At 03:47 Abu Dhabi time a Jin10 flash hit my phone: all three major A-share indices had turned positive. Shenzhen Component +1.37%. ChiNext +2.24%. Nearly 2,300 individual stocks up.
I didn't read the headline. I did arithmetic.
Mainland A-shares list somewhere between 5,300 and 5,400 companies. Twenty-three hundred divided by 5,350 is 43%. Roughly three thousand names were flat or red while the tape printed green. The index was up. The market was not.
That is the most transferable lesson of the week, and it has nothing to do with China. I run a $2 million multi-chain yield book across Arbitrum, Optimism and Base. I read breadth screens every morning the way a cardiologist reads an EKG. And the crypto version of this exact print happens constantly โ a green total-market-cap candle sitting on top of a tape where six of ten tokens are bleeding out.
Alpha isn't the headline number. Alpha is the denominator nobody publishes.
What the flash actually contained
Strip the framing and there were three values and one state description. Shenzhen Component +1.37%. ChiNext +2.24%. Approximately 2,300 advancers. And the phrase "turned positive" โ meaning the indices had been negative at some earlier point in the session.
That is the entire information set. No turnover figure. No limit-up count. No index level. No margin balance. No sector breakdown. No Shanghai Composite print.
And no year.
The date stamp reads October 23. The year is absent. This is not a pedantic detail โ it is the difference between two completely incompatible macro universes. October 23, 2024 sits inside the window opened by the late-September 2024 policy package, in the middle of a risk-appetite repair driven by expectations of incremental fiscal expansion. October 23, 2025 sits in a different rate cycle, at a different index level, with a different turnover structure and a different foreign-flow disclosure regime. Every single inference about policy direction flips depending on which one it is.
An undated market print is not a data point. It is a rumour with a decimal.
I want to be explicit about why a DeFi yield strategist is reading Chinese equity tape at four in the morning. It isn't macro tourism. It's because breadth arithmetic is venue-agnostic. The same weighting mechanic that lets 2,300 advancers produce a green Shenzhen print also lets 40% of tokens produce a green total-market-cap chart. If you cannot read breadth on one tape, you cannot read it on the other. The failure mode is identical and it is expensive in both places.
Index math: the denominator nobody publishes
A market-capitalisation-weighted index is a concentration device disguised as a thermometer. It answers one question โ did the largest constituents go up โ and it answers nothing about whether the average listed company went up.
This is not a flaw. It is the design. It is also why the design is so frequently misread.
Run the arithmetic on the October 23 print. If 2,300 of roughly 5,350 names advanced, then approximately 3,050 names were flat or negative. The advance-decline ratio was roughly 0.75. Anything below 1.0 is a negative-breadth session by construction. Yet the headline read "all three major indices turn positive."
Both statements are true. Only one of them describes the market most participants actually own.
Now consider what ChiNext is. It is not a broad growth basket. It is a concentrated index whose weight sits heavily in the power-equipment and new-energy chain, pharma, electronics and TMT. When ChiNext prints +2.24% on a thin-breadth day, what has usually happened is that a handful of very large constituents moved, and the index arithmetic did the rest. The 2,300 advancers were not the reason the index was green. They were the backdrop against which a small number of heavyweights dragged the weighted average upward.
The distinction matters because it determines what you do next. If 2,300 names rising meant broad participation, you would buy a basket. If 2,300 names rising means a concentrated index moved, you buy the index and avoid the basket. Opposite conclusions from the same headline.
Breadth below 50% converts an index print from a signal into a financing event for the largest constituents.
The 87 basis point spread is the only forward-looking number
ChiNext +2.24%. Shenzhen Component +1.37%. Spread: 87 basis points.
That is the number I actually care about, and it is the only figure in the flash with predictive content.
The mechanism is duration. Growth equities derive most of their present value from cash flows that arrive in later years. When you discount a back-loaded stream, the discount rate dominates the valuation. A 50 basis point move in the risk-free rate that barely touches a utility's valuation can move a long-duration growth index by multiples of that. So when growth outperforms value by 87 basis points in a single session, the market is telling you something about the discount rate, not about earnings. Earnings do not reprice in a day. Discount rates do.
This is the cleanest read available from a four-number flash, and even it is contested. There are at least four competing explanations for a growth-over-value spread that all produce the same tape:
Liquidity expectation. Discount rate expectations fall, long-duration assets reprice first. This is the classic read.
Thematic catalyst. A single policy signal or sector event โ new-energy capacity rules, an electronics order cycle, a pharma approval โ lifts the heavy index constituents and nothing else.
Oversold rebound. If the preceding sessions were a sustained drawdown, the highest-beta names bounce hardest. This is mean reversion, not information.
Passive flow. Broad-based ETF creation in a concentrated growth index mechanically bids the constituents regardless of anyone's view.
All four produce +2.24% on ChiNext. The flash contains nothing that discriminates between them. Anyone who told you on October 23 that they knew which one it was, is selling you a narrative, not an analysis.
But the spread still has value: it tells you where the marginal buyer went. And in crypto, the identical spread exists and is measurable in real time.
Same illusion, different chain
Total crypto market capitalisation is a cap-weighted index with a 55-60% single-name concentration. It is, functionally, a leveraged BTC proxy with decorative altcoin seasoning.
This means the most-quoted number in the asset class โ total market cap โ is structurally incapable of telling you whether the market is healthy. It tells you whether Bitcoin is up.
So I built a different screen. Three measurements, run every morning against the top 200 tokens by circulating market cap:
Percentage of the top 200 trading above their 20-day moving average. Percentage above their 50-day. And the gap between the cap-weighted 7-day return and the median 7-day return.
That third number is the one that maps directly onto the A-share print. Consider a recent session from my own log: total market cap +1.8%, BTC +2.6%, tokens above 20-day moving average at 38%, median token 7-day return at -2.1%.
Read that again in Chinese-equity terms. The cap-weighted aggregate printed green. The median asset was down over 2%. The breadth reading was 38% โ worse than the 43% in the Shenzhen session. The green candle was a Bitcoin candle wearing a market-wide costume.
While the headlines screamed that crypto was up 1.8%, the actual distribution of returns said four in ten tokens participated and the typical token lost money over the week.
You don't need a macro thesis to trade that. You need to know that a 38% breadth reading on a rising cap-weighted index is a distribution event, and that distribution events have a characteristic half-life.
How the screen is actually built, and why the oracle problem nearly broke it
I don't trust a breadth number I can't reconstruct from first principles, so here is the construction and its failure mode.
To compute "percentage of top 200 above their 20-day moving average" you need, for each token, a price at each daily close โ 200 tokens times 20 days. The prices come from oracle feeds. Chainlink and Pyth are the practical options.
Which means the breadth of the entire market is being measured through a data layer that is itself a small committee of reporters with a heartbeat deviation threshold. The decentralisation story and the operational reality are two different products. A feed that updates on deviation or on a heartbeat is not a price. It is a periodic opinion about a price.
In March 2023, during the USDC depeg, I watched this break in real time. Several feeds held stale values while the deepest DEX pools had already repriced. My screen flagged 61 tokens as trading above their 20-day moving average that were, in fact, 8-14% below it. The breadth reading jumped from 31% to 58% in ninety minutes without a single token actually rallying. I sized a rebalance off that screen and spent the next six hours unwinding it.
The fix cost me two weeks of engineering. Every breadth input is now reconciled against the volume-weighted mid-price from the deepest on-chain pool for that asset, with a staleness flag that quarantines any token whose oracle timestamp deviates from block time by more than three minutes. Tokens with less than $250,000 of daily pool volume are excluded entirely, because their "price" is a quote, not a market.
The lesson generalises past my book. If your breadth measure depends on an oracle, your breadth measure is partly an oracle-quality measure. And in a bear market, where pool depth has thinned across the board, the share of tokens with unreliable pricing rises โ which means breadth screens systematically overstate participation exactly when you need them most.
Stablecoin supply is the honest breadth, with a catch
Price breadth is contaminated by discount-rate expectations. Stablecoin supply is not. Issuance tracks settlement demand, not narrative.
Which is why I track net stablecoin issuance by chain as a second, orthogonal breadth measure. If aggregate supply is expanding while price breadth is deteriorating, someone is accumulating dry powder โ that divergence is informative. If both are falling, the market is genuinely contracting and the green candles are noise.
But there is a catch that most supply-growth bulls get wrong, and it's the reason I discount the headline number heavily.
Most of the durable growth in stablecoin float is not crypto-native demand. It is inflation-driven survival demand in emerging-market corridors. Nigerian, Argentine and Turkish users are not adopting stablecoins because they believe in decentralisation. They are adopting them because the local currency is losing purchasing power faster than the dollar, and a dollar-denominated token on a cheap chain is the least-bad savings vehicle available. The naira devaluation cycle and the Argentine peso's trajectory did more for USDT float than any DeFi incentive programme ever did.
So the correct interpretation of a rising stablecoin supply is a macro read on emerging-market currencies. Only after stripping the EM corridor growth do you get a crypto-sentiment reading. And the chain distribution tells you who is doing what: Tron for payment rails, Ethereum for settlement and collateral, Base for retail flow, Arbitrum for leveraged DeFi.
I flagged this in my own book last quarter. Headline stablecoin supply was up meaningfully. Strip the corridor growth and crypto-native float was essentially flat. Had I read the headline as risk appetite, I would have added duration into a tape that was quietly distributing.
Selective disclosure is itself a data point
The Jin10 flash listed Shenzhen Component and ChiNext. It did not list the Shanghai Composite.
That omission is not neutral. Financial flash services are ruthlessly consistent: when the main board moves materially, they print it. When it doesn't, they sometimes don't. The absence of a Shanghai number implies that main-board heavyweights โ banks, liquor, energy โ were approximately flat while the growth indices ran.
Which corroborates the entire structural read: concentrated growth leadership, value lagging, index green on thin breadth.
I apply the same forensic instinct to crypto disclosures, because the industry has turned selective omission into an art form. A protocol tweets "TVL all-time high" without mentioning that TVL is denominated in USD and the token is down 40% โ the number is technically true and behaviourally meaningless. An exchange reports "record volume" without a spot/perp split; perpetual volume is a leverage indicator, spot volume is a demand indicator, and conflating them is the oldest trick in the book. A bridge announces "cumulative volume" without disclosing whether the assets are still there.
I don't trade a number I can't decompose. If I can't split it into the components that would produce different trading decisions, it's marketing.
The contrarian read
The consensus interpretation of a session like October 23 is simple: risk appetite repaired, growth leadership confirmed, add exposure.
I don't buy it. Here is the counter-read.
An intraday V-shaped reversal โ indices negative, then positive โ with sub-50% breadth is not a sentiment event. It is a liquidity event. A sentiment event produces broad participation, because sentiment is a mass phenomenon and mass phenomena have wide participation by definition. A liquidity event produces a narrow, cap-weighted green print, because what happened is that forced selling met a bid, and the bid did not need to be broad. It just needed to be large enough to absorb the sellers in the heaviest names.
These two event types have completely different forward profiles. Sentiment events persist and broaden. Liquidity events decay. The absorbing bid finishes absorbing, the marginal seller is gone, and with no broadening participation the index drifts back toward the median constituent โ which, remember, was flat or down.
My working rule, from having traded both sides of this in crypto for years, is that a thin-breadth reversal has a decay window of three to five sessions unless volume expands materially. Volume is the only confirmation that the reversal was demand and not just the absence of supply.
And there's a second, sharper contrarian point. The 3,050 names that were flat or red are not the noise. They are the market. The 2,300 advancers are the index. Retail buys the index and reads the headline. The desk measures the red count and asks what is being distributed into the green print.
That asymmetry is the whole game. In crypto it shows up as retail buying a green total-market-cap chart at 38% breadth while the median token has been bleeding for a week. Same trap, different venue, same people falling into it.
The final contrarian note is about attribution itself. Building a monetary-policy or fiscal narrative on a four-number flash with no year attached is not analysis โ it is fiction with footnotes. The most valuable professional output on a data set this thin is a precise statement of what cannot be said. I have watched analysts construct elaborate easing-cycle theses from a single undated ticker. They were not wrong about the direction; they were wrong about having a basis.
What I'm doing with an actual book
Abstract breadth talk is cheap. Here is how it translates into position sizing on $2 million split across three L2s.
First, I run a personal breadth check before any rebalance. I hold fourteen positions. On a green day I count how many sit above their entry basis. If the answer is five, my portfolio is a cap-weighted index of my own mistakes โ one or two positions are carrying the P&L and the rest are drag. That is the moment to reduce the count, not add to it.
Second, breadth regime determines instrument choice. Below 50% breadth, I buy beta โ the index, the liquid major, the broad exposure โ because single-name selection at sub-50% breadth is a coin flip against a hostile tape. Above 60% breadth, single-name selection starts paying because participation is real and dispersion is reward rather than punishment.
Third, cross-chain rebalancing is itself a breadth problem. My working float is deliberately small, because moving size between Arbitrum, Optimism and Base means either paying for third-party bridge risk or accepting native-bridge latency. Cumulative bridge losses across the industry have passed $2.5 billion and the sector still routes the majority of cross-chain value through the same class of contract. I use native rollup bridges for the bulk and third-party only for small tactical size. The cost is capital inefficiency โ a seven-day native exit is effectively a seven-day option I have written against my own book โ and I price that drag at roughly 40-60 basis points annualised. That is the insurance premium for not being a bridge statistic.
Fourth, I keep a rules log. My 2025 AI agent lost $30,000 of a $100,000 allocation in two weeks, and roughly a third of that came from a governance parameter change the agent had no model for. The remaining $70,000 of net profit came from speed the agent had and I didn't. Both halves of that result are in the log, and the log is why I now treat automation as an execution layer and never as a risk layer. A bot can find breadth faster than I can. It cannot decide what thin breadth means.
The signals I'm actually watching
Four checks on the equity side, all low-cost and high-information.
Turnover. A trillion-plus in single-day turnover with a 20% expansion over the prior session converts the reversal from a squeeze into a trend. Below that threshold, treat it as decay-prone.
Advance-decline ratio. Above 60% participation and the rally has broadened. Persistent sub-50% while the cap-weighted index grinds higher is the classic distribution signature.
The ChiNext-to-Shanghai excess spread over five to ten sessions. If the 87 basis point spread keeps widening, the growth/duration regime is confirmed. If it compresses, the marginal buyer rotated and the read was wrong.
Margin balance. Two consecutive weekly increases confirm that leveraged money is entering rather than exiting. In crypto the analogue is perp open interest expanding alongside spot โ that's fresh leverage. Open interest expanding while spot is flat is a crowded trade.
On-chain, the mirror set: percentage of top 200 above the 20-day moving average crossing 60%; the cap-weighted-versus-median return gap closing toward zero; stablecoin net issuance measured ex-EM-corridors; and net bridge flows by destination chain. Those four together tell me more about whether a green candle is real than any single aggregate will.
Here is the question I'm sitting with. If a cap-weighted index grinds higher for five consecutive sessions while breadth stays under 50% and the median asset bleeds, what exactly is being distributed โ and to whom?
You don't need a model to answer that. You need the denominator.
Go find it before you size the trade.