# What the KXCO Ontology Exposed While the Market Stood Still

Ten names up against a flat index since 25 August. How a live ontology of 869 sourced claims surfaces candidates, and why the people at the end of it still decide.

Canonical HTML: https://www.livetradingnews.com/what-the-kxco-ontology-exposed-while-the-market-stood-still
Last modified: 2026-09-11

---

By Shayne Heffernan · 2026-09-11
Tags: $NVDA, $TSM, $ASML, $INTC, $GOOGL, $SPCX, $CBRS, $ORCL, $ARM, $BABA, $BIDU, $META, $AAPL, $AMD, $MSFT, $AMZN, $PLTR
Signed: ML-DSA-65, anchored on Armature L1.
Nothing in this article is investment advice.

On 25 August our ontology stamped a fresh set of rows across the artificial intelligence sector. Over the sixteen days to the 10 September close the Nasdaq Composite went nowhere, down 0.27%. Ten of the names in that record went a long way: Intel up 14.68%, Meta up 13.04%, SpaceX up 7.42%, Oracle up 5.65%.

That is the case for building an ontology instead of another dashboard, and it is what this article is about.

The system does not pick stocks and it does not allocate. It assembles a sourced, dated, typed picture of a sector and puts it in front of people, and the people argue with it and decide. Sometimes the decision agrees with the ratings. Sometimes it overrides the most attractive thing on the page, which is what happened with the Chinese names and is the most instructive part of this quarter.

## What this is and is not

KXCO maintains a live ontology of the AI sector, public at [kxco.ai/ontology-live](https://kxco.ai/ontology-live/). As of 11 September 2026 it holds 393 entities and 869 typed claims, 815 of them carrying a source URL, alongside 46 findings.

It is not a signal service. It does not publish a model portfolio, it does not tell you position sizes, and nothing in it should be read as a trade. When we write that a claim was recorded on a date, that is a statement about a research record and nothing else. A print is not a position. Anyone who blurs those two is selling you something.

What the system produces is an assembled, sourced, dated picture of a sector, with the evidence grade attached to every line. Humans then argue with it. Over the last few weeks that argument has led us to start buying some of the names in this article, to sell others, and to keep waiting on a third group where the businesses are strong and the politics around them are not ours to control. We are not going to publish the book, because the book is not the point and telling you what we own is not research.

The point is that the tool works, and the evidence for that is in what it exposed before the market agreed.

## How a name gets from the graph to a decision

It is worth describing the loop, because the interesting part is not the technology.

The graph produces ratings and findings. Ratings are the consensus layer: for each listed name we restamp price, market capitalisation, trailing multiple, analyst rating and the twelve-month consensus target, every row re-read at a close rather than carried forward from the last time we looked. Findings are our own: forty-six of them at present, each ranked by severity, each naming the entities it rests on and the evidence that would overturn it. The claims underneath both are typed, sourced and dated.

None of that is a decision. It is a brief.

What happens next is people reading it and arguing. A wide consensus upside is a reason to look, not a reason to buy, and most of the work is establishing why the gap exists. Sometimes the answer is that the market has not caught up with a contracted order book, which is roughly the Oracle case. Sometimes it is that the market has caught up perfectly well and the multiple is simply high, which is roughly Arm at 259 times trailing earnings. Sometimes the gap exists because of something the graph does not contain at all, which is China.

Positions follow from that argument, and they have gone both ways over the last few weeks. Some names in this article we have been buying. Others we have sold. We are not publishing the book, because what we own is not research and a list of holdings tells you nothing about the method that produced it. The method is the thing on offer here.

The failure mode we have watched other people fall into is treating the output as an instruction. A system that assembles evidence beautifully and then gets handed the allocation decision is not an upgrade on a bad analyst, it is a faster one. The entire design of this record assumes a person at the end of it who is willing to disagree with what is on the screen.

## What it exposed

The analyst rows in the ontology were last restamped on 25 August and again on 11 September. Take the 25 August stamp as a fixed, checkable starting line and measure to the 10 September close. Over those sixteen days the Nasdaq Composite went nowhere, down 0.27%.

Against a flat index, the names the record carries moved a long way:

- Intel, up 14.68%
- Meta, up 13.04%
- SpaceX, up 7.42%
- Oracle, up 5.65%
- Apple, up 5.38%
- Arm, up 5.22%
- AMD, up 5.10%
- Cerebras, up 4.03%
- TSMC, up 2.54%
- Nvidia, up 2.49%

And at the other end, the three we declined:

- Baidu, down 3.04%
- Tencent, down 3.08%
- Alibaba, down 9.11%

Ten names up against an index that did not move, and the three carrying the widest theoretical upside on the whole table are the ones that fell. That is the machine and the people each doing their job. The ratings pointed hardest at China. The judgment said no. Sixteen days later the judgment is ahead.

We are not claiming those returns. We did not own all of those names, we did not own them for that whole window, and anyone who counts a list of price moves as performance is doing arithmetic on a fiction. What we are claiming is narrower and more useful: the record put those names in front of us with the reasons attached, early enough to act on, and the reasons are still sitting in the graph with their dates on for anyone who wants to check them.

Two of those reasons are worth showing, because they were written before the fact.

On 25 August the Nvidia row in the record read, in full: "AI-GPU monopoly: 93% of revenue is data-center and still compounding at scale. FQ2 on 26 August is the sector's single largest scheduled datapoint." It did not predict the number. It said which number in the sector mattered, the day before it landed, and then FQ2 came in at $96.2bn of revenue, up 106%, with Data Center at $89.0bn.

On the same day the Palantir row read: "Fastest-growing public name, Q2 revenue up 93%, but 154x earnings, priced to perfection." Three days later it was restamped to say the stock had run through its own consensus target with 3% left. Palantir is down 3.98% since that first stamp while the index was flat, and it now trades below the target it ran through. That is the system flagging a name to leave alone, which is a less exciting output than a buy idea and considerably more valuable.

## Oracle, and why the record is built the way it is

The single clearest demonstration of the method is a name where the price and the business went in opposite directions for three months.

Oracle closed at $201.26 on 10 June. By 10 September it was $152.94, down 24%. A screen showed a broken chart. Our record showed something else, because it does not carry Oracle's contracted backlog as a number. It carries it as a dated series.

$455bn at Q1 FY2026. $553bn at Q3 FY2026. When the new figure arrived it did not overwrite the old one. The old claim was closed with an end date and the new one marked as superseding it, so both remain queryable and the shape of the series is visible rather than implied.

On the evening of 10 September, Oracle reported FQ1 FY2027. Remaining performance obligations came in at $664bn, up $209bn year on year and roughly $33bn above consensus. Cloud infrastructure revenue was $7.4bn, up 121%, against total revenue up 30% to $19.3bn. The company delivered 850MW of additional datacentre capacity and more than 300,000 GPUs in the quarter, and guided to at least $90bn of revenue for the year. The shares rose about 7% in extended trading.

We took every one of those figures from Oracle's own 8-K exhibit filed with the SEC rather than from the coverage of it, which is why our record grades them filed rather than reported. That grading is not decoration. It is the difference between a number a company put its name to and a number a journalist typed, and over a few hundred claims the difference compounds.

The useful part is what the structure made visible. For three months the stock was being marked down while the contracted order book underneath it compounded, and because the record held both facts with their dates attached, the divergence was legible while it was happening rather than after it resolved. That is the entire argument for building an ontology instead of another dashboard.

![Figure 1. Three-hop supply-chain reach for thirteen entities in the KXCO AI Sector Ontology, showing that the most connected name is not the hardest to replace.](https://livetradingnews-media.nyc3.digitaloceanspaces.com/media/2026/09/11/cmpgg3-d1352c989830cf44.svg)

## Who is actually irreplaceable

Everyone in this sector uses the word chokepoint, and almost everyone uses it to mean large and important, which is a different thing.

Because our claims are typed, the question can be asked properly. Figure 1 shows undirected three-hop reach across supply-chain claims only: starting from each company, how many of the 393 entities in the graph you can reach in three steps of a supply relationship. Nvidia leads at 130 entities, 33.1% of the sector. TSMC reaches 30.3%, SpaceX 28.5%, Oracle 27.7%, Synopsys and Cerebras 27.2%. ASML comes eleventh of the thirteen measured, at 72 entities and 18.3%.

ASML is nonetheless the chokepoint, and Nvidia is not.

What separates them is degree. The nine entities the graph flags as single points of failure carry a mean degree of 5.2 connections. The largest hubs in the same graph average 47. A hub is a company everybody deals with, and when a hub fails there are generally two or three other companies to deal with instead. Nvidia has AMD and a listed wafer-scale competitor on this very page. A chokepoint is a company with no alternative at any price, and it tends to look modest in a connection count precisely because there is only one route through it. ASML is the sole source of the EUV lithography that prints every leading-edge AI chip on earth, and there is no second supplier, no substitute process and no timeline on which one arrives.

The same measurement applied to Intel produces the most interesting number in the figure. Intel reaches 26 entities, 6.6% of the graph, the lowest of anything measured and by a factor of nearly three. For the company that was the centre of this industry for three decades, that is not a statistic so much as a thesis in a form you can check: Intel has been disconnected from the AI supply chain.

Which is why the asymmetry is worth attention rather than avoidance. A reconnection event of any kind, a foundry customer of scale, a packaging win, a government anchor, has to move a node carrying 26 edges rather than a mature hub carrying 47. The same headline does far more work on Intel than on Nvidia. Intel has risen 14.68% since 25 August and is still 28.8% below its highest close of the last three months.

One more reading from the same record, and it is the only signal of its kind in our short interest data. Across the settlement periods covering the second half of August, short interest fell at Oracle, SpaceX, Palantir and TSMC. At ASML it rose 12.71%. Somebody was adding to a short position into a drawdown in the one company in this supply chain that cannot be replaced.

![Figure 2. The sixteen circular-financing claims in the KXCO AI Sector Ontology. Nvidia is the source of nine of them.](https://livetradingnews-media.nyc3.digitaloceanspaces.com/media/2026/09/11/cmpgg3-36d79f4c5d972375.svg)

## The obligations that have not happened yet

The second thing an ontology does that nothing else does is hold relationships no single filing contains, because they span several parties and none of them has any incentive to draw the whole shape.

We keep sixteen claims in a group labelled loop, meaning capital that returns to the party that supplied it. Nvidia is the source of nine. Thirty billion dollars into OpenAI, which buys Nvidia GPUs. Up to ten billion into Anthropic, which buys Nvidia GPUs. Two billion each into CoreWeave, Nebius and Nscale. Up to $2.1bn committed to IREN, which builds the data centres. An investment in Figure AI, which buys the chips.

Then the two that dwarf every equity cheque in that list and are not equity at all. Nvidia has guaranteed up to $105bn of OpenAI's lease and power obligations, dated in our record to 17 August 2026, and is in talks to backstop up to $250bn of OpenAI's debt. A guarantee is not revenue and it is not an ordinary liability. It is a contingent obligation that does nothing whatever to an income statement, right up until the moment it does everything to one.

The remaining seven loops run between Microsoft and OpenAI in both directions, Amazon and Anthropic, AMD and Anthropic at up to $5bn of equity against roughly two gigawatts of hardware, SpaceX and Nvidia, and Tencent and DeepSeek. Two of the sixteen are graded derived, meaning they are our reading of the structure rather than anything a party published. They are marked as such on the figure and in the data, because a record whose whole proposition is that claims carry their provenance cannot quietly carry claims that do not.

None of that is an allegation. The narrow point is the one worth holding: a meaningful share of the demand signal in this sector is financed by the party that books the revenue from it, the largest commitments in the chain are guarantees rather than cash, and no stock screen in existence has a field for who else is on the hook when a counterparty misses. It does not tell you to sell anything. It tells you how large a position you should be comfortable holding, which is a different and more practical question.

![Figure 3. Twenty-eight claims connecting US export control to Chinese domestic substitution to the three listed names, in the KXCO AI Sector Ontology.](https://livetradingnews-media.nyc3.digitaloceanspaces.com/media/2026/09/11/cmpgg3-f371e7207a78589b.svg)

## The Chinese names, and the one variable that is not in the graph

This is the most instructive case in the record, and it is worth setting out properly rather than in a headline.

Start with what the companies have actually done, because it is formidable and it is badly reported in the West. In April 2026 DeepSeek trained V4 entirely on Huawei Ascend silicon. Not partially, not for fine-tuning: a frontier model trained end to end on domestic hardware, by a team working under export restrictions that were specifically designed to make that impossible. Whatever anyone thinks of the politics, that is one of the more impressive engineering results of the last two years, and the market has largely declined to price it.

What followed is what you would expect. Tencent and ByteDance began ordering Ascend 950PR parts. iFlytek moved $337m of compute onto Ascend. SMIC now fabricates domestically for Ascend, Cambricon, Biren and Moore Threads. Alibaba is simultaneously buying Nvidia hardware, shipping a competing part through its T-Head subsidiary called Zhenwu M890, and open-sourcing a stack named Sail whose stated purpose is breaking Nvidia's CUDA lock-in. That is four hedges against four different futures, executed at once, by a company trading on 24.78 times earnings. It is not the behaviour of a sector that has been contained.

Figure 3 shows the twenty-eight claims that connect American export control to Chinese domestic substitution to the three listed names. The control regime is real and precisely dated. Huawei went onto the Entity List in 2019. Nvidia A100-class exports were banned in October 2022, which our graph links directly to Alibaba's $52.5bn domestic AI commitment. In 2025, fresh curbs forced Synopsys and Cadence to halt work in China, cutting Huawei off from the design automation tools every chip on earth passes through. In 2026, AI-chip exports were formally limited to Alibaba and ByteDance among others.

Against that, in July 2026 the United States cleared H200 sales to ten Chinese firms, roughly 400,000 units on the order of $50bn, naming Alibaba, Tencent and ByteDance. Tencent has pledged to more than double AI spend to over $5bn in 2026. Alibaba is simultaneously buying Nvidia hardware, shipping a competing part through its T-Head subsidiary called Zhenwu M890, and open-sourcing a stack named Sail whose stated purpose is breaking Nvidia's CUDA lock-in. Those are four hedges against four different futures running at once, from a company on 24.78 times earnings.

Then the claim that reorganises the rest. In April 2026, DeepSeek trained V4 entirely on Huawei Ascend silicon. Not partially. Tencent and ByteDance both began ordering Ascend 950PR parts afterwards, iFlytek moved $337m of compute onto Ascend, and SMIC now fabricates domestically for Ascend, Cambricon, Biren and Moore Threads. The PRC has anchored a $295bn state AI plan on Huawei.

There is a further claim in that cluster that says more about the regime than any speech has. Under an arrangement our record calls the China Licence Revenue Share, established by the United States government, Nvidia pays 15% of its China H20 revenue and AMD pays 15% of its China MI308 revenue directly to the Department of Commerce. Export control in this sector is no longer a simple prohibition. It has become a toll road, and both governments are now financially participating in a trade they each describe as a security matter.

And one further claim, which is the one that decided it for us: the Chinese government restricts overseas travel for top AI staff at Baidu, Alibaba Cloud and DeepSeek.

Read that sequence as an analyst and it is constructive. Capable engineering teams, state backing, cleared access to Western silicon, a working domestic alternative, and multiples pricing none of it. The consensus upsides of 72%, 64% and 56% are the market's own arithmetic agreeing.

Read it as an owner of capital and a second variable appears that has nothing to do with the businesses. These companies sit between two governments that are both using them as instruments of policy, and the record documents both sides doing it. Washington restricts what they can buy, then charges a 15% toll on the sales it permits. Beijing anchors a $295bn programme on their supply chain and restricts the travel of their senior engineers. Neither of those is a fact about Alibaba, Tencent or Baidu as operators. They are facts about where those companies are standing.

That is the variable we are not willing to size around yet, and the honest word for our position is not yet rather than no. The analysis is not wrong. It simply does not contain the deciding input, and it never will, because no source publishes what two states intend to do next. The same risk sits on the American names from the other direction, which is why we hold it as a question about the situation rather than about the country.

Since that decision Alibaba is down 9.11%, Tencent 3.08% and Baidu 3.04% while the index was flat. Sixteen days is not vindication and we are not treating it as any. The point is narrower: the machine narrowed a broad question down to one specific thing a person had to decide, which is the only claim we have ever made for it.

We will keep watching them closely, and we expect to own them. Every claim in Figure 3 is already assembled and dated, so if the political risk resolves rather than compounds, the work is done and the position can be taken quickly. The names have not failed a test here. They are waiting on a variable that is not theirs.

## What the graph cannot hold

The China decision is the clearest case, but it is not the only one, and it is worth being precise about the boundary because vendors in this space rarely are.

Our record holds what is claimed, by whom, on what evidence, valid when. It is very good at supply relationships, capital flows, contractual obligations, ownership, stated policy and anything that appears in a filing or a credible report. It is good at the shape of things: who depends on whom, what has no substitute, which commitments are contingent.

It holds nothing about sovereign intent. It cannot tell you whether a government that has fenced in its AI engineers will direct those companies, or when, or in whose favour. It holds nothing about management quality beyond what people have publicly said and done, and a chief executive's judgment under pressure is not a claim anybody has published. It holds nothing about timing, because the record is a description of a state and states do not come with a schedule. And it holds no view whatever on the price you should pay, which is why every figure in it is a number rather than a recommendation.

Thirty-five claims in the graph touch a Chinese state entity, and they are unusually concrete. The National AI Fund subsidising SMIC at $8.2bn. The Cyberspace Administration regulating model registration at Alibaba Cloud, Baidu and ByteDance. SenseTime supplying surveillance systems, iFlytek holding education and smart-city contracts, 01.AI providing sovereign AI. The state funding a domestic substitute for ASML's DUV immersion machines. Counter-sanctions against Lockheed Martin and Anduril. The travel restrictions on AI staff at three separate companies.

Not one of those thirty-five is a claim about what the state will do next, because no source publishes that and inventing it would be the precise failure this record exists to prevent. We mark our own analytical readings as derived so nobody mistakes them for the other kind, and forty-four of the 869 claims carry that mark.

So the boundary is clean. Everything the machine can establish, it establishes with its evidence attached, and it declines to guess about the rest. What crosses the gap is a person taking a view, and that view being recorded as a decision with the brief it was taken against, so it can be reviewed later by somebody who was not in the room.

## A number that will not stand without its method

One more object from inside the record, because it shows what a finding looks like when it is built properly.

The graph tracks $2.91tn of capital raised by the entities in it. Next to that sits a figure of roughly $10tn of announced AI capital commitments worldwide through 2030, and the finding carrying it opens by stating in capitals that this is an analyst estimate and not a sourced figure, and that it measures something different from the tracked total beside it.

It then shows the build with overlaps removed rather than summed. AI-specific data centre capex through 2030 is $5.2tn of McKinsey's $6.7tn total, the remaining $1.5tn being traditional IT and excluded. Hyperscaler capex sits inside that figure and is not counted twice, though its trajectory is the reason to believe it: the four largest guided $410bn for 2025 and $725bn for 2026, with $1tn projected for 2027. Semiconductor capacity is not data centre capex, so it is added, running about $200bn in 2026 with cumulative leading-edge and memory capex through 2030 between $1.0tn and $1.25tn. Generation and grid beyond the fence is added net of the share already counted, since American investor-owned utilities have announced $1.4tn through 2030 driven primarily by data centre load, of which roughly $1tn is genuinely additional. Sovereign programmes add about another $1tn, and private equity into AI companies roughly $0.5tn.

That sums to about $8.8tn on the mid case, with a defensible range of $8tn to $11tn depending on how the power overlap is treated. The finding then says the quiet part in plain words: $10tn is the round number inside that range and it is a judgment, not an arithmetic result. It closes by naming what would falsify it, which is announced capex being cut rather than raised at the next two guidance cycles, or the power constraint capping the build below 219 gigawatts.

A headline number, attributed to a named analyst rather than an institution, marked as an estimate rather than a fact, shown with its arithmetic and its double-counting removed, bounded by a range instead of asserted as a point, and published alongside the test that would break it. Those are structural properties of the record rather than stylistic choices, which means the four hundredth finding will carry them too.

## Go and interrogate it

The showcase is not a slide. The page carries a question box wired into the graph, and it answers from the claims rather than from a model's memory of the internet. Three questions asked on 11 September, verbatim.

Asked what Oracle's contracted backlog is and how it has changed, it returned $664bn at Q1 FY2027, noted the increase from $553bn and $455bn as history rather than as current fact, and cited the SEC filing.

Asked about short interest in Palantir and Oracle and what it does not tell you, it gave the settlement figures and then volunteered the limits unprompted: that the data publishes twice a month in arrears and cannot time anything, and that a falling number is equally consistent with shorts covering into weakness as with no new shorts arriving.

The third is the one to show a sceptic. Asked what percentage of TSMC's float is short, it answered: "The percent of TSMC float that is short is NOT STATED. The short interest is the US ADR line only, where one ADR represents five ordinary shares. The share count is the whole company, most of which trades in Taipei under a separate short-selling regime. A percentage of that denominator would understate by roughly an order of magnitude and would not be comparable with the US names."

It had every input required to produce a number and declined. Any general-purpose model will divide the two figures and hand you 0.13% with complete confidence. The difference is not the model, it is that the record knows which of its own numbers are not comparable and carries that knowledge as data, so the refusal is structural rather than a matter of the machine being in a careful mood that day.

Everyone demonstrates the answers their system gets right. Watching one decline to answer something it cannot support, on a question where a wrong answer would have looked perfectly reasonable, is worth considerably more.

The whole record is machine-readable. The [data.json](https://kxco.ai/ontology-live/data.json) export carries the complete graph, and the [Cypher export](https://kxco.ai/ontology-live/ontology.cypher) will load it into your own Neo4j instance in about forty-five seconds. All three figures here came from Cypher queries run against exactly that file after a wipe and reload verified at 393 entities and 869 claims, so the arithmetic can be reproduced rather than trusted.

Publishing it in a form a competitor can load is deliberate, and it costs us something. It is also the clearest statement we can make about where the value actually sits, which is not in the data. Public information assembled carefully is a demonstration. The engine that assembles it, keeps it honest across years and points it at something only you can see is the product.

## Round Table, pointed at your own book

The engine underneath is Round Table, KXCO's proprietary ontology engine, and it is a different category of tool from what most desks are being sold. The market is full of systems that deliver more data faster and leave your people to assemble the picture alone. Round Table closes a different gap: it delivers situational awareness and reasoning alongside your people, so the machine and the human work the same picture at the same time. Agents propose, a named person decides, and the decision is recorded with the evidence that supported it in a form that still reads in ten years.

Against a real book that stops being eighteen public tickers. It becomes your positions joined to your counterparties, your diligence files, your correspondence and your portfolio companies.

Every holding carries its thesis as dated claims rather than a memo somebody wrote once, so when a position falls 24% you can establish in minutes whether the reason you bought it changed or only the price did. That is the Oracle question in this article asked about your own portfolio. Every counterparty carries an exposure map, so a single point of failure three hops from a position surfaces before it matters: Figure 1 is that calculation over thirteen public companies, and over a book it answers a harder question about how much of your capital depends on one supplier nobody has on a watchlist. Every claim carries who asserted it and when, so an investment committee can reconstruct what was known on the day a decision was taken rather than what the file says now, which for anyone with limited partners, a regulator or a board is the whole of the audit. And every number that stops being true is superseded and dated rather than overwritten, so the history of your own marks survives their correction.

Round Table is available to funds, family offices and institutions. It runs against private books of record, deployed and pointed at a specific book rather than handed over as software, which is why there is no self-service tier and no public version.

If you allocate capital and would rather have this running over your own holdings than over eighteen names everybody can see, [start the conversation with us](https://kxco.ai/contact). We will show you what the engine finds in your own data before you commit to anything.

## What we are watching

Oracle's next backlog print, and whether delivered capacity keeps pace with contracted revenue, since 850MW and 300,000 GPUs in a quarter is the constraint $664bn has to pass through. The 15 September short interest settlement, and whether ASML's rising short position continues. SMIC's Ascend production rate, and whether any laboratory outside China reproduces frontier training on non-Nvidia hardware. Any reconnection event at Intel, where one foundry customer of scale moves a node carrying 26 edges a very long way. And the political risk around the Chinese names, which is the one variable that would let us act on work that is already finished.

Stocks mentioned in this article: $NVDA, $TSM, $ASML, $INTC, $GOOGL, $SPCX, $CBRS, $ORCL, $ARM, $BABA, $BIDU, $META, $AAPL, $AMD, $MSFT, $AMZN and $PLTR.

The ontology behind every figure is public and queryable at [kxco.ai/ontology-live](https://kxco.ai/ontology-live/).

Shayne Heffernan, Ph.D., is the founder of Live Trading News, the KnightsBridge Group, Knightsbridge Law and the KXCO.ai ecosystem spanning post-quantum cryptography, identity, attestation and enterprise ontology.

This article is commentary and research, not investment advice. Nothing in it is a record of a trade, a recommendation or a position. Every price is a close-to-close figure between the dates stated, and every position figure is a regulatory settlement number carrying its own date. Prices move. Do your own work.

---

This Markdown mirrors https://www.livetradingnews.com/what-the-kxco-ontology-exposed-while-the-market-stood-still. The HTML page is canonical.
Site index for AI clients: https://www.livetradingnews.com/llms.txt