# QIP-6 proposes a $5 seat cap floor for up to 14 sessions

Draft. No ballot is open and no floor is active.

## In short

Each QUOTUM seat has an inference cap for the session: a maximum it can use, not a guaranteed grant. The contract sets that cap from measured input, roughly a day of net NVDA inflow into the sink, split across the seats at opening. QIP-1's $10 floor covered sessions 27 to 40 and ended at the 4 November bell; session 40 also ran as QIP-2's first max day. Since then, caps follow measured input again: session 41 opened with $42.20 of measured input for 17 seats, about $2.48 a seat.

QIP-6 asks whether quotum should raise opening input so that each seat's cap is at least $5 per session for up to 14 sessions, when safety checks pass. The cap can be lower if the supply-value bound applies or seats join just before opening. The extra use, and the buy-and-burn of cap nobody uses, are paid from NVDA the sink already holds. It is QIP-1 again at half the floor and with a lower reserve buffer. The $5 is a cap for the seat, shared across the seat's keys, not a payment. On days when measured input already gives each seat $5 or more, the cap is unchanged; that session still counts as one of the 14.

The vote signals a preference. It does not execute anything by itself. The team publishes execution receipts or a specific refusal, and passage cannot bypass the funding, accounting, provider or reserve checks.

## Why now

In sessions 27 to 39, the floor sessions before that max day, seats used $313.56 of inference, $24.12 per session on average, and $1,426.57 was bought and burned. Use fell over the pilot: sessions 34 to 39 averaged $9.60 of inference per session for all seats together, against $36.57 in sessions 27 to 33. About 82% of what those sessions settled was burn rather than inference. At 17 seats, a $10 floor would open $170 a session. These figures are a sample of the pilot, not a full account of its reserve use: other approved policies also drew on the reserve in that period.

Session 41's cap was about $2.48 a seat. QIP-6 tests a $5 cap target, $85 a session at 17 seats, still well above recent use and half of a $10 floor's gross opening input. The supply-value bound can make a cap lower, and provider-funding safeguards can limit access further. QIP-6 promises no particular workload and no $5 grant to any seat.

The free reserve read on 3 November was $3,428.07. Subtracting only bell 40's $831.92 burn gives about $2,600, before later inflow, provider obligations and NVDA price moves. This is an illustration, not a current read. At that balance and 17 seats, QIP-6's first check, $2,190, would pass, while a 14-session $10 floor under QIP-1's checks would need $4,380. A fresh reconciled reserve read is published before each opening, and every check runs again each time.

Measured input is an approximate-day net inflow, after excluding transfers straight from the curve and pool and subtracting OPS claims. Manual transfers and claim timing can move it. An OPS claim inside the window lowers measured input and so raises the uplift; each session record lists the OPS claims and manual transfers inside its measurement window. It is not an audited daily fee total.

## What changes

For an opening eligible under the pilot's activation and session rules, while no stop has latched and every check passes, the keeper passes the larger of measured input and $5 times its count of seats before open into TaxSink.open. The contract divides that input across the live seats, so each seat's cap is at least $5 unless the supply-value bound is lower or seats join between the keeper's count and the open transaction. Otherwise, outside a QIP-2 max day, the keeper passes measured input, with zero uplift. The on-chain tax24hUsd field stores this effective input, not measured input. Every session record shows measured input, effective input and the uplift between them. The uplift is not new revenue and not a separate transfer. It raises the input used for caps.

QIP-2 is unchanged. Its remaining candidate is the session that opens Tuesday 10 November at 20:00 UTC. QIP-2's own check runs first, from the same reserve read: at 17 seats it needs $2,850, and QIP-6 neither adds to that check nor lowers it. If it passes, the max day targets $50 per seat under QIP-2, unless the supply-value bound is lower, QIP-6 adds nothing to that session, and if the session falls within S0 through S0+13 it counts as one of the 14; a max day before S0 counts as none. If it fails, that max day does not run; it is QIP-2's last candidate, so QIP-2 has no max days left, and that session opens on measured input. QIP-6 adds nothing on a QIP-2 candidate session either way, and its first session can never be that one. QIP-6 sets no reserve aside for that max day, so QIP-6's spending can make QIP-2's check less likely to pass. The reverse also holds: if the max day runs, its $850 can make a later QIP-6 check fail. Before S0 that only delays the start, possibly until approval expires; after S0 it ends QIP-6.

## What stays the same

The $50 per-seat ceiling, the 0.25% supply-value bound, and the claim, report and bell routes. QIP-2 and its own checks. OPS keeps its existing 40% of ordinary claims, which is outside this vote, and the uplift creates no extra OPS claim. Seat 1 still gets inference access under the same seat rules as every seat. There are no payouts, returns or fee sharing for holders. This is not a token price floor and not a payment to holders. QIP-6 starts no new backing round under QIP-4 or QIP-5.

## Cost

At the bell, recognized use by seats present becomes debt to the provider, and unused cap, even cap never handed out, buys and burns QUOTUM where free NVDA and price guards allow, as today. Raising caps can therefore raise burns without more inference use: in sessions 27 to 39 burns were $1,426.57 against $313.56 of inference. Approving QIP-6 approves more of that spend from the sink. It is a cost of the pilot, not a price promise.

For a floor-only example with 17 seats throughout, effective input of $85 each session, no tighter supply-value bound and successful settlement, recognized use plus successful burns consume $85 gross per session, $1,190 over 14. Each new seat adds $5 per session at a $5 cap. Measured input is not new money on top: holding it at session 41's $42.20, the uplift is about $42.80 a session, about $600 over 14. That is input support, and it becomes the reserve's net draw only if real inflow of that size arrives every session, with unchanged prices, seats and liabilities. As an illustration only, with no new inflow, about $2,600 of free reserve would end near $1,400. If QIP-2's max day also runs inside the pilot, it replaces one $85 session with $850 at 17 seats, $1,955 over the 14 sessions. These are not forecasts or a promise of 14 boosted sessions.

Seats that join after opening get the same cap, so exposure grows with the seats at the bell. With 100 seats at the bell and a $5 cap it is about $500 for that session before settlement guards; at a higher cap, that cap applies.

## When it stops

Before each boosted session, free reserve must be at least $1,000 plus $5 times the keeper's seat count before open for every pilot session still left, including this one. At 17 seats with all 14 sessions left that is $2,190; with one left it is $1,085. Equality passes, and the check does not lock reserve funds. QIP-1 used a $2,000 buffer; at the illustrative balance and 17 seats that buffer would not let a 14-session $5 floor start ($3,190), so QIP-6 sets it at $1,000. That is a weaker check than QIP-1's, it does not lock funds, and at the illustrative balance the starting margin is about $400. After opening, the keeper compares the cap in the Open event with the cap it admitted. If the event cap is higher, uplift stops for the rest of the pilot. Seats leaving before open executes are one possible cause. That session's cap and settlement stay in force. Stale price or reserve data, unresolved provider accounting, a provider shortfall or the team's emergency stop also end the uplift. If input measurement fails, the session opening is postponed.

Before the first boosted session, a failed check only delays the start, within the approval window. After it, any stop is final for this pilot and resuming needs a new proposal. A stop never cuts the cap of a session already open.

## How the vote works

Voting weight is liquid QUOTUM plus attributable staked QUOTUM, including queued stakes, counted once. It is read at snapshot block B, the latest finalized block before this proposal is published, so nobody can buy weight after reading it. Moving tokens after B does not change a ballot's weight. OPS on seat 1 is the only team wallet left out. Protocol accounts such as the sink, pools and staking escrow do not vote, and tokens held in pools give no one voting weight. Every other eligible wallet votes as an ordinary holder, without labels.

Voting is a gasless signature on quotum.org/governance. It is not a transaction and approves or moves no tokens. Only ordinary addresses with no code at B can vote under the fast rules. App wallets that use EIP-7702 delegation, such as FOMO, and contract wallets such as Safe cannot vote on QIP-6. The charter has the exact policy. QIP-6 passes if For weight exceeds Against weight, final eligible ballots across For, Against and Abstain carry at least 1% of supply at B, and at least five distinct eligible positive-weight addresses have a counted final ballot. Addresses are not people, and large holders can decide the result.

QIP-6 is the first proposal in the charter's standard class: voting runs 6 hours, then 6 hours for objections to the count and 6 hours to check the proofs, so the earliest final result comes 18 hours after opening. The count is published when voting closes. Two zero-value transactions on chain 4663 record the hashes of the receipt log and the tally, one after voting closes and one with the final result. The final result is the actual published record, not the earliest time it is allowed. The team sponsors it from seat 1; OPS's weight counts in no vote.

Passing is not starting. The pilot starts only if the ledger, objections and every check pass and a scheduled 20:00 UTC boosted opening succeeds at least 2 hours and no later than 168 hours after the final result. Otherwise approval expires. If the final result is published Friday 6 November at 09:00 UTC and every check passes, the first boosted session can open that day at 20:00 UTC, and if all 14 run daily on time the last closes at the bell on Friday 20 November. Failed checks can move the start within the approval window, and the activation record publishes the actual dates. Discussion runs in one public discussion thread linked from this page. The full voting rules are in the charter, and common questions are in the FAQ.

## Words used below

Session: one daily cycle, from open to the bell.
Measured input: approximate-day net NVDA inflow into the sink, in USD.
N: the keeper's count of seats before it sends open.
M: the number of pilot sessions left, including the one being checked.
Effective input: what the keeper passes to open. For an eligible pilot opening while checks pass and no stop has latched, the larger of measured input and $5 times N; otherwise measured input.
Uplift: effective input minus measured input.
B: the snapshot block that fixes voting weight.
S0: the first boosted session. The pilot covers S0 through S0+13.
OPS: the team wallet on seat 1.

Exact rules

These rules are the proposal itself. If anything above reads differently, these rules apply.

## Ballot package

| Field | Value |
| --- | --- |
| Venue | quotum.org/governance, own gasless EIP-712 ballots and public deterministic tally |
| Class and durations | standard class, v = o = p = 6h |
| Sponsor | the team from seat 1; OPS's weight counts in no vote |
| Ballot URL and discussion thread URL | https://quotum.org/governance#qip-6 and https://github.com/qvotvm/qip-1-receipts/issues/6 |
| B number, hash, totalSupplyWei, readable supply, quorum and sponsorship amounts | in the ballot package: `snapshotBlock`, `snapshotHash`, `totalSupplyWei`; quorum is 1% of `totalSupplyWei` and sponsorship 0.25%, each rounded up |
| Frozen strategy, exclusions and reconciled ledger manifest | in the ballot package: `exclusions`, `ledgerSha256`, `ledgerManifestSha256` and the ledger-strategy artifact; the ledger is published with the package at quotum.org/governance/qip-6/ledger.json with the same hash |
| Proposal hash | `proposalHash` in the ballot package |
| Publication, voting, objection, proof-check, final and expiry UTC timestamps | in the ballot package and its publication record: `votingOpens` (V), `deadline` (D = V+6h), objections [D,D+6h), proof check [D+6h,D+12h), final eligible from D+12h; activation expiry F+168h |
| Collector identity and receipt proof/mirroring policy | as for QIP-1 to QIP-5: collector Ed25519 public key `MCowBQYDK2VwAyEAetjCl1JO350n7s/Df/GIS9GOwd+DPE/cXLS+M61jVZM=` (DER SHA-256 `5d15f2b83a89b476dd3091388b69593d323f15d7755e970ba6315b0210549ce7`); independent mirror Ed25519 key with DER SHA-256 `0367d223ac5b79b8fe59b459aa90f4e912be08e70494ce4b1e7c4ad5e613df87`; mirroring and recount at least every 15 minutes, at D and at or after F |
| Anchor wallet address | `0xBbbC8F2aa7C59136AA630E2195c504907844C822`, an ordinary account funded only for gas, distinct from the keeper signer and OPS |
| Typed-data origin/version and canonical hash encodings | domain `quotum-FAST-ballot-v1`, chain ID 4663, as in the receipts document of the ballot package |
| Account policy | `eoa-only-v1`: only addresses with empty code at B, by ECDSA signature of the address's own key; EIP-7702 delegated addresses and any other code are not accepted; account evidence hashes are in the ballot package |
| Objection operator and resolution/finalization procedure | quotum team (@quotumorg) in the discussion thread, under the challenges document of the ballot package |
| Reserve valuation, freshness and provider/accounting criteria | fixed in the ballot package before voting opens |
| Fallback acceptance and restart rules | the collection and receipt policies of the ballot package, as for QIP-1 to QIP-5 |

Resolve ballot-critical fields before collection. Publish SHA-256 hashes of the exact vote files: token/staking addresses, chain, B/hash, ledger JSON/CSV, strategy, OPS-only team exclusion, separate protocol exclusions, schema and receipt/conflict/objection rules. Deterministic replay must reproduce the package. Discussion uses one linked public discussion thread; X only links to it. Ballots bind domain quotum-FAST-ballot-v1 and chain ID 4663. Switching venue after collection begins requires a new frozen package and restarted vote.

## Schedule and activation rules

Freeze the account policy, the B-selection rule and holder instructions before publication. B is the latest finalized block on chain 4663 returned by the official endpoint immediately before first publication, strictly earlier than publication; its hash is agreed by two independently run readers, and the publication record in the package states the selection. The data manifest, the ledger at B, exclusions as applied and account evidence with their hashes, is assembled after B, independently reproduced and published before voting opens; it cannot be frozen before B. If unreliable, delay voting and republish dates. A B hash mismatch at any re-check invalidates the package; never silently replace B.

V follows publication of the complete final B package. B's header timestamp and the first public publication of this proposal must each be at least two hours before V. QIP-6 is in the charter's standard class. Voting accepts receipts on [V,D) with D = V+6h; every signed deadline equals D, and the close tally is published at D. Count objections run on [D,D+6h) and proofs are checked on [D+6h,D+12h). The earliest final result Fmin is D+12h. F is the time of the actual published final result, after every objection has a published decision, the final checkpoint is independently verified and the B checks pass; nothing finalizes at Fmin by time alone. After D, the close anchor is sent from the anchor wallet: a zero-value transaction to itself on chain 4663 with the close prefix, the proposal hash and the SHA-256 hashes of the signed receipt-log head and close tally. At or after F, the final anchor records the final head and tally hashes, even when unchanged. Both anchors and the passing B checks are published in the final record before activation. Every activation bound uses actual F. Approval expires at F plus 168 elapsed hours, inclusive for a successful opening. The proposed timestamps are publication Thursday 5 November 2026 at 11:00 UTC, V at 15:00 UTC, D at 21:00 UTC, and the earliest final result Friday 6 November at 09:00 UTC; the package fixes the actual values.

S0 is the earliest scheduled 20:00 UTC opening at least 2 hours after the final result, no later than expiry, with passage, completed objection decisions and all implementation, data, provider and reserve checks passing before opening. Failed candidates defer to the next eligible opening within expiry. If none succeeds, approval expires. Publish the next on-chain session ID and reviewed policy record before sending open; verify the successful Open event. After activation, any stop is permanent for this pilot: uplift cannot restart and no new S0 can be chosen.

The signed proposal commits to the activation rule and frozen economic policy. Publish S0, its opening date and derived end in a later linked activation record, outside signed proposal bytes. The interval ends at integer session S0+13 inclusive, at most 14 openings; failed ticks, delays, unboosted sessions and QIP-2 max days add no sessions. No renewal.

Use an immutable versioned policy package at /governance/qip-6/policy.json and later records at /governance/qip-6/sessions/<session>.json, linked from the QIP page. Keep activation/session records outside the signed proposal, bound to policy hash and actual Open event/transaction, with mirrors and revision history. They cannot change approved rules, budget, duration or stops. Freeze canonical proposal bytes and hash encoding before signing, excluding self-referential hash display and later records.

## Question

Should quotum run this pilot with existing sink reserve, supplying max(measuredNet24hUsd, 5 * N) to TaxSink.open, targeting a $5 per-seat inference cap for eligible sessions, which can be lower if the supply-value bound applies or seats join between the keeper's count and execution, while preserving the $50 per-seat ceiling, 0.25%-of-supply-value bound, QIP-2 and the claim, report and bell routes?

This input sizes inference caps. It is not a minimum token price or a payment to holders. The supply-value bound can put a seat's cap below $5. All eligible seats participate, including seat 1. No fixed cohort or guaranteed wallet grant. Late seats can enlarge exposure. Consumption remains genuinely reported. Unused cap, including unissued grants, may buy and burn QUOTUM under existing guards. Raising input can increase burn without more inference use. Disclose failed burns.

Before each uplift, require free reserve of at least 1000 + 5 * N * M USD, with N > 0; with no seats there is no uplift. Before voting opens, fix the exact reserve valuation, freshness checks and provider reconciliation criteria, and the reviewed policy build that implements these checks. Failed checks, unresolved accounting, provider shortfalls or a published emergency stop disable uplift. Failed measurement postpones opening. Early stops require a new proposal to resume; stopping cannot cancel current caps or bell. N here is the keeper's count before sending open. If the cap in the successful Open event is above the admitted cap, for example because seats left before open executed, uplift stops for the rest of the pilot.

At a QIP-2 candidate, QIP-2's unchanged gate is evaluated first from the same fresh reconciled reserve snapshot; QIP-6 adds no term to it and does not lower its buffer. If QIP-2 runs, it supplies max(measuredNet24hUsd, 50 * N) and QIP-6 supplies no uplift; that session consumes one of QIP-6's 14 only if its session ID is within S0 through S0+13, and a max day before S0 consumes none. If QIP-2's check fails, that max day does not run and the session opens on measured input. On any QIP-2 candidate session QIP-6 supplies no uplift, writes no admission and cannot choose S0; QIP-2's rules alone decide that opening. QIP-6's gate does not include QIP-2's requirement and sets no reserve aside for it.

Session records publish measured/effective input, uplift, caps, grants, reserve/oracle reads, liabilities, provider reconciliation, stop reasons and actual burn/forward receipts. tax24hUsd stores effective input, not measured fees. OPS keeps its existing 40% of ordinary claims, outside this vote. Uplift adjusts opening input, creates no new claim and gives OPS no extra share.

## Disclosure terms

QIP-1's disclosure terms, its sections Input and access, Caps and settlement, Records and stops, and Exposure and accounting limits, apply to QIP-6 with these substitutions: $5 for $10, 1000 + 5 * N * M for 2000 + 10 * N * M, /governance/qip-6/ for /governance/qip-1/, and about $500 for about $1,000 at 100 seats. In particular: a seat's cap is the lowest of $50, effective input divided by the seats at opening, and 0.25% of QUOTUM supply value per seat where the contract applies that bound. The policy fixes the conservative admitted cap used for the post-Open comparison, and each record shows it next to the event cap. A cap is not a grant: grants stay within session-budget and provider-funding safeguards shared across a seat's keys. Provider consumption, recognized sink spend, uncovered use and recorded and unresolved liabilities are reported separately, and unresolved accounting blocks uplift. Use by seats that leave before the bell can stay outside recognized sink spend and is never assigned to another holder. Seat 1's consumption is labelled separately.

## Choices and outcome

Choose For, Against or Abstain. In raw units, quorum is ceil(totalSupplyWei/100) at B, without subtracting exclusions. For must exceed Against; at least five unique eligible positive-weight final voters across all choices must participate. Abstain counts for quorum and addresses. Sponsorship and exclusions follow the charter.

OPS, read OPS() on the sink contract 0x42AbC7b64eD47dB6899b4F01Aa6c765c600a53a2, alone is team-excluded from outcome, quorum and address participation. Other wallets are ordinary holders without labels. Protocol accounts are separate nonvoters; attributable stakes, including queued stakes, count once for eligible stakers.

Without a complete reconciled B ledger, no ballot or activation. If QIP-6 does not pass for any reason, the policies otherwise in force, including QIP-2, stay in force. Publish each session's reconciliation after its settlement and a complete pilot reconciliation after the last covered session or an early stop, before proposing continuation. If no activation succeeds by F plus 168 hours, publish an approval-expiry record with every attempted candidate and its failure reason. Approval expiry limits starting; it does not shorten an interval already activated. Inference access follows existing stake, seat and cap rules. No holder payments, returns or fee sharing; no arbitrary reserve transfers.

## Questions

Draft. These answers describe a proposal, not something already running. The proposal and charter hold the full rules, and they apply if an answer here differs.

### What am I voting on?

Whether the keeper should target a cap of at least $5 per seat per session, for up to 14 sessions and only while safety checks pass. The cap can be lower if the supply-value bound applies or seats join before the open transaction. The extra use and burns this allows are absorbed by NVDA the sink already holds.

### Where do I vote?

Votes happen only on quotum.org/governance, and we never ask you to vote or connect a wallet anywhere else. A post asking you to vote for a listing or to connect a wallet on another site is not from quotum.

### Is that $5 paid to every seat?

No. The $5 is the seat's inference cap for the session, shared across its keys. It is not a distribution and nothing is paid out. The keeper sets the opening input to at least $5 times the seats, using its seat count before sending open. The actual cap can be below $5 if the supply-value bound is lower or seats join before the open transaction executes. It is shared across the seat's keys and guarantees no grant or payment.

### Why $5 and not QIP-1's $10?

Recent use is far below $10 a seat, and the reserve is smaller than when QIP-1 started. At 17 seats a $5 floor opens $85 a session against $170 for $10. Under QIP-1's reserve check, $10 for 14 sessions at 17 seats would need $4,380, well above the illustrative $2,600.

### Why a $1,000 buffer?

QIP-1 required $2,000 plus the floor for every session left. At the illustrative $2,600 and 17 seats, a 14-session $5 floor needs $3,190 under that rule and could not start. With a $1,000 buffer it needs $2,190. The buffer is lower, so near the end of the pilot, with few sessions left, the reserve can fall closer to $1,000 before a check fails. At the start, with 14 sessions left and 17 seats, the line is $2,190.

### What happens on a QIP-2 max day?

QIP-2 is unchanged and its own check decides first. For the 10 November candidate it needs $2,850 at 17 seats. If the max day runs, it targets $50 a seat under QIP-2 and QIP-6 adds nothing; it counts as one of the 14 only if it falls within S0 through S0+13. If its check fails, that max day does not run, QIP-2 has no candidates left, and that session opens on measured input; QIP-6 adds nothing on that day either way. QIP-6 sets no reserve aside for the max day, and if the max day does run, it can make a later QIP-6 check fail: before S0 that delays the start, after S0 it ends QIP-6.

### Can it stop early?

Yes. Before each boosted session, free reserve must be at least $1,000 plus $5 times the keeper's seat count before open for every pilot session still left, including this one: $2,190 at 17 seats with all 14 sessions left, $1,085 with one left. An Open event cap above the admitted estimate, bad price or reserve data, unresolved accounting, a provider shortfall or the team's emergency stop also stop uplift. After activation a stop is final for this pilot. No stop cuts the cap of a session already open.

### When does it start and end?

QIP-6 runs in the charter's standard class: 6 hours of voting, 6 for objections and 6 for proof checks, so the earliest final result comes 18 hours after opening. If the final result is published Friday 6 November at 09:00 UTC and every check passes, the first boosted session can open that day at 20:00 UTC; if all 14 run daily on time, the last closes at the bell on Friday 20 November. Failed checks can move the start within the approval window. Skipped sessions do not extend it, and it does not renew.

### Does my vote pay me or control funds?

No money. Inference access still follows the existing stake, seat and cap rules. The vote creates no payment, fee share or ownership. It cannot move reserve funds freely or change contract constants. The team publishes receipts or a specific refusal.