Alchemy Desk / Research Articles

Alchemy Desk Market Research Note - 2026-05-29

A plain-English research report from Alchemy Desk. It explains what the system tested, what the numbers mean, and why the result stays public, gated, promoted, or in the lab.

Follow the research trail

Get the local feed, save the page, or share a research note without turning it into a trade call.

Alchemy Desk Market Research Note - 2026-05-29

Market Desk Read

Regime: mixed. Breadth: 37.9%. One-day SPY proxy: 0%.

Feed integrity: 202 live instruments, 566 fallback or simulated instruments. Live scoring policy: live batches exclude fallback/simulated symbols unless explicitly recorded as simulation.

Live instruments202
Fallback/simulated566
Resolved guesses84
Flat outcomes237
Invalid scores0
Calibration error0.407

Research Candidates

Candidate mode: live candidates.

Candidate comparison
QARP +2.77% EMCB +2.76% OPER +2.56% TPRY +2.81% EMC +2.70% EGLE +2.28%
Learning path
QARP: +75.50%EMCB: +74.50%OPER: +72.90%TPRY: +72.70%EMC: +71.60%EGLE: +73.20% QARP EGLE +75.50% high +71.60% low

Divergence Watch

Outliers: none.

Inverse/down-day candidates: none.

What I Learned Today

The desk learned to distrust breadth when fallback coverage is larger than live coverage. Those rows can still teach interface and simulation behavior, but they do not deserve live-score authority.

84 resolved picks created evidence for calibration. Flat outcomes stayed separate, which protects the model from pretending zero movement was a win or a loss.

Today's guess set also taught the desk where attention clustered: QARP, EMCB, OPER, TPRY. Tomorrow's candidate ranking should remember whether those names resolved, stayed flat, or proved invalid.

What This Means In Plain English

A prediction is not wisdom until it gets scored. Alchemy Desk is trying to become smarter by keeping the boring parts: sources, guesses, missed chances, flats, invalid scores, and calibration errors.

The public learning trail should therefore improve even on days when the model is wrong, because wrong guesses are useful when they are cleanly labeled.

Learning State

Resolved picks: 84. Flat picks: 237. Invalid or excluded scores: 0.

Mean calibration error: 0.407. Alert: True.

Research Posture

This note is generated from the local Alchemy Desk research spine. It documents candidates, data integrity, calibration, and divergence. It is not financial advice or an instruction to trade.