fingerprint is 319912662565143840648537829274096628662; the Gate-1 prefix and session-extension tri-cost fingerprints remain 761344322944648516915419303163561564 and 232930896964031914744951243466209908657. The access fingerprint extends the reap-A token-index record, preserving its NREV and TIDX samples, with the fixed slash admission row, the eight-tranche/256-unit BlockStart page envelope, and the permissionless RevealAuction lock/mirror/FIFO reservations. A paid offender reserves Rk(82)+Wk(61)+5*Scan(1), including its index read, independent of retained delegation count. Payout fan-out is reserved only for the payable cohort; unpaid dissenters cannot force a full delegation payout reservation. RevealAuction lock acquisition reserves Rk(3)+Wk(3), its exact-routing page reserves Rk(1+K)+Wk(1), its apply page reserves Rk(7+8K)+Wk(5+5K), and its final tail reserves Rk(89)+Wk(69). On this combined tree, the nine-runner/six-reveal container fixture measures 129R/60W with 264 reserved without TEE, and 131R/64W with 270 reserved with TEE, excluding the signer envelope. The exact tuples preserve SC’s operational-failure supplement and remove principal-lock reads from earned balance credits. The service reservation bounds work per turn, not completion time: a persistent Data fault stalls the global queue until operator repair. Corrupt registration fields or selection-snapshot overrides remain outside the recovery guarantee and can still fail payerless BlockStart (CIP-13 §3.6.1). The 1,089-unit BlockStart total remains fixed; Timeout gains two empty-queue discovery units from Container (542 to 544 and 77 to 75). This pslash fingerprint record is historical and superseded by the rules-v2 regenesis record below.
For the rules-v2 regenesis, in the reap-B lifecycle schedule (2026-10-09), table 25 combined the paged delegated-slash pslash schedule above with admission-owned lifecycle service. For that historical schedule, the executable-access fingerprint was 243794661460798170916120920526295940897, the Gate-1 prefix tri-cost fingerprint was 168145511714895572341926247020650150790, and the session-extension tri-cost fingerprint was 32749786728156956193380083196789874576. The LIFE samples enroll spill reads, expiry release, the ordinary service-clock read and the Council union-clock row. Commitment, crash-attestation and cancellation rows change; the fixed first-fit JobSubmit class and pslash page/rescue pricing remain as stated above. The terminal-fit sample also includes the already prepaid lifecycle release byte work: its maximum signed terminal bound is 11,364,702 Cycles, below the 16,000,000 Runner lane. This changes the session-extension fingerprint without changing that tariff. These replace the pslash fingerprints for this schedule. RULESET_REVISION remains 1 and the runtime fingerprint is unchanged. Those three fingerprints identify the historical rules-v2 lifecycle schedule (2026-10-09), before the per-stripe CBFS amendment. The historical version-25 rules-v2 per-stripe schedule (2026-10-13) has executable-access fingerprint 43898731206593650534591137215127141622, Gate-1 prefix tri-cost fingerprint 90373281343040047390234500419638149607, and session-extension tri-cost fingerprint 203813918996922481701798039249637615344. These supersede the reap-B lifecycle fingerprints for the combined per-stripe schedule. The combined tree also includes the separate table-25 route-policy cap amendment (2026-10-13): route MAX_POLICY_BYTES falls from 65,536 to 64,512 bytes, changing the policy-allocation and pin-bound tri-cost samples while leaving the executable-access fingerprint unchanged. The separate table-25 CIP-10/CIP-15 amendment (2026-10-13) removes the UpdateRouteManifest row (Rk(2)+Wk(1)) and its tri-cost classification, and changes UpgradeSelf from OptionalPoint to Composite(POINT_WRITE) with two route-slot writes/deletes; the combined fingerprints also include these changes. The per-stripe CBFS amendment retains table 25 and RULESET_REVISION 1: inventory and rent assignments move to stripe rows, validated rows are reused during mutation, and the lowest object-chunk tier becomes 32 MiB (§5.5; CIP-9 §14). Any later rules-v2 change to an access row MUST update this §5 regenesis record.
The admission-owned lifecycle calendar in CIP-2
reserves expiry and reap before creating a job, refusing a full spill window with
E1213. It uses the Timeout allowance: 544 on odd real turns and at least 288 after
priority pslash work on even turns. The current bounds are BlockStart 1,823 and
Container 75.
Admission conservatively relies on 288 per unpaused turn; CIP-2 derives
the shared clock-domain capacity, stage bounds and 416/128 = 3.25/block cheap ordinary
completed-job ceiling. The old residual-only best-effort reap is replaced.
The common job-family helper still covers attachment, seal and Storage Manager
token rows; ordinary lifecycle’s inline one-runner/no-attachment reap measures 31
structural accesses including its four wrapper operations. Native financial
finalization has a separate bounded stage. Native nonempty mount admission keeps
its token-index write and bounded byte-work charges.
The lifecycle amendment keeps the first two admission probes inside JobSubmit’s
existing fixed class, with one additional committed read per spill probe (at most
14). Cancellation uses 42R/33W plus 2R/2W expiry release; with the signer envelope,
87 structural operations charge 118 at the maximum scalar. Commit uses 9R/5W,
crash attestation 5R/1W, and each committee reveal adds one service-clock read.
Council pause, extend and unpause on a job actor add 4R/1W for the union clock.
The historical baseline rows in Appendix A retain their named provenance.
For DeleteSecretVersion, the args-only handler row is fixed(5, 8). The common pause gate adds one committed read, giving the displayed phase-one Rk(6)+Wk(8) before dynamic evidence operations. DeleteSecret has a fixed(2, 2) handler row, displayed as Rk(3)+Wk(2) with pause, then reserves V*(Rk(3)+Wk(6)) for the committed active versions and the exact indexed evidence operations for each override. Appendix A.2 imports these complete descriptive rows.
The artifact retirement row comes from the Public Store table (version 13); the three registration
rows come from the permanent registration table (version 16) and the hire and revoke rows from the hire-lifetime table (version 14). The other changed RAS rows are stated in §5.5 and A.5.
These changes are not part of the version-11 pin or either baseline fingerprint.
Under the historical table-25 GC rules, native inventory GC fault isolation adds metered freeze/ownership reads, a next-slot occupancy read before new enqueue, and bounded quarantine publication within the existing EpochSweep/System allowance (1,000 structural units, 32 items, 128 work units per item plus 24 quarantine headroom). It changes no transaction access row or client mirror and retains table 25 and its three fingerprints. Payerless sweep access observations and completion block boundaries can change; quarantine intentionally changes state roots and zero-amount attribution-event bytes.
5.0 Primitives (landed)
The access meter accumulates structural units. The gate charges each distinct cold
R, distinct cold W, NestedFrame, and each aggregate Scan/Copy/Fanout preflight (its base and per-item units summed) in raw structural weight, with no scalar applied at the gate. The access_weight_scalar_ppm scalar is applied once at a meter boundary. No host operation or allowance applies it individually:
access_limit is floor-converted to a structural meter capacity; at settlement the structural total is ceiling-converted once into the charged accesses_used written to the receipt. The per-block aggregate ceiling, the System-selection sub-lane, and each deferred child are separately owned meters that convert at their own boundaries. All arithmetic is unsigned 128-bit; charged_accesses maps structural zero to charged zero and any non-zero structural total to at least one charged unit. Implementations MUST NOT apply the scalar per operation charge: a per-charge ceiling (e.g. ceil(1·ppm) + ceil(1·ppm)) disagrees with the boundary conversion ceil(2·ppm) and is a consensus divergence. access_weight_scalar_ppm = 1_000_000 is the identity (structural ≡ charged); the genesis live scalar is 1_350_000, so a structural total s settles to ceil(s · 1.35) charged.
5.1 PVM HostApi
Each host call decomposes intoR/W/Scan/Copy/Fanout/NestedFrame operations plus its W_HOST_CALL crossing credit. Their weights follow §5.0. The complete per-method decomposition is the HOST_API_ACCESS_ROWS table, reproduced verbatim in Appendix A.1 and bound by the executable-access fingerprint above. Representative rows:
Timer host-call declarations carry a per-fire access authority. The persisted
Timer record carries access_limit_per_fire: u64 adjacent to gas_limit_per_fire. Decoders MUST accept the current version exactly and MUST NOT infer a default for any other record shape. schedule_timer stores the classic default DEFAULT_TIMER_ACCESS_PER_FIRE = 160; schedule_timer_ex may store a larger opt-in request (§5.3). The coordinated CIP-27 fork_actor row prices physical storage copy:
Copy(n) uses the syscall-time read-your-writes snapshot’s maintained kv_count,
including preceding staged inserts/deletes. The copy, byte charge, cap check and
birth accounting MUST use that same snapshot. Do not enumerate storage to obtain
this precharge count. This is a required implementation change from the pinned
prototype’s committed-counter rule, not a claim that the new ABI already ships.
The row/fingerprint and snapshot-bound tests must change together in the CIP-27
release; historical table fingerprints above remain historical. The snapshot’s
n bounds the logical source image and copied output; raw-store dispatch must
reject an over-bound image before extra work. CIP-30’s O(1) storage-root fork is
not consensus-live; its activation MUST specify its replacement Access row
before removing physical-copy charges.
5.1.1 call_actor system subroutes
These rows replace the ordinary call_actor actor/code/pin route when intercepted (EventSub get_rank/get_orderbook/get_min_bid/update_bid/topup, SKM list_account_keys/initialize_paid_stream_config/register_content_keys/revoke_account_key). SKM register_content_keys is Rk(4 + 2N) + W(out-keys) with N <= CIP7_MAX_CONTENT_KEYS_PER_CALL = 256. The user-signed System row set (one entry per wire variant) is reproduced in Appendix A.2 (USER_SYSTEM_ACCESS_ROWS) at the current table 43.
5.2 Scalar-safe allowances
Gate charges, the deferred-child and Timer tiers of §5.3, and the native reservation rows of §3/Appendix A stay in structural units inside their meter; thecharged_accesses / structural_access_capacity conversion of §5.0 is applied once per meter boundary (transaction settlement, the aggregate ceiling, each selection/child boundary), never per charge. The authorities USER_TX_ACCESS_LIMIT, ORDINARY_SYSTEM_TX_ACCESS_LIMIT, DEFAULT_ACCESS_LIMIT, the lane budgets, and the aggregate ceiling use charged units. On entry, the meter floor-converts these to structural capacity without re-scaling them. At the identity scalar structural and charged coincide; at the 1_350_000 genesis scalar a structural total s settles to ceil(s · 1.35) charged.
Client envelope sizes, class ceilings and the maximum governed scalar come
from shared protocol parameters. Size fixed defaults at 1,350,358 ppm and
charge the live scalar (launch 1,350,000):
ceil((native + 9 + 4·additional_accounts) · scalar / 1,000,000).
State-dependent work still requires estimation and a new signature.
The current empty signed actor setup is 9 + 6 + 99 = 114 structural / 154
launch-charged accesses, before handler work. The six setup reads include the
actor’s KV-byte counter, which rent bills.
Runner capacity at maximum scalar is
floor(1,158 × 1,000,000 / 1,350,358) = 857 structural accesses.
The route row is 9 + 69 + 6·[hired] + (6 + 16M if M > 0 else 0);
47 hired mounts use 842 while 48 need 858 and refuse; 48 unhired mounts use 852
while 49 need 868. These are access-only bounds, not a promise about the other
meters or policy validation.
5.3 Natural-Timer and actor-minted deferred envelopes
The deferred-child tiers are the calibrated C0 constants:
The complete general-child tier corresponds to
scale(162, p) charged
accesses: 162 / 219 / 219 at identity / launch / maximum scalar: a two-unit
gate plus a 160-unit execution tier, with the required reserve equal to the
complete tier (margin 0).
A minimal delivering deferred actor needs 10 + 99 + 1 = 110 structural accesses; a one-subscriber async event fire needs 4 + 40 + 75 + 1 = 120. The 160-unit child execution allowance therefore delivers work but cannot itself admit a nested actor call (109 + 115 = 224 > 162; D36). At the genesis scalar, charged(215) = 291 and charged(375) = 507, so an 887-unit Timer phase admits exactly one actor-Intent opt-in fire per block.
The actor-minted deferred System class includes an outgoing message to the Job Dispatcher for JobSubmit and intent_deposit/intent_withdraw/intent_broadcast[_ex] for IntentDeposit/IntentWithdraw/IntentBroadcast. It is classified from the child’s provenance/sender read; it then reserves its fixed native row from the child’s meter before the handler’s dependent store work, in full, never refunded; if the reservation does not fit, execution returns OutOfGas (code 1005) before any raw-store dispatch and commits no partial state. Conditional nested hook frames are charged live, only when they materialize. The hooked IntentDeposit/IntentWithdraw native structural models are 202/200 structural; the fixed reserved rows are Rk(8)+Wk(4) / Rk(7)+Wk(3) with the two frames live-only (ACTOR_SYSTEM_ACCESS_ROWS). JobSubmit does not scan the runner registry for metering. Its actor-minted child row is Rk(1) sender ingress (ACTOR_SYSTEM_ACCESS_ROWS), while its selection cost is a fixed+variable 43 + 5N + 6M row (the bounded-layout fixed term) reserved from the System-selection sub-lane. This second consensus meter is separate from the transaction/child access meter and drains even if the transaction later fails. N is the bounded frontier-plus-tail candidate pool min(eligible, MAX_SELECTION_CANDIDATES = 69) and M the committee; the MAX_REGISTERED_RUNNERS = 10_000 registration cap bounds the set but is not the metering row.
5.3.1 Bounded payerless native Timer work
Native timer bookkeeping has a separate structural allowance from the payer-funded handler requests above:NativeTimer registry row reserves the full 300 structural units as the Timer-native term of P_other, within the OutsideShared pool used to derive the aggregate ceiling in §8. Its successful completion scales and records that allowance exactly once. The native meter separately aggregates its operational structural use and applies the live scalar once to TimerReport.native_accesses_used and TimerReport.native_accesses_limit.
This payerless allowance MUST NOT debit fee_payer, enter handler escrow, burn, tip, refund, receipt accesses_used, or consume the separate 887-access natural-Timer handler lane. It MUST NOT feed ordinary access-basefee demand U_a. Natural timer handlers retain their live-basefee settlement and their existing exclusion from ordinary block U_a.
5.4 Execution envelope and deferred work
Since the table-20 wallet-identity amendment,DeployActor and fork_actor also cold-read the destination wallet marker through the ordinary transaction host gate before claiming an Actor address. This distinct key probe is additional to the historical shapes below; it follows transaction-scoped cold/warm accounting.
The ExecuteActor, DeployActor, and deferred event-fire rows use the §5.0 weights (top frame 99, nested 40 + 75, scan 5 + ceil(n/8)). Every library-pin scan in call_actor, fork_actor, ExecuteActor, and DeployActor MUST charge the fixed worst-case Scan(MAX_LIBS_PER_ACTOR) bound because there is no O(1) pin count. The engine-minted deferred event-fire batch, unsubscribe_event/force_unsubscribe_event bookkeeping-topic re-fire, and Copy(n) fork rule use those weights.
5.5 Signed RAS / CBFS rows
Signed RAS uses the 1,507 charged / 1,116 maximum-scalar structural ordinary-System ceiling. Complete single-account rows includeE=9; each
additional account adds four structural units (CIP-3 §2.5).
An attached authorization reserves its indexed
nonce work and the delegation wallet’s AccountAuth read, even for redundant
self-authorization: 6R + 3W = 9 additional structural accesses. The complete
floors below round once at maximum scalar.
Stage and StageRootless reserve a complete 174 structural / 235 charged
minimum once. This is the shared complete row; token/workload
flags add their declared reads before rounding. Rootless uses the volume-write ceiling; clients compute token/workload
combinations from the complete structural row before rounding.
Current staged-manifest funding
The current public staged-chunk cap is 290. The derivation pins 1,116 structural ordinary-System capacity and this cap; with the maximum relay-delta set and no shards, 290 chunks cost 1,115 while 291 cost 1,118. Current signing and Stage preflight MUST fund the complete finalize structural reservation, including rent funding and the committed pending-epoch and native-route additions. The historical rules-v2 formula below does not set these bounds. The current reservation isF′ below. Fixed 117 = 84 + 33: E=9 plus
the 53R + 16W + Scan(1) phase-one row (75), then 33 for the two unqueued rent
sources, routing and the owner auto top-up quote. Each added stripe costs
1R+2W inventory plus 7 rent assignment; each removed stripe costs 3R+4W plus 7.
Each queued epoch K adds one snapshot read and one 11-unit funding source; J=1
for a native-route stage adds its two authority reads. Signers and Stage preflight
use the full F; the node checks the same sum.
The cap alone does not establish that a complete shard/context row fits.
Stage rejects an over-cap declaration with QuotaExceeded before staging; its cumulative reservation must also fit the structural capacity.
Historical manifest audit (pre-auth-v3 table 25)
The following audit retains its older source rows and arithmetic for provenance. It is not the current capacity table: use §8’s E=9/lane/scalar bounds and the current reservation for signing. In particular its two-scan PoR row, 1,504 ceiling, 1,114 capacity, 292-chunk audit and former Stage floor are historical.SettlePorChallengethird-miss initial row meters41R + 20W + 2·Scan(4,096), with each scan costing517, plus the genesis rent snapshot’s2R+1W: 1,098 native / 1,106 signed structural, or 1,494 charged at both launch and maximum scalars. The constants retain a conservative sizing bound of52R + 20W + 2·Scan(4,096) = 1,106native /1,114signed. In that historical schedule, the retained bound determined the 1,504 cap/reserve and maximum scalar; it was not yet aligned with the metered row. The client minimum isE+1098. The sizing boundE+1106does not set that minimum; the configured default is 1,504. At launchcip31.cbfs.por_slashing_enabled = 0, PoR settlement is alarm-only.- When a slash-distribution branch runs,
SettlePorChallenge,SubmitDrainAudit, andReportUnreachableRelayreserve and perform the whole2Nphase in the same transaction, after discoveringNother relay profiles and before per-recipient store work. Their native initial rows including rent evidence are respectively1,098,1,072, and562; complete signed rows are1,106+2N,1,080+2N, and570+2N. The whole row MUST fit: at launch or maximum scalar, a drain audit with 18 other relays needs1,080+2·18=1,116 >1,114structural units. First refusing counts at identity / maximum scalar are respectively 200/5, 213/18 and 468/273. Full-cardinality paths cannot execute until staging/cursors exist. CommitManifestFinalizehas the rent-inclusive signed structural boundF′ = 109 + 8·[Ac≠Rc] + 3H + 2D + 10·Ac + 14·Rc + 2⌈A/512⌉ + 2⌈R/512⌉, with+12K+2Jat stage preflight, against 1,114.Hcounts staged chunks,Drelay deltas,A/Radded/removed shard refs, andAc/Rcdistinct added/removedshard_idvalues (stripes). The whole-stripe rule makesA=w·AcandR=w·Rc; event-page ceiling terms still count refs. Fixed109includesE+48R+14W+Scan(1)=76plus 33: 29 for both unqueued rent-funding sources and routing, and 4 for the worst-case owner auto top-up quote (3R+1W, reserved once when the owner is not the sender and a wallet draw is wanted). Inventory costs are1R+2Wper addition and3R+4Wper removal; each stripe additionally reserves 7 for rent history, giving coefficients 10 and 14. A net stripe-count change reserves 8 for PoR weight updates.Khere counts queued storage epochs, not erasure data shards; each adds one snapshot read and up to 11 rent-funding accesses.J=1for a native-route stage. Actual execution can be cheaper; admission funds the bound. Finalize’s pending queue can grow after Stage, so its context can exceed the preflight context and require restaging.- Direct commits and Finalize charge an upper bound at 1 Cell per logical persisted key+value byte: all state growth, plus full rewrites of rows sized by the transaction inputs. With
Pa=⌈A/512⌉,Pr=⌈R/512⌉, stripe widthw, and the ref/stripe/delta counts above, the application charge isC_delta = (312+107w)·Ac + (312+115w)·Rc + 109·(A+R) + 59·(Pa+Pr) + 166·D + 2394 + 260·w + 334·KCells. HereKis the durable queued-epoch count already discovered by settlement, as in the access preflight; it is not the erasure data-shard parameter. Native execution and client pricing use the same bound. Direct and staged paths pay the same application charge for the same delta and settlement context, independent of page boundaries; Stage ingress and calldata are separate. Native application reserves the additional334·Kbefore funding or manifest writes without another access. Clients use fresh pending-epoch context; for an empty delta, which reveals no width, they conservatively reservew=32while native execution charges the stored width. Width 32 remains the Cells-metering and row-decode bound; the creatable erasure grid is2 ≤ K ≤ 16,1 ≤ M ≤ 8(K+M ∈ 3..=24). - Shared read-only settlement reserves queued rent Cells immediately after queued-row accesses and before writes for
DepositVolumeEscrow,WithdrawStorageFunds(Some(volume)), manifest application and escrow-paid write admission other thanCommitManifestStage. Escrow-paid Stage admission deliberately retains its zero queued-rent Cells charge even when it writes rent rows, preserving the existing Stage client budget. For the charged paths, the first catch-up in an epoch charges334 × unsettled_rows. A later same-epoch catch-up, whether a separate transaction or a second pass inside one transaction, charges334 × 0 = 0, because it writes no rent rows. The queued rent rows are written once, by whichever catch-up reaches them first, so that catch-up pays for them. In the native application formula,Kisunsettled_rows; clients conservatively budgetpending_rowsbecause settlement can change between signing and inclusion. Account-reserve withdrawal without a volume does not settle rent. With tight Cells authority, the reservation can refuse before post-settlement terminal-state, balance or withdrawal-floor checks (E_STORAGE_WITHDRAW_BELOW_FLOOR, including its maximum-withdrawable diagnostic). A volume already terminal at ingress is still refused before settlement. OnOutOfCells, the full Cells limit counts as used, as with an out-of-gas transaction on Ethereum, so nothing is refunded; the attempted rent and manifest writes are rolled back. Both the receipt and charged Cells can change. - Application row basis. An inventory stripe/ordinal pair costs
216+32wbytes. Each stripe incarnation has one binary rent history row with a 92-byte key; the key supplies volume, shard and incarnation, leaving4+75wbytes for an open value or4+83wfor a closed value. An addition writes one open history row; a removal reserves one closed history row and the possible inventory swap, even for a tail removal. Re-adding a removed stripe uses a fresh inventory incarnation event sequence and leaves its prior history unchanged. Event pages cost 109 bytes per ref plus 59 per page, separately on each side. The166·Dterm combines 78-byte high-water writes and up to 88 bytes of commitment growth per relay delta.2394+260wcovers the new relay snapshot (281+260w, covering up to2wprimary/alternate relays), event summary (342), ack (59), event head (48), snapshot counter (78), inventory revision (66), account summary (365), scalar commitment growth (336), PoR growth (57), platform/pool rows (94), and two active/live funding sources (2·334). Each extra queued source reserves334 = 112+158+34+30for full Funding, Epoch, queue-entry and tail writes; this also covers the smaller deferred-recipient or legacy-account branch. - Only the in-place remainder of
StorageCommitment, the global stage index and existing PoR rows stays under base; their positive growth, including any new keys, is charged. Every other manifest row, including snapshots, events, funding, epoch, deferred and pool rows, counts full key+value writes, including overwrites. Physical storage/Merkle framing and ordinary sender nonce/fee persistence are outside this manifest-specific charge. Deletes append tombstones with no logical-value Cells surcharge; physical reclamation occurs later. The Cells reservation is separate fromF′and the 1,114 structural-access limit. The access fingerprints do not cover Cells coefficients; encoding and client-parity tests guard these bounds. - Signed Cells authority. Clients add the two existing 500-Cell base rows and the exact existing JSON-calldata intrinsic for the instruction they sign to
C_delta(orC_stagebelow). Computed direct/Finalize signers refreshKafter their last await before signing and reserveK+1, adding 334 Cells for an epoch captured before inclusion. Native application charges use the actual queued-epoch count; this signing margin does not change the tariff. Finalize pricing requires the complete staged delta. Manual gas overrides are caller-owned and MUST include their ownK+1margin. The native maximum atw=32,Ac=96,Rc=0,D=3,H=6,K=0is 705,070 application Cells; the same shape maximizes computed Finalize authority at 706,701 Cells, including 1,297 base/calldata Cells for the maximal 64-byte payload and 334 inclusion-headroom Cells (generic length-only bound: 706,706). Client pricing adds calldata and inclusion headroom. This is belowBLOCK_CELLS_TARGET = 4,000,000, a fee target, not a hard block limit. Larger padded direct payloads can exceed that target; the relayer’s separate request and Cells policy is specified in CIP-9 §11.1. - Computed direct/Finalize CLI and runner signers read
Kimmediately before and after/estimate_gas, then again at the final signing await. Any change in either direction discards the estimate and repeats the check, up to three attempts; continued changes return a retryable “queued epochs are changing” error without a final signature. A stable estimate supplies access authority and Cells at that sameK+1. The signed access limit ismin(ceiling, max(suggested, used + reserved) + client_access_limit(12)). The extra inclusion epoch converts 12 structural accesses to charged units at the maximum governed scalar. RPC/percentage padding and inclusion headroom are optional and capped at the lane ceiling. Epoch-growth arithmetic never causes local refusal; the node’s bound estimate and ordinary signed-limit validation determine whether the transaction fits. - The relayer checks the cumulative staged delta against the node’s 1,114 structural-access preflight and 99 combined distinct added/removed IDs (
Ac+Rc) before funding or signing; refusals determined entirely by the request occur before RPC. Fresh stages use the lower bound atK=0,J=0; resumed stages useK=0and the committed header’s exactJ. A fresh shape within2Jof the limit can still be funded and then refused by the node. Only complete byte-identical Stage instructions for the same volume, stage and ordinal are de-duplicated in the cumulative delta. Any difference, including header, authorization or token fields, is refused before RPC. Direct commits have no equivalent node args-only admission predicate. - At the historical staging capacity, with no extra route/epoch terms,
F′≤1,114setMAX_COMMIT_STAGE_CHUNKS=292:H=292,D=64,A=R=Ac=Rc=0costs 1,113, while 293 chunks cost 1,116. With no deltas/shards, 335 chunks cost 1,114 and 336 cost 1,117; the public 292 cap applied in that historical schedule. Over-budget Stage rejects withQuotaExceededbefore staging, and Finalize retains bounded reservations. Boundaries now depend on stripe width and event pages as well as stripe count. At the default widthw=6,D=64, and no context surcharge, 85 added stripes (510 refs, one stage) cost 1,100, while 86 cost 1,115; 61 removed stripes (366 refs, one stage) cost 1,104, while 62 cost 1,118. At widthw=4andD=64, 86 added stripes fit one stage at 1,110, while 87 cost 1,120. A bulk import splits 4,096 added refs (1,024 stripes) into 12 completed commits: 11 of 344 refs (86 stripes) and 1 of 312 refs (78 stripes, cost 1,030). Adding stages cannot bypass the finalization bound. The node enforcesMAX_SHARDS_PER_COMMIT=4,096added plus removed refs across a complete commit: Stage checks the aggregate for each new page before any stage write, and Finalize checks each decoded page before accumulating refs. Direct commits and individual Stage pages have the stricter 512-ref cap, as do single-stage sponsored commits; node cap violations reject withQuotaExceeded. The relayer mirrors the direct cap. These bounds supplement the funding predicate; a legal whole-stripe SDK delta that fits the predicate cannot reach 4,096 refs. An object’s shard set remains atomic; the chunk-tier, depth, split and queued-epoch conditions for a 1 GiB mutation are specified below. CommitManifestStage/CommitManifestStageRootlesseach has a 159 native / 167 complete structural floor, or 226 charged at the maximum scalar. Both reserve the conservative2R+1Wowner-nonce allowance at ingress, including cheaper owner/runner/escrow paths. Rootless forwards this allowance once. Rooted escrow Stage adds one authority read for a nonempty token; rooted defaults are 226–227 charged, and rootless uses the volume-write ceiling. The former 48-access default is insufficient.DeleteVolumehas 18 native / 26 complete structural without authorization, 21 / 29 with authorization; maximum-scalar charged floors are 36 / 40, respectively.- Historical Stage ingress Cells. Each Stage separately charges
C_stage = 2186 + 109·(A+R) + 40·D + 3268·[H>1], whereA/Rare page-local refs,Dcounts header relay deltas andHis the declared total chunk count. It reserves full key+value writes for the body (139+109·(A+R)), header (427+40·D), uploaded ordinals (1,256), and, forH>1, accumulated totals (3,268), plus 364 for global-index insertion/growth. Metadata rewrites count in full even if prior pages already wrote them. In that historical schedule, uploaded ordinals reserved all 292 entries; totals reserve the exact preflight maximum of 99 combined distinct IDs.H=1never writes totals; everyH>1page reserves them because its arguments alone do not establish completion. Only the global index’s in-place remainder stays under base. Stage’s base and calldata Cells are additional; Finalize charges application once. Client pricing prices Stage ingress independently. - Direct
CommitManifestwithout attached owner authorization is59+O+T+K+2D+10·Ac+14·Rc+2⌈A/512⌉+2⌈R/512⌉+8·[Ac≠Rc]+F_rentnative, plusE=8. Attached authorization adds three structural accesses.O=1denotes a workload-effect binding andT=1a nonempty token. The pending queue contributesKsnapshot reads; actual rent funding/routing, including the owner auto top-up quote, is bounded byF_rent<=33+11K. Estimate state-dependent work and reject unfundable shapes. An over-budget direct commit fails closed with access exhaustion; manifest/instruction effects are not applied. An otherwise admitted signed transaction retains its charged failure, nonce and wallet-identity effects (§8 and CIP-3 §2.5). Direct commit and Finalize retain the active Public-volume pin check, even for the current root.TransferVolumehas 34 owner / 37 authorized complete structural accesses, with maximum-scalar charged floors 46 / 50. - Direct manifest, Stage, Finalize and DepositVolumeEscrow each reserve one point read plus
Scan(1)(seven structural units) before dispatch. All use one bounded discovery of authoritative keyed captures, including when the active cursor or range header is absent or stale. Incomplete transaction catch-up fails closed as specified in CIP-31. Deposit is 41 native / 49 signed structural, plusKsnapshot reads and actual rent funding/routing, including the owner auto top-up quote, bounded by33+11K. - RegisterRelay, UpdateRelay, SetRelayStatus, SelfDisableRelay, SubmitDrainAudit, ReportUnreachableRelay and SettlePorChallenge include the
2R+1Wrent snapshot. RespondPorChallenge includes 15 reads and 7 writes of rent proof evidence plus its relay-auth read: 55 native / 63 complete structural, maximum-scalar default 86. Appendix A.5 lists all eight changed rows.
w=3..24), manifest depth 1–3 with no DAG repartitioning, D ∈ {w, 2w, 64}, and no queued epochs or route surcharge, the single-path model gives these worst cases for path-local updates, with one rewritten manifest node per level. The formula is monotone in D, and consensus enforces D ≤ MAX_RELAYS_PER_VOLUME = 64, so these cases bound every permitted relay-delta count:
For then-current-tier objects at widths up to 6 and
D=64, a depth-4 replace with no DAG repartitioning costs 1,108 structural accesses before context surcharges, leaving 6 units and admitting no queued epoch. In the table-25 storage-funding schedule (2026-10-13), a depth-3 replace with two extra added DAG nodes cost 1,112 and fit that schedule’s 1,114 capacity. A depth-3 replace with three extra added DAG nodes costs 1,122, and a depth-4 replace with one extra added DAG node costs 1,126; both exceed 1,114 with no context surcharge. Depth 4 can occur within the object-count cap, and manifest boundaries depend on the descriptor as well as the path, so replacements can split nodes. The client checks the complete delta and refuses an unfundable mutation before signing; the owner can delete and recreate the object in separate commits only if both previews fit, with a period when the object is absent.
In volumes holding many multi-chunk objects, forced byte cuts can make one insert or delete repack every following manifest leaf and exceed one commit even at shallow depth; clients refuse unfundable deltas before signing.
The object-chunk tiers in CIP-9 §14 in that storage-funding schedule were part of this historical bound. The depth-1–3 worst replace without repartitioning admits at most two queued epochs, at 12 units each; it still fits with the two-unit native-route surcharge. With three or more queued epochs, a near-cap replace waits for catch-up and a new preflight. Large deep-manifest replacements at wide grids need multiple stages through the bearer/relayer-key path; bearerless sponsored commits remain limited to one stage of at most 512 refs.
For the historical RAS / CBFS records superseded by rules-v2, the following formulas use the five-access signed envelope and the named tables only. They preserve the terms associated with the historical fingerprints in §5; current funding uses the current auth-v3 reservation.
-
Table-13 finalize and cap provenance (Public Store table 13, 2026-09-23, retained in registration table 16). The signed predicate was
65 + 3H + 2D + 2A + 2R + 2⌈A/512⌉ + 2⌈R/512⌉ <= 1,111. WithD=64,A=R=0, 306 chunks cost1,111and 307 cost1,114, deriving the public 306-chunk cap. Version 12 already asserted a fixed term of64and cap 306; version 13’s Public-artifact pin read made that term65. The earlier CIP figure60 + X_fhad omitted four pre-pin accesses as well as the new pin read. The former 307, 948 and 1,024 caps were superseded. -
Direct commit in registration table 16. The executable native row was
52+T+K+2X_m, and the complete signed row was57+T+K+2X_m, whereX_m=D+A+R+⌈A/512⌉+⌈R/512⌉. At the1,500identity limit, the first failing signed row hadK+2X_m = 1,444−T. The client table for registration table 16 mislabeled57+T+K+2X_mas native; a later client correction changed it to the native/signed pair recorded here. -
Historical hire-lifetime pin (hire-lifetime table 14, 2026-09-23, retained in registration table 16). CIP-33 §2.1.2 requires a Public volume reference for each non-
Revokedhire pinned to a Store artifact. Table 14 added a count update at Shared hire creation and irrevocable revocation. ItsTradingPostHirephase-one row wasRk(12)+Wk(6)plus the exact selected splitP*(Rk(2)+Wk(1));TradingPostRevokeHirewasRk(9)+Wk(3). These are the historical rows retained in A.2, not new table-21 measurements. -
Pinned-volume deletion (pinned-volume deletion table 15, 2026-09-24).
DeleteVolumegained the Public Store pin read, changing9R+8Wto10R+8W:18native /23signed structural with the then-currentE=5. -
Devnet-merge repricing (merged table 16, 2026-09-25). Native shard inventory changed additions to
2R+2Wand removals to the non-tail maximum4R+4W; a net count change added2R+1Wfor the PoR directory revision. The inventory revision added1R+1Wto direct commit/finalize, and a workload-effect binding added one read to direct commit/stage (reserved unconditionally by finalize). Finalize’s fixed native row remained46R+14W, giving signed65+X_f, whereX_f=3H+2D+4A+8R+2⌈A/512⌉+2⌈R/512⌉+3·[A≠R]; the 306-chunk cap atD=64,A=R=0remained unchanged. Direct commit was signed56+O+T+K+2D+4A+8R+2⌈A/512⌉+2⌈R/512⌉+3·[A≠R]; at the1,500identity limit, the first failure occurred when the terms after56+O+Tsummed to1,445−O−T. The client table for the merged table still carried the formulas from that later client correction.
5.6 Bank writer guards (rules-v2 table 25)
The writer-guard measurements and native rows in this table are historical. The current complete E=9 transaction totals follow the audit below. The table-25 Bank audit measures the existing card-mutating handlers before and at expiry, with full owner/agent indices, third-party native refunds, Council pause checks on every guarded write, and account vesting-floor checks. It also extends those measurements to maximum legal issuance policy/index fixtures, all default-pointer branches, due unfreeze, and fiat mint/settlement. Self-owned/locked cards are included. Council-paused admission consumes one read and no writes for every Bank variant. These reservations include the dispatch pause read; transfer-hook frames and host calls reserve their own work separately.
The historical table-25 rows above omitted writer checks even before expiry. A successful due
Deposit/Withdraw or Close adds one guarded expiry write: one read and one write.
Due SetPolicy/Freeze/TransferOwnership reject after that write; the transaction
savepoint rolls it back. Successful pre-expiry ownership transfer determines
its maximum read reservation. Transfer retains seven reserved writes against
six measured writes: the expiry write cannot co-occur with a successful transfer,
as due cards reject, just as for SetPolicy/Freeze. SetPolicy and Freeze each
retain two reserved writes against one measured write. These three rows each
retain exactly one spare write; the access oracles assert equality with the
measured maxima plus only this documented slack. Close at expiry consumes all
ten reserved writes and both 64-entry index scans. Issuance includes six Bank
writer pause reads; V2 adds the consent-marker guard, while its CUSD-hook
allow-pair write is outside the Council-pausable system-address band. Fiat
mint/settlement each include three guarded writes (replay marker, mint record,
and token balance); neither invokes a transfer hook. Card encoding, write
allowances and scan bounds are unchanged.
The table-25 rows above are also the current native rows. With the current
E=9 single-account envelope, the unhooked complete structural rows are 23
for Deposit/Withdraw, 64 for Close and 40 for TransferOwnership. At maximum
scalar 1,350,358 ppm, they require 32, 87 and 55 charged accesses.
IssueCard, IssueCardV2, MintFromFiatVoucher and SettleProvider have complete
rows 30, 35, 22 and 21, requiring 41, 48, 30 and 29. The current assembled
schedule is table 43.
6. De-conflation
TheStorage KV Read and Storage KV Write rows leave CIP-3 §2.2.1’s cycle schedule. cycles carries no per-call read or write term. Read-side bytes are priced in cycles by measured bandwidth using read_cycles(bytes) = 7 * ceil(bytes / 1024) (STORAGE_READ_CYCLES_PER_KIB = 7). The meter tracks these bytes as storage_read_bytes in GasCategory::Storage. state_get charges the returned value bytes; state_scan_prefix charges the sum of key and value bytes across the returned rows. Before either read touches the store, it requires the cycles of its largest possible result, a function of the request alone: read_cycles(MAX_ACTOR_KV_VALUE_BYTES) for state_get, and read_cycles(min(limit, MAX_PREFIX_SCAN_LIMIT) × (ACTOR_STORAGE_SCANNABLE_MAX + MAX_ACTOR_KV_VALUE_BYTES)) for state_scan_prefix (448,441 cycles at 1,000 rows). A request whose reservation does not fit the remaining cycle budget fails as an out-of-cycles charge without reading; otherwise only the returned bytes are charged, so a successful read costs what it did before. A read persists nothing, so it consumes no cells, and the byte charge does not enter the cold/warm access set. Persisted write bytes continue to consume cells. (Sources: CIP-3 §2.2.1–2.2.2; Technical Whitepaper §§17.3–17.4.)
After the change, cycles measures pure compute plus read-side byte bandwidth, accesses measures store I/O structure, and cells measures persistent bytes. memory_cells MUST remain a hard per-execution liveness bound and MUST NOT enter fee settlement.
7. Transaction fields and wire encoding
Transaction version 3 retains these four Access fields in the canonical
positions below. CIP-40 §7.1 owns the complete v3 field order:
The fresh-genesis leading
CURRENT_TX_VERSION is 3, and the four fields participate in Write, Read, EncodeSize, signing preimages, signatures, and strict decode in the frozen order above. The earlier Access-meter implementation used version 2; that historical pin is not an accepted encoding in the v3 release. origin_remaining_access is carried on the receipt/response types. (Sources: Technical Whitepaper §§2.1, 2.5.)
Deferred transactions MUST carry access parity: origin_remaining_access is optional exactly like origin_remaining_cycles/origin_remaining_cells, and a child MUST satisfy access_limit <= origin_remaining_access. Like its siblings, a deferred transaction MUST carry max_fee_per_access == 0 and max_priority_fee_per_access == 0.
The existing Transaction.access_list (EIP-2930 scheduling/prefetch) and dashboard/CBFS accessMode (a storage read-cap) are unrelated and are not reused by this meter.
8. Basefee, fees, access lanes, and governance
As specified in WP §4.5, a deferred transaction (a message, acreate_deferred_tx child, an actor-minted System child, a runner callback, or an async event-subscriber child) is paid by the actor whose code queued it, its from, through the actor’s balance and then its owner’s under FallbackToOwnerBalance, at the current lane-effective basefees with priority zero. DeployActor grants the new actor FallbackToOwnerBalance on its signer by default, with no cap, unless the deploy opts out or sets a lifetime cap; the owner can revoke or re-grant it at any time. An actor created by fork_actor receives no such grant. One exception to actor-first: a deferred JobSubmit child escrows its job price from the actor alone, so when the actor’s entitled owner can cover the child’s whole maximum fee (grant constraints and spendable balance), the owner pays that fee and the actor’s balance stays for the price; otherwise the actor-first order applies. Before it executes, its maximum cost (its cycles and cells limits and the charged conversion of its structural access limit (CIP-43 §5.0), priced exactly as settlement prices them, at those basefees) is reserved from that waterfall; if the reservation cannot be covered, the transaction does not execute, commits no state and no descendants, and its receipt records DeferredPayerUnfunded (structured code 1015). After execution the actual cost is charged from the reservation and the remainder refunded, owner’s share first. Timer fires keep their scheduler pre-charge (CIP-5 §6.3) and event-fire batches keep their subscription budgets (CIP-29). The originating transaction’s remaining gas is authority for its descendants, never a monetary escrow.
Message identity and frame rollback follow WP §3.3 and CIP-6 §9.2. Settlement applies only to retained work: a caught failed frame’s create_deferred_tx children, including those queued by on_fork, are discarded; intent-authorized deferred work and async event-fire requests retain their specified survival rules.
access_basefee is denominated in base units (1 CBY = 10^9 base units) and uses the same independent EIP-1559 update family as the other meters:
U_a = successful_bucket + min(failed_bucket, T_a) over the ordinary transactions in the block transaction list. The successful (uncapped) bucket holds both Success and OutOfGas transactions; only an ExecutionError transaction goes to the failed (capped) bucket. An OutOfGas-heavy block is therefore demand-uncapped. Each bucket adds the transaction’s own accesses_used and the System-selection access it caused, credited to the same bucket as its parent status. Natural-Timer transactions execute in a fresh end-of-block phase and are excluded from U_a (and U_c); the dedicated timer basefee remains cycle-priced, and timer accesses pay the live access price without contributing to a later block’s U_a. The payerless per-block allowances (block-start and epoch sweep) add charged usage to System-lane capacity reporting but are excluded from payer-congestion demand U_a (D19). The inherited constants are alpha = 96, and a ±1/96 per-block clamp. The cycle and cell floor is MIN_BASEFEE = 10 base units. The access curve clamps to the governed min_access_basefee; MIN_BASEFEE does not set its floor; launch genesis seeds it, and the genesis access basefee, to LAUNCH_MIN_ACCESS_BASEFEE = 35,000 base units per access.
The current capacity contract is the auth-v3 release. All access authorities below
are charged unless explicitly structural. Historical table-21/25 examples in
§5 and Appendix A do not override these bounds.
The three payer lanes sum to
2,051 + 1,158 + 887 = 4,096; adding the System
lane gives nominal 8,192, while actual shared aggregate remains 4,096. Scalar conversion happens once per owned meter, never once
per operation. A complete signed row uses E=9+4k; pure preflight’s identity
subset uses 3+4k and has primary-only floor 5, as defined in CIP-40 §7.3.
For a callback, own work is E + 68 = 77 structural; the callback tier
is 149 + 162 = 311. With charge(s,p)=ceil(s·p/1,000,000), its suggested
complete authority is ceil((charge(77,p)+charge(311,p))·11/10):
427 / 577 / 577 at identity / launch / maximum scalar. Thus two require
854 / 1,154 / 1,154 and retain 304 / 4 / 4 accesses in the Runner lane. The
current invariant explicitly pins strict headroom at both nonidentity scalars.
TEE, delegation, offenders and larger committees require their additional
rows.
The proposer seeds BlockStart with a
parent-state bound 1139+2n, 0≤n≤342, whose maximum is 1,823 structural.
At maximum occupancy that charges 1,823 / 2,462 / 2,462 at identity /
launch / maximum scalar, leaving aggregate payer authority 2,273 / 1,634 /
1,634 before a Timer hold. This is a maximum-occupancy calculation, not a
fixed charge for an empty due page.
A full 887-access Timer hold leaves 1,386 / 747 / 747. Two callback
suggestions plus two maximum User stamps require 2,902 / 3,202 / 3,202;
lane fit alone therefore does not establish aggregate full funding. The
parent-state Timer bound and selection use the current block implementation;
no selected-count promise is inferred from these subtractions. Timer tiers
remain 160 / 322 / 375 structural: five/two/two complete fires at identity and
four/two/one at launch/maximum, before other aggregate work.
Signed class ceilings remain charged authority;
engine-produced deferred work uses authenticated origin authority and its
separate complete-reservation rule. Live authority includes lane/sub-lane and
aggregate remainders after held finalization work. The signed per-transaction
floor, validation order and refusal behavior follow.
Tail floor. A signed transaction’s floor depends only on its bytes:
k is additional_signers.len(), n the instruction count; 2 is the
sender-row read/write between steps. The per-instruction dispatch table is:
A plain
Transfer therefore has floor 13 (18 charged at 1,350,000 ppm). The
floor is converted once with charged_accesses at the block’s committed scalar.
Validation order. For each block row the STF, before any lane, deferred or
executor state is touched: (1) refuses access_limit above the class ceiling
with E1900 and no execution; (2) computes
live = min(lane remaining, System sub-lane remaining, aggregate remaining − held finalization); (3) if live ≠ 0 and live < charged(floor), the
block is invalid; (4) if live = 0, refuses the transaction with E1900
(insufficient lane budget or insufficient block aggregate access authority) and no execution; (5) otherwise executes with authority
min(access_limit, live). Deferred rows have floor 0 and instead make the
block invalid when their complete reservation exceeds live. An honest
proposer never selects a signed transaction whose lane or aggregate remainder
is below max(charged(floor), client admission floor); it re-queues it and
holds that sender’s later nonces for the block.
CBSS key-release sub-lane (dormant). From the chain’s recorded
CBSS_RELEASE_LANE_ACTIVATION_HEIGHT (a CIP-50
not-before gate, NOT_SCHEDULED on every network), CBSS key-release traffic
has its own System sub-lane. A row is in it when it is signed (no origin,
Delivery::Transaction) and carries exactly one instruction that is a
KeyDelivery of any kind or the RAS RequestServiceVolumeDekRelease.
Deferred and multi-instruction rows are not release rows; they keep their
existing tx_lane classification (a zero-origin deferred row is Timer, and a
mixed instruction list can be User or Runner). The STF, proposer
and RPC admission all use this one classifier.
A release row’s “System sub-lane remaining” in the validation order is the
release sub-lane’s. It never draws on the ordinary 1,507 reserve, selection or
anchors, and they never draw on it. The shared aggregate (4,096), BlockStart
seed, natural-Timer and finalization holds and the tail floor are unchanged
and bound the sum, so the sub-lane adds no block capacity. Release cycles have
a hard cap: a release row runs only if the release cycles already used plus its
signed
cycles_limit fit within 16,400,000; otherwise it is refused with E1900
before execution. Its cycles also debit the unchanged System lane and block
totals, so other System work keeps at least the other 16,400,000 before the
System lane is spent. The cap is not a reserved share. An honest proposer
considers release candidates before other candidates and reserves their signed
cycle limits against the cap and against the System lane’s remaining cycles
(the signed limits of System rows already selected), except that a sender’s later candidate never
moves ahead of its earlier non-release one; validators cannot enforce this
order. Below the recorded height release rows use the ordinary reserve
and every receipt, root and lane report is unchanged.
The constant-envelope-only guard in the earlier §8 text is superseded by this
rule. Execution fees/refunds and failure rollback remain CIP-3 §2.5’s
rules; unused lane authority is not reassigned.
For fee-policy hardening (D21 / D24), the scalar and tri-basefee are loaded from BasefeeConfig/system:basefee_config at startup, submission, estimation, finalization, and proposal selection. The sole governance persistence path validates before writing; node-local consumers refuse an invalid committed config without panicking. Engine startup returns that refusal to main, which logs and exits nonzero. RPC interfaces that depend on committed fee state (/basefee, /governance/params, /estimate_gas, submission, node-authored runner transactions) fail closed with HTTP 500 when the committed tri-basefee or BasefeeConfig is malformed; clients MUST treat this as fee policy unavailable / do not sign and MUST NOT fabricate a default fee policy.
The access fee remains accesses_used × (access_basefee + tip_a), where
tip_a=min(max_priority_fee_per_access,max_fee_per_access−access_basefee).
Unused access authority refunds at the signed maximum fee; access basefees are
fully burned. The sender charging/rollback rules remain CIP-3 §2.5.
The codec range is separate from governance’s accepted
[1,000,000, 1,350,358] ppm interval. UpdateBasefeeConfig and SubmitProposal
carry both access target and scalar under the Tier-0 basefee configuration.
Per-row schedule changes require the coordinated release in CIP-3 §4. Maximum scalar is
floor(1,507 × 1,000,000 / 1,116) = 1,350,358: the maximum signed staged-finalize
bound determines the ordinary-System fit. At that maximum the row costs 1,507;
one ppm above costs 1,508. BlockStart, Timer, deferred and callback fits are
checked independently.
The complete one-runner four-transaction Tier-0
recovery includes E=9 once in each row and rounds each separately at maximum
scalar.
Registration requires 65,600 Cycles, 5,500 fixed Cells, 4,096 calldata Cells and
127 accesses; governance adds 60,000 Cycles, 4,280 Cells and 115 accesses.
The complete tri-meter authority is 139,718, yielding the funded-account
common basefee ceiling 954,708,770,523 atomic units (or 954,708,770 when
accounting for the maximum System multiplier). The independent u64
non-overflow ceiling is 132,028,400,590,543, not an account funding guarantee.
A split Treasury does not guarantee that any single account can fund this path;
the ceiling assumes one schedule-free account holds the complete Treasury bucket.
9. Exhaustion and receipts
ExecutionStatus::OutOfGas carries accesses_used. Access exhaustion uses that existing status and does not introduce a new structured-error code, because structured-error bytes participate in receipt_root.
- A transaction whose
max_fee_per_accessis below the block’saccess_basefeefails with the existingBASEFEE_TOO_LOWstructured error code1008. Its context appendsmax_fee_per_accessandaccess_basefeeaftercell_basefee, in that order. - A deferred child that cannot reserve its complete structural tier fails with
DeferredChildOutOfAccessescode1005(E1005), contextused,limit,child_kind,child_access_tier_structural, rolling back the requesting transaction atomically (D41).
accesses_used, remaining_access, and cumulative_accesses_used adjacent to their cycle/cell siblings in canonical Write, Read, EncodeSize, and RLP encodings, and receipt-root computation commits the resulting bytes; RPC receipt responses carry the same fields. The D37 receipt-leaf semantic property digest binds tx_hash and remaining_access so a divergent remaining_access cannot pass unnoticed.
For nested exhaustion (B7), an omitted nested-call access allowance follows the EIP-150 caller-reserve shape: reserve = ceil(remaining / 64), and the nested call receives remaining − reserve (the all-but-1/64 rule), so a caller always retains 1/64 of its remaining access authority. Exhaustion of a smaller nested cap is an isolated subscriber/callee failure and cannot suppress later work while the live lane still has headroom; provenance is carried on the receipt.
For estimate-then-sign (D22 / D24), POST /estimate_gas runs the same child-affordability formulas, ordering, and partial-prefix policy as execution and returns access_limit_suggested so a client can estimate, read access_limit_suggested, and sign. Static daemons that cannot round-trip an estimate MUST sign explicit, bounds-validated limits; a fee-policy-unavailable response (§8) means do not sign.
10. Determinism and safety
Every charge MUST derive only from committed or input-visible quantities and MUST NOT depend on observed latency, hardware cache warmth, validator model, wall-clock time, or timeouts. Charging MUST occur before dispatch; exhaustion MUST fail closed into the committedOutOfGas outcome, while a wall-clock reap remains a non-committing infrastructure backstop. Zygote is the normative production engine; its committed accesses_used, receipt bytes, and state roots are canonical. The in-process engine is a test-only reference oracle. Cross-engine differences MUST be recorded and resolved in Zygote’s favor, with Access-equality conformance and demonstrated comparison interfaces specified in §12. Access accounting MUST use the same transaction snapshot boundary as state and messages. The settlement inequality holds by denomination (§5.0): structural meter usage never exceeds the structural meter capacity, and the charged accesses_used written to the receipt never exceeds the corresponding charged effective authority. The direct wire-field inequality accesses_used <= access_limit applies to signed transactions, whose access_limit is charged. An engine-minted deferred transaction instead carries a structural access_limit, so a fully consumed 162-structural child reports 219 charged accesses at the genesis scalar. This exceeds its wire 162 while remaining within its structural capacity. The unused limit is refunded per §8.
11. Activation
The meter is enforced from block 0.ACCESS_WEIGHT_TABLE_VERSION = 4 documents the historical baseline schedule. The Public Store table (2026-09-23) has table version 13 with Public artifact validation and an Active-artifact volume pin. The hire-lifetime table records table version 14 with the Shared-hire lifetime correction (§5.5); the pinned-volume deletion table records table version 15 with the pinned-volume deletion guard; the permanent registration table (2026-09-25) records table version 16 with the permanent volume_id reverse index. Version 16 and its §5 fingerprints identify only that measured registration table; they do not identify a final launch table. Table numbers after 13 on that line collide with devnet numbering, and the devnet merge (2026-09-25) already changed the version-16 rows and fingerprints (§5, §5.5). The existing occupied-ID rule and GC commitment tombstone block deterministic ID reuse without a new pin-specific creation row. CIP-33 §2.1.2 also leaves rent-driven garbage collection of a pinned volume and perpetual OneShot availability unresolved; any selected consensus guard would require measured rows. The final table version, rows, and fingerprints MUST be measured against the complete implementation and selected together before the next coordinated regenesis. Validators on that network MUST use that selected table from block 0. Old transactions cannot be replayed under the new table, including Private registrations. A hot-merge is out of policy.
Table 21 and the three table-21 fingerprints (2026-10-05) in §5 identify the historical table-21 schedule. Tables 22 (2026-10-06, role-row retirement) and 23 (2026-10-07, OpenAuction Rk(16)+Wk(8)) are subsequent amendments in the same release. Those table-25 pins are historical. The current release uses table 43 and the fingerprints in the current-auth-v3 amendment above. This specification does not authorize a reset or claim deployment. The historical per-stripe CBFS amendment extended that version-25 schedule without a version bump.
The following D50 continuation/deferred gate record is historical: these gates were flipped to height 0 (regenesis-active) at the baseline landing:
DEFERRED_RECEIPT_AUTH_REMOVAL_ACTIVATION_HEIGHT = 0JOB_CALLBACK_ACTOR_BINDING_ACTIVATION_HEIGHT = 0RAS_GC_IS_TERMINAL_ACTIVATION_HEIGHT = 0SDK_CONTINUATION_PERMISSION_ACTIVATION_HEIGHT = 0SDK_CONTINUATION_DECODE_REJECTION_ACTIVATION_HEIGHT = 0SDK_CONTINUATION_RESUME_PERMISSION_ACTIVATION_HEIGHT = 0
PVM_ZYGOTE_ACTIVATION_HEIGHT = u64::MAX): the in-process engine was the live consensus engine and the zygote path the conformance oracle. Other dormant activation constants were unchanged by that baseline. The table-21 amendment removes ACTOR_CREDIT_TO_SENDER_ACTIVATION_HEIGHT and makes sender delta settlement unconditional; it does not introduce a replacement activation height.
12. Conformance vectors: gas-vectors.v2
cip-39-gas-vectors-v2.json has schema cowboy.cbqs.gas-vectors.v2. It removes state_read_cycles/state_write_cycles, adds one R per charged read and one W per charged write at the identity scalar, and records accesses = charged_reads + charged_writes. The normative SHA-256 of the v2 artifact is:
key_validation_cycles = 600 and each row’s key_validations count. Cycles gain
600 × key_validations; branch coverage, Cells and Accesses are unchanged.
The CBQS-v2 family (wire tag 218, version 2) is an executing, state-writing,
fee-charging interface. Its current native rows and E=9 complete signed totals
are tabulated in Appendix A.4. Its reservation is charged before
admission and never refunded. The shared CBQS artifact is dispatch-scoped and
excludes the transaction envelope, so updating E does not change its bytes.
Cross-engine conformance compares the test-only in-process reference oracle with production Zygote. For equivalent host-call traces under the same metering policy and access weights, the host-call access corpus MUST produce identical accesses_used, including alias-key, nested-revert, and non-identity-scalar cases. Receipt and state-root comparisons MUST record any differences and resolve them in Zygote’s favor; universal byte equality with the reference engine is not an established invariant.
The recorded reference verification (2026-10-07) demonstrated these surfaces: unchanged Cells and Accesses for the compared fork cases, and complete state-root, receipt-root, and encoded-receipt equality for the recorded reference vector. These checks do not establish full receipt/state equality for the entire host-call corpus. The historical access-integration consensus golden corpus was 55 re-blessed rows (21 added, 34 modified); devnet sync-3 moved no golden vectors and only re-derived the two fingerprint constants.
13. Security considerations
The access table and its version are consensus-critical. Any individual weight change requires a coordinated release under CIP-3 §4, and all clients MUST implement the exact same table, identified by the current table-43 fingerprints in the Status amendment above; §5 retains historical table records. Onlyblock_access_target and access_weight_scalar_ppm are live-governable, within [1_000_000, 1_350_358] ppm for the scalar.
Missing a store path permits unmetered I/O denial of service; charging after dispatch makes the halt depend on wall-clock work already consumed; non-canonical key aliasing makes cold/warm results diverge; an unbounded native reservation or refundable native charge creates a content-dependent refund oracle. Implementations MUST therefore enforce a non-bypassable gate, charge before dispatch, reserve native and deferred rows in full without refund, and conformance-test alias keys, nested-tree sharing, speed invariance, cross-engine Access equality (§12), and write-flush cadence. (Sources: CIP-3 §2.2.1.1; Technical Whitepaper §§Determinism, 12.1, 13.1.)
Appendix A: Baseline access rows with CBSS, Public Store, and RAS amendments
A.1 and A.3 reproduce the baseline row schedule from baseline table 4 (2026-09-07). A.2 and A.4 are the current table-43 rows. A.5 retains its explicitly historical table-25 E=8 audit; the current RAS floors are in §5.5. The baseline fingerprints in §5 identify the baseline only.A.1 PVM HostApi rows
HOST_API_ACCESS_ROWS:
HOST_ENVELOPE_ACCESS_ROWS:
A.2 User-signed System instruction rows
Entitlement role wire tags 33–35 are permanently retired and have no access rows (table 22, 2026-10-06). BankActor instruction definitions and theMAX_CARDS_PER_OWNER_BANK = 64 bound are retained in the BankActor data reference. Pricing an instruction here does not enable it; SubmitRegisterBankProposal remains rejected by the inspected node.
USER_SYSTEM_ACCESS_ROWS, one row per wire variant with its Rk/Wk/Scan/NestedFrame shape. This table contains the CBQS-v2 row (Cbqs) and the RAS sub-variant pointer (Ras); the CBQS-v2 numeric rows are in A.4, the current RAS floors in §5.5 and the historical table-25 RAS sub-variant rows in A.5.
These are the current table-43 rows (2026-11-02), one per wire variant.
A.3 Actor-minted deferred System rows
ACTOR_SYSTEM_ACCESS_ROWS, reserved in full before the first raw-store dispatch and never refunded:
A.4 CBQS-v2 decoded rows
The current native rows are the five codec-owned reservations, and each complete one-account transaction addsE=9. Round the complete row once at
launch 1,350,000 / maximum 1,350,358 ppm. Rooms carry no CBQS rent, so there
are no account rows.
A.5 Historical table-25 signed RAS / CBFS sub-variant rows
TheRas entry in A.2 dispatches to the exhaustive RAS sub-table. These historical selected rows include the rules-v2 rent amendment and the per-stripe inventory/assignment amendment (§5.5), with the 1,504 charged / 1,114 maximum-scalar structural ordinary-System ceiling. N counts all other relay profiles reserved for distribution, including zero-payout entries; O and T indicate a workload-effect binding and nonempty token; K counts pending epochs; A/R count refs and Ac/Rc count distinct added/removed stripes. Every +2N phase must fit one transaction; staging/cursors remain outstanding.