Methodology

The instrument

One DIEM entitles its holder to $1 of Venice.ai API credit per day, refreshed daily, indefinitely, while the holder stakes it. DIEM is minted by locking staked VVV (sVVV) at an on-chain mint rate and burned to unlock that collateral. The desk treats DIEM as a stream of daily compute credit and VVV as the collateral asset behind it.

In plain words

Think of DIEM as a coupon book. Whoever holds one DIEM gets a fresh coupon every day, good for one dollar of Venice's AI service that day. Use it and it was worth using; skip a day and that day's coupon dies. Someone who cannot use the coupons can rent the book out and collect a share of what the renters actually use. The desk's question for DIEM is simple: what would a careful person pay today for that endless stream of daily coupons? The answer turns on two things: how much of each coupon really gets used, and how confident you are that Venice is still here in five years. Those are the $u$ and the $r$ in the formulas below, and the desk shows you both dials instead of hiding them.

VVV is the key that makes new coupon books. Lock staked VVV in the machine and it mints DIEM at a posted rate. That gives VVV one provable value: the coupons one key can mint, valued the desk's way. That number is a floor, roughly what the mint window alone justifies, and VVV trades far above it. The extra is what buyers pay for future burn growth, the chance the rules improve, and plain optimism. The desk can now measure VVV sent to the zero address, but no chain reading tells us how long that flow will last or what multiple it deserves. The desk does not call the extra wrong; it declines to turn those forecasts into its own target, and it shows the gap every day.

Everything on this site follows one rule: measure, do not guess. Every number states its formula and where each input came from. When an input is missing, the chart shows a gap instead of a guess. When the desk must assume something, the assumption is written down next to the number it touches.

Why this is not like Terra/Luna

Terra's stablecoin promised a fixed dollar and printed fresh LUNA for anyone who redeemed, so a run minted trillions of new tokens and crushed the price. DIEM makes no dollar promise: its price floats, and if you bought DIEM on the market you cannot redeem it at all; only the original minter can. Burning DIEM mints nothing; it only unlocks collateral that was already sitting in the vault. The loop that killed LUNA needs a fixed promise plus printing on demand, and this system has neither. That is a fact about the plumbing, not a promise that the tokens will hold their price.

Desk metrics

Implied yield

In plain words: if the daily coupons were certain to keep coming forever, the price you paid works out to a yearly interest rate; this is that rate.

$P$

DIEM price in USD: GeckoTerminal token-level aggregate, DexScreener fallback.

The face value of the stream is $365 per year, so $r^{\ast}$ is the gross annual yield the market demands to hold the claim. No haircut, no assumption: a pure restatement of price.

Spreads and survival half-life

In plain words: how much extra yearly yield DIEM pays over two safer choices, and, read pessimistically, how long the market is acting like Venice has left.

$r_f$

Aave v3 USDC supply APR on Base, a live on-chain risk-free-ish benchmark.

$\text{APR}_{\text{stake}}$

sVVV staking APR, read from the staking contract at the pinned block.

The Desk charts both spreads: the yield DIEM offers over simply staking VVV, and over parking USDC. The survival model reads the Aave spread as a hazard rate on a perpetuity and converts it to a half-life:

Read $t_{1/2}$ as an upper bound on market pessimism about the platform, not a survival forecast: the spread bundles hazard, liquidity, and risk premia. Domain policy: when $s_{\text{Aave}}$ is negative the half-life is undefined and shown as n/a; when it is exactly zero the half-life is infinite.

Mint ceiling

In plain words: what it would cost to make a brand new DIEM yourself; nobody sensible pays much more for a coupon book than the machine charges to print one.

$P_{\text{VVV}}$

VVV price in USD, same sources as P.

$m$

marginal mint rate (DIEM per sVVV), read from the staking contract at the pinned block.

$0.20$

locked-emissions share: a documented modeling parameter matching the contract's veniceEmissionsPercentageWhenLocked.

Above the hard ceiling anyone can stake VVV, mint DIEM, and sell. The carry variant adds the staking yield forgone on the locked collateral, treated as a perpetuity and capitalized at $r^{\ast}$. That capitalization is an explicit modeling assumption, which is why the Desk shows the ceiling as a band rather than a line.

Redemption wall

In plain words: what the original minters would pay to buy their coupons back and unlock their collateral, stacked up price by price.

$\text{sVVV}_i$

collateral locked in position i, rebuilt from raw mint and burn events; the event log is canonical.

$\text{DIEM}_i$

DIEM outstanding in position i, from the same rebuild.

$h$

2% haircut for VVV price risk over the 7-day unstake cooldown, an explicit assumption.

Only original minters can redeem, each at their own average mint rate, so the floor is not a line but a demand curve. Each position has its own breakeven $P_{\text{be},\,i}$; sorting positions by breakeven and walking down in price stacks the cumulative DIEM buy-back demand into the wall. The Desk also reports the aggregate moneyness of these conversion options: a position is in the money when its unlock value per DIEM exceeds the market price, and the aggregate intrinsic value is $\textstyle\sum_i \max(\text{unlock}_i - \text{market}_i,\, 0)$.

Utilization-adjusted fair value

In plain words: the desk's own answer to what a coupon book is worth: count only the coupons that actually get used, take a small fee off, and discount the stream for the risk that it ends.

$u$

share of the daily $1 credit actually consumed.

$f$

pricing factor, at most 0.98 by the settlement mechanics; the sensitivity view fixes it at that cap.

$0.98$

broker-fee keep rate: the lender receives 98 cents per settled dollar. A different constant from the f cap, even though the two happen to share a value.

$r$

required yield; slider, default 0.25.

The desk ships this as sliders for $u$ and $r$ plus the iso-price curve inside the desk model card rather than a single number: the desk does not publish a price target. Two companion numbers come from the same mechanics:

Consumer value $V_c$ counts only what a holder actually spends; credits that reset unused are worth zero. Breakeven utilization $u_{\text{be}}$ inverts the fair value at the market price and traces the iso-price curve; points where it exceeds 1 are omitted rather than clamped, because a clamped point would no longer satisfy V = P.

The fixed $r$ of 25% can be read in two parts: a few points for what safe dollars earn (the live Aave rate on the desk page), and the rest as the market's insurance charge against the stream ending. Treat that residual as a hazard equivalent, not a literal probability: it bundles survival with liquidity and risk premia, the same caveat as the half-life metric above. At today's risk-free rate the residual implies a half life of roughly three and a half years, and the desk page computes that decomposition live rather than freezing it here.

Purchasing power index

In plain words: does one day's coupon buy more AI work than it used to, or less? This index is that, with the shopping basket frozen so nobody can quietly swap the groceries.

The basket is locked by explicit model ids and versioned (v1, locked 2026-06-12: 1M input plus 1M output tokens on claude-opus-4-8 plus 10 images on z-image-turbo). Rising PPI means a DIEM-day buys more inference than at the base date. A day with an incomplete basket is skipped, never substituted. Changing the basket requires a new version chain-linked at the overlap; silent re-basing is forbidden.

Word on the canal: the street models

What it is

Community valuation models for DIEM and VVV, archived from public posts and shown with full attribution. The desk supplies the apparatus (the market line, yields, on-chain series); the authors keep their opinions. The desk model appears as one equal row among them and can never be privileged.

Provenance

The shared framing here did not come from the desk. In Venice's launch week, the founder, Erik Voorhees, described DIEM as valuable by a perpetuity formula on its daily compute credit, and VVV as having no obvious ceiling, in his post. Most street models archived here, and the desk model, build on that first framing: a daily coupon discounted to present value. The desk did not originate the idea, only the measurements.

Archival policy

Each model is archived the day it is catalogued: full post text, engagement counts, capture timestamp, and link. The archive is canonical; the page never depends on the live post. If an author edits or deletes the original, the entry is flagged but never rewritten. Archived quotations are rendered verbatim. Registry entries are immutable and versioned: any change to content, parameters, or formula is a new version, and history keeps the old one.

Re-run, never adjusted

A number in a published post is one of three things, and which one it is decides whether it may move. Some figures are observations: the author read a moving quantity off the world on the day they wrote, like a mint rate of about 100 VVV to mint one DIEM, or current annualized buy-and-burns of $6.563M. Re-anchoring one of those to what the desk measures is re-running the model, not rewriting it. Some figures are judgment: a 35% discount rate, a four-year horizon, a 65% capture haircut, a 50x multiple. Those never move, in any line on this site. And some are forecasts, like a projection of 2027 burns. A forecast never moves directly either, because the author's claim is not the level, it is the RATIO between the level and something they observed. Austin published both his current burn run rate and his 2027 figure, so his claim is the 9.66x between them; the desk carries that ratio forward on the run rate it measures and leaves his multiple and his supply forecast alone. This classification is the desk reading the posts. It is editorial, it is revisable, and it is not part of the archive.

Recomputable models are re-run daily with the author's exact published formula. A parameter bound to a live input, the on-chain mint rate or a dependency on another model's value, uses the live value when it carries signal and the author's published value otherwise; the parameter table marks which is which. One exception is deliberate: the floor component of the desk's VVV estimate never substitutes a published value for a missing input; a day without a live mint rate or a desk DIEM value is a gap in the series, because a floor computed from a guess would not be a floor. The desk re-runs each author's formula with live inputs and updates any parameter that is measurable over time, utilization, the mint rate, market prices, with its on-chain or market series, marking each as a desk-measured input on the parameter table and export. Two guards hold. The desk never swaps a different measured series in for a named quantity: its secondary-market realization is NOT fed into any author's credit-realization assumption, because liquidating a credit and using one are different economics. And genuinely non-measurable judgment, a private discount rate or a flat dollar target, stays exactly as the author published it. Models whose stated conclusion has no dollar level (a multiple, a framework) are archived and shown but excluded from the tape chart.

Utilization is the clearest case. Two authors state a utilization figure as a personal base case, the share of the daily credit a holder expects to use, not as a reading of live data. Where a model exposes that figure the desk updates it with the network's on-chain utilization oracle when the oracle carries a fresh signal, and falls back to the author's published figure otherwise. This is disclosed, not a claim that network utilization and one holder's usage are the same quantity; the parameter table and export mark it live on the days it applies. The oracle currently posts zero with a stale timestamp, which the desk treats as missing signal rather than zero usage, and a positive reading that has not refreshed within a couple of days is treated as stale for the same reason, so today every utilization figure still shows its author's published value.

Line weight carries how much measurement is in a line. A hairline means the formula ran and nothing measured moved, so the value is the author's constants. A normal weight means at least one measured input was live that day. The heaviest weight is the desk's conditioned re-run. The market tape alone is dashed amber, and each model keeps one hue whatever its lines are doing. One model is kept off the shared chart entirely: a fifth author colour could not be made distinguishable for colourblind readers on this background, and cycling a colour would make two models look like one, so its re-run lives on its card instead and the card says why.

Two desk proxies for utilization

The utilization oracle posts zero, so the desk measures the quantity from both sides instead and publishes the bracket rather than a point. The upper bound is the share of DIEM staked for API access, read straight off the token contract: staked DIEM is eligible to draw the daily credit, and unstaked DIEM certainly draws nothing, so real usage cannot be higher. The lower bound is the share of DIEM actually monetized through the Gondola venue, which sees one marketplace and never sees a holder consuming their own credit, so real usage cannot be lower. Neither is any author's utilization figure and neither is presented as one. They feed the desk's conditioned line only, never a published one, and when either is missing or the two cross over, there is no bracket and the line is a gap rather than a guess.

The conditioned line

Every card can carry two numbers. The first is the model as published. The second is the desk re-running that same formula with the author's observations re-anchored to measured series, and everything they judged left exactly as written. The card lists what was substituted, names the series it came from, gives a one-sentence reason for each, and states plainly what was left frozen. The published block is untouched: the conditioned re-run carries its own version, so changing a substitution never edits an archived parameter.

A conditioned line has no fallbacks. If a measurement it needs is missing or stale, that day is a gap, because a re-run that quietly reverts to the author's constants is just the published line drawn in a heavier stroke. Two archived models state a framework and no dollar target at all; for those, nothing of the author's is plotted and the only line is the desk's, captioned to say so. One of them, the peer-multiple comparison, reaches a per-token figure through two of the author's own stated assumptions applied to measured burns, which is two inferences deep and labeled that way. Kraken's discount ladder gets no conditioned line at all: his fair value is a fixed dollar level by construction, so there is nothing in it for a measurement to move.

What the desk will not do

Some of the most tempting parameters here are measured by inverting the market price. A discount rate quoted as market-implied is 365 divided by the price. A going-concern premium is that same figure minus a risk-free rate. Feed either into a fair-value formula and the answer collapses onto the price it came from: the model would track the tape perfectly and measure nothing at all. The desk therefore shows those figures and never lets one reach a formula. This is enforced rather than promised: the code that evaluates a model cannot see a market price, and a test re-runs every model with both prices tripled and fails if any value moves.

As of, and when a number goes stale

Every live value on a card carries the date it was observed and how old it is. Each source has a maximum age: two days for the utilization oracle, three for the on-chain mint rate and the marketplace series, four for the Gondola day, five for the Treasury yield, which publishes on business days only. Past that age the number is still shown, with its date and its age, and the card says outright that the formula used the author's published value instead. Hiding a stale reading would be worse than labeling it. Past thirty days it is dropped entirely, because a number that old has stopped describing anything.

A model that chains onto a scenario also carries a filled band on the tape, drawn live: the desk re-runs the model's own formula across the parent's scenario endpoints with the live measured inputs, so the band moves as the mint rate and prices move. It is disclosed as a re-run band, the model's math at a stated scenario times live inputs, not a range the author typed. When the utilization oracle carries a fresh signal the scenario collapses and the band narrows to that measured point. Point-target models, which state a level without a formula, keep a single line and no band.

The desk models, one per asset

The desk model for DIEM is $V(u,\, 0.98,\, r)$ with $r$ fixed at 25% (the street's median required yield when the page launched) and $u$ at the 0.5 scenario midpoint until utilization data carries signal; its band spans $u$ from 0.25 to 1.0. These are documented scenario parameters, not a forecast, and they only change with a version bump recorded here.

On the VVV side the desk publishes one card, and the distinction inside it is the point. The floor, $m \times V_{\text{DIEM}}$, where $m$ is the live marginal mint rate and $V_{\text{DIEM}}$ is the desk DIEM value above, is contract-enforced today and needs no assumption about the future. It is a floor, not a target, and it never draws a line: it is the verifiable component of the desk estimate, shown as a caption line on that card.

The desk estimate (added 2026-07-30), in plain words: Venice buys and burns VVV with real revenue, and the desk measures that flow exactly on chain. The estimate assumes today's burn pace simply continues, pays about 4 years of current burns up front (a 25% required yield), subtracts the drag of new VVV being minted (measured supply growth), spreads the result over every token, and adds the mint-window floor.

As a formula that is $m \times V_{\text{DIEM}} + \dfrac{B}{(r_b + g) \times S}$. $B$ is the measured 30-day annualized burn spend (every VVV transfer to the zero address, reconciled to the balance delta, marked at desk prices), $g$ is the measured 30-day annualized growth of totalSupply, which is how emissions dilution reaches the chain, and $S$ is the fully diluted supply read daily. Each piece wears its label: $B$, $g$ and $S$ are observations; holding the burn pace constant in perpetuity is a forecast, stated as such, because the schedule is company policy rather than a contract term and can move in either direction on an announcement; $r_b$ is a desk judgment that only changes with a version bump recorded here. The estimate is never presented as a bound; the premium the market pays above the floor is still shown rather than explained away.

Data sources

Prices: GeckoTerminal token-level aggregate (primary) with DexScreener fallback, recorded per snapshot with per-pool detail. On-chain: all reads through a single finalized block pinned per day on Base via Alchemy; contract addresses are documented facts, verified against source. Venice catalog: the public models endpoint, snapshotted daily per model. Every value carries its source label into the UI and the CSV exports.

VVV buy-and-burns

The desk indexes every VVV Transfer to the zero address on Base for each completed UTC day. Because the VVV contract has no public method that reduces total supply, the transfer sum is reconciled against the zero address's balance increase over the same blocks. The USD figure is an estimate: the exact VVV quantity multiplied by the desk's canonical same-day VVV price, not the exact dollars spent by a buyer. The live comparator on a model card annualizes only a complete 30-day window, and stays blank if a date or price is missing. It measures the current flow; it does not update an author's forecast or valuation multiple.

Gondola

Gondola is a DIEM inference marketplace: DIEM holders lend idle Venice API capacity and buyers pay in USDC. The desk reads its public market data daily as a secondary cross-check on the primary Surplus series below, and unlike Surplus it publishes settled, venue-wide numbers with history. Three figures are recorded, each labeled Gondola and none fed into any fair-value formula: a settled realization, the USDC a lender nets per $1 of Venice list face consumed (lender received over DIEM consumed), which loosely cross-checks the desk's primary realization; a discount versus Venice list, the median across active text models of the best marketplace quote against Venice's own list price, a trailing 24 hour observation of quotes rather than executed fills; and the venue's daily volume with its active lender and buyer counts, carried alongside every figure because the venue is still thin. Only the realization is directly comparable to Surplus; the discount uses a different denominator and the volume is a single venue, not the whole market.

Surplus marketplace

Surplus Intelligence is the desk's primary and canonical DIEM secondary-market source. Its public market data is ingested daily as the DIEM secondary market in inference credits: sellers there are overwhelmingly Venice-API keys funded by DIEM. Three metrics are recorded, each over a sampled top-20 order books by volume, never marketplace totals: executed 24h volume on Venice-host offers; the median QUOTED discount of the best Venice offer vs each book's direct-provider reference price; and an estimated realization, the USDC received per $1 of Venice-credit face consumed, matched against Venice's own catalog prices. Realization and discount have different denominators (Venice list prices carry a markup over direct providers), so the two numbers are not comparable, and neither is fed into any author's model.

VeniceStats cross-check

VeniceStats is an independent community analytics site built by @gekko_eth, not affiliated with Venice.ai or with this desk. The desk uses it strictly as a secondary cross-check, never as a canonical input: every metric published here is computed from the desk's own on-chain reads and event indexer, and historical series are self-derived from chain events. A cross-check script diffs the desk's numbers against VeniceStats' public API to catch divergence, and the results are labeled as a check, not a source. Where the desk cites a figure from VeniceStats, it is attributed to VeniceStats with a link. Per their terms, VeniceStats' own disclaimer applies to that data: it may contain inaccuracies and should be verified independently.

Verification and change control

The Phase-1 numbers passed an independent recomputation gate on 2026-07-18 (three dates, every derived metric, the basket, and position breakevens; recorded in the repository). The daily snapshot writes only fields whose step succeeded, so a retry never overwrites good history; failed steps are recorded with their errors in the row's audit trail. CSV exports of every series exist for hand verification at /api/export (metrics, positions, models, street).

FAQ

Disclaimer

Analytics, not financial advice. The desk observes; it never transacts on chain. Street models are community work reproduced under attribution for commentary; their conclusions belong to their authors. Nothing on this site is a recommendation to buy or sell anything.