SCINTILLA · ALLOCATION

What changed on 9 Oct, later — how it gets to the number is a picture (DB2). Alan, 9 Oct, 12:20: "This dashboard tool — how it gets to 75.9 — I don't know, man. Can we do something a little bit more visual on this kind of thing?" The four lines under the formula on the first screen are now one picture (study/db2/how.mjs); the formula line stays above it and every caption is one of those lines, word for word. The account bar runs from 0 to 100% of the account: the solid block is the part that is always in (70%), the track after it is the part that swings with the market (30%), and the track is filled to the number, with a bright mark and the number where the fill ends. The wedge under the track opens onto the scale below it: the track from end to end is the market reading from 0 to 100, so a market reading of 20 fills a fifth of the track. The rows under it are how the market reading is made, all on that one scale. It starts at a typical day (49). Each counted part has a bar that starts where the row above ended — the up colour adds, the down colour takes away, and the dark bar to its left is the level so far. A part that adds nothing (the VIX under 20) is drawn as a bar of no width, never left out. While cash is raised, a bracket spans the level before it with its mark at the half, and the half that is freed is drawn in outline. The last row is the market reading the rows come to, in the same white as the fill above. A sum that runs past the end of the scale is cut there and pointed, and a row says where the market reading stops. The picture is drawn again with every read and with every what-if chip, from that move's own parts. Each figure in a caption is rounded on its own, so three captions can read one more or one less than the total beside them (49 + 2 − 12 reads 39 where the total says 40); the bars are drawn at the exact values, and resting the pointer on a row shows that row's own sum to one decimal. Two small things were put right in the same lines: on a day the VIX adds, its add is no longer said a second time as "from the parts you switched on"; and a move that rounds to nothing no longer carries a sign (the credit line read "−0.0%" at the close of 9 Oct).

What changed on 9 Oct — the first screen is the rebalancing dashboard (DB1). Alan, 9 Oct, 09:50: "ok agree on dashboard but it needs to be wired to the live prices of whatever determines it — SPY, QQQ, VIX and HYG … I need to see it move throughout the day … pies would be a better look, and the total money invested should be the big item … a bit of a summary of how it recommends 78.4% — small, simple and executive … remove the you hold and such." The first screen is the dashboard, live on the tool's own reading: version 3 of the market reading (DS3's recommended version — the VIX counted by Alan's own rule, a close at 20 adds 10 points and at 23 adds 20, a touch counts half; and a treasury rally may only raise the credit part — on top of version 2's two changes), read every minute while New York trades and every five minutes outside it, the age beside the number. The big item is the total money invested at the size the viewer picks ($300K to $1.5M in $50K steps, $500K to start, kept in this browser and nowhere else); nothing on the first screen reads the account — no holdings, no room, no tickers, no levels, no orders. Two pies, whole slices: the whole account at the number (conviction · core · kept in cash) and fully invested (conviction · core), every slice with its share and its dollars at the size. The balance slider splits what is invested between conviction and core (25 : 75 to start, "1 in 4"); Reset puts the size back to $500K and the balance to 25. The summary says how the number is made — the held part plus the tactical part times the market reading, then one line per counted part in plain words. The what-if chips are the tool's own dip engine, SPY and QQQ moving together with credit selling (or rallying) as it usually does with such a move; a chip shows the number and the pies at that move. Everything the tool showed before is in THE LONG VERSION under the dashboard, closed to start; nothing was removed. Version 3's two switches sit under ASSUMPTIONS beside version 2's; with both off the number is version 2's, with all four off version 1's. The number here is this minute's own; the money panel in the long version also shows the average of the last hour, which can differ by a few tenths.

What changed on 7 Oct, later — one forward P/E, and the core candidates' comps. Alan: "Why is there inconsistencies everywhere on the forward P/E … if that's the right one, it should be everywhere", "I need to see Google, Amazon and Broadcom — what the comps would look like … I need it reflected in portfolio allocation … I need to go through the workflow", "debt should matter." The decision cards this page reads were re-priced on the Hub dashboard's own forward P/E: the price over the next four quarters of expected earnings. The cards had used a blend of two fiscal years (21.2 times for Alphabet) and the comps the fiscal year in progress (16.9); the dashboard said 24. With a fresh live price the page divides that price by the card's earnings, so the number here moves with the dashboard's; without one it stands on the card's close and says so. Growth beside it is the four quarters after those, over those four, and P/E ÷ growth uses both.

Where it shows. One line under the brief names the ten core candidates with their forward P/E and how far the price is from the centre of their comps range. Step 5b lists them: price, forward P/E, growth, P/E ÷ growth, the comps range, the peers that set the price, net debt over EBITDA and the Geiger's place in its own year; each opens to its peers' own rows. The knockout's pick cards carry a VALUE line. Each pick in step 6 has two more readings, the forward P/E and the debt, and the peers that price it. The core candidates are a list kept with the page, not a rule. The peers that set a price are the ones that do the same business: for Alphabet, Amazon, Broadcom, Nvidia, TSMC, Vistra and Micron they are named from the 7 October brief; Micron's now include SK hynix, Samsung and Kioxia, whose figures come from a dated file and who are on no other screen.

The blend first; growth two ways; the checks (the 11:20 steering). Step 5b and each pick lead with the comps range from every yardstick together, and open to each yardstick's own row: the company's multiple, its peers' median, the price that median implies and its weight. Forward P/E is one of six. Earnings growth is shown two ways because one number misled on Micron: the last twelve months as reported → the next four quarters (+135%), and the next four → the four after (+22%). The growth credit is said whether it applies or not. Every set is checked automatically — how many of the kept peers share the business, and how far the pricing peers are from the company's size — and a flag is printed with the card's other warnings. Micron and its line are read three ways: US-listed peers only, with SK hynix, and with Samsung and Kioxia too. Amazon is priced with Alibaba, JD and PDD in, their yuan converted. Where the reviewed lines carry all three rails of a long-term channel (Micron, Amazon, Western Digital) the pick says where the price in use sits in it.

Debt. Net debt over the last twelve months' EBITDA, with a word for it, and the years of free cash flow it would take to repay all the debt. Interest cover reads "not on file": the statements the Hub stores carry no interest line. In the Hub's knockout study debt takes points off a name only among those that already passed on fundamentals; this page's own knockout does not use it yet.

A stored price has an age. When the live price service cannot be reached this page used to fall back, without saying so, on a stored price table. On 7 October at 11:08 ET that table was last written on 18 August: Micron was priced at 931 against a 6 October close of 1,046, and every level and comps range was measured from there. A stored price older than four days is no longer used (the Hub's own rule). A name with a decision card is then read on its card's close and says so; a name without one shows no price.

What changed on 7 Oct — the money at the top. The first panel is the deployment system's pie: the shape at this minute's reading outside, the account today inside, then slice by slice, the dips and every assumption (see "The deployment system, live" below). It took the place of the five-market bars that stood there through 7 October — first as placeholders, then reading the second version of the deployment matrix under the words "next round in progress". This is that next round. Neither the first version's numbers (33, 22, 44, 88 and 84% for the five markets) nor the second version's rows are drawn any more, and no scenario file is read. The code that drew the bars is still in the page, unused, and is listed for removal.

The two input sections. Every dial, toggle, button and stored setting was moved, one at a time, to see what it changes: the % invested, the mix, the names, or nothing. INPUTS is laid out from the answer. HOW MUCH TO INVEST holds only what moves the %: the ladder's five steps, the two ends of the VIX and 10-year voters, and (on each voter's own row in step 1) the voters' weights. WHERE IT GOES — SECTOR COMPARE holds what moves the mix and the names and never the ladder: the five weights of the blended sector reading, the ranking's three weights, the smallest sleeve, a conviction name's own size, and the dials for how many names. One honest exception is said beside it: the blended sector reading is also one voter in the heat, so its five weights reach the % through that row — the panel prints how far, measured as it stands. WHERE YOU ARE holds the one dial that only sets the gap in step 7. LISTED FOR REMOVAL holds what changes nothing: nothing was removed, each is still stored, and step 9 prints the whole table.

Conviction by name. There is no conviction share. The conviction names are the approved list (kept with this page, in Alan's words and dated) and any name kept in the knockout. Each has its own size: the figure typed on its row in step 6 when there is one; else the rule written beside it in the approved list — for Micron, two fifths of what is invested and never more than 30% of the account, with 20% the full build; else nothing, so a kept name takes no money until it is given a size. If the names together would take more than is invested, each is cut in the same proportion and says so.

Sleeves in rank order. What is left of the invested part is the core. Every candidate sleeve — the eleven sectors, crypto, metals, oil and the tree's AI cohorts — answers three questions, each as its place among the candidates that day (first 1, last 0, equal answers sharing the middle of their places): GROWTH, the middle expected sales growth of its companies (next fiscal year against the year before, from the stored analyst estimates; three companies with an estimate are needed); REGIME FIT, what the sleeve itself is doing — improving, leading, pulling back or breaking down, in that order; OPPORTUNITY, how washed out its blended reading is. The score is the three places weighed, growth 50 and the other two 25 each by default. Sleeves are funded in rank order and sized by score. There is no cap on how many: the last of the ranking is left out while any funded sleeve would be smaller than the smallest sleeve allowed, and the table says once where the core stops and, on each sleeve left out, what holds it back. A sleeve with no growth reading (a metal has no sales) counts as the middle on growth and says so. The market's own regime does not yet change which sleeve regime is preferred; that is a choice still to make.

Where the core parks. A funded sleeve's money sits in its branch's parent fund until the knockout places it in names: the sector's own fund for a sector, the fund the tree names for a cohort (the memory fund for memory and storage), the metal or the coin itself. The broad equal-weight fund holds the core only when no branch is funded.

The picks read the decision cards. A name with a decision card shows the card in step 6 and on its knockout card: the comps range per share drawn as a band from low to high with a mark at the centre and a needle at the price, and the upside to the centre with the card's own warnings (thin when few peers price it, fragile when one peer moves the centre, peers without figures); growth as sales over the next twelve months against the last twelve — sales, because a one-off gain sits in other income and never in sales; the Geiger's place in its own year; and the nearest reviewed line under and over the price, by the Lab's own label. In the knockout a carded name's growth, forward and trailing P/E and price to sales are the card's figures in place of the feed's; margins and leverage are still the feed's. A name without a card reads the feed and says so. The feed is checked as it is read — of the names it shows earnings for, how many it gives a forward P/E — and the sources line marks it while that is under half.

A name's ladder. The ladder under a conviction name is its confluence zones below the price: two or more named levels within 1% of each other, as of the close the zones file is dated — reviewed lines by the Lab's own labels, and the 21-, 50-, 100- and 200-day averages. Each zone is a band on one strip with the price at the right, and is listed with its members to the cent and its distance from the live price. A line standing alone between two zones is drawn as a tick and listed as standing alone, never as a zone.

The deployment system, live (7 Oct). The answer to "how much, in which sleeves, at which levels", read from live prices. Its reading, its parts, the levels, the reset tracker and the stretch rules sit under the ladder in step 2, where the matrix's line was; its pie sits in the money panel at the top. Steps 2 to 7 still run on the ladder: nothing else on this page uses the reading yet. The reading runs from 0 to 100; 50 is a typical day. It is 50 plus the points of the parts that vote, and it moves one thing only: how much of the tactical money is deployed. The parts. Fifteen readings were tried in five families — SPY and QQQ together (RSI, distance from the 21-, 50- and 100-day, the daily Geiger's trend and momentum), the VIX (its close and its intraday high, each placed in its own past year), breadth (the eleven sector funds and the equal-weight fund above their 50- and 200-day), credit (HYG's RSI and Williams %R with its payouts added back, and credit's own ten-session move with the rates leg taken out — half of the 7–10 year treasury fund's move), and the leaders (how many of the six core candidates hold their 21- and 50-day). Each was fitted on the days from 10 March 2009 to the end of 2017 and replayed on 2018 to 2026 against what the 50/50 blend of SPY and QQQ did 20 and 60 sessions later, on top of the parts kept before it. Two earned a vote: the two indices' RSI (on 60 sessions) and credit's own move (on 20). The other thirteen are lights: each row shows what it would add, drawn as an outline, and a tick box counts it if you want it counted — each light was measured one at a time, so ticking several counts overlapping evidence twice. The odds under the reading are from days the model had not seen (fitted to 2017, read on 2018 to 2026), beside any day's. The pie. Three kinds of money: the core, held through ordinary pullbacks; the conviction names, each with its own slot; and the tactical part, which the reading deploys. The outer ring is the shape at this reading; the inner ring is the account today at live prices; what is held fills a name's conviction slot first, then its core share. Inside the core the sleeves carry rank weights in comps order. Its sizes are its own inputs: it does not yet read step 3b's conviction sizes or its ranking, so the two can differ until they are joined. The dips are a further fall in SPY and QQQ together from this minute's prices. The headline number holds credit where it is (a calm dip); the number beside it lets HYG sell off the way it usually has (0.2% per 1%). The VIX, the funds and the names move the way they usually have (the VIX about 7.8% per 1%, each name by its beta to QQQ); "reaches" names the deepest add level each name would touch. Levels. For each name: its 21-, 50-, 100- and 200-day, the nearest named pivots under the price from the Lab's reviewed lines as extracted on 6 October (TSMC and Oracle have none stored), and where its own usual cool-off would take it — each with how far away it is and the QQQ and SPY level it maps to through the name's 120-day beta to QQQ; a level that lands within 1% of one of QQQ's own averages says so. The reset tracker. A push is a close at a 60-session high with the RSI at or above the name's own hot mark (the 75th percentile of its past year); it starts a leg, and the cool-off runs from the leg's top until a close back above it. For each gauge the shaded zone is that name's own past cool-offs — it begins where the shallowest quarter stopped and darkens where the usual one reached; the dot is now and the small triangle is the low since the top. Nebius has too little history and reads against the pooled leaders. The stretched end. Each rule is scored as one trade: sell at the signal's close, buy back at the first close 2% or more lower inside 20 sessions, otherwise at the close of session 20. What is live. Prices are the chart API's quotes and its macro board (the VIX, with the session's high and low); today's high and low per symbol come from its 30-minute bars; the daily closes are its last 540 sessions; the Hub's own seven-timeframe Geiger is shown beside the daily one. It reads on load, every minute while New York is open, every five minutes before the open and after the close, and once from 16:10 for the settled bars — never while the tab is hidden. The age of the reading ticks every second. A read that fails leaves the last reading in place with its age and the cause; a symbol with no live price keeps its last close and is named; a part with no reading does not vote and says so. What is not live. The model, the odds, each name's usual cool-off, the named pivots, HYG's payout list and FRED's high-yield spread are a dated file rebuilt by the study; the spread is shown with its date. Every assumption is an input with its baseline in brackets, kept on this device. The account is not kept in this page's code: it starts from a made-up example, marked as one, until cash and shares are entered under ASSUMPTIONS. The measurement, every part's test, the trim odds, the replay and what could be wrong are on the study page, study/ds1/DS1.html.

What changed late on 6 Oct — pictures in place of paragraphs. THE BRIEF is one sentence, one gauge, two bars and at most six bullets. The gauge is the market's heat on its usual scale with the ladder's five steps drawn as bands (the edges at −0.5, −0.2, +0.2 and +0.5 are the lines between them), today's needle, the % invested under each band, and under it the two edges that would change the step with how far away each is. The two bars are the two camps: every voting row above zero against every voting row below, each with its weighted average and the share of the vote it carries. The sentences those pictures replace are kept, folded, under "the same in words" in step 1.

The voters, by family. Every row sits in one of five families: INDEX (the S&P, the Nasdaq, small caps, semis, the eleven sectors blended), BREADTH (equal-weight against cap-weight in every pair, advance/decline, our companies above their 200-day and 50-day, concentration, small against large, TRIN, up against down volume), RATES & CREDIT (the 10-year, the curve, high yield, long bonds, and oil, which votes as inflation pressure), SENTIMENT (the VIX, its term structure, the put/call, bitcoin as risk appetite), DEFENSIVES (staples and utilities, gold). A family is drawn as one bar — the weighted average of its voting rows — with its share of the vote; press it for its rows. A row's share of the vote is its weight divided by the weight of every voting row, and that number says whether the row is used: a row at 0% is a reading only, and a row with no reading or one 30 days old sits in the folded list below with its reason. The heat is the sum of each family's share times its average — the same weighted average as before.

What votes now that did not. (1) Breadth reads every pair: the equal-weight fund's Geiger minus the cap-weight fund's for the S&P (RSP − SPY), the Nasdaq-100 (QQQE − QQQ) and the eleven sectors, averaged; it took the weight the single S&P pair carried, and that pair stays under it as a reading. Inside ±0.05 a pair counts as level. (2) The share of our own companies — the ones in the tree, about 450 — whose last close is above their own 200-day average, and the same against the 50-day: half above reads 0, all +1, none −1; weight 0.25 each. It is our universe, not the whole market. (3) The Interactive Brokers put/call: its place among its own measured sessions, turned so a low put/call (little fear) is the hot side; while it has fewer than 20 other sessions the vote is scaled by sessions ÷ 20, and under three it does not vote; weight 0.25. On a device that already holds saved dials only those weights were changed, once.

Sectors that know the tree. Step 3a shows the eleven sectors, crypto and metals as before, and under them the cohorts of the tree's AI heading — the tree is read from its two tables as it stands, so a revision of the tree shows by itself. A cohort is read three ways, each on the Geiger scale, averaged with equal weight: its members each counted the same, its members by market value, and the fund the tree names for it. Metals are read from the metal: the bullion trusts for gold and silver on the chart API's Geiger are the reading; the futures row from the Hub's second source is shown beside it and never averaged in, because it is a different computation and the two can disagree widely; copper has no commodity line in the estate and is named as missing; the miners are the folded detail in the metals branch and never enter the reading. "How the blend is made" draws the five methods of every sector side by side with the blend, the spread from the highest method to the lowest as a bar, and the number of companies that sit in every method's set (the fund's holdings, the names we serve, the tree's close tier, the membership the ranking averages). A cell more than 0.20 from the blend — half the middle step's width — is ringed, and a spread of 0.40 or more, a whole step, is marked "they disagree".

Branches. Pressing a row in 3a opens its branch: the funds to track (the State Street fund, the Invesco equal-weight twin, and the Vanguard and iShares near copies the tree lists; for a tree cohort the funds the tree names; for metals the trusts) and the names that pass the comps filter — at least two measured fundamentals readings and a fundamentals score of 0.50 or more against the name's own peers, price outliers out first — best first. Each name carries its growth (the consensus revenue of the next fiscal year against the reported year before, from the stored analyst estimates; the comps feed's growth is not shown because it changes from call to call), its Geiger, and its nearest reviewed line above and below the price with the Lab's own label. The lines come from the Lab's extract of 6 Oct, kept as a dated file beside the page (19 names); a sloped line moves a little each session, so every level carries that date. When the dated lines file has not the name, the levels its decision card carries are used.

The shape of 6 Oct, replaced on 7 Oct. That night the money had a fixed shape: conviction a fifth of what is invested, the equal-weight index half the core, six sleeves at most and none under 4%. Three of those four are gone (see the 7 Oct paragraphs above); the smallest sleeve stays, and the three dials are listed for removal in INPUTS.

Small caps or breadth. If no branch is funded the core waits in the equal-weight S&P, not in small caps: the bet is that the average large stock catches up with its index (breadth returning). The two reads are shown side by side from the same daily bars — the Geiger gap to the S&P, the distance off the high of the last 225 sessions, the 50-day and 200-day, the RSI. The count in the last row is the regime study's, dated 5 Oct: of the sessions of the last two years that looked like today (middling heat, narrow leadership), how often the S&P beat the equal-weight S&P over the next 60.

The regime step's mechanics (6 Oct 2026). PUT / CALL is the Interactive Brokers collector's own count (puts ÷ calls over the served names, one row per session): the row shows the newest session that measured anything, with its date, how many names were measured and its age; a session where the gateway was logged out measures nothing and is named. (That morning it carried weight 0 and no level; since that night it votes on its own sessions, scaled down while they are few — see above.) A row with no reading, or whose newest reading is 30 or more days old, cannot vote: it sits in the folded "not voting — why" list under the voters with its reason and date, and its weight is kept. The sentence under the heat splits the voters into the camp above zero and the camp below, each with its weighted average — hot weight × hot average plus cold weight × cold average, over all the weight, is the heat. "Against its own past" replays the same formulas over the stored readings: every voter with a stored past, only on days when all of them are there, and separately the four with a full year; today is measured on the same rows it is compared with, the count of days is always given, and under 60 days the line says THIN. The sector count in the heat row and in the brief is one function: the eleven State Street sectors, each on its blended trend and momentum.

The sources line (6 Oct). Under the title, one line says how many of the page's sources read live, how many are old and how many could not be reached; "details" opens the list. A source is LIVE when it was read just now and is within the age that source is allowed (the Geiger thirty minutes, the quotes fifteen, a profile a week, the sector ranking three hours); OLD when the read worked but the newest row is older than that; UNREACHABLE when the read failed, and the row then says what the page uses instead (an older table, or a voter that abstains). Nothing is faked: an unreachable source never shows a number.

The knockout. Built to the written knockout concept on the Hub and to nothing else. Who may face whom: only two companies in the same business line — the comps method's business lines (imported from the Hub as they are) over each company's revenue by segment and the company profiles, each company placed in the line that carries most of its revenue; one hand rule added here (AMAT, LRCX, KLAC, ASML are semiconductor equipment). How a duel is scored: four readings (price = P/E or P/S for a loss-maker, revenue growth, net margin, leverage), each the name's position between the first and third quartile of the line's field (the family when the line has fewer than five measured names; the card says which), the peers priced strangely out of the field first; the score is the average, a missing reading counting one half; a duel is won on fundamentals when the scores are at least 0.02 apart, otherwise the Geiger decides the timing. Seeds by score inside the line, first plays last, rounds until one is left; a lone name is champion unopposed and says so. The champions stand on a podium by score and never duel; the podium's first names are the picks. The fundamentals for every served company are loaded so each line has its field.

3a and 3b (as built on the morning of 6 Oct; the rails, the pie and the ranking rule it mentions have since been replaced, see above). Step 3 is two screens. 3a SECTORS says what each sector is doing and nothing else: the blended reading per sector (five readings averaged, the weights in INPUTS), drawn in the Hub's own Geiger chip — the same flat bar the Hub's board uses, the centre line at zero, the fill to the right when positive (overbought, stretched) and to the left when negative (oversold, washed out), green and red as on the Hub. The cap-weighted chip on the left and the equal-weighted chip on the right meet at the blend: that is the bow tie. The words are the ladder's own five steps. 3b HOW THE MONEY SPLITS takes those readings and nothing else and shows, column by column, how the invested part is split: coldness = (1 − reading) ÷ 2 (never below 0.05); × the sector's turn (improving ×1.35, pullback inside an uptrend ×0.85, breakdown ×0.6, leading ×1); normalised; each sector held between 2% and 35% and renormalised; × the step's % invested. The last column is the % of equity the pie and the picks use.

What changed on 5 Oct. Two breadth voters: ADVANCE / DECLINE is the line over the names the Hub serves to this tool — one up or down per name per completed session from the chart API's daily bars, summed over the last 20 sessions, plus today's session from the live quotes; the vote is that line's slope, scaled so two thirds of the names rising every day reads +1 (default weight 0.5). CONCENTRATION is the equal-weight twin (RSP) minus the cap-weight index (SPY) over the same 20 sessions, scaled so a five-point spread reads ±1, with the share of the served set's cap-weighted move that its ten largest names carried (default weight 0.5). The VIX votes live from the chart API's macro board; the VIX term structure (spot ÷ 3-month) uses the VIX table's last close for the 3-month leg because no live 3-month VIX exists in the estate, and says that date on its dial.

The sector method dial. TARGET MIX reads each sector one of four ways: STATE STREET SECTORS (the sector fund's own Geiger), HUB SECTOR COMPARE (the fund and its equal-weight twin, averaged), MARKET BOW TIE (the average served stock in the sector, every name equal-weight, live from the chart API, with the bow tie against the cap-weighted read), TREE READINGS (the tree's close tier, 5,316 common stocks and ADRs, rolled up by sector — 652 of them carry a sector in the estate today). The default is a mix, 40 / 20 / 20 / 20 in that order; the weights are dials in INPUTS. The pie, the moves and the ranking all follow the dial. Crypto, metals and oil have no fund, so every option reads them from their own constituents.

The short list. For the five sectors the tool favours most (sound sectors first, then by share), the ADD names and the two best of each one's comps field — the business-first peers from the peer table, same sector and within ten times its size, ranked by fundamentals — as the approved cards. KEEP / DROP appends one row to the comps decisions table through the public write path the Hub's comps tab uses; the latest row per company and peer is the state. An ADD tag needs at least two measured fundamentals.

What goes on the Hub. A pick, a name on his favourites, or a KEEP decision is tagged NEEDS THE LIVE HUB; a comps peer without those is FUNDAMENTALS ONLY — OFF-HUB IS ENOUGH. The list is written as control/ALLOCATION-HUB-GUIDANCE.json (ticker · tag · reason · date) for the served-set decision.

The fold. Every section is folded to one screen (1680 × 1050: about 840 px of body, scrolling inside) with a sticky section bar; THE MONEY and THE BRIEF are open on the first screen; INPUTS opens as a drawer over the page. Nothing was removed; the fold state is remembered on this device.