TENSORSWING · iOS
Seven models. One question, asked every Sunday.
Two in-house forecasters and five open foundation models each call the five-day move of every name in a fixed universe of 119. All 833 calls are hashed and publicly timestamped before Monday's open, graded against Friday's close, and kept forever in a public record you can check yourself.
AI OUTPUT · INFORMATIONAL ONLY · NOT ADVICE · NO TRADES ARE PLACED IN THIS APP
AI OUTPUT · INFORMATIONAL ONLY · NOT ADVICE · NO TRADES
Stocks
AI PREDICTIONS · W38 · 833 CALLS · SEALS FRI
WATCHINGNOWBY FRIAGREEMENT
NVDANVIDIA178.90183.20
XLEEnergy Select Sector91.3689.80
SPYSPDR S&P 500648.12650.40
TOP STOCKS THIS WEEKSORT · TOP PICKS
1 · AVGOBroadcom352.10361.90
2 · TLTiShares 20+ Yr Treasury88.4089.70
3 · JPMJPMorgan Chase301.55296.10
SEE ALL 119 CALLS ›
Agreement shows how far the models agree on the direction, weighted by how right each has been on this stock lately. It is not a probability that the price gets there. Not advice.
WEEKS GRADED
2
W36 · W37 · W38 SEALED
CALLS GRADED
1,640
7 MODELS × 119 NAMES × 2 WEEKS
BEST MODEL SO FAR
55.9%
PATCHTST · 128 OF 229
UNIVERSE
119
100 STOCKS · 19 ETFS
HOW A WEEK WORKS
The hash goes out on Sunday. The label comes out on Friday.
Every model calls every name. Nothing is edited or withdrawn in between, and a week that misses its own deadline is written into the incident log rather than quietly fixed.
01
SUN 18:00 ET · SEAL
Seven models call all 119 instruments. Each call is hashed with a salt, and one Merkle root fixes the set of 833. The root is stamped on Bitcoin and by two RFC 3161 authorities.
hashes/<model>/<week>/<ticker>.sha256
02
MON–THU · SEALED
The app tracks the live move against each call. Mid-week a call reads ahead or behind, never win or loss, because nothing is resolved yet.
<week>.manifest.ots
03
FRI 18:30 ET · LABEL
Every call is graded on Friday's close and published as a label: win, loss, or scratch when the week rounds to exactly 0.00%. A model that declined to call is recorded as an abstain.
labels/<week>.json
04
SAT 00:00 ET · AUDIT · KEPT
The drand beacon's round at midnight picks at least 20 calls to open with their salts, plus every scratch and abstain. The full archive is time-locked to open by itself 104 weeks after seal.
audits/<week>/<model>/<ticker>.json
INSIDE THE APP
Four tabs. Nothing that nags.
Stocks, Analytics, News & Events, Models. No badges, no counts, no marketing. Two notifications a week, the publish and the resolution, plus any alerts you switch on yourself.
This Week
SEP 14–18, 2026 · 833 CALLS SEALED BEFORE MONDAY'S OPEN
Search symbol or company
119 STOCKSSORT · TOP PICKS
#NOWBY FRIAGREEMENT
1 · AVGOBroadcom352.10+0.8%361.90+2.8%
2 · TLTiShares 20+ Yr Treasury88.40−0.2%89.70+1.5%
3 · JPMJPMorgan Chase301.55+0.3%296.10−1.8%
4 · XLVHealth Care Select142.70+0.1%144.30+1.1%
5 · AMDAdvanced Micro Devices164.05−1.1%159.80−2.6%
6 · MSFTMicrosoft512.30+0.4%515.10+0.5%
7 · GLDSPDR Gold Shares318.90+0.6%320.10+0.4%
Tap a stock for its calls, levels and record.
THIS WEEK
The whole field in one table. NOW is the market, BY FRI is the consensus Friday close, and the five capsules show how far the models agree, weighted by each one's record on that stock. No number is shown that could be read as a promised probability.
NVDA
W38 · PREMIUM · CHOOSE YOUR CONSENSUS
TA CONSENSUS · BEST TWO▲ RISE · 183.20
The default is this stock's two best models by record. Tick any row to recompute.
PATCHTST▲ RISE184.107 / 11
GRANITE TTM▲ RISE182.306 / 10
CHRONOS▲ RISE181.005 / 10
N-HITS▼ FALL176.405 / 11
TOTO▲ RISE180.205 / 10
TIMESFM▼ FALL177.904 / 10
CHRONOS-2▲ RISE179.604 / 10
HISTORY · LAST 8 GRADED WEEKSON NVDA
PATCHTST
GRANITE TTM
The record is never gated. Every model's past is open to every tier.
STOCK PAGE
Every model's call on the stock, with its Friday target and its graded record on that name. The consensus defaults to the stock's two best models, and you can tick any set you like. Free reads the house models; Premium reads all seven.
News & Events
UNIVERSE · 41 HEADLINES · UPDATED 12 MIN AGO
58%
Crowd bullish · all stocks · this week
WED 17FOMC rate decisionMACRO
THU 25COST earnings, after closeEARNINGS
Chipmakers slide as export-rule draft circulates (Chinese)
FULL ANALYSIS › · SOURCE ↗
NVDADraft rules would cap the H20 successor; the read is mildly bearish into Friday.
AVGOLittle direct exposure; treated as noise for this week's call.
AI-written reads of public headlines. Not advice.
NEWS & EVENTS
Public headlines digested into one AI read per affected stock, with the crowd's mood and the next two weeks of scheduled events alongside. Translated headlines keep their original text a tap away. Premium unlocks the per-stock reads.
Screens illustrate the current build (app 1.2). Prices, capsules and reads are examples, not live data.
THE RECORD SO FAR
Two weeks graded. This is not yet evidence of anything.
It is shown anyway, in full. Across 1,640 graded calls the seven models sit between 47% and 56% on direction, and none of them beats the naive forecast on size by a margin worth naming. The verified record begins at 2026-W36; W35 was lost to a pipeline migration and is logged as incident 001, not deleted.
A coin-flipping model posts 55.9% over 229 calls about one time in 25. Read the table as a baseline being established, not as a result.
GRADED CALLS · 2026-W36 – W37
1,640 CALLS · 7 SCRATCH · 19 ABSTAIN
MODELW36W37HIT RATE÷ NAIVE
PATCHTSTTENSORSWING · HOUSE
57.0%
54.8%
55.9%
0.96
GRANITE TTM r2IBM · PRETRAINED
58.1%
50.0%
54.0%
0.97
CHRONOSAMAZON · PRETRAINED
57.6%
45.4%
51.5%
1.01
N-HITSTENSORSWING · HOUSE
61.9%
40.9%
51.3%
1.06
TOTO 2.0DATADOG · PRETRAINED
48.3%
53.8%
51.1%
0.93
TIMESFM 2.5GOOGLE · PRETRAINED
46.6%
49.6%
48.1%
1.05
CHRONOS-2AMAZON · PRETRAINED
46.6%
47.1%
46.8%
0.93
MAJORITY4-OF-7 VOTE · TIES SIT OUT
62.7%
49.6%
56.1%
—
HIT RATE = WINS ÷ (WINS + LOSSES), DIRECTION ONLY, BEFORE COSTS. ÷ NAIVE = MEDIAN ABSOLUTE ERROR OVER THE ERROR OF FORECASTING NO MOVE; BELOW 1.00 BEATS GUESSING ZERO. RECOUNTED FROM labels/2026-W36.json AND labels/2026-W37.json IN THE PUBLIC RECORD.
VERIFY IT YOURSELF
No account, no app, and no trust in the operator.
Every week's Merkle root is anchored on Bitcoin by OpenTimestamps and stamped by Sectigo and DigiCert. Friday's labels are stamped again before the beacon fires, so they could not have been written after the audit sample was known.
✓833 calls existed before Monday's open, and the root says which 833
✓No call can be added, dropped or reworded without breaking the root
✕It does not reveal the sealed calls before Friday
✕It says nothing about whether the calls are any good
Verifier status, 16 September 2026. Merkle roots, label coverage, Bitcoin and RFC 3161 stamps, and the beacon audits pass for every graded week: W36 opened 57 and W37 opened 50 sampled, scratch and abstain leaves against their commitments and labels. One check fails today and is shown rather than hidden: W37 and W38 were sealed after Monday's open, and the protocol asks for a late-seal incident entry that the operator has not filed. The calls themselves were stamped when sealed and are unchanged.
$ git clone https://github.com/YermekIbrayev/tensorswing-record
$ cd tensorswing-record
$ python3 scripts/verify.py 2026-W37
[ PASS ] merkle root · 833 leaves
[ PASS ] label coverage · 0..832
[ PASS ] OTS stamp · Bitcoin height 965996
[ PASS ] RFC 3161 · sectigo · digicert
[ PASS ] labels before beacon target
[ PASS ] audit set · 50 leaves opened · drand
[ FAIL ] seal time · after Mon 09:30 ET
$ ots verify 2026-W37.manifest.ots \
-f manifests/2026-W37.manifest.json
// merkle root 412e276c…a4d207ff
// sealed 2026-09-08 00:27 UTC · protocol 1.0.1
// sealed 2026-09-08 00:27 UTC · protocol 1.0.1
THE UNIVERSE RULE
The models never choose what they are allowed to call.
A perfect record on hand-picked names is worthless. Membership is computed from public data by a rule anyone can run, committed to the public repository before it takes effect, and recomputed on the first of January, April, July and October.
V1 RULES
EFFECTIVE 11 AUG 2026
NEXT REFRESH 01 OCT 2026
READ THE FULL RULE →
COMMON STOCKS100
01US-listed, ranked by median daily dollar volume over the trailing 90 calendar days
02Share price above $5
03Market capitalisation above $10B
ETFS — FIXED LIST19
INDEXSPY QQQ IWM DIA
SECTORXLK XLF XLE XLV XLI XLY XLP XLU XLB XLRE XLC
MACROGLD TLT
INTLEEM FXI
EXCLUDED FROM V1
Leveraged and inverse ETFsA TQQQ call is a QQQ call with borrowed conviction.
Crypto and crypto ETFsRound-the-clock sessions break Friday-close scoring.
Non-US listings and FXWrong sessions, wrong benchmark.
WHAT IT COSTS
Free to follow the record. $39 a month for every model.
Free reads the house models' live calls on every stock, keeps four names on a watchlist with one alert, and sees the last eight graded weeks. Premium reads all seven models, twenty names and five alerts, the full archive, the proof pages and the AI news reads. The graded record itself is never gated. No trial, cancel any time.
DOWNLOAD ON THE
App Store
FREE TIER · PREMIUM $39/MO · CANCEL ANYTIME
Premium
$39 / MONTH
✓Every model's call on every stock, the moment it publishes
✓Choose which models form your consensus, per stock
✓20 watched names, 5 alerts, predicted-level and price-move pushes
✓Full archive, proof pages and per-stock AI news reads
✕No brokerage connection, ever
QUESTIONS
Is this investment advice?
No. TensorSwing publishes informational research commentary produced by statistical models with no human review of individual calls. It is not personalised to you, and it is not a recommendation to buy, sell or hold anything.
Does it connect to my brokerage?
Never. The app does not place trades, connect to a broker, or link to your financial accounts, and it is not personalised to you. There is no brokerage login and nothing to link.
What counts as a win?
A call is a win where Friday's close moves in the called direction, and a loss where it closes against it. A week whose move rounds to exactly 0.00% is a scratch: reported in full, and excluded from the hit rate rather than counted as either. Grading is on direction alone, before trading costs.
Which models make the calls?
Seven. N-HITS and PATCHTST are trained in-house every Sunday. CHRONOS, CHRONOS-2, TIMESFM, TOTO and GRANITE TTM are open foundation models from Amazon, Google, Datadog and IBM, run zero-shot. Every model calls every name in the universe; a model that declines a name is recorded as an abstain and opened in the audit.
How many notifications will I get?
Two a week for the cycle itself: Sunday evening when it publishes and Friday evening when it resolves. Optional alerts (watched names, followed models, price moves) add more only if you turn them on, and every one has its own switch. No re-engagement nudges; marketing only if you opt in.
Can a bad week be removed later?
No. Every week ever sealed stays in the public record, in full, and nothing is edited after Friday. The Merkle root covers the whole set, so a call cannot be quietly dropped either. A missed obligation is written into a hash-chained incident log, not fixed in place.
The next cycle seals Sunday evening.
Seven models, one question, and a record that cannot be edited afterwards.
DOWNLOAD ON THE App Store