A Track Record for Every Management Team

Score Management Against What They Said They'd Do

A 40 to 60-name coverage list produces hundreds of forward-looking statements a quarter. Guidance Scorecard captures each one when it is made, links it to every later restatement, and checks it against the reported result when its period closes. The output is one record per commitment and one scored track record per company, comparable across your coverage.

The report below is Tesla's current record as of its 2Q-2026 call: the three metric commitments in force, the seven events management has put a date on, and for each the filing it was last stated in and how it has moved.

Two rows show what linking restatements does. FY-2026 capex was introduced at more than $20B in January 2026, raised to more than $25B in April, and reaffirmed at that level in July. Optimus production at Fremont was planned for before the end of 2026 in January, narrowed to late July to August in April, and widened back to July to December in July. Each reads from a single row with the history attached: one guide raised once and then held, one milestone narrowed to a two-month window and then let out again.

Baik, Farber, and Lee (2011) found that managers who issue frequent and accurate forecasts run firms that outperform peers on both operating and stock performance. A score built from management's own commitments can be checked row by row in an IC discussion.

One Commitment, Tracked Across Every Restatement

A commitment is one metric or one anticipated event for a given segment, target period, and currency basis. Every later statement about it links back to that record, even when management moves the date, widens the range, or changes the wording. Unsupervised FSD for personal use was "before the end of this year" in April 2025, "by the end of this year in certain geographies" in July 2025, and "probably in the fourth quarter" of 2026 in April 2026. Guidance Scorecard reads those as one commitment pushed out by a year, with all three quotes attached.

A commitment carried forward and reported on every quarter reads differently from one mentioned once and quietly dropped, and a target met at the third attempt reads differently from one met as first guided. Linking the restatements is what makes both differences visible, and they are what the scores below measure.

Official Guidance Matters, but the Edge Is in the Informal Forecasts

Official guidance is carefully managed and widely tracked. Revenue ranges, EPS bands, and the formal outlook are already covered by every major data vendor. What they do not cover is the operational commentary around it: production volumes, user growth, factory openings, customer adoption, milestone timing. Executives speak more freely about operations, and that is where their confidence shows.

Guidance Scorecard covers official guidance and every informal forward-looking statement an executive makes across earnings calls, press releases, investor presentations, interviews, and capital markets days, all drawn from the filings, press releases, and earnings calls we ingest. In the Tesla record above, the capex line is sourced to the quarterly report, six rows to the shareholder update, and three to the earnings call. The capture reads the entire document, including the Q&A, and separates items that share a sentence, so two geographies or two products named together become two commitments you can track independently.

Two Statement Types, Both Testable

• Metric-based: quantitative targets for financial or operational measures (revenue, margins, production output, unit economics), at the company level or by segment, product line, or geography.
• Event-based: milestone predictions, externally visible (product launches, factory openings, regulatory approvals) or internal but strategically important (reorganizations, hiring plans, technology rollouts, cost-reduction initiatives).

Each statement is standardized with a clear metric or event, timeframe, scope, and target.

Vague Language, Made Testable

The hard part of this work is translating loose phrasing without losing what management said. Guidance Scorecard does it with documented assumptions and the original quote attached to each interpretation:

• "low to mid single digits" becomes a 1 to 5% range
• "flat q/q" becomes a tolerance band around the prior period
• "similar to last quarter" becomes a relational test against the prior result
• "around the middle of next year", said on the 2Q-2026 call about AI5 volume production, becomes June 2027
• "probably around the end of this year or early next year", said on the same call about self-driving on the Tesla Semi, becomes 1 October 2026 to 31 March 2027

Reviewers can see how each translation was resolved and challenge it if they disagree.

A Score You Can Take to an IC Meeting

When the period arrives, each commitment is resolved against what happened. Metric commitments are labeled beat, met, or missed against the reported result, on the basis management guided to, with adjusted vs. as-reported preserved. The check reads the reported financials directly. Event commitments are resolved on a schedule informed by their expected dates and labeled occurred, delayed, pending (the window is still open, so it is not yet a miss), or cancelled. They are validated against company communications and external reporting, since launches and factory openings are often visible publicly without a formal filing.

Those outcomes roll up into two scores per company, each on a 1 to 5 scale:

• Accuracy: how surprised you would have been holding the latest guidance. An outcome outside the guided range counts in either direction, since guidance routinely beaten by a wide margin tells you as little as guidance that is missed, and so does a cut or raise outside the range previously given. Surprises are weighted by size and materiality: a small beat on a minor metric barely registers, a headline target cut three times does not.
• Discipline: whether management keeps giving the guidance it started. Recurring guidance such as revenue or margins should be carried forward every period, and one-off commitments such as a factory opening tracked until they resolve. Guidance issued once and dropped, withdrawn, or no longer reported against pulls it down.

The two are independent. A team can guide every quarter without fail while its numbers keep surprising, or land its numbers while quietly dropping the guidance it would rather not report.

The Tesla report at the top of this page is the scored version for one print: accuracy 3 of 5, discipline 3 of 5, assessed at the 2Q-2026 call on 22 July 2026. Free cash flow came in at -$1,092 million against guidance for a negative figure, and Texas training compute more than doubled as guided. Of the 30 expected events in its table, 16 had occurred (the Cybercab and Semi production starts, the Model S and X line shutdown, the Robotaxi launches in Dallas, Houston, Miami, Orlando, and Tampa), four had slipped past their window (Robotaxi in Phoenix and Las Vegas, the Optimus Gen 3 unveiling, EU-wide FSD approval), and ten were still open. Each row keeps the date the commitment was first provided, linked to the filing, so a reader can check the slip against the original wording.

Where the Record Shows Up

Both reports on this page are Deep Research Agent outputs. Three agents sit on the company Guidance page. They share the same source record. Pick by the question you are asking.

• Current Guidance for the live guide: metric targets whose period has not ended and events whose window has not closed, each with the restatement history attached. The Tesla table at the top of this page is this agent. Outcomes are not scored.
• Guidance Track Record for the historical record: closed commitments labeled beat, met, missed, occurred, delayed, or cancelled, rolled up into accuracy and discipline scores.
• Guidance Due at Earnings for one print: the commitments that came due at that earnings release, scored against what was reported. When an earnings review runs, this is the evaluation that appears next to the current call commentary. The Tesla scores at the top of this page (3 of 5 and 3 of 5) come from this agent.

Scoping Current Guidance to an earnings period lists the guidance that print issued. Guidance Due at Earnings scores the guidance that print put to the test. Those are opposite slices of the same print.

You launch any of them the same way you start any Deep Research Agent. Results land back on the Guidance page. The same evaluations are queryable in AI Analyst Chat for ad-hoc questions, and the full restatement history for each commitment is browsable as structured tables in the Marvin Labs platform. For the end-to-end management-quality workflow, see the Management Quality Assessment solution.