Analyst Desk · Live paper account
The Paper Desk
Every call the Analyst Desk publishes becomes a paper trade on real market prices, under fixed risk rules, starting from five thousand dollars. Every result is scored here in public, losses left in.
Updated each trading day · Prices: Alpaca market data · Started Oct 17 · Paper trading, not advice
Paper account
$0
−1.4% from $5,000
S&P 500, same period
+0.0%
the benchmark to beat
Win rate
0%
2 closed trades
Avg win / avg loss
— / −4.1%
per closed trade
Worst drop from a peak
−0.0%
Oct 17 → Nov 4
The execution cycle
last run Nov 4, 21:26 UTC
1 · Translator
no new calls
nothing to translate
reads each published call and names the one instrument that tracks it, or none
2 · Second opinion
nothing to check
a second model answers the same question; a trade needs both to agree
3 · Risk officer
stops checked
nothing to size
plain code sizes or refuses: caps, stops, correlation, the opening bell
4 · Executor
no orders
resting stops in place
market orders at the paper broker, fills booked to the cent, stops rested
5 · Scorekeeper
2 positions marked
ledger $4,931.30
marks every position, writes the day’s ledger row, benchmarks against SPY
Account value vs the S&P 500
Where every call went
17 calls · 4 traded · 13 not traded, with reasons
Declined means no listed instrument tracks the call, or the two models disagreed. Blocked means the risk officer refused a proposal: a cap, a stop out of range, a twin already held. Every reason is published below.
Open positions
Shown 7 days after entry · Intel Suite subscribers see them the day they open
TBF direct
−$15.36−3.1%
The 10-year stays high through Dec 9: the 10-year par yield on the US Treasury daily par yield curve is 4.85% or higher on Dec 9 2026 (5.09% on Oct 16)
- Hot September CPI and PPI keep the Fed holding in October and leaning toward a December hike, so the 10-year Treasury yield stays at or above 4.85% through Dec 9
- TBF
The 10-year Treasury par yield, 5.09% on Oct 16, must still be 4.85% or higher on the Dec 9 2026 Treasury daily par yield curve, with long yields not rallying away. TBF should hold roughly flat or gain through Dec 9; a 10-year close below 4.85% (about a 4% drop in TBF) invalidates the call.
◆ fingerprint 9a30cb783a2fa… · Oct 19
$500.04 · Dec 9 · stop −4%
CPER direct
−$12.22−2.4%
Copper stays high on mine supply: the most-active COMEX copper contract settles at or above $6.00/lb every trading day from Oct 19 through Dec 31 2026
- Mine supply shortfalls (Chile output at a 15-year low, Centinela strike risk, falling LME stocks) keep the most-active COMEX copper contract at or above $6.00/lb
- CPER
The most-active COMEX copper contract must settle at or above $6.00/lb every trading day from Oct 19 through Dec 31 2026, up from $6.531 on Oct 16, so CPER holds its level or rises. A settlement below $6.00 means the floor has failed, which is roughly an 8% fall in CPER, so the stop sits at the 8% limit just above that line.
◆ fingerprint f5e60a761241d… · Oct 19
$500.01 · Dec 31 · stop −8%
| The call | The trade | Size | P&L so far | Exit by |
|---|---|---|---|---|
The 10-year stays high through Dec 9: the 10-year par yield on the US Treasury daily par yield curve is 4.85% or higher on Dec 9 2026 (5.09% on Oct 16)
The 10-year Treasury par yield, 5.09% on Oct 16, must still be 4.85% or higher on the Dec 9 2026 Treasury daily par yield curve, with long yields not rallying away. TBF should hold roughly flat or gain through Dec 9; a 10-year close below 4.85% (about a 4% drop in TBF) invalidates the call. ◆ fingerprint 9a30cb783a2fa… · Oct 19 | TBF Inverse long-term Treasuries, unleveraged (rises when long yields rise) direct | $500.04 | −$15.36−3.1% | Dec 9 · stop −4% |
Copper stays high on mine supply: the most-active COMEX copper contract settles at or above $6.00/lb every trading day from Oct 19 through Dec 31 2026
The most-active COMEX copper contract must settle at or above $6.00/lb every trading day from Oct 19 through Dec 31 2026, up from $6.531 on Oct 16, so CPER holds its level or rises. A settlement below $6.00 means the floor has failed, which is roughly an 8% fall in CPER, so the stop sits at the 8% limit just above that line. ◆ fingerprint f5e60a761241d… · Oct 19 | CPER Copper futures direct | $500.01 | −$12.22−2.4% | Dec 31 · stop −8% |
Closed trades
CRAK direct
−$25.57−5.1%
The NYH ULSD crack against WTI, from EIA daily spot prices, averages above $90/bbl in both November and December 2026
- NYH ULSD crack vs WTI stays above $90/bbl through Nov and Dec 2026 as US distillate stocks stay tight
- CRAK
The EIA spot-basis NYH ULSD crack against WTI must average above $90/bbl in both November and December 2026, with US distillate stocks staying well below the five-year average. If that holds, refiner margins stay elevated and CRAK should hold or rise through December 31.
◆ fingerprint 32dd7de685202… · Oct 19
$499.99 · Oct 27 · stopped out
GLD direct
−$15.55−3.1%
Gold's floor holds: the World Bank monthly gold price averages at or above $4,050/oz in both November and December 2026
- Gold's floor holds: monthly average stays at or above $4,050/oz in November and December 2026
- GLD
Record ETF holdings and continued central-bank buying should absorb selling driven by higher real yields, keeping the World Bank monthly gold average at or above $4,050/oz in both November and December 2026. The call fails if gold falls and stays below about $4,050, roughly 2.7% under the Oct 16 spot of ~$4,164, long enough to pull a monthly average under the line.
◆ fingerprint 2a17e8ce1db4d… · Oct 19
$500.01 · Oct 23 · stopped out
| The call | The trade | Size | Result | Closed |
|---|---|---|---|---|
The NYH ULSD crack against WTI, from EIA daily spot prices, averages above $90/bbl in both November and December 2026
The EIA spot-basis NYH ULSD crack against WTI must average above $90/bbl in both November and December 2026, with US distillate stocks staying well below the five-year average. If that holds, refiner margins stay elevated and CRAK should hold or rise through December 31. ◆ fingerprint 32dd7de685202… · Oct 19 | CRAK Global oil refiners (refining margins, diesel and gasoline cracks) direct | $499.99 | −$25.57−5.1% | Oct 27 · stopped out |
Gold's floor holds: the World Bank monthly gold price averages at or above $4,050/oz in both November and December 2026
Record ETF holdings and continued central-bank buying should absorb selling driven by higher real yields, keeping the World Bank monthly gold average at or above $4,050/oz in both November and December 2026. The call fails if gold falls and stays below about $4,050, roughly 2.7% under the Oct 16 spot of ~$4,164, long enough to pull a monthly average under the line. ◆ fingerprint 2a17e8ce1db4d… · Oct 19 | GLD Gold bullion direct | $500.01 | −$15.55−3.1% | Oct 23 · stopped out |
Calls the desk declined, and why
Every verdict is kept. A decline is a judgment too, and it can be wrong.
- no tradeUSDA NASS Prospective Plantings (scheduled March 31, 2027) reports 2027 U.S. corn planted-area intentions (all purposes) below 95.0 million acres, as first published.
The call is about a non-price outcome, the corn acreage intentions in a NASS report due March 31, 2027, which falls outside the 89-day window, so there is no way to hold a position through it. Any link to front-month corn prices before late January would be a stretched chain, not the call itself.
Oct 27 · agriculture
- no tradeChina keeps its October 2025 rare-earth package suspended past Nov 10: MOFCOM publishes a notice extending or replacing Announcement No. 70 by Nov 12 2026
The call is about whether MOFCOM publishes a policy notice extending the rare-earth suspension, not about a price, and no listed instrument tracks rare earths or this legal outcome; the brief itself says it cannot be expressed as a single price.
Oct 17 · metals · the brief
- no tradeThe Red Sea return stays selective: the PortWatch Bab-el-Mandeb 7-day average stays below 35.0 transits/day through Dec 31 2026 (28.3 for Oct 5–11; 2026 range 23.7 to 42.6)
The call is about a transit count staying below a threshold, which is a non-price, status-quo outcome; no listed instrument tracks Bab-el-Mandeb traffic, and BWET's TD3C rates are driven by the Hormuz war rather than the Red Sea, so any chain to a price would be stretched.
Oct 17 · maritime · the brief
- no tradeAsia-Europe container spot resumes easing as Suez capacity returns: Drewry Shanghai-Rotterdam prints below $3,522/40ft on Dec 17 2026
The call is on Asia-Europe container spot rates (Drewry Shanghai-Rotterdam), and no listed instrument tracks container freight: BDRY covers dry bulk, BWET covers crude tankers, and neither can be shorted to express easing rates.
Oct 17 · maritime · the brief
- no tradePhosphate stays tight: World Bank DAP averages at or above $780/t in October and November 2026, because the constraint is sulfur, not the tariff
No listed instrument tracks DAP phosphate prices. MOO is a broad agribusiness basket dominated by equipment, seed and animal-health names, and a call that DAP merely holds above $780/t would barely move it, so routing the call through MOO would stretch the chain.
Oct 17 · food-fertilizer · the brief
- no tradeUS egg CPI stays below year-ago on the November 2026 print despite the fall HPAI season
No listed instrument tracks US egg prices: DBA, CORN, SOYB and MOO do not hold eggs or layer economics, and an egg CPI call does not move any of them in a predictable direction.
Oct 17 · food-fertilizer · the brief
- no tradeFAO Food Price Index is below 140.0 on the Dec 4 2026 release (November index), as first published
The call caps the FAO Food Price Index below 140, which is a bearish-to-neutral view on food prices, and no listed instrument can express it because there is no inverse agricultural fund and DBA, CORN, WEAT, SOYB and MOO can only be bought long.
Oct 17 · food-fertilizer · the brief
- no tradeUS food-at-home CPI is below 2.5% y/y on the November 2026 print
This call is about how one BLS statistic rounds (retail grocery CPI landing below 2.5%), and no listed instrument tracks it. Expressing low grocery inflation would mean shorting agricultural commodities, and the list has no inverse ag instrument.
Oct 17 · food-fertilizer · the brief
- no tradeBitcoin's real-yield sensitivity, narrowed to what survived: the Pearson correlation of daily bitcoin log returns (FRED CBBTCUSD) with daily changes in the 10-year real yield (FRED DFII10), over consecutive DFII10 observation dates Oct 16–Dec 31 2026, is below zero
The call predicts only the sign of the daily correlation between bitcoin returns and real-yield changes, not the direction of either, and no listed long-only instrument pays off on a correlation.
Oct 17 · monetary-crypto · the brief
- no tradeHold in October, hike by December: the FOMC leaves the target range unchanged at its Oct 28 decision, and the range's upper bound is 4.25% or higher after the Dec 9 statement.
The call is about the timing of a Fed policy decision (no move on Oct 28, a hike by Dec 9), and no listed instrument tracks the fed funds target range; the long-end yield trade through TBF is scored separately under MONETARY-006, so using it here would stretch the chain.
Oct 17 · monetary-crypto · the brief
- no tradeNo signed US-Iran agreement that lifts the U.S. blockade of Iranian ports and reopens the Strait of Hormuz is publicly confirmed as signed by both the U.S. and Iranian governments on or before Nov 3 2026
The call is a diplomatic non-event (no jointly confirmed signed US-Iran deal by Nov 3) and explicitly does not call price direction; the stalemate is already the status quo, so no listed instrument expresses it without forcing a chain.
Oct 17 · geopolitical · the brief
- blocked CRAKUS distillate inventories stay at least 8% below the five-year average in every weekly EIA print released Oct 21 through Dec 31 2026
already holding CRAK
Oct 17 · oil-energy · the brief
- no tradeHormuz AIS-visible transits stay below 20/day on every trailing 7-day average dated Oct 17 to Dec 31 2026
The call counts AIS-visible ships through Hormuz staying near the current level, not a price. The brief itself says Gulf oil is still reaching market near prewar volumes through bypasses and dark convoys, so a continued low ship count gives no new move in crude, tanker freight or any other listed in
Oct 17 · geopolitical · the brief
Calls with no fund that expresses them go to the event book below when a public contract settles on the same fact; otherwise they are scored on the Record only.
The event book
Paper contracts at Kalshi and Polymarket public prices · $1,000 book · staked by the translator's edge over the price · shown 7 days after entry
Event book
$1,030.38
+3.0%
Open markets
1
all shown
Settled
2
100% won
Declined
14
no contract matched
- noKalshitranslator 95%−$0.22
Will the 7-day moving average of transit calls through the Strait of Hormuz as reported by the IMF PortWatch be above 60 before January 1, 2027? (Before Jan 1, 2027)
111 × 91¢ paid ($101.34) · now 91¢ · pays $1 if right · closes Jan 5
from the call: Hormuz AIS-visible transits stay below 20/day on every trailing 7-day average dated Oct 17 to Dec 31 2026
◆ b91592208d368…
- yesKalshi+$16.24
Will the Federal Reserve Hike rates by 0bps at their October 2026 meeting? (Fed maintains rate)
116 × 86¢ paid ($99.76) · settled YES Oct 28 · won
from the call: Hold in October, hike by December: the FOMC leaves the target range unchanged at its Oct 28 decision, and the range's upper bound is 4.25% or higher after the Dec 9 statement.
◆ e4ebef70dd9a6…
- noPolymarket+$14.36
US-Iran Hormuz Agreement by November 2? (US-Iran Hormuz Agreement by...?)
114 × 87¢ paid ($99.64) · settled NO Nov 3 · won
from the call: No signed US-Iran agreement that lifts the U.S. blockade of Iranian ports and reopens the Strait of Hormuz is publicly confirmed as signed by both the U.S. and Iranian governments on or before Nov 3 2026
◆ 16056510449e1…
Result by market
- Rates · 1 trade−$15.36
- Energy · 1 trade−$25.57
- Metals · 2 trades−$27.77
Direct calls vs chains
A direct call trades what it names. A chain trades what the event moves next, at a smaller size.
- Direct
- 0%
- win rate · 2 closed
- Chain
- —
- win rate · 0 closed
How the desk trades
1 · Analyst
The Analyst Desk publishes a dated call with a measurable outcome and a resolution window.
2 · Translator
An AI agent traces the chain from the call to a price and proposes one instrument, exit date and stop, with its reasoning.
3 · Risk officer
Fixed rules in code accept or reject every proposal. No agent can change them.
4 · Executor
Approved trades fill at market prices in the paper account and close at their stop or exit date.
5 · Scorekeeper
The account is marked each trading day against the S&P 500 and published here.
The rules
- · Exchange-traded funds, plus bitcoin and ether at half size with wider stops (8–13%, Alpaca holds them around the clock), all bought outright: no leverage, no shorting
- · At most 10% of the account in any one trade
- · At most 61.8% invested at once, across up to 8 positions
- · Two models must name the same instrument before any trade; a disagreement is published as a decline
- · Chain trades sized down: 62% one step removed, 38% two steps
- · At most 2 positions resting on one world view (the Record's correlation group)
- · A fund whose returns have tracked an open position (correlation ≥ 0.79 over 55 sessions) trades at 62%; a third such fund is refused
- · Every trade carries a stop 3–8% below entry, resting at the broker for the whole shares
- · Three single stocks (MOS, CALM, ZIM) stand in only where no fund tracks the variable: counted as at least one step removed (62%), stop 5–8%, at most 2 open
- · Up one stop-width, the stop lifts to entry and then trails that width under the high; it never moves down
- · No fund entries in the first 34 minutes after the opening bell
- · Every trade carries an exit date 5–89 days out
- · Down 3% in a day: no new trades until the next session
What the desk can trade
Energy
- USO
- BNO
- UNG
- CRAK
- XLE
- XOP
- URA
- UGA
- OIH
- XLU
Metals
- GLD
- SLV
- GDX
- COPX
- XME
- CPER
- REMX
- LIT
- SIL
- SLX
- PPLT
- PALL
Rates
- TLT
- TBF
- IEF
- TIP
- HYG
- SJB
Currencies
- UUP
- UDN
- FXE
- FXY
Agriculture
- DBA
- CORN
- WEAT
- MOO
- SOYB
- CANE
- MOS
- CALM
Crypto
- BTC/USD
- ETH/USD
Equities
- SPY
- SH
- KRE
- PSQ
- SMH
- FXI
- KWEB
- EEM
- EUM
- EWJ
- EWZ
- JETS
Real assets
- VNQ
- XHB
- DBC
- KRBN
Shipping
- ZIM
- BDRY
- BWET
- SEA
Defense
- ITA
The thinking behind the rules
What this desk is
A public test of the Analyst Desk's judgment. Every call a published brief makes is turned into at most one paper trade on real market prices, under fixed rules, from a $5,000 book. The point is not the money; it is a record nobody can edit after the fact: each decision is fingerprinted before the outcome exists, every loss stays on the page, every call the desk declined is listed with its reason, and open positions reach the public after 7 days.
The pipeline, run by run
- Translator (the one AI step, with a free-model fallback) reads a published call plus the opening of its brief and answers one question: which listed instrument tracks the call's own variable, through how many causal steps, with what stop and exit date. It may answer "none", and often should. Every answer is one model's judgment, so a second model answers the same question and a trade goes forward only when both name the same instrument; a disagreement is recorded as a decline, with both sides quoted, not traded.
- Risk officer (plain code, no AI) sizes or refuses the proposal: one position per instrument, at most 2 per correlation group (the Record tags each call with the world view it rests on, so one view cannot fill the book), a fund whose daily returns have tracked an open position (correlation at or above 0.79 over the last 55 sessions) sized like a one-step chain and a third such fund refused, a single stock (three are listed, each the only vehicle for its commodity) counted as at least one step removed from the call, with a 5–8% stop and at most 2 open, 10% of equity for a direct trade, smaller for chains, stops and exit dates inside their ranges, total exposure under 61.8%, at most 8 positions, no entries after a 3% daily drop, nothing under $13.
- Executor places market orders at the broker's paper endpoint (the client cannot reach any other host), books fills to the cent, rests a stop at the broker for the whole shares of each fund position and the whole of each coin position, and checks the fractional remainder each run.
- Scorekeeper marks every open position, writes the daily ledger row and the S&P benchmark.
- Event desk takes the calls no fund can express (a Fed decision, a policy notice, a monthly average above a line) to a separate $1,000 paper book against Kalshi's and Polymarket's public contract prices (both venues are searched; the translator sees both lists): up to 10% of that book per market, staked by edge (the translator states its own probability that the side pays; the stake is 38% of the Kelly stake that probability implies against the price, never below 2%, and nothing when the probability does not clear the price), at most 8 markets, no side priced outside 5–95¢, settled when the exchange finalizes. No account, no key, no order.
Why the sizes step down with each causal hop
A call about diesel margins traded through a refiner fund is one inference. A call about a shipping chokepoint traded through tanker freight adds a second: the chokepoint must bind, and freight must respond. Each step is a place to be right about the world and still wrong about the price. So a direct trade gets full size, a one-step chain 61.8% of it, a two-step chain 38.2%, and three steps are refused. The hop count is read from the chain the translator writes, never from its label, and can only round up: a mislabeled chain makes a trade smaller, never larger.
Stops, holds and the exposure cap
Stops sit 3–8% under entry for funds (tight for direct calls, wider for chains) and 8–13% for coins, which swing harder. Once a position is up one stop-width, its stop lifts to entry and then trails that width under the highest price seen, so a winner cannot round-trip into a loss; the stop never moves down, and the resting order at the broker is replaced each time it lifts. New fund entries wait 34 minutes after the opening bell, where spreads are widest; a proposal held at the bell is taken at the next window. Holds run 5–89 days and end on the call's own window, not its next data point. The 61.8% cap keeps more than a third of the book in cash at all times, so a bad week cannot become a ruinous one, and the daily-loss halt stops the desk from adding risk into a drawdown. Coins are capped at 13% of equity and 5% per trade. Everything is long-only and unleveraged; a bearish view is expressed through an inverse fund from the same list.
Two kinds of correlation
The Record tags each call with the world view it rests on, and the desk holds at most 2 positions per view. Price history catches what the tags miss: before a fund is bought, its daily returns over the last 55 sessions are compared with every open position's, and a correlation at or above 0.79(more than three fifths of the day-to-day variance shared) counts the two as one trade in two wrappers. The second seat is sized to 61.8%; a third is refused. A fund that moves against the book is never penalized: a hedge is not a bet repeated. The desk's first four positions (refiners, inverse Treasuries, gold, copper) sat in four different Record groups and expressed one macro view; this rule exists because the tags alone did not catch that.
Provenance
Each proposal carries a SHA-256 fingerprint of its public fields (call, instrument, hops, exit date, stop, thesis, decision time), fixed when the decision is made and published with the trade, so anyone can recompute it from the scorecard API and confirm nothing changed after the result was known. The plan is to anchor these fingerprints with the Record's on-chain pipeline, so a decision can be proven to predate its outcome without trusting this site.
What comes after paper
About 20 closed paper trades or 6 weeks, with the desk ahead of the S&P 500 over the same period, before the same agents are allowed near a real account, and then only with a hard capital limit, a kill switch, a daily loss stop, human approval above a threshold, and keys that cannot move money.
What the record will answer
- Whether direct calls beat chained ones, by win rate and by expectancy.
- Which markets the desk reads well and which it should stop trading.
- Whether the translator declines the right calls: every "no trade" verdict is published with its reason, so a reader can disagree on the record.
- How much of a result is the call and how much is the instrument: ETF tracking error, spot-versus-futures basis and roll costs all live between a correct call and a losing trade.
- Whether event contracts score the desk's judgment better than funds do for policy and threshold calls.
Known limits
- Paper fills are market orders at the quote and event contracts are bought at the posted ask with no fees; real fills would carry spread, slippage and fees, so a live desk should expect a slightly worse number.
- Fund stops rest at the broker for whole shares only; the fractional remainder is checked at run time, so a gap through the stop can cost more than the stop implies.
- Scheduling is best-effort: runs land inside their hour, and the run history shows every gap.
- Two models agreeing is not two independent opinions: both read the same brief and can share a blind spot. The record of disputed calls shows how often they differ.
- A batch of calls from one week shares a world view; early results are correlated, so the first weeks say less than their count suggests. The per-group cap and the measured-correlation rule limit the damage, not the correlation itself, and a correlation measured on the last 55 sessions can change tomorrow.
Reading: the calls ledger, the briefs.
Fingerprinted before the outcome
When a trade is proposed, the desk fixes a SHA-256 fingerprint of the decision: the call, the instrument, the chain length, the exit date, the stop, the thesis and the time. The fingerprint is published with the trade, and the scorecard API (/api/paper-desk) carries every field it was made from, so anyone can recompute it and confirm nothing changed after the result was known.
Paper trading on real market prices. Research and decision support, not investment advice; past paper results do not predict future results.
Read the briefs behind every trade in the Intel Suite →