Usage guide

Start with an existing report

1. Open a free NVDA sample. Check its report type, data date and cutoff.
2. Read the conclusion, evidence and failure conditions in that order.
3. Open the method section. Separate observations from assumptions.
4. Use the reports library to compare the same ticker, type and dates. Reading an existing report does not use generation allowance.
5. If you use an AI platform, follow its connection guide. Ask it to keep the date, source link and limitations.

Before requesting a new version

Check your plan and remaining allowance on your account page. A preview is not a full report. Availability depends on the report type and the service status.

What it is and when to use it

Overnight developments and the assumptions to check before the US session.

Before the session, to prepare a list of questions.

Read it in three steps

1. Check the data cutoff and the expected session.
2. Read the overnight developments and affected tickers.
3. Write down the conditions that would change the reading during the session.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

Find NVDA only if the brief covers it. Note the cited overnight event, its timestamp and the condition to recheck at the open. If absent, do not infer a view.

Three common misuses

• Treating an overnight move as a forecast for the session.
• Combining cutoffs from different sessions.
• Reading an omitted ticker as a neutral conclusion.

Three questions for your AI platform

• Find an available NVDA Pre-Market Brief and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

It is a dated briefing, not a continuously updated price feed.

What it is and when to use it

A session update on what has changed since the pre-market brief.

During the session, to revisit the morning assumptions.

Read it in three steps

1. Match the session date with the pre-market edition.
2. Compare the changed facts and unchanged assumptions.
3. Check which earlier conditions still apply at the new cutoff.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

If NVDA appears in both editions, compare the same claim and its timestamps. A missing afternoon update does not prove the morning claim still holds.

Three common misuses

• Treating no update as confirmation.
• Comparing it with another trading day.
• Reading a session change as a lasting company change.

Three questions for your AI platform

• Find an available NVDA Midday Brief and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

It covers information available at its cutoff and can miss later developments.

What it is and when to use it

A review of the completed session and questions for the next one.

After the close, for a dated research note.

Read it in three steps

1. Check the session and final data cutoff.
2. Compare the outcome with the assumptions recorded earlier.
3. Keep the open questions and next review conditions.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

If NVDA is covered, compare its closing discussion with the morning condition. Separate a price observation from an explanation for it.

Three common misuses

• Calling a same-day explanation a causal result.
• Treating closing information as known before the open.
• Dropping unresolved questions from the record.

Three questions for your AI platform

• Find an available NVDA Closing Wrap and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

A daily review does not establish why a move happened.

What it is and when to use it

A dated record of model readings and disagreements.

To see whether different model rules give different readings.

Read it in three steps

1. Check each model name, version and data date.
2. Read the conditions behind each reading and any disagreement.
3. Open the relevant model explanation and failure conditions.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

Look for NVDA only where a model includes it. Compare the rules and dates before comparing the readings. A missing record is not a position.

Three common misuses

• Counting correlated models as independent votes.
• Calling a model reading an executed position.
• Combining rebuilt history with forward records.

Three questions for your AI platform

• Find an available NVDA Model Desk Daily and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

A rule-generated reading is not an executed trade. Historical and forward records have different evidential roles.

What it is and when to use it

A weekly reading of market structure and competing explanations.

To set the week's research questions.

Read it in three steps

1. Check the week, publication date and data cutoff.
2. Read the main view alongside the contrary case.
3. Record the conditions and review date that would change it.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

Use the weekly context when reading an NVDA assessment. If NVDA is not covered, do not turn a market view into a company conclusion.

Three common misuses

• Reading a weekly view as a daily forecast.
• Applying a market view to every ticker.
• Ignoring the contrary case after a matching headline.

Three questions for your AI platform

• Find an available NVDA House View and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

It is a weekly research perspective; later facts can change its premises.

What it is and when to use it

A single-stock reading of structure, model evidence and failure conditions.

Before drawing a conclusion about one ticker.

Read it in three steps

1. Confirm the ticker, data date and assessment horizon.
2. Read the conclusion and its failure conditions together.
3. Check the method, sample and missing inputs before comparing versions.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

Open an existing NVDA assessment. Write down its data date, the stated reading and the exact failure condition. This walkthrough supplies no current price or result.

Three common misuses

• Treating a scenario level as a personal instruction.
• Reading an older data date as current.
• Dropping the failure condition when summarizing.

Three questions for your AI platform

• Find an available NVDA Stock Model Assessment and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

The reading applies to the stated data and horizon. It is general research and is not tailored to your circumstances.

What it is and when to use it

A company thesis presented alongside the contrary evidence.

To examine what could make a company thesis fail.

Read it in three steps

1. Read the thesis, scope and data date.
2. Pair each claim with its evidence and counterargument.
3. Record what observation would invalidate the thesis.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

For an existing NVDA report, choose one claim, find the counterargument and record what evidence would settle the disagreement. Do not invent missing financial figures.

Three common misuses

• Reading only the thesis.
• Treating repeated sources as independent support.
• Calling an unresolved disagreement settled.

Three questions for your AI platform

• Find an available NVDA Company Research · Thesis & Counterarguments and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

A coherent thesis can still be wrong. The report is limited by the evidence and cutoff it states.

What it is and when to use it

A company review of its business, financial statements and valuation assumptions.

To understand the inputs behind a company valuation range.

Read it in three steps

1. Check the financial period, currency, units and cutoff.
2. Separate reported financial facts from valuation assumptions.
3. Read the sensitivity and limitations before comparing ranges.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

In an existing NVDA fundamentals report, identify the financial period and one valuation assumption. Compare the range only under the same units and assumptions.

Three common misuses

• Comparing mismatched financial periods.
• Reading an estimate as a reported fact.
• Treating one valuation range as a certain future price.

Three questions for your AI platform

• Find an available NVDA Company Research · Fundamentals and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

Valuation depends on assumptions and incomplete information; a range does not predict an outcome.

What it is and when to use it

A dated examination of a verifiable event and comparable historical observations.

To distinguish an event record from a claim about its effects.

Read it in three steps

1. Confirm the event source, event time and study window.
2. Check the benchmark, sample and treatment of overlapping events.
3. Read uncertainty, alternative explanations and failure conditions.

Walk through NVDA

Reading example only. Use an available report and its stated date; this is not a current assessment.

For NVDA, use a documented event only when a report is available. Record its timestamp, comparison window and competing events. No event return is supplied here.

Three common misuses

• Treating correlation around an event as causation.
• Selecting only favorable events or windows.
• Ignoring overlapping announcements and repeated tests.

Three questions for your AI platform

• Find an available NVDA Event Study and give its date and original link. If none exists, say so.
• Which claims are observations, and which depend on assumptions? Keep the report's wording and dates.
• What would invalidate this reading, and what remains unknown?

Limits

Available now, subject to your plan’s generation allowance and service availability. A verifiable event in the last 90 days is required; otherwise no event study is generated.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Read beta and alpha with the benchmark, look-back window and uncertainty.

How to read it

Confirm the benchmark and window; read the interval; check whether the conclusion changes under another benchmark.

Limits

Past exposure does not fix future exposure. A beta estimate is not a loss limit.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Compare model conditions and disagreements for a ticker.

How to read it

Choose the ticker and data date; compare the model rules; open each model's limitations.

Limits

Related models are not independent votes. An absent reading is not evidence of a neutral result.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Examine a factor with a stated hypothesis, comparison and validation window.

How to read it

Check the frozen hypothesis; inspect placebo and multiple-testing treatment; read the out-of-sample and cost assumptions.

Limits

A favorable fitted result alone does not validate a factor. Missing evidence stays missing.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Examine concentration, co-movement and sensitivity under stated inputs.

How to read it

Check weights and units; compare concentration and exposures; read the scenario assumptions.

Limits

A historical correlation can change under stress. Do not submit account credentials or sensitive personal data.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Turn a research idea into a testable statement with a contrary case.

How to read it

State the claim and horizon; compare the base rate; name the observation that would refute it.

Limits

A plausible story is not a measured effect. Keep personal conviction separate from evidence.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Read a market classification with its inputs and dated definition.

How to read it

Check the classification rule; inspect the input dates; review the uncertainty near a boundary.

Limits

A regime label describes a rule applied to data; it does not establish the next market state.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Separate benchmark-related movement from the unexplained part.

How to read it

Match the dates and benchmark; read the decomposition; consider alternative events.

Limits

A decomposition is not causal proof. The unexplained part may reflect omitted variables.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Follow changes in explicitly chosen research conditions.

How to read it

Read the condition and timestamp; check the originating report; review the latest available data.

Limits

An alert is a prompt to review evidence. Do not assume immediate or uninterrupted delivery.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Calculate consequences under assumptions you specify.

How to read it

Record the inputs; change one assumption; compare the result and limitations.

Limits

A scenario is conditional, not a probability or a forecast unless the method explicitly establishes one.

Availability

Coming soon. This guide describes how to read the board when it becomes available; no result is supplied here.

Purpose

Keep a hypothesis, failure condition and review date in one record.

How to read it

Write the hypothesis before checking outcomes; set the review rule; record the result without rewriting the original claim.

Limits

A stored plan does not establish an immutable preregistration or independent verification.

Keep the claim and the evidence separate

1. State the question, population and decision horizon.
2. Freeze the hypothesis, outcome, comparison and failure rule before seeing results.
3. Record the data dates, exclusions and missing inputs.
4. Run the planned comparison; account for costs, uncertainty and all tested variants.
5. Use a held-out or forward period and an independent check where required.
6. Record the result, contrary evidence and next review date without hiding failed tests.

Limits

Following these steps is a research practice, not proof that a particular report has passed them. Check that report's evidence.

Connection address

https://axiocairn.com/mcp/chatgpt

Connect step by step

1. Open Settings → Connectors → Add custom connector.
2. Name it Axiocairn and paste the address.
3. Choose OAuth and sign in with Google or GitHub.
4. Start a chat, enable Axiocairn and ask the test question.

Check the result

Ask: What reports does Axiocairn have right now? Keep the report dates and original links. Platform menus and plan requirements can change.

Connection address

https://axiocairn.com/mcp

Connect step by step

1. Open Customize → Connectors → Add custom connector.
2. Name it Axiocairn and paste the address.
3. Sign in with Google or GitHub and allow access.
4. Enable Axiocairn in a chat and ask the test question.

Check the result

Ask: What reports does Axiocairn have right now? Keep the report dates and original links. Platform menus and plan requirements can change.

Connection address

https://axiocairn.com/mcp/grok

Connect step by step

1. Open grok.com/connectors → New Connector → Custom.
2. Name it Axiocairn and paste the address.
3. Sign in with Google or GitHub.
4. Ask the test question.

Check the result

Ask: What reports does Axiocairn have right now? Keep the report dates and original links. Platform menus and plan requirements can change.

Connection address

https://axiocairn.com/mcp/grok

Connect step by step

1. Open Settings → MCP → Add new server.
2. Paste the address as a streamable HTTP server.
3. Sign in when Cursor opens the browser.
4. Ask the test question in chat.

Check the result

Ask: What reports does Axiocairn have right now? Keep the report dates and original links. Platform menus and plan requirements can change.

Connection address

https://axiocairn.com/mcp/muse

Connect step by step

1. Ask Muse to add a custom connector named Axiocairn.
2. Paste the address when Muse asks for the server.
3. Sign in at axiocairn.com and create a one-time code.
4. Paste the code into Muse within 10 minutes.

Check the result

Ask: What reports does Axiocairn have right now? Keep the report dates and original links. Platform menus and plan requirements can change.

Check every summary

1. Ask for the data date and original report link.
2. Keep the failure conditions and missing information in the answer.
3. Compare versions only after matching ticker, report type and data date.

Scope of access

The website and MCP use the same account permissions and generation allowance. An AI summary can omit context; the original report remains the reference.

Reconnect in four steps

1. Remove Axiocairn and add it again. Use the URL shown for your platform. Axiocairn does not use API keys. If a connect screen asks for an API key or only receives a domain, cancel, then add it from the connector / MCP option using the full URL (for example https://axiocairn.com/mcp/muse) and sign in with Google or GitHub to authorize.
2. Use the same sign-in account. Sign in with the Google or GitHub account you used to activate Desk.
3. Refresh the tool list. Reopen the tools in your AI agent.
4. Ask a test question. What is the latest Stock Model Assessment for NVDA?

Messages and next action

• identity_required: Not signed in: sign in with the same Google or GitHub account that you used to activate Desk.
• forbidden: Your plan does not include this content: check your current plan on your account page.
• quota_exceeded: Monthly generation allowance used up: read existing reports, check your reset date or plan on your account page.
• market_closed: Market closed: check the report schedule and try again at the next scheduled update.
• service_unavailable: Temporarily disabled: try again later; if it still fails, send Feedback with your platform, failure time and time zone.
• connection_failed: Check the connection steps and messages below.
• rate_limited: Too many requests: try again {wait}.

Missing tools or a changed connection

Refresh the tool list, then reconnect using the full URL for your platform. If it still fails, report the platform, time and time zone without passwords or access codes.

What uses allowance

Desk can generate 30 new report versions per calendar month. Desk Pro can generate 120. Stock Model Assessments, company research and event studies share this allowance. Company research and event studies together have a sublimit of 10 for Desk and 40 for Desk Pro.

A new version is identified by ticker, report type and data date. Only a new version whose generation your account triggers counts. Opening an existing report, reopening it or reading a version generated by another account does not count.

The allowance resets at 00:00 America/New_York on the first day of each month. Samples, daily briefs, Model Desk Daily and House View do not count. Fair-use rate limits still apply.

Reader and availability

Reader can read every sample in full and previews of other reports. Paid access follows the plan. Desk and Desk Pro can generate company research and event studies when those features are available. Company Research and Event Study generation: Coming soon. Stock Model Assessments remain available on request.

Each new company research report or event study uses one generation from the total allowance and one from the company and event sublimit. Failed generations are refunded. Viewing reports that have already been generated uses no generation allowance and follows your plan’s access rules.

Use the same account

1. Open your account page with the sign-in method used to activate the plan.
2. Check the plan, expiry or renewal date and remaining generation allowance.
3. Enter an access code only in the account redemption field. Check the result before retrying.

Codes and invitations have different purposes

Each account can redeem one access code. If it already has a paid plan, the code is not used up. Do not post the code in a public question or issue report.

An invitation gives the invited person 14 extra free days. After that person's second successful payment, the inviter receives one month of Desk, up to three times. Check your account for the actual invitation and reward status.

Read the time labels

• Data date: the observation period or date used by the report.
• Cutoff: the latest information included in that version. Publication time can be later.
• Version: the same ticker and report type can have different data dates or corrections. Keep the original link when comparing.

Missing and corrected information

Check missing fields, assumptions and corrections in the report itself. A content fingerprint can help compare content; it does not prove the underlying data or conclusion is correct.

This guide supplies no live price, sample size or model result. Use the method and evidence attached to the report you are reading.

Risk exposure notation

In a single-benchmark regression, r_i,t − r_f,t = α + β(r_b,t − r_f,t) + ε_t. Here r_i is the asset return, r_b the benchmark return, r_f the reference cash rate specified by the model, α the intercept, β the fitted exposure and ε the residual. Check the report for frequency, window and units.

For the fitted coefficient, t = estimate / standard error. Its interpretation depends on the error model and sample. A 95% confidence interval is a procedure-based uncertainty statement, not a 95% probability that a fixed parameter lies in the observed interval.

Event-study notation

AR_t = r_i,t − E[r_i,t | model]; CAR[a,b] = Σ AR_t over the stated window. The expected-return model, estimation window, event timing and overlapping events affect the result. This notation does not establish causality.

Multiple testing

For Holm adjustment, order the m p-values p_(1) ≤ … ≤ p_(m), then compare p_(k) with α / (m − k + 1), stopping at the first non-rejection. The testing family and all variants must be stated before interpreting a corrected result.

These are explanatory formulas, not a claim that every board or report implements this exact method. Read the method attached to each result.

Terms used in reports

• Benchmark: A stated comparison series or method.
• Beta: Estimated sensitivity to a chosen benchmark over a chosen window.
• Alpha: The fitted intercept under a stated model, not proof of skill.
• Confidence interval: An interval produced by a stated uncertainty procedure.
• Sample size N: The number of observations after the stated exclusions.
• Out-of-sample: Data not used to choose or fit the tested rule.
• Forward record: An output recorded as time passes, separate from rebuilt history.
• Placebo: A comparison designed to test a non-target explanation.
• Base rate: The observed frequency in a defined reference group.
• Failure condition: A stated observation that would make the reading no longer apply.
• Preregistration: A recorded plan fixed before outcomes are examined; inspect its timing and binding evidence.
• Content fingerprint: A content identifier for comparison, not proof of truth.
• CAR: Cumulative abnormal return over a defined event window.
• Holm correction: A multiple-testing procedure over a declared family.

Read the full policies

The linked policies contain the full conditions. This guide does not replace them. Read the disclaimer alongside each report.

Refund conditions

Protect your account information

Read the privacy policy for collection, processing and deletion details. Do not include passwords, payment-card details or access codes in a research query or issue report.

How to use reports

1. Open the reports page and pick a report.
2. Check the data date and cutoff time at the top.
3. Read the reading first, then the evidence and the failure conditions.
4. Compare two reports only when you know both dates.

How to use the model pages

1. Open the model pages and pick one of the five models.
2. Read what it looks at and when it does not apply.
3. Open today's Model Desk Daily and find that model's status.

How to use the bench

1. Open the model pages.
2. Pick a model and read when it does not apply.
3. Compare it with today's Model Desk Daily. The model screener is coming soon.

Connect an AI agent for the first time

1. Pick your platform on the connection guide.
2. Copy the Axiocairn URL shown for that platform.
3. Add it in the platform's connector settings.
4. Sign in once when prompted.
5. Ask a test question: What is the latest Stock Model Assessment for NVDA?

How to choose a plan

1. Compare the plans in the plan table.
2. Choose monthly or annual billing.
3. Sign in and check out through Stripe.
4. Open your account page and confirm your plan.

How to use your account page

1. Sign in with the account you used to activate your plan.
2. Check your plan, its status and the renewal date.
3. Enter an access code or manage your subscription when you need to.

How to manage your data

1. Read the privacy policy.
2. Check Connected AI agents on your account page.
3. Delete your account on the account page, or email contact@axiocairn.com.