A decision market gives you two numbers and a rule. The two numbers are what the organisation is worth in each of two futures; the rule is that the future the market prices higher is the one that executes. Everything else — the clock, the liquidity split, the threshold, the wallet-level tape — exists to make those two numbers hard to fake and legible after the fact.
But a verdict is only an opinion about a balance sheet, and the balance sheet is the part most readers skip. A proposal that asks for $200K out of a $40M treasury and a proposal that asks for $200K out of a $300K treasury are the same headline and two entirely different questions. So reading a decision market is two readings held together: the machine that produces the verdict, and the treasury the verdict is drawn against. This post is both, in that order, because that is the order the market itself runs them.
Part I — The machine
Pass edge and the threshold
When a proposal opens, the protocol mints two conditional markets over the same token: a pass branch and a fail branch. Both trade for the length of the market. When it resolves, one branch becomes real and the losing branch reverts — positions there are unwound as if the trades never happened. The FutarchyAMM shares liquidity between the two sides, which is why both prices move when one is pushed, and why you cannot read one branch without the other in view.
The whole verdict compresses into one number, the pass edge:
passEdge = ((passTWAP − failTWAP) / failTWAP) × 100It is compared against a per-proposal threshold. On MetaDAO's template that is −3% for team proposals and +3% for non-team proposals; across observed proposals the range runs roughly −3% to +4.1%. A negative threshold inverts the burden of proof: the proposal passes unless the fail branch beats the pass branch by that margin. Teams spending their own treasury are asked to clear a lower bar than outsiders asking for it — the mechanism encodes, in a single signed number, who is trusted with the benefit of the doubt.
What the pass edge actually prices is a conditional expectation: what is the token worth if we do this, versus if we do not? That framing is worth holding onto, because it is why the direction of an individual trade tells you so little. The market is not voting. It is quoting two valuations and letting the spread between them stand as the decision.
The clock is the mechanism
The trading period runs about 72 hours, split into two unequal halves:
| Window | Length | What it does |
|---|---|---|
| Pre-TWAP | ~24h | Price discovery on the record, but excluded from the verdict |
| TWAP window | ~48h | The only prices that count |
Because the verdict is a time-weighted average over 48 hours, a trade placed in the final ten minutes carries roughly 0.35% of the weight. Late money is loud and nearly weightless. That is the design, not a flaw: a verdict averaged over 48 hours prices conviction held over two days rather than the last person to touch the book, and it is what makes the mechanism robust to a flash loan. You cannot borrow a decisive price for a block and give it back; you have to hold it against the clock.
It also means a snapshot of the live spread is not a verdict. It is a partial sum. A pass edge that has been positive for six hours and a pass edge that has been positive for thirty are the same instantaneous number and very different states of the market. When this desk reads a live proposal, the countdown is not decoration — it is the denominator on how much the current print is allowed to mean.
Direction is not the signal
The most common misreading is treating a buy as support. With two markets, every position has two expressions:
| Trade | Economic alignment |
|---|---|
| Buy pass | Pass-aligned |
| Sell fail | Pass-aligned |
| Sell pass | Fail-aligned |
| Buy fail | Fail-aligned |
Selling the fail branch is a bet the fail branch will never exist. It is identical in economics to buying pass, and it looks like selling. Any read that sorts trades by surface direction gets the market backwards — and because the two pools share liquidity, a large sell on one side is mechanically a push on the other. To recover the actual argument you have to net the four legs into a single quantity, net pass volume tilt:
tilt = [(passBuy + failSell) − (passSell + failBuy)] / totalVolumeRun the same formula on trade counts instead of dollars and you get trader tilt. The gap between the two is informative on its own: dollar tilt without trader tilt means one wallet is carrying a side, and a side carried by one wallet is a side that leaves when that wallet does.
Thin markets are not automatically weak
The ecosystem is small enough that a $2M MetaDAO allocation once passed on around $3K of volume. The reflex is to call that meaningless. Sometimes it is, and sometimes it is fast consensus — Solomon's DP-00002 resolved on about four independent trades once an arbitrage bot was stripped out of the tape.
The question is never how much volume there was. It is how much it would have cost to move the price:
- Slippage on the trades that actually printed
- Depth to 1% — the dollars needed to shift each branch a percent
- Single-trade fragility — would removing the largest trade change the verdict?
- Snap-back — did a spike hold, or did it decay within minutes?
- Information environment — was there anything to disagree about?
A market with $50K of depth and no contested information can be a genuine consensus. The same market with a live treasury request is cheap to push. Both are true at once, which is why depth belongs next to every pass edge, and why a pass edge quoted without the depth beneath it is a number doing less work than it appears to.
Strip the bots before you read the tape
Arbitrage keeps the two branches consistent and contributes almost no opinion. It is recognisable: regular cadence, paired pass-sell and fail-sell, clusters of near-identical ticket sizes, and a near-neutral net position at the end. Wallet trade frequency sorts the rest — one or two trades reads as conviction, three to ten as high engagement, twenty-one or more as arbitrage or high-frequency activity. None of this is a moral judgement about the wallets; it is a filter, and the filter is the difference between a tape that shows you an argument and a tape that shows you a market-making loop.
Strip those wallets, then re-run the tilt. What is left is the actual argument.
The market is a negotiating counterparty
Futarchy is usually described as a verdict machine. In practice teams treat it as a live counterparty. Solomon's "Gigabus" proposal — $2.7M across 2,276 trades — was pricing toward failure until the team amended the deal mid-window, adding Proph3t as a signer and committing to return about $2M. The market re-priced against the new terms, not the old ones, and passed the amended deal.
That is the part that does not appear in a pass/fail column. The market moved the proposal before it judged it — which is the strongest argument that the number is worth reading in the first place. A machine you can negotiate with is a machine that is being listened to.
Part II — The balance sheet
Everything above tells you what the market decided and how firmly. None of it tells you what the decision was about. Every proposal is an ask against a balance sheet, and before the market's opinion is worth anything, one number has to be established: how large is this ask relative to what the DAO actually holds?
Umbra's proposal was $1.5M. On its own, that is just a figure. As 62% of the treasury, it is the entire story — and no pass edge, however clean, changes what a market is being asked to sign off on.
NAV, and why the DAO's own token is not in it
Net asset value is the treasury's non-native holdings minus liabilities.
NAV = non-native treasury − liabilitiesThe exclusion is the point. A DAO holding a large balance of its own token does not hold assets in any sense that survives contact with selling — the balance is denominated in the thing whose price it would move. Counting it inflates every downstream ratio, and inflates it most in exactly the situations where the treasury is under stress. The moment you would need the money is the moment counting the native token lies to you hardest.
Adjusted supply
Per-token figures need a denominator that reflects what can actually trade:
adjustedSupply = total − protocolControlled − locked − unvested
NAV/AS = NAV ÷ adjustedSupply
premium = (spot ÷ NAV/AS) − 1A positive premium says the market prices the organisation above the cash in its treasury — it is paying for the operating business. A discount says the market values the DAO below its own bank balance, which is a governance signal before it is a valuation one: it usually means holders do not expect to see that money.
NAV/AS is an estimate, and is labelled as such by the compilers who publish it. Locked and unvested schedules are disclosed with varying precision, and the denominator inherits every one of those uncertainties. A ratio built on an estimated denominator is a range wearing the costume of a point, and it should be read as the range.
Runway is a range, not a number
Runway is normally quoted as a single figure. For a DAO holding protocol-owned liquidity it has to be quoted twice — with POL and without — because spending POL is not free.
The AMM those tokens sit in is the same liquidity that decision markets price against. Thinning it lowers the cost of moving the market that governs the treasury. A DAO can extend its runway by six months and halve the cost of attacking itself in the same transaction. Both effects belong in the same sentence, and a runway figure that mentions only the first is not a shorter truth — it is a different one.
Burn versus dilution
Two different things reduce value per token, and they leave different fingerprints:
| Action | NAV | NAV/AS |
|---|---|---|
| Spending from the treasury | Falls | Falls |
| Supply growth (unlocks, new issuance) | Unchanged | Falls |
If NAV/AS is falling while NAV holds, nothing was spent — the denominator grew. Projecting either forward means reading the schedule, not the balance: monthly allowances, cliff dates, approved raises, and expected revenue. The balance sheet tells you where the DAO is; only the schedule tells you where it is going.
Buybacks are a percentage, not a dollar figure
A buyback's dollar size says nothing. The metric that carries information is its size as a share of non-LP treasury — what fraction of the spendable balance was converted.
On that measure Loyal spent 79.8% buying its own token back, against Solomon's 17.1%. Solomon's aged better, and the reason is structural rather than a question of timing: a buyback converts spendable assets into the DAO's own token, which, as above, does not count toward NAV. A large buyback is a decision to stop being able to spend. It is a legitimate decision, and it should be read as the balance-sheet transformation it is rather than as a signal of confidence.
The ICO cohort shows up later as voters
Treasury analysis usually stops at the balance. The distribution that filled it carries into governance, and this is the seam where Part I and Part II meet.
ICOs are reported as committed versus accepted, with an oversubscription ratio, and the commitments follow a power law: Ranger's top 50 wallets were 42.7% of $86.4M committed. Those cohorts do not disappear at launch. When Ranger's decision markets ran, its largest pass-aligned wallets were ICO-era backers who had lost confidence in the project — the original buyers choosing to shut it down. The concentration you read off the raise is the concentration that shows up in the tape you learned to strip in Part I.
A concentrated raise is a concentrated electorate on a delay.
What Chromatic computes from this
Every proposal Chromatic reads is reduced to the same surface from both halves: pass edge against threshold, margin of pass or fail, net pass volume tilt, trader tilt, depth, and the wallet-level composition behind each side — set beside an ask readout of the requested amount, the treasury it is drawn against, and the share that represents. Live pages decode the market state from confirmed Solana program accounts; historical cases are reproduced from published forensics; treasury snapshots carry their date.
The discipline is the same on both sides. Where an input is missing, the rating is withheld rather than guessed. When the ask or the treasury is unknown, the readout says unknown — a proposal with no visible ask is not a small ask. Catalog entries and treasury snapshots are not proposals, and are never presented as though a proposal had been scored against them; a stale snapshot is labelled stale rather than shown as current.
The reasoning behind every one of those rules is in where Chromatic's decision-market numbers come from.