How do you review your first strategy in Thesis Lab?

Before you start

Thesis Lab is currently a free closed beta with finite included AI usage. New accounts need a single-use invitation link or code. Existing members can continue signing in. The content hub is public reading material; the application currently opens on its existing beta hostname, linked below.

This is a hypothetical walkthrough, not an observed customer result. For the first exercise, use an invented company and a non-sensitive principle. Your concrete output is one saved, approved version that matches your intended wording, with unresolved issues still visible. An invitation gives access to the beta; it does not establish that the product is suitable for a particular financial decision.

Sign in and verify unlock

Use the offered Google or email/password sign-in flow. Complete email verification where required. Authentication identifies the account; a separate vault method opens the encrypted research. Follow Secure with a passkey and confirm the compatible browser and credential provider can actually unlock the vault.

A recovery key is an optional emergency method. If you already have a recovery-key vault, keep its original working method while connecting a new passkey. Support cannot replace lost decryption material. Do not put a recovery key, password, or provider credential into a principles conversation or feedback message.

Describe one principle and inspect Unknowns

Try an explicit hypothetical statement: I want the agent to explain which reporting period supports each operational claim and pause when the required evidence is absent. Inspect the system’s interpretation. Does pause mean ask for a source, mark unknown, or stop the simulation? Does reporting period mean the period measured or the time of publication?

Resolve the ambiguity that matters for your intended test. Approve/save the version only after its wording matches your intended rule. Keep unresolved issues visible. A saved version is a record of an interpretation; it is not evidence that every clause has an executable test.

Optional next step: inspect a supported dated case

Choose an available company/date and review the evidence cutoff before the decision. Read the source passage and distinguish a reported fact from the agent’s explanation. Ask which specific rule was exercised and which parts of the strategy remain outside the simulation.

The current price dataset covers 3 January 2022 through 22 October 2026 (2022-01-03–2026-10-22) for 20 named surviving equities plus SPY. It is retrospectively selected, split-normalized closing-price history excluding dividends. SEC evidence consists of dated Microsoft, NVIDIA, and Apple snapshots. These resources do not amount to comprehensive live research or an all-market point-in-time dataset. A missing company, date, or evidence item should remain a limitation rather than be filled with a fabricated observation.

Finish with a short review record

Finish the first task by checking that the saved version is the wording you approved. If you also inspect a simulation, record the version, case and cutoff, supporting source, one disagreement, and next question. Keep simulated output separate from real account activity. There is no connected brokerage order execution, and a result is not independently validated investment performance.

Included AI can consume multiple underlying calls during a team or backtest workflow; free beta does not mean unlimited inference. Selected AI context is processed transiently through Thesis Lab and OpenAI, while saved research is encrypted before cloud persistence.

Method, sources, and limits

The steps were checked against the released beta documentation on 26 October 2026. This guide did not create a production account, issue an invitation, make a model call, or execute a trade. The example is a workflow exercise. Recheck the application’s current limits and supported choices when you use it.