MALI

Track record

Every call MALI has made, scored against SPY over the same window. Calls are hashed and chained when they are made, so a later edit or deletion shows.

What the record shows

Too few scored calls to say anything yet: 1 directional calls have been scored against SPY, and at least 30 are needed before a hit rate is stated.

Not investment advice. Updated Wed, 18 Nov 2026 22:07:31 GMT.

Cumulative
Calls
1377
7 counted · 1 scored · 0 open · 6 not scoreable
Hit rate vs SPY
—
n = 1 · 95% interval 20.6–100.0%
Brier score vs SPY (coin = 0.250)
—
n = 1 · 95% interval —
Mean excess return per call (signed by direction)
—
n = 1 · 95% interval —
Hit rate on the raw return (secondary)
—
n = 1 · 95% interval 20.6–100.0% · counts a market rise as a bullish hit
Named stock beat SPY
—
n = 1 · what always saying bullish would have hit

Fewer than 30 scored directional calls: the headline figures are held back until there are 30, and no conclusion is drawn. The counts and 95% intervals above show what has been scored so far.

By engine
EngineCounted callsScoredHit rate vs SPY95% intervalBrier vs SPYRaw hit rate
Debate engine00————
Forecast engine71—20.6–100.0%——
Calibration

Stated probability against how often calls were right

Each dot is a band of stated probability; its height is the share of those calls that beat SPY in the stated direction, with its 95% interval. Dots on the diagonal mean the probabilities meant what they said.

0%25%50%75%100%50%60%70%80%90%100%Stated probabilityRight vs SPY
As a table
BandnStatedRight vs SPY95% interval
50%–60%150.3%100.0%20.6–100.0%
60%–70%0———
70%–80%0———
80%–90%0———
90%–100%0———
Every call
Newest 500 of 1377; the CSV has every one.Download CSV
Issued (UTC)EngineTickerCallProbabilityHorizonStockSPYExcessOutcome vs SPYRawHash
2026-10-16 00:31ForecastTSLABullish59%21 sessions———Not scoreable—2167bce4eb… *
2026-10-16 00:31ForecastTSLABearish58%21 sessions———Not scoreable—e27fb6d2b2… *
2026-10-16 00:31ForecastTSLABearish53%21 sessions———Not scoreable—bbee47b72e… *
2026-10-16 00:31ForecastBTC-USDBearish54%21 sessions−15.0%−4.1%−10.9%Right✓2ef23f86a2… *
2026-10-16 00:31ForecastBTC-USDBearish51%21 sessions−31.1%−4.1%−26.9%Right✓75d1322092… *
2026-10-16 00:31ForecastBTC-USDBullish54%21 sessions−45.4%−4.1%−41.2%Wrong✗d663e6a258… *
2026-10-16 00:31ForecastBTC-USDBearish51%21 sessions−31.6%−4.1%−27.5%Right✓6683ff2bfa… *
2026-10-16 00:31ForecastBTC-USDBullish53%21 sessions−3.8%−4.1%+0.3%Right✗aa9b29cfd3… *
2026-10-16 00:31ForecastBTC-USDBearish50%21 sessions−5.7%−4.1%−1.6%Right✓c77f5d20ef… *
2026-10-16 00:31ForecastBTC-USDBullish56%21 sessions+50.5%−4.1%+54.6%Right✓ec558efdf1… *
2026-10-16 00:31ForecastBTC-USDBearish53%21 sessions+147.0%−4.1%+151.1%Wrong✗75d2f17de2… *
2026-10-16 00:31ForecastBTC-USDBearish51%21 sessions+113.9%−4.1%+118.1%Wrong✗2100c6f989… *
2026-10-16 00:31ForecastBTC-USDBullish51%21 sessions+290.3%−4.1%+294.4%Right✓7c77d1f943… *
2026-10-16 00:31ForecastGLDBearish55%21 sessions———Not scoreable—60ea505cf4… *
2026-10-16 00:31ForecastGLDBearish54%21 sessions———Not scoreable—105eceecb2… *
2026-10-16 00:31ForecastGLDBullish54%21 sessions———Not scoreable—fa5c6b59ab… *
2026-10-16 00:31ForecastGLDBullish54%21 sessions———Not scoreable—87cd3c5193… *
2026-10-16 00:31ForecastGLDBearish50%21 sessions———Not scoreable—3fa72cbacd… *
2026-10-16 00:31ForecastGLDBullish53%21 sessions———Not scoreable—8a28ba5a4d… *
2026-10-16 00:31ForecastGLDBullish53%21 sessions———Not scoreable—a532b7a85b… *
2026-10-16 00:31ForecastGLDBearish55%21 sessions———Not scoreable—5ad2abcecb… *
2026-10-16 00:31ForecastGLDBearish55%21 sessions———Not scoreable—720f3cd3e5… *
2026-10-16 00:31ForecastGLDBearish50%21 sessions———Not scoreable—4d7c32b27b… *
2026-10-16 00:31ForecastTLTBullish52%21 sessions———Not scoreable—ca36e26abc… *
2026-10-16 00:31ForecastTLTBearish52%21 sessions———Not scoreable—6d5bda7610… *
2026-10-16 00:31ForecastTLTBearish57%21 sessions———Not scoreable—549f794924… *
2026-10-16 00:31ForecastTLTBullish56%21 sessions———Not scoreable—6b6429833f… *
2026-10-16 00:31ForecastTLTBearish52%21 sessions———Not scoreable—5b978356a5… *
2026-10-16 00:31ForecastTLTBearish53%21 sessions———Not scoreable—35a34a7af6… *
2026-10-16 00:31ForecastTLTBullish52%21 sessions———Not scoreable—b0f2f21599… *
2026-10-16 00:31ForecastTLTBearish54%21 sessions———Not scoreable—66fc121081… *
2026-10-16 00:31ForecastTLTBearish56%21 sessions———Not scoreable—5b1d483558… *
2026-10-16 00:31ForecastTLTBullish53%21 sessions———Not scoreable—90dad51300… *
2026-10-16 00:31ForecastSPYBearish51%21 sessions———Not scoreable—bafc502232… *
2026-10-16 00:31ForecastSPYBearish52%21 sessions———Not scoreable—8586d2ebb7… *
2026-10-16 00:31ForecastSPYBearish51%21 sessions———Not scoreable—f0278c28e0… *
2026-10-16 00:31ForecastSPYBullish55%21 sessions———Not scoreable—539f2dce2b… *
2026-10-16 00:31ForecastSPYBearish50%21 sessions———Not scoreable—584f3bc1ea… *
2026-10-16 00:31ForecastSPYBullish52%21 sessions———Not scoreable—6280754657… *
2026-10-16 00:31ForecastSPYBullish54%21 sessions———Not scoreable—8c1923b6e7… *
2026-10-16 00:31ForecastSPYBullish50%21 sessions———Not scoreable—a687a96374… *
2026-10-16 00:31ForecastSPYBearish56%21 sessions———Not scoreable—879823168a… *
2026-10-16 00:31ForecastSPYBullish53%21 sessions———Not scoreable—ddc9516d39… *
2026-10-16 00:31ForecastAAPLBullish50%21 sessions———Not scoreable—224a6cdf96… *
2026-10-16 00:31ForecastAAPLBearish51%21 sessions———Not scoreable—eb72243247… *
2026-10-16 00:31ForecastAAPLBullish51%21 sessions———Not scoreable—be9f5c04a9… *
2026-10-16 00:31ForecastAAPLBullish50%21 sessions———Not scoreable—97dc2926b4… *
2026-10-16 00:31ForecastAAPLBullish57%21 sessions———Not scoreable—3af1d81c1b… *
2026-10-16 00:31ForecastAAPLBearish52%21 sessions———Not scoreable—dddd2ebccb… *
2026-10-16 00:31ForecastAAPLBullish52%21 sessions———Not scoreable—3c8d61d82d… *

Faded rows are repeat calls on the same name, horizon and day: listed, but counted once in the statistics. * chained more than two hours after the call was made (calls made before the chain existed): the chain attests to them only from when they were added.

Tamper evidence
The chain verifies: 2754 entries, each hash matching its content and the one before it.
Chain head: #2754 d6c1dadf3f22b3ac0f3c43006265ea4ca840e394a26c68516f5f5d7c6cacd090
Rule: hash = sha256 of seq|prev_hash|payload; the first entry's prev_hash is 64 zeros. Keep a copy of the head: if an earlier entry is ever changed, the head you kept no longer matches.
Check it yourself
import hashlib, json, urllib.request
entries = json.load(urllib.request.urlopen("https://mali-trw.com/api/python/track-record/chain?limit=5000"))["entries"]
prev = "0" * 64
for e in entries:
    assert e["prev_hash"] == prev, e["seq"]
    h = hashlib.sha256(f"{e['seq']}|{prev}|{e['payload']}".encode()).hexdigest()
    assert h == e["hash"], e["seq"]
    prev = e["hash"]
print("chain verifies; head", entries[-1]["seq"] if entries else 0, prev)
Methodology

What counts as a call. The debate engine: every thesis its agents debate becomes a call with a direction, a stated probability of being right and a horizon in trading sessions. The forecast engine: every daily forecast whose last point carries a probability of the price being higher; its call is the side that probability leans to, with that probability. A forecast at exactly 50% takes no side and is not a call.

Scoring against SPY. Each call's return is measured from the close it was made on to the first close on or after its due date, and SPY's over the same sessions. A bullish call is right when the stock beat SPY, a bearish one when it lagged SPY, a neutral one when the gap stayed inside half a one-sigma move. A rise every stock shared is not a bullish hit. The raw return, which does count it, is kept as a secondary figure.

Point in time. Only prices that existed when a call was made enter its entry: the debate engine's SPY entry is read at issue time; the forecast engine's is SPY's last completed close when the forecast was issued, so its window can start up to a session apart from the stock's intraday anchor. Exits are closes printed after the due date.

Statistics. Hit rates come with Wilson 95% intervals; Brier scores and mean excess returns with normal-approximation 95% intervals. Below 30 scored directional calls no conclusion is drawn. Repeat calls on one name, horizon and day count once. Calls with overlapping horizons are correlated, so the intervals are, if anything, too narrow. The debate engine's agent weights are recomputed on this same basis.

What is excluded and why. A call on SPY itself has no excess return. A call whose exit close or SPY's is still missing five days after it was due is marked not scoreable, never filled in. Nothing is removed from the list.

Limits. The hash chain shows edits made after a call was chained; someone with database access could rebuild the whole chain, which only a head kept outside it reveals. Calls on names a user asked about are selected by that user. Past results do not predict future ones, and none of this is investment advice.