Preface¶

In agent hosts like DSH that emphasize “everything is a plugin,” research-oriented plugins can easily stop at “generating a seemingly complete analysis.” For investment research, the harder question is not whether the text is complete enough, but whether the evidence is sufficient, where the inference boundaries lie, and whether conclusions can still be updated by subsequent facts.

Resanity is a Skill for retail investor research, focusing on evidence search, economic exposure chain mapping, and load-bearing claim auditing. It decomposes investment decision-making into “what is observed, what can be inferred, what cannot be inferred, and what impact it has on decisions,” and provides mechanical capabilities around this framework, including DSH plugins, anchor checkups, and optional data credentials.

What This Is¶

The current official code version of Resanity is 0.2.1, maintained by Thhoho, licensed under MIT. The GitHub address is: https://github.com/Thhoho/reSanity.

The typical problems it solves are:

• Is a message an official statement, market hearsay, or further inference?
• Have policy, order, or capacity changes already reached delivery, revenue, gross margin, and cash collection?
• What expectations have already been priced into the market?
• Which fact should update one’s previous judgment?

Resanity is not a stock recommendation tool, not a market data terminal, and does not place orders, set positions, or promise returns on behalf of users. It focuses on the evidence boundaries and judgment updates within the research process.

Core Features¶

Investment Decisions Decomposed into Four Questions¶

Resanity decomposes judgments that would change investment decisions into:

1. What is observed
2. What can be inferred
3. What cannot be inferred
4. What impact it has on decisions

The key point of this decomposition is separating “information that looks relevant” from “evidence sufficient to support a conclusion.”

Investment Research as the Default Target Scenario¶

Investment research is Resanity’s default target scenario, with primary capabilities including:

• Sourcing primary materials;
• Mapping economic exposure chains;
• Annotating evidence and inference boundaries;
• Auditing conclusion strength via load-bearing claims.

When evidence is insufficient, WATCH_ONLY, NOT_EVALUABLE, or “take no action for now, wait for a specific verification” are all valid conclusions.

Non-investment experiments must be explicitly invoked by the user and are not triggered automatically. Medical diagnosis and legal judgment require separate protocols and are beyond the scope of current general experiments.

Cognitive Anchor Lifecycle¶

Resanity uses cognitive anchors to preserve updatable and reviewable judgments. The anchor lifecycle is:

active / refuted / realized / archived

The reminder only reads date triggers from active anchors. In other words, at the system level, the main functions are reminders and read-only checks—it does not automatically rewrite judgments on behalf of the user.

DSH Plugin Capabilities¶

The DSH plugin provides:

• A bundled Skill provider;
• /resanity-check anchor checkup;
• Optional Tushare credential commands.

systemNotifications in the configuration defaults to false, and OS notifications are only invoked when explicitly enabled by the user.

Tushare is not part of the core research protocol, but when credentials and dependencies are available, it is the default highest-priority source for automated daily A-share data collection; collection failures do not fall back to other sources silently.

0.3 In-Development Capabilities¶

The 0.3 in-development capabilities noted in the materials include:

• Configuration-gated anchorTimer / commands / dataTools / webFetch;
• Model data tools;
• webFetch providers;
• Six-stage research discipline;
• Compact announcement tool output;
• Anchor expiration pre-collection;
• Invalidation tracking;
• Delivery compilation gates;
• Lightweight single-arm evaluation suite;
• New commands /resanity-audit and /resanity-init;
• Predefined regression gates.

These capabilities are implemented and have passed testing, but require a DSH restart to take effect and are not included in the 0.2.1 release tree.

Installation and Enablement¶

Install from a local tarball to a specified profile:

dsh plugin --profile headless add /absolute/path/resanity-0.2.1.tgz

Before installation, it is recommended to review the source code and the MIT license. DSH plugins run with the permissions of the current dsh process, so the command source, package version, and local directory path should be confirmed before installation.

Typical Usage¶

Investment research can be asked directly:

Is this company's connection to a hot topic a conceptual mapping, or has it already formed attributable revenue and cash?

As of today, which segment from product validation to revenue and cash has been closed by primary evidence for this company?

Where are the truly scarce links and profit pools in this industry? What expectations have already been priced into the market?

Non-investment scenarios must be explicitly invoked by the user, for example:

Please use Resanity to draw a possibility map for this product plan and audit three load-bearing claims.

Such invocations are experimental generalizations and should not be understood as validated general-purpose capabilities.

Applicable Scenarios and Cautions¶

Suitable for:

• Developers who need to break down the investment research process into evidence, inference, and decision impact;
• Users who need to mount an investment research Skill within DSH;
• Those who want to use anchor reminders, read-only checkups, and mechanical checks to support their judgment process.

Not suitable for:

• Requests for direct stock recommendations;
• Requests for automatic order placement or position setting;
• Requests for guaranteed returns or Alpha judgment;
• Treating a single piece of output as a validated market conclusion.

The current version has passed engineering and mechanical validation, but there is no research effectiveness benchmark test for the same version completed independently; these checks do not prove Alpha, returns, or PMF.

The 0.3 in-development capabilities are not yet in the 0.2.1 release tree and require a DSH restart to take effect. Users should distinguish between the capabilities of the current official version and those under development in the repository.

Conclusion¶

The value of Resanity lies in separating the “information, inference, and conclusions” that are easily mixed together in investment research, and using evidence boundaries, load-bearing claims, and cognitive anchor reminders to determine when review is needed. It does not make investment decisions for users, but makes the research process easier to inspect and update.

GitHub: https://github.com/Thhoho/reSanity

Community directory page: Verified materials do not provide a confirmable URL.