Burst Monitor · the read from the Friday, November 13 close
The S&P 500 gained 2.5% last week and four of every five stocks finished higher. It was a real rebound from the worst day since April 2025. Underneath, it was mostly one group's rebound. The chip and AI-hardware stocks that were hit hardest on November 5 bounced back together, while the home, consumer and property stocks that have led since October stood still as interest rates jumped. An instrument built to watch how the five hundred move together saw each of those groups moving more tightly, as a group, than at any point this autumn. In the hardware group's case, more stocks moved as a single unit than on any day in its 26-year record. Here is what it saw, in plain terms, how firm each part of it is, and what to watch with Nvidia reporting on Tuesday.
the typical (median) return of each group since Aug 13, with the S&P 500 and the typical stock of all 500 · the two groups were fixed by name on Oct 21 · the dashed line is the Oct 6 record · shaded = this week
If you are new here · the story so far, in four lines
The read in one paragraph. The board does not show a market breaking down together, and it does not show one that has healed. It shows a market organized into two blocs that take turns. The bloc that led October, home builders, retailers, landlords and payment companies, stopped rising the day interest rates jumped. The bloc that fell hardest, chips and AI hardware, bounced as a unit and is still well below where it started. The reading that would mean real trouble is the two blocs starting to fall together; the reading that would mean repair is the hardware bloc continuing to recover while the leaders hold. Neither has happened yet. The analyst's judgment is caution: there are signs of real stress, and not yet the signs of a market that has tipped. Chapter 5 lists what would tell the two apart.
1 · The week
Monday did nearly all of it. The index rose 2.35% and 84 of every 100 stocks rose with it. The other four sessions left the index almost exactly where Monday put it, and they were not quiet underneath: on Thursday three stocks in four fell while the index barely moved, and on Friday the index fell while 63 of every 100 stocks rose.
An index and its typical stock can disagree because of arithmetic. The S&P 500 is cap-weighted: each company counts in proportion to its market value, so its ten largest listings, well over a third of the value, can pull the index away from what most of its members did. "Typical" here means the median: the stock in the middle when all five hundred are lined up by their returns.
each session's return: the index, the typical stock of all 500, the rally group and the AI group · the label over each day is the share of the 500 that rose
Hold the two groups fixed and the week sorts itself. The AI group gained 3.77% at the median, but made all of it on Monday and Wednesday and gave ground on Thursday and Friday. The rally group gained 1.95%, most of it on Monday, and then went sideways. Friday was the familiar alternation in miniature: the rally group up half a percent, the AI group down two-thirds of one.
| group (held fixed by name since Oct 21) | names | Monday | Friday | the week | % up, week | since Oct 6 |
|---|
Each row is the median of its members' own returns, not a portfolio. The two groups are the instrument's October 21 selections, not industry categories, and they do not cover the whole index. The ten are the ten largest listings shown: nine companies, with Alphabet trading as two lines.
Two details keep the rebound in proportion. First, it was not a uniform AI recovery. Nvidia lost 12.7% on the week and is 32% below where it stood at the record, while Broadcom gained 18% and Micron 13%. Second, the sectors that led October were the ones that stood still. Energy gained 6.5% with all 21 of its names higher and industrials 3.6%, while the typical consumer-cyclical stock gained 0.1% and the typical real-estate stock 0.2%.
Since the October 6 record the picture is the one this series has described for a month: the index is down 3.3% while its typical stock is up 7.3%, with 73 of every 100 members higher. The rally group is up 11.2%. The AI group is down 7.7%.
2 · Two blocs
The instrument's simplest reading takes every stock's daily return over the last 21 sessions and asks how many separate things are going on. The story count is the effective number of independent patterns in those returns. Twenty would mean twenty separate stories. A count collapsing toward three or four means one move has taken over, which is what the market looked like on the way into 2008, March 2020 and the spring of 2025. Since mid-October the count has fallen from 13.6 to 6.0, and it fell again this week. The question is what it is falling into.
Two more words. The first pattern is the single biggest shared movement in those returns. In an ordinary market it is simply "the market": on an up day nearly every stock rises. The second pattern is the biggest movement left once the first is removed. In this market the first pattern is carried by the rally group and the second by the AI group, each largely absent from the other's.
21-session reader (ink) and 63-session reader (gold), by morning · each morning's reading uses returns through the session before, so the hollow markers are the windows that end on Friday itself
If one move were taking over, the first pattern would dominate. It does not. Its share of the market's day-to-day variation rose a little this week, to 34%, which is close to its long-run middle. The second pattern's share rose too, to 19%, the fifth-highest reading in more than 6,700 sessions, against a usual 8.5%. The count is falling because a second story is unusually strong, which is a different condition from one story swallowing the rest.
This week the analyst noticed something on the board that sharpens it. The instrument keeps a tally of stocks that have been captured by the first pattern, meaning that pattern alone accounts for more than half of the stock's own movement over the last three months. That tally had been climbing for two weeks: 25 names on October 30, 56 a week later, 68 on Friday. The board keeps no such tally for the second pattern. Run the same test there and it shows 18 names, then 43, then 58.
a stock is captured when one pattern accounts for more than half of its own movement over the last 63 sessions · gold = the first pattern · blue = the second · each morning's reading uses returns through the session before
The names in each list are not a mystery. The first is home builders (D.R. Horton, PulteGroup, NVR), home improvement (Home Depot, Lowe's), landlords and property trusts (Realty Income, Simon Property), payment and ratings companies (Visa, Mastercard, S&P Global, Moody's) and household staples (Procter & Gamble, Colgate). There is not one technology company in it. Call it the home-and-consumer bloc. The second is chip makers and their equipment suppliers (Broadcom, AMD, Micron, Lam Research, KLA), AI hardware (Arista, Dell, Vertiv) and the electrical and power-equipment companies building data centers (Eaton, GE Vernova, Quanta). Call it the hardware bloc. Those labels are the analyst's, from reading the lists. The instrument sorts by co-movement only; it does not know what any company makes.
| the home-and-consumer bloc | the hardware bloc |
|---|
Returns are medians of the members' own returns. So that the week's own returns play no part in choosing the names, the returns use the lists as reconstructed for Friday, November 6, from data through November 5. The lists as of this Friday give +0.28% and +7.26%.
The two counts are not equally unusual, and both lean on a few large days.
In size, the first is ordinary and the second is a record. In most years the first pattern captures more of the market than it does now, about one stock in six; October's 25 was the unusually low figure, and 68 is a climb back toward normal. What is notable is who is in it. These are the stocks that carried the rally, they are up about 15% since the record, and this week they stopped. Fifty-eight stocks captured by a second pattern is the highest figure in the 26 years of record; the previous high was 45, around the turn of 2017, in the rotation that followed the 2016 election.
Both counts are amplified by big sessions, on different days. The hardware count owes the most to two: November 5, when the bloc's typical member fell 9.4%, and Monday, when it rose 6.2%. Take those two days out of the three-month window and its count falls from 58 to 6. The home-and-consumer count has its own pair. Take out November 4, the day after the election, when its typical member rose 5% while the hardware names fell 4%, and October 22, a day of the same shape, and it falls from 68 to 25. In neither case does the pattern itself go away. The same stocks still lean the same way; fewer of them clear the more-than-half bar. A count like this turns a matter of degree into a yes or a no, so it moves in jumps.
That is worth knowing for two reasons. It says what kind of group the hardware bloc has been: one that falls together and rises together on the big days, which is exactly the property that matters to anyone who holds several of its members and thinks of them as separate positions. And it says how to read the count from here. Those sessions stay inside the three-month window into early February, so a high count by itself is not fresh confirmation. The thing to follow is the names, and what the same stocks do on new days.
One name already stands out. Of the 43 stocks on the hardware list a week ago, 42 are still on it and 16 more have joined. The one that left is Nvidia. The share of its movement that the second pattern accounts for fell from 51% to 35% in a week. The bloc's rebound went on without its largest member, which is increasingly on a path of its own.
Two smaller readings on the board point the same way. The instrument also tracks how loud or quiet each stock's own swings are against its own history. On Friday 42 names were unusually quiet, the most this year, and 19 unusually loud, about twice the usual number. Only four of the 42 were on the quiet list a week earlier, so this is a condition spreading through changing names, not a fixed group. And the spread between the quiet and loud ends printed at 1.67, among its widest prints of the autumn. That print should be held loosely: the last several mornings' prints of the same measure have each been revised down by 0.2 to 0.3 once later data arrived.
The market did not become one thing this week. It became more plainly two.
Outside the instrument · the analyst's layer, kept separate
The instrument reads prices only, by design. Its job is to say where the pressure and the coordination are, not why. Everything in this box is the analyst's: dated facts, an explanation they are compatible with, one view stated before the outcome, and what would weaken each.
The facts, dated against the sessions. Sunday Nov 8: Nvidia's board announced an independent investigation concerning its chief executive. Tuesday Nov 10, 8:30 am ET: consumer prices for October rose 0.4%, against 0.3% expected, with core prices up 0.3% for a third month; a Federal Reserve governor had said the evening before that a half-point increase should be considered in December if core prices came in that high. The two-year Treasury yield, the market's shorthand for where the Fed is headed, closed at 4.73% from 4.56% on Monday, its largest one-day rise since April 2025. Wednesday Nov 11: Nvidia's board accepted the resignation of Jensen Huang; the company announced it on Thursday and named its finance chief, Colette Kress, interim CEO. Friday Nov 13: the two-year yield closed at 4.88% and the ten-year at 5.15%, from 4.51% and 4.86% a week earlier. The Fed next meets December 8–9. Nvidia reports earnings on Tuesday, November 17, after the close.
A compatible explanation. Home builders, landlords, retailers and payment companies are among the businesses most exposed to borrowing costs and to household budgets. Their bloc rose with everything else on Monday and then lost ground on each of the next three sessions, beginning the day the inflation report landed. The hardware bloc followed a different path: a rebound from the November 5 selloff that ran across the group, with Broadcom and Micron in front, while Nvidia lost its chief executive and an eighth of its value.
The analyst's read of the rebound: a caution. Taken together, the week's readings point to real stress. The stocks that led stopped rising as borrowing costs jumped. The stocks that fell are moving in lockstep on the big days, and the largest of them has gone its own way. New structure is forming at the edge of the market's memory. None of that yet amounts to a market that has tipped: its down days are still not shared ones, the first pattern is no stronger than usual, and the typical stock remains well above where it stood at the record. So the rebound is best held as a caution rather than a call in either direction, a condition on the radar and being watched closely, with the markers in chapter 5 as the things that would change it.
The analyst's view, stated Tuesday November 10, before that day's close. There would be no ordinary post-midterm bounce this time: households have already used the credit and savings that normally cushion them, and nothing that was pressing on them before the election has changed. On the instrument that view predicts the rally group, consumer names in front, rolling over and the two crowds converging. The history it argues against is strong: the index has been higher twelve months after every midterm since 1950. So far: from Tuesday's close to Friday's the index gained half a percent, and the home-and-consumer names did stall. But the market's down days have not yet become shared ones, which is what the view requires.
What would weaken each. Against the rates reading: staples and financials rose more than 2% in the same week, and the home-and-consumer bloc gained on Friday as yields made new highs. Against reading the hardware bloc through Nvidia: the bloc rose nearly 7% in the week Nvidia fell 13%. Against both: neither explains the split itself, which began in July.
3 · The map
The second picture does not use returns at all, which is what makes it a useful check on the first. It asks how much of each stock's movement is the market's shared movement, and whether that stock's share of it rises and falls in step with everyone else's.
The motion map places each stock by two things. Vertically, its pressure: how caught up it is in what the whole market is doing, against that stock's own normal. The white line is the floor; below it a stock is under more shared pressure than usual, above it the stock is moving on its own. Horizontally, its coupling: how closely that stock's pressure follows the marketwide pressure over the last six months. Far right is the crowd, stocks breathing in time with the market. Far left are stocks keeping their own time. Being low or left on this map says nothing about whether a stock's price rose or fell.
Watching this map as a film through the week, the analyst's reading was that the field has two gathering regions, with stocks breaking toward one or the other: the crowd at the right-hand edge, and a group out on the left. Checked by name, that is the same division again. Over the week the rally group's typical member moved 0.11 toward the crowd, on an axis that runs from −1 to +1, and 82 of every 100 of its members moved that way. The AI group's typical member moved the other way. The number of stocks inside the crowd rose from 70 to 107 in five sessions.
the same positions as the screen above · gold = stocks captured by the first pattern (home-and-consumer) · blue = captured by the second (hardware) · grey = everyone else · hover for names
Color the map by the two blocs from chapter 2 and they fall on opposite sides. Of the 59 home-and-consumer names on the map, 23 are inside the crowd and one is on the far left. Of the 52 hardware names, 23 are on the far left and one is in the crowd. Plenty of both sit in between, so this is a corresponding division seen by a second tool, not the identical one. It is still two measurements built from different things, returns in one case and pressure in the other, drawing the same line through the market.
the typical member's position on the map's horizontal axis, six-month lens · gold = the rally group · blue = the AI group · above zero is toward the crowd
The longer view shows how new this is. Through the summer the two groups rose and fell on this axis together, the AI group the more tightly coupled of the two. In October they separated. The AI group has dropped out of step with the market's shared rhythm, while the rally group, after falling with it, turned back toward the crowd three weeks ago and is still going.
One caution belongs with any reading of this map. Its lens is a rolling six-month window, so a stock's position can also shift because an old stretch of sessions has dropped out of the back of the window rather than because of anything new. This week's move was checked for that: with the window's start held in place, the rally group's shift is unchanged. The same check will matter more from about November 20, when a dense stretch of sessions from last spring begins leaving the window and the whole map will move for that reason for several weeks.
A third tool agrees in its own terms, and it comes with a picture. The instrument sorts stocks into rooms by who moves in rhythm with whom, and draws the result as a river: each column is one session, each block in it is a room, and the ribbons show where the stocks in a room went next.
Count the blocks first. At the start of the month a column held nine rooms; Friday's holds seven. Nearly half of all stocks, 48%, changed rooms over the past two weeks in the company of at least four others, up from 37% a week earlier: fewer rooms, and more coordinated traffic between them.
Then follow the lit ribbon. For 44 sessions the utility stocks had a room named for them. On Friday the ribbon swings up into a larger room and the purple is gone. The utilities did not scatter: 23 of the 24 are still together. The room they now share took in enough financial companies to be named for them instead, 25 to 24. A room's label can change hands by a single stock. The group underneath is what persists, and it is the group that should be followed.
4 · The memory
The last reading has a time dimension. Every morning the instrument asks how far back the market is remembering: over what lookback its bursts of shared movement are most tightly organized against the index's distance from its high. Last week's edition flagged something new arriving at the edge of that picture. It is still there, it is larger, and it has company.
The picture is called the surface. Up the side is the lookback, from about six months at the bottom to more than two years at the top. Across is timing: on the left half the shared movement follows the index's drawdown, on the right half it leads it. Each cell is colored by the relationship at that lookback and timing: red where the two move together, blue where they move against each other. Red and blue are signs of a relationship, not market up and market down. For most of this year the picture has been dominated by the large red structure on the left: the market organized around an old memory, with the bursts trailing the drawdown by seven or eight months.
The reading here is the analyst's, from watching reconstructed films of this surface that cover more than five years of market history. The place to watch is the bottom edge, and the moment that matters is a transition: a well-formed structure of the opposite kind arriving along that edge while the established one still holds its shape. Two weeks ago the lower right was nearly empty. Now a blue structure fills about a third of the zone marked B, and a red one nearly half of the zone marked A.
share of cells showing a strong relationship in three fixed zones of the surface · ink = the established red structure on the left · red = zone A · blue = zone B, the two boxes marked on the screens above
Three things keep that in proportion. The established structure is not fading: by area it is within a point of the largest it has been all year. The middle of the picture, the short lags and leads on either side of zero, is almost empty of either color; what structure there is sits at the two ends. And a full zone is not the same thing as a lasting structure. The chart shows zone A filling in August and emptying within three weeks, while the index went on to its record. Looked at closely, that red did not vanish. It weakened and slid sideways out of the box. Fixed zones are convenient to measure, but the structures move, so the question for the weeks ahead is whether these two keep their strength and form, wherever they travel.
One change did complete this week. For a month two lookbacks had been nearly tied as the market's working memory, one of about eleven months and one of about twenty. On Wednesday the shorter one took the lead, and by Thursday it had held it long enough to become the instrument's settled reading. The margin between them remains thin.
So the surface says a transition is under way at its edge, and does not say which way it resolves. A plain candidate explanation is worth stating because it can be checked: both arrivals may simply be the pullback since October 6 entering the short lookbacks. If so, they should fade if the index retakes its high, and keep growing if the pullback extends.
5 · What would change this read
Last week's edition ended with five things to watch and three ways the market could go. Here is what each marker did, which branch the week fit, and what to watch now.
| what was watched | a week ago | what happened |
|---|---|---|
| Whether the largest AI names had really finished falling in their own swing | seven of the ten largest read as past the bottom | Mostly held. On the re-readings early in the week, six or seven of those seven stayed past the bottom. |
| Whether a broad day would shrink the second pattern | second pattern 18% of the motion | It did not. Monday lifted 84 stocks in 100 and the second pattern's share rose, to 19%. |
| Stocks climbing back toward the top of their own swing | 15 of 500 | Began to fill, then stalled: 30 on Monday, as many as 41 on Thursday, 30 on Friday. Of the AI group's 37 technology names, two are climbing. |
| Which memory holds | two lookbacks nearly tied | The shorter one took the lead on Wednesday and the settled reading on Thursday. |
| The arrival at the edge of the surface | one new structure, a week old | It persisted and grew, and a second arrived beside it. |
Against the three branches, the week fit the first, still taking turns, on every count. The first marker of a turn appeared midweek, stocks beginning to climb back, and did not hold through Friday. Nothing in the week looked like one market.
Both of Friday's lists are saved. Do they keep their members, and what do the same stocks do on new days? A high count alone is not fresh news; the names are.
Does the hardware bloc move as a unit again, and which way? Does the home-and-consumer bloc move opposite, as it has, or with it? The instrument first reads Wednesday's close on Thursday morning.
Does the bloc keep its distance from the hardware names, or begin to fall alongside them? That is the difference between branch 1 and branch 2.
Do the arrivals keep their strength and form, inside the marked zones or beyond them, or weaken and drift as August's red did? Does the established structure finally give ground?
The instrument's words
Reading notes, sources and construction
comms/fable_20261115/edition_build/analysis.py and capture_history.py write data/results.json; the week's checks are build/w4_my_walk.py, w5_exit_split.py and w6_nested_boxes.py. The two groups are comms/sol_20261025/map_dial_named_endpoints.csv. An independent second read reproduced the capture counts, the roster overlaps and the window-exit result.How this edition was made. Jason Connerty built the instrument, ran its chain on the Friday close, and walked the board; the observation that the captured count was climbing, the two gathering regions on the map and the reading of the surface are his. An independent second read checked each observation against the record, tested the edition's central claim, and reviewed the draft. Claude (Fable 5.2) ran the numbers and wrote the edition.