The N/A Report: A Failed Analysis Pipeline and the Settlement Test for Crypto Research

Last month I read a document that should not exist. It was a second-stage analytical report on a blockchain project. It ran to roughly four thousand words of structured prose, organized into nine dimensions, with tables, confidence markers, and a risk matrix. Every substantive field in it read the same way: N/A — information insufficient.

I have spent twelve years reading crypto research. I have read documents that were wrong, documents that were fraudulent, documents written by people who did not understand what a merkle proof was. I had never read a document that was empty and still insisted on its own architecture. That is what made me keep it.

The report was not a hack. It was not a joke. It was the honest output of a two-stage analytical pipeline in which the first stage — the extraction of structured information points from a source text — returned a null set. The second stage, which is supposed to consume those information points and produce technical, tokenomic, market, ecosystem, regulatory, governance, risk, narrative, and supply-chain analysis, did exactly what it was built to do. It produced all nine dimensions. It filled them with nothing.

What interested me was not the failure. Failures are common and mostly boring. What interested me was the shape of the failure. The pipeline did not crash. It did not refuse. It did not revert. It emitted a complete, formatted, apparently authoritative document whose central claim — repeated eleven times across nine sections — was that it could not claim anything at all.

There is a term for this in protocol design. We call it a stale oracle. A price feed that has stopped updating but has also stopped reverting will keep feeding a lending market a number that used to be true. The market will keep liquidating against it. Nothing in the system will scream. The system is working exactly as specified. That is the horror.

This is an article about research infrastructure, not about price. But I want to be clear about where I am standing before I go further. Liquidity is a mirage; only settlement is real. I have watched four cycles teach that lesson to people who refused to learn it. I am not interested in writing the fifth version of the same warning. I am interested in what happens when the apparatus we use to understand markets begins to fail in the same structural way the markets themselves fail.

The N/A report is a specimen. I want to dissect it.

The architecture nobody audits

To understand why a pipeline would produce four thousand words of nothing, you have to understand how modern crypto research is actually built. Almost nobody outside the industry does, because the industry prefers to present research as a product rather than a process. A report appears. A thesis is published. A dashboard publishes a number. The machinery underneath is invisible, and invisibility is a business advantage.

So let me describe the machinery, because I have worked inside it.

The dominant architecture is a two-stage decomposition. Stage one is extraction. A source text — an announcement, a whitepaper, a governance post, a news article, a thread — is parsed into what practitioners call information points. An information point is the atomic unit of analytical meaning. It is a single falsifiable claim lifted from the source with enough context to be independently cited. "Protocol X raised $40M at a $400M valuation led by Fund Y." That is an information point. "Protocol X is building the future of modular settlement." That is not. It is a marketing assertion with no verifiable referent. A well-built extractor distinguishes the two. A badly built extractor treats them as equivalent because both are strings.

Stage one produces a set — ideally a non-empty one — of these atoms. Stage two consumes the set and produces analysis across a fixed taxonomy of dimensions. Technical positioning. Token economics. Market structure. Ecosystem position. Regulatory exposure. Team and governance. Risk. Narrative. Supply-chain transmission.

The taxonomy is not arbitrary. It is the residue of a decade of post-mortems. Each dimension corresponds to a category of failure that has destroyed real capital. Token economics exists because of unlock cliffs that vaporized retail. Regulatory exposure exists because of the SEC and its many imitators. Ecosystem position exists because projects that looked self-sufficient turned out to depend on a single liquidity venue that later delisted them. The nine dimensions are scar tissue. That is what gives the framework its authority.

But notice the dependency. The entire structure rests on stage one. Every dimension is derived. Nothing in stage two is primary. Stage two is a function whose input is the information point set. If that set is empty, the function has no domain. There is no analytical maneuver that recovers information that was never extracted.

This is the same relationship that exists between execution and settlement in a blockchain. Execution is the layer where state changes are computed. Settlement is the layer where those state changes become final and irrevocable. Execution can be fast, probabilistic, reorganized. Settlement is where truth is decided. And the iron law of that architecture is this: if settlement fails, execution produces nothing of value, no matter how elegantly it runs.

Research pipelines have a settlement layer. It is stage one. Almost nobody treats it that way. Almost nobody audits it. Almost nobody puts a validity gate in front of stage two. Which is why a pipeline will happily run an empty settlement through nine dimensions of execution and produce a document that looks like knowledge.

What the nine dimensions actually did

I want to walk through the report's sections, because the failure is not uniform. Some dimensions degrade honestly. Some degrade into theater. The distinction matters, because it tells you which parts of crypto research are mechanical and which are narrative-dependent — and narrative-dependent analysis is exactly what a bull market cannot afford.

The technical section behaved correctly. Asked to evaluate innovation, maturity, security assumptions, and performance metrics with no technical input, it returned N/A for each and stated plainly that no technical positioning was possible. It noted that it could not determine whether the subject was a layer-one, a layer-two, an application, or infrastructure. It could not identify zero-knowledge rollups, optimistic rollups, directed acyclic graphs, sharding, parallel EVM execution, or modular design. It flagged no risk, because there was nothing to flag.

The token economics section did the same. No token name, no supply schedule, no allocation table, no unlock calendar, no value-capture path. It declined to identify a Ponzi flywheel, and it was right to decline — flywheel detection requires knowing where the incentives come from and where they go. Without a flow map, calling something a Ponzi is itself an unfalsifiable claim. The section resisted the temptation. That is discipline.

The market section is where I first noticed something interesting. Given no asset, no venue, no float, no funding rate, no positioning data, it returned N/A across every field. But it also added a line I did not expect: it noted that it could not determine whether the (unknown) catalyst had already been priced in. That is a sophisticated instinct. It is also the instinct that most crypto research inverts. The default posture of a bull market analyst is to assume the catalyst is not priced in, because assuming otherwise ends the conversation.

The ecosystem section produced an empty dependency graph. Upstream, subject, downstream — all blank. No contributor counts, no contract deployments, no daily active users, no retention. It could not assess developer health or user stickiness. Correct.

The regulatory section ran a Howey test against four elements — money invested, common enterprise, expectation of profit, efforts of others — and returned N/A for each, with no composite judgment. No jurisdiction. No registration status. No KYC posture. Correct.

The governance section returned N/A for team capability, industry experience, stability, voter participation, top-ten concentration, and proposal quality. It could not evaluate a single investment round. Correct.

The risk section is where the report became, briefly, genuinely useful. Unable to assess technical, market, operational, regulatory, competitive, or narrative risk — because there was no object — it identified exactly one risk with high confidence. Process risk. The risk that a downstream decision-maker would consume an information-free document and act on it.

That is not a placeholder. That is the sharpest sentence in the entire report. A system designed to enumerate the ways capital can be destroyed, given no object to analyze, turned its attention on itself and found the failure. Then it did something more: it recommended a circuit breaker. It proposed that the pipeline should suspend itself when the information point count falls below a threshold, and that a downstream validity gate should reject any stage-two output derived from fewer than N atoms.

Read that again. The broken system diagnosed its own break and prescribed an engineering fix. That is what a properly designed analytical framework does under starvation. It does not hallucinate. It reverts.

The narrative and supply-chain sections behaved differently, and this is where I want to slow down, because it gets to the center of why I am writing this.

Narrative analysis, given no narrative, returned N/A for heat cycle, durability, fundamental support, delivery verification, and sentiment. Fine. But look at the expected-versus-actual table. It contains three rows — user growth, revenue, technical delivery — and every cell reads N/A. The framework was built on the assumption that narrative can be decomposed into expectations and realizations, and that the gap between them is the tradable signal. That is a 2021 assumption. It treats social heat as a fundamental. It treats the ratio of chatter to substance as a measurable quantity. In a market where the chatter is manufactured by paid distributors and the substance is a GitHub commit history that can be squatted, that ratio is not a signal. It is a product being sold to you.

Supply-chain transmission behaved the same way. An empty transmission graph, from mining and infrastructure through protocols and DeFi to end users, with every node blank. It correctly declined to analyze transmission that had no origin.

So here is the picture. Six of nine dimensions degraded into honest silence. Two degraded into structural theater — they maintained the shape of analysis while having nothing to analyze. And one turned on the machine itself and found the real risk.

That distribution is not random. It maps directly onto which dimensions are falsifiable and which are interpretive. Technical claims can be checked. Unlock schedules can be checked. Howey elements can be argued. But sentiment, narrative durability, and expectation gaps cannot be settled by any ledger. They are the parts of crypto research that function as narrative instruments rather than measurements, and narrative instruments do not need inputs to produce output. They only need an audience.

Liquidity is a mirage; only settlement is real. The same sentence applies here. Whatever cannot be settled against a source is not analysis. It is atmosphere.

The pipeline has no heartbeat

Now I want to get technical, because the report's most important contribution was accidental, and I do not think its authors understood what they had found.

The report recommended a validity gate. It did not specify how to build one. I have built these. Let me describe what actually goes wrong, because the failure mode here is not unique to research pipelines. It is the same failure mode that has repeatedly destroyed DeFi protocols, and the parallel is exact.

Start with the oracle analogy, because it is precise and not decorative.

A lending protocol needs a price. It gets that price from an oracle. The oracle is a dependency. The protocol's solvency is a function of the oracle's honesty. Now: oracles fail in two distinct ways. They can report a wrong price, which is loud. Or they can report a stale price, which is silent. The wrong price triggers liquidations that everyone can see and argue about. The stale price triggers liquidations against a market that no longer exists, and the first person to notice is usually the person being liquidated.

Every serious oracle design since 2019 has been an attempt to make staleness loud. Heartbeat thresholds. Deviation triggers. Circuit breakers that halt a market when the feed stops updating. The entire engineering discipline of oracle design is the conversion of silent failure into loud failure.

Research pipelines have no such discipline. There is no heartbeat. There is no deviation threshold. There is no circuit breaker. A stage-two pipeline will accept an empty input set and emit four thousand words without a single warning propagating to the consumer. The document is formatted normally. The tables are rendered. The confidence markers are printed. The consumer has no way to distinguish a report derived from forty information points from a report derived from zero, unless they read the fine print — and in a bull market, nobody reads the fine print, because there is too much of it and reading it costs you the trade.

This is not a bug in one pipeline. This is the default condition of institutional crypto research in 2026.

I want to be concrete about how it happens, because I have watched it happen at scale.

When capital is abundant and asset prices are rising, the demand for research is not a demand for truth. It is a demand for coverage. A fund needs a view on forty assets. A desk needs a diligence memo on a token before the unlock. A media outlet needs five pieces a day. The throughput requirement is set by portfolio breadth and publication cadence, not by how much is actually knowable about any given asset.

So the pipeline gets optimized for coverage. And coverage is a throughput objective, which means the binding constraint is time per asset, which means the validity gate is the first thing to go. A gate that rejects empty inputs is a gate that rejects assignments. An analyst who returns N/A is an analyst who returns nothing. In an organization measured by output volume, returning nothing is the one unacceptable answer.

I have seen this before. In 2021 I audited the compound interest and liquidity mechanics of two of the largest lending and stablecoin protocols, and I wrote a long internal document arguing that the TVL figures driving the entire sector's valuation were not measurements of capital. They were measurements of incentive flow. Deposits that existed only because they were being paid to exist, double-counted across recursive wrappers, marked at a token price that was itself a function of the deposits. TVL was a coverage metric wearing the costume of a fundamental.

Nobody wanted to hear it, because the coverage metric was the reason the sector had a valuation at all. The number was load-bearing. So the number stayed.

The N/A report is the same phenomenon at a different layer. The pipeline produces documents because documents are the product. An empty document is still a document. It can be filed, cited, forwarded, scored. The alternative — returning nothing — breaks the production line. And the production line is not evaluated on whether the analysis is true. It is evaluated on whether it is delivered.

There is a profound lesson buried in this, and it is the lesson this whole cycle keeps re-teaching. Systems fail not when they produce wrong answers, but when they are structurally incapable of producing no answer at all. A blockchain that cannot halt is not resilient; it is brittle in a specific and dangerous way. A market maker that cannot widen spreads is not efficient; it is a liquidation cascade waiting for a trigger. A research pipeline that cannot return N/A is not productive. It is a narrative machine with a bibliography.

Where the atoms come from

The report's authors made one further observation that deserves more attention than it got, and I want to expand it because it is the deepest problem in the entire stack.

The report noted that the first stage may have failed at the input source — that the raw text itself might have been empty, or that extraction might have crashed on a malformed document. It marked this inference with medium confidence. Then it stopped. It treated the question of where information points come from as out of scope.

It is not out of scope. It is the whole game.

An information point is a derived object. It exists because someone read a source and decided a claim was sufficiently structured to be lifted. That decision is not neutral. It is an editorial act. And the epistemology of the entire downstream analysis is inherited from the epistemology of that act.

Consider the most common information point in crypto research: a protocol's total value locked. Where does it come from? Usually a press release, a dashboard, or a team-authored blog post. The dashboard aggregates what contracts report. The contracts report what users deposit. The users deposit because they are being paid to deposit. The payment comes from a token whose price is supported by the existence of the deposits.

Follow that chain and you find that the number everyone cites as a fundamental is a four-step derivation from a marketing decision. It is not false. It is worse than false. It is derived, and its derivation is unaudited. A figure like that can pass through stage one extraction as a clean atom — "TVL of $1.2B" — and then be consumed by nine dimensions of stage-two analysis as though it were a measurement of something real. Every downstream conclusion inherits the unverified nature of that one number, and no confidence marker anywhere in the document is calibrated to account for it.

This is the analytical equivalent of an unaudited bridge. The bridge holds because people keep walking across it. The fact that people are walking across it is cited as evidence that it is structurally sound. When it fails, the post-mortem discovers that the collateral was rehypothecated, the custodian was a multisig with four signers, and the attestations were audited by a firm that audited nothing.

I have a bias here and I will name it. I spent six months in 2019 tracking fifty wallets by hand to determine how much of Uniswap V1's reported liquidity was real economic activity and how much was transient manipulation by a small number of actors. The answer, roughly, was that about eighty percent of the liquidity was not there for the market. It was there for the appearance of the market. The pools were real. The capital was, briefly, real. The economic function was not. It was a set of positions taken to produce a metric that attracted other positions.

That audit changed how I read everything afterward. It taught me that in crypto, the number is almost always the product and the market is almost always the byproduct. And it taught me that the most dangerous information point is not the obviously promotional claim — the "revolutionary" and the "paradigm-shifting" that any competent extractor discards. The dangerous information point is the numerical one. Numbers carry an unearned air of settlement. They look like facts because they have units.

So when a stage-one extractor pulls a TVL figure, a raise amount, a user count, a transaction-per-second benchmark, or an annualized yield out of a source text, it is not extracting knowledge. It is extracting a claim about knowledge, with the provenance stripped off. And the pipeline cannot reattach the provenance later, because provenance is not derivable from content. It exists only at the moment of extraction, and it is lost the instant the number becomes an atom.

Liquidity is a mirage; only settlement is real. Settle the number or don't cite it. There is no third option that survives contact with a bad cycle.

Why the bull market makes it worse

Everything I have described gets worse in exactly the conditions we are in now, and I want to state the mechanism precisely because "bull markets are frothy" is not an explanation.

The mechanism is throughput pressure. Let me lay it out as a chain of constraints.

Asset prices rise. New capital enters. Capital allocators expand their mandate from the ten assets they could diligence properly to the hundred they now need exposure to. Analysts are hired or, more often in this industry, pipelines are deployed. The pipeline's coverage objective rises. Time per asset falls. Extraction quality falls with it, because thorough extraction requires reading the whole source and resolving ambiguities, and there is no time for either.

At the same time, the supply of source material explodes. Every project with a treasury commissions content. Every narrative attracts a layer of commentary. The ratio of text to verifiable claim collapses. Stage one now has to extract atoms from a corpus that is mostly atmosphere, and under time pressure it will extract the atmosphere because the atmosphere is easier to parse. Marketing language is grammatically simpler than technical language. "We are building the trust layer for global settlement" is a cleaner string than a paragraph about a sequencer's dispute window.

So the pipeline degrades in two directions simultaneously: fewer atoms per source, and lower-quality atoms per source. And the output volume goes up, because the coverage objective went up.

That is exactly the configuration that produces an N/A report and does not flag it.

I have seen the institutional version of this. In 2024 I worked with three researchers on a report examining the inflow data of a large spot Bitcoin ETF against traditional gold ETFs, trying to isolate what actually drove institutional entry. The headline finding was that regulatory clarity — not any technological breakthrough — was the dominant variable. But the finding that stuck with me was subtler. The flows were not a sentiment measurement. They were a compliance measurement. The capital moved when the legal structure permitted it to move and stopped when it did not. The instrument was indifferent to narrative. It could not be persuaded.

That is a settlement-grade signal. It has provenance, it has custody, it has an audit trail, and it cannot be manufactured by a content operation. It is one of the very few things in this industry that behaves like finality.

And notice what it implies about research. If the most reliable signal in the market is a compliance-driven flow, then the analytical apparatus should be organized around legal and custody structures, not around sentiment and narrative. But the apparatus is organized the other way, because sentiment is cheap to measure and legal structure is expensive to understand. The cheap measurement won. It always wins under throughput pressure.

There is a regulatory-macro parallel here that I think is underappreciated, and it comes from where I now spend most of my time.

I moved from speculative analysis to central bank digital currency research between 2022 and 2023, after the collapse of a large algorithmic stablecoin and the winter that followed. I did it because I had watched billions of dollars of capital flow into instruments that had no reconciliation function, and I wanted to understand what a system looks like when reconciliation is the design constraint rather than an afterthought.

Central bankers do not start with a narrative. They start with a balance sheet that must balance. A wholesale CBDC pilot in Southeast Asia, of the kind several central banks in the region have been running, is not primarily a technology demonstration. It is a reconciliation test under stress. The question is never whether the system can move a token. The question is whether, when the token is stuck, the two sides of the ledger agree on what happened and who owes what. If they do not, the pilot has failed regardless of how fast the transactions settled.

The comparison to crypto research is uncomfortable and precise. Crypto research has no reconciliation function. There is no mechanism by which a claim made in a report must be settled against a source of truth before the report is consumed. There is no reconciliation under stress, because there is nothing to reconcile against. The report is issued, cited, and forgotten. When it is wrong, the cost is borne by whoever acted on it, and the analyst bears none of it.

That asymmetry is the actual problem. Not bias, not incompetence, not even incentive — though all three are present. The problem is that the analytical layer of this industry has no settlement, and therefore no finality, and therefore no accountability. It is pure execution with no base layer.

The most honest document of the cycle

Here is where I have to say something that will read as perverse, and I mean it.

The N/A report is the most honest document produced in crypto research this cycle.

Not because emptiness is virtue. Because the report, faced with an information vacuum, refused to fill it. It had every opportunity to fabricate. It could have inferred a plausible project type, guessed at a token model, applied sector medians to an unknown asset, and produced a document indistinguishable from the thousands of others circulating right now. Nothing in its architecture prevented that. Many pipelines do exactly that, and their outputs are consumed with more confidence than the N/A report, precisely because they contain more substance to be confident about.

The N/A report did the opposite. It declined at every step. It declined to identify technical positioning, declined to model supply, declined to judge securities status, declined to name a team. It made exactly one positive assertion — that process risk was high — and it was correct.

That is a settlement-grade act. It is the analytical equivalent of a node refusing to sign an invalid block. It cost the system throughput and it preserved the ledger.

Now watch what the industry will do with it.

The industry will read the N/A report as a failure of input, not a failure of architecture. It will conclude that the fix is better source acquisition — more scraping, more feeds, more sources, a larger corpus, a more aggressive extractor that pulls more atoms per document. It will treat emptiness as scarcity and respond by manufacturing supply. It will not add a validity gate, because a validity gate reduces output, and output is the metric.

That is the wrong lesson and it is the predictable one. We have watched this exact correction happen in DeFi. When a protocol failed because its oracle was stale, the sector did not conclude that oracles need circuit breakers and diversification. It concluded that oracles need more data sources — which is how we ended up with oracle designs that aggregate dozens of feeds, most of which are republished from the same three venues, and call the result decentralization. The node count went up. The independent source count did not. The illusion of robustness increased and the actual robustness did not move.

I hold the same view about that as about everything else in this stack. An oracle that solves a decentralization problem by running a network of permissioned nodes is not a solution. It is a relabeling. The count is not the property. The property is independence, and independence is not a number you can print.

The same applies here. A pipeline with a million information points extracted from the same press releases as every other pipeline has not achieved coverage. It has achieved duplication, at scale, with a confidence interval that does not reflect it.

What the correct lesson would look like: a research pipeline that treats extraction as a settlement layer, versions its extracted atoms, tags provenance at the moment of extraction rather than afterward, refuses to emit derived analysis below a minimum atom count, distinguishes structurally between "we could not classify this" and "we did not process this," and propagates the confidence of its weakest atom to the confidence of its strongest conclusion.

That last one is the hard one. It means a report whose headline conclusion rests on an unaudited TVL figure must be flagged as having the confidence of an unaudited TVL figure. In practice almost no report would survive that rule, because almost every report rests somewhere on a number that came out of a blog post. That is the honest state of the field. The flag would make it visible. The industry would rather the flag not exist.

The parallel failure nobody is watching

I want to widen the frame before I close, because the pipeline failure is a symptom of a broader structural condition, and the broader condition is where the real risk lives.

Three years ago I began work on the intersection of artificial intelligence and blockchain, driven by a specific problem: verifying that a model was trained on the data it claims to have been trained on. I spent four months interviewing AI engineers and crypto economists across Singapore and Manila, and produced a framework for what I called trustless verification of model provenance. The core insight was that blockchain's useful contribution to AI is not payments and not token incentives. It is an immutable record of what data was used, when, and by whom — an audit trail for the epistemic claims of a model.

That work and the N/A report are the same subject. Both are about verification under adversarial conditions. Both are about the difference between a claim that is asserted and a claim that is settled.

Here is what worries me. The industry is currently building an enormous volume of analytical and AI-generated output on top of an information base that is itself unverified. Models are being trained on crypto content that is ninety percent marketing. Analytic pipelines are being fed sources that are themselves outputs of analytic pipelines. Research is being produced at a rate that no verification mechanism can keep pace with. The result is a closed loop in which claims circulate, get cited, get re-extracted as atoms, and get re-emitted as analysis with the provenance fully laundered off. By the third iteration, an unaudited number from a blog post has become a consensus fact with a citation history.

This is not a hypothetical. I have traced specific figures through three layers of secondary sourcing and found the origin to be a single promotional post from 2023. The number is now in sector reports. It has a life. It will outlive the project it described.

That is what a research pipeline without settlement produces. Not error, exactly. Something worse: the systematic erosion of the ability to distinguish an assertion from a measurement.

And this is why the bull market matters. In a bear market, nobody produces research, because nobody pays for it. The information base is thin but relatively clean, because the only people writing are the ones who care. In a bull market, research is produced at industrial scale, and industrial scale is incompatible with verification. The volume of claims rises faster than the capacity to settle them. The gap between assertion and measurement widens. And the market prices the assertions, because the assertions are what is available.

Liquidity is a mirage; only settlement is real. I keep returning to this because it is not a slogan. It is a description of where value actually lives in this industry, at every layer. Price action is execution. TVL is execution. Social volume is execution. Narrative is execution. All of it is reorderable, all of it is probabilistic, all of it can be reorganized by a large enough participant with a large enough incentive. The only things that do not reorganize are settlement facts: who holds custody, what a contract says, what a court will enforce, what a balance sheet records after reconciliation. Those are the load-bearing elements. Everything else is weather.

Where this leaves the analyst

I am not going to end with a price target or a cycle call, because that would be exactly the kind of output the N/A report was right to refuse.

I will end with the thing I actually believe, which is that the binding constraint on this market over the next three years will not be liquidity. Liquidity arrives when narrative is strong. It leaves when narrative breaks. It is not scarce. It is not a moat. It is, in the technical sense, a mirage — visible, directional, and made of nothing solid enough to stand on.

The binding constraint will be verification capacity. The institutions that entered this market in 2024 came in through a legal structure, and they will stay only as long as the structures they are exposed to can be reconciled. A balance sheet does not accept narrative. A custodian does not accept sentiment. A regulator does not accept a dashboard. Every layer of institutional capital that enters this market drags a reconciliation requirement behind it, and the analytical apparatus of this industry currently has no capacity to satisfy one.

So the question I would put to anyone building research infrastructure right now is not how many assets they can cover. It is this: if your pipeline were handed an empty document, would it tell you? Or would it produce four thousand words and a risk matrix, and let you find out in the drawdown?

I already know the answer for almost all of them. I read one that told the truth. It produced nine dimensions of nothing, and then it diagnosed itself, and then it recommended a circuit breaker that nobody will install because installing it would reduce output.

That document is sitting on my desk. Every week, another one just like it is produced somewhere, except filled in. And every one of those is harder to trust than the empty one, because the empty one is the only one in the stack that settled.