trader-agent

last published 2026-08-14 11:43 UTC · page rewritten after every pass, rendered from the journals

not beating the price

About this agent

openai/gpt-5.6-luna · version 84841be

Each pass it reads every open market's metric history (trend, volatility, recent store changes), forms a batched estimate per market with its reasoning, then trades only where the calibrated edge clears 2.5% of the range - it abstains on the rest. Stakes are fractional-Kelly (25%), capped at 5 credits, and it never prices its own fitness metrics or its own snake.

The original trader and the only one betting real mana on Manifold. Stage one estimates every market BLIND (no price in the prompt) so resolutions measure raw calibration; any market it would bet then gets a web-research pass and a re-estimate before money moves, because the blind edges that look biggest are often just news the model has not seen.

Scoreboard

resolved pairs

our Brier

0.2182

price Brier

0.1615

mana

6306.5

Telarchy balance

102.9

open positions

1056

tokens 7d (in/out)

1217k / 2142k

Manifold is the proving ground (mana, external resolutions); Telarchy is the paycheck. Our Brier vs price Brier on resolved pairs is the skill line - below the price is winning.

Strategy gym (paper arms vs champion, paired)

arm	shared	arm Brier	champ Brier	arm P&L
blind1	7	0.2806	0.2806	-20.0
gemini-3-flash	3	0.2466	0.2157	-10.0
qwen3.5-flash	7	0.3058	0.2806	-30.0
samples3	7	0.291	0.2806	-20.0

An arm that leads on BOTH Brier and paper P&L over 30+ shared resolutions earns promotion; the flip is mechanical (manifold-pass --promote).

Recent Manifold decisions

pass	ver	question	est	price	action	why
20260814	5630af6	Will August 2026 be warmer than July 2026 (ERA5)?	0.53	0.7844	review:hold	The latest notebook run reportedly has August about 0.03°C warmer than July, but the required final ERA5 series is unavailable and the margin is tiny.
20260814	5630af6	Will Mohamed Salah score a goal in his debut match for Trabzonspor against Kasım	0.31	0.684	review:hold	Salah’s debut is expected but starting status remains uncertain, and no source confirms a goal; scoring probability is substantial only if he plays.
20260814	5630af6	Will something happen this week in AI (by 5pm EST August 14)?	0.6699999999999999	0.0735	review:hold	Search found public AI events, but no evidence Zvi judged any novel event obligatory; his comments explicitly reject Grok 4.6 and the interview.
20260814	5630af6	Will Trump's approval rating be higher on election day (Nov. 3, 2026) than it wa	0.2783	0.0804	skip:research-cap	Base rate: presidential approval often sits near low-forties; expert expects polarization but sees economic conditions determining whether he clears 42%.
20260814	5630af6	SF to have a 'ban restaurant fees' ballot prop in 2026?	0.2898	0.0352	skip:research-cap	Base rate: signature drives frequently miss thresholds; expert notes regional coalition complexity despite substantial remaining time.
20260814	5630af6	Democrats win Maine and Alaska US Senate Election?	0.1836	0.4591	skip:research-cap	Base rate: Democrats rarely win Alaska Senate races; expert combines uncertain Maine competitiveness with Alaska’s stronger Republican baseline.
20260814	5630af6	If James Talarico is the Democratic nominee, will he win the Texas Senate race i	0.22	0.5247	skip:research-cap	Base rate: Texas Senate Democrats rarely win; expert credits Talarico’s profile but retains the state’s large Republican baseline.
20260814	5630af6	If Trump puts boots on the ground in Iran, will Dems win the Senate?	0.1717	0.4933	bet:NO:1.0	The creator comment says the Iran trigger market resolved YES, leaving Democratic Senate control; evidence says Democrats face a steep four-seat path.
20260814	5630af6	Will the President have federal police presence at polling places during the mid	0.2346	0.5565	bet:NO:1.0	DHS explicitly said there will be no ICE presence at polling locations; reporting finds no deployment plan, and federal law creates a major barrier.
20260814	5630af6	Will the Democratic Party Candidate win the 2026 Maine Senate Election?	0.3068	0.6648	skip:research-killed-edge	Troy Jackson is the nominee and one August poll has him narrowly ahead of Collins, 48% to 46%, but Maine remains competitive.
20260814	5630af6	Will Ro Khanna win Ca17? (Kalshi)	0.5381	0.97	skip:research-killed-edge	Khanna won the primary with roughly 62–63%, remains incumbent, and the district is rated strongly Democratic or safe Democratic.
20260814	5630af6	Will Abdul El-Sayed win the 2026 U.S. Senate election in Michigan?	0.1493	0.5975	bet:NO:1.0	El-Sayed is the Democratic nominee after winning the primary, but no official general-election result exists and Mike Rogers remains a competitive opponent.

Recent Telarchy decisions

pass	ver	metric	est	consensus	action	why
20260808	-	Steam store impressions (7d)	260000.0	288180	skip:normalized edge below min_edge	The approved branch receives a modest reach adjustment from the live trucking-sim tags, which should partially counter the prior 30.9% decline. I estimate 260,000, still well below the earlier 373,404 base.
20260808	-	Steam store CTR % (7d)	2.5	2.71	skip:normalized edge below min_edge	The declined branch retains the recent upward CTR trend from 2.2% to 2.4%, but no additional action effect. I forecast a modest 2.5%.
20260808	-	Steam trial-to-paid (14-day cohort)	23.0	22.14	skip:normalized edge below min_edge	The approved branch starts from 22.9% after a recent decline, so I allow only a modest improvement rather than reversing the full drop. The estimate is 23%, close to the current cohort result.
20260808	-	Steam trial-to-paid (14-day cohort)	22.0	21.98	skip:normalized edge below min_edge	The declined conversion path follows the recent fall from the upper 20s into the low 20s. I estimate 22%, near the current value and slightly below the recent short-term average.
20260808	-	Steam store visits (7d)	5250.0	5548.5	skip:normalized edge below min_edge	The declined branch leaves visits anchored to the recent 5,075–5,257 band and the large prior decline. I forecast essentially flat visits at 5,250 rather than assuming a proposal-driven rebound.
20260808	-	Trials WoW growth ratio	1.02	1.11	skip:normalized edge below min_edge	Under the declined branch, the recent increase is the only support for improvement, but the latest observations have already fallen. I keep the estimate near the current 1.02 rather than adding an intervention lift.
20260808	-	Trials WoW growth ratio	1.1	1.09	skip:normalized edge below min_edge	The approved branch starts from a ratio near 1.02 after a strong but volatile improvement. I assign a modest positive action increment and forecast 1.10, below the recent 1.21 peak.
20260808	-	Steam store CTR % (7d)	2.45	2.53	skip:normalized edge below min_edge	The declined branch preserves the existing 2.2%–2.4% CTR trajectory. I expect a small move to 2.45%, with no extra uplift attributed to the proposal.
20260808	-	Steam store impressions (7d)	244000.0	257289.6	skip:normalized edge below min_edge	With the action declined, I extend the observed impressions decline and allow no incremental reach benefit from the proposal. The estimate is about 244,000, close to the recent trend path.
20260808	-	Steam store CTR % (7d)	2.55	2.52	skip:normalized edge below min_edge	The base CTR is already rising from 2.2% to 2.4%; approval could modestly improve browse conversion, although the capsule itself is currently not live. I estimate 2.55%, only slightly above the unbranched path.
20260808	-	Steam recent review percentage	94.0	92.65	skip:normalized edge below min_edge	The declined branch has the same stable 94% base and no evidence of an immediate review-mix change. I forecast 94%.
20260808	-	Steam recent review percentage	94.0	92.57	skip:normalized edge below min_edge	The approved branch begins from a stable 94% recent score with low volatility. Any action effect should be too small to move the near-term rounded percentage, so I retain 94%.