Caftra Labs.
Building AI assistants.
We’re developing AI assistants with distinct tasks, one project at a time. Research and journaling are our first area of work.
Independent lab · First project in development
Maven.
Our first assistant is intended for traders who document their own market observations and want a clearer record of their reasoning. The first release is planned around organising notes, assumptions, and post-trade review.
The explorer below shows manually written examples of that planned workflow.
View the projectOne lab, distinct tasks
Caftra Labs is a home for assistants with distinct tasks. Our first project focuses on trading research and journaling; further products will be shaped around other user problems.
Future directions
Document research and workflow automation are areas we may explore. Their product scope and timing remain to be defined as we learn from our first project.
How we develop our tools.
About Caftra LabsMaven.
Research, recorded.
For traders who keep their own research records, the planned first release brings observations, assumptions, and review notes into a structured journal. This explorer illustrates one part of that workflow.
Hypothetical observations, written to explain the process.
Location before direction.
A move means more when its location is explicit. Start by separating the market setting from the interpretation of a single observation.
Recorded observation
Price is testing the upper boundary of a prior trading range.
Still unresolved
The observation alone does not establish whether price will sustain outside that range.
Question to carry forward: what would distinguish a sustained move from a return into the range?
Make the expectation testable.
An upward move outside the range creates a continuation hypothesis. Write down what should follow and what would weaken that explanation.
Expected follow-through
Subsequent observations remain outside the range and sustain the upward move.
Weakening condition
Price returns into the prior range instead of sustaining outside it.
Question to carry forward: does the next evidence support the original expectation?
Compare the expectation with the evidence.
In this written example, subsequent observations remain outside the range. The original hypothesis has support, while uncertainty remains.
Observed in this example
Price remains beyond the prior upper boundary and sustains the initial upward move.
Interpretation to review
This is consistent with the continuation hypothesis. It does not establish a future outcome.
In this written example, price returns into the range. The evidence weakens the initial continuation hypothesis.
Observed in this example
Price moves back inside the prior range rather than sustaining outside it.
Interpretation to review
The expected follow-through did not occur. An opposite hypothesis would need its own evidence.
Question to carry forward: what changed in the evidence, and what changed in the thesis?
Keep the conclusion traceable.
Preserve the setting, the original expectation, and the observation that supported it. A human reviewer still evaluates the interpretation.
Research note
Continuation was supported by the observations in this illustrative case; uncertainty and weakening conditions remain explicit.
Human review
Check whether the recorded evidence supports that wording and whether relevant context is missing.
Retain the initial thesis alongside the evidence that weakened it. Revising an interpretation should leave a visible reasoning trail.
Research note
The continuation thesis weakened when price returned into the range. No opposite conclusion is established by that fact alone.
Human review
Check whether the revision follows from the evidence and whether a new hypothesis needs separate validation.
Question to retain: could someone else reconstruct this reasoning from the note?
This explorer contains manually written examples. It is a concept illustration, not a live assistant, AI response, market feed, or trading recommendation.
Planned first release
A research and journal assistant for organising a trader’s own observations, recording assumptions, and reviewing past reasoning. This initial public scope centres on documentation and review.
Planned Claude integration
Input: user-supplied research notes, journal entries, and written observations. Output: structured notes that retain the supplied facts, assumptions, missing information, and review questions.
How we plan to evaluate it
Compare each generated note with its source record, check for invented or omitted facts, and flag unsupported conclusions. AI involvement and unresolved questions should stay visible.
Where the project stands
The project is in development. This website contains a manual concept explorer; the planned assistant and Claude integration are not available here.
Evidence before
conviction.
These questions guide our first product’s development. Browse a topic and open a note to see the reasoning behind it.
Can an observation become a testable thesis?Connect the market setting to an expectation that can be checked.
A useful thesis connects context, a directional move, and an expected follow-through. It also states what would weaken or invalidate the interpretation.
We want the research process to preserve that sequence rather than reduce it to a single label. The workflow explorer illustrates how the same initial expectation can meet supporting or weakening evidence.
The question we’re exploring
Can a research note make the expectation and its failure condition clear enough for another person to review?
Explore the thesis stageCan higher-timeframe context stay intact?Keep the original setting visible when inspecting more detailed observations.
Lower-timeframe observations should support or challenge the original thesis without silently replacing it. A local move can look different when separated from the market structure around it.
Our research explores how to keep context and more detailed observations distinct and traceable. Both should remain available when the interpretation is reviewed.
The question we’re exploring
Can more detailed evidence be examined without losing the location and structure that gave it meaning?
Explore the context stageCan AI help review without hiding uncertainty?Ground explanations in supplied evidence and keep human review explicit.
We plan to evaluate Claude for turning user-supplied research notes and journal entries into structured records. The output should preserve the original observations, assumptions, missing information, and review questions.
Evaluation will compare generated notes with their source records and check for invented facts, omissions, and unsupported conclusions. The integration remains planned; the examples on this website are written manually.
The question we’re exploring
Can AI organize an explanation while retaining its assumptions, missing evidence, and unresolved questions?
Explore the review stageThese are development questions and illustrative explanations. No performance results or completed studies are being presented.
Small beginnings.
Deliberate development.
A lab for focused AI products.
Caftra Labs is an independent, early-stage AI product lab. We are developing our first assistant and exploring further product directions around distinct user problems.
Our first project focuses on trading research and journaling, for people who want to organise their own observations and review their reasoning. Further products may address document research or workflow automation; their scope and timing are still being explored.
A few useful answers.
Will Caftra Labs build other products?
That is our longer-term direction. Our first project is in development, and further assistants would serve different tasks. Future products are exploratory and do not have announced release dates.
Are the products available?
Our first assistant is in development and is not yet available as a public service. The explorer contains manually written concept examples.
Does the explorer generate AI analysis?
No. It lets you inspect illustrative research stages and compare predefined examples. Planned AI use for notes, explanations, and review still requires development and evaluation.
Are there published trading results?
We are not presenting public performance results, backtests, or promises of returns. This website describes the project and its intended research approach.