Meta and Equifax lead because they combine recognizable businesses or substantial reactions with saved assessments from both systems. One is a directional hit and the other a miss. These selected cases illustrate the grading; they do not establish season-wide superiority.
We compare the AgentSmyth Earnings Agent with the baseline, the earnings system we used before this redesign. Assessment score and evidence-adjusted score refer to the Earnings Agent. The baseline keeps its own calculation and score bands. We share structured summaries of the assessments, rather than their complete prose or internal generation process.
Meta: both systems saw the spending risk
Meta reported after the close on October 21. Both assessments were saved before the release. The Earnings Agent's questions separated advertising performance and holiday revenue, the affordability of infrastructure investment, and legal exposure within the annual expense outlook.
| Topic | Baseline | Earnings Agent | What happened |
|---|---|---|---|
| Advertising | Anticipated a modest revenue beat against analyst expectations. | Expected Q3 revenue of $63.8–$64.1 billion, with advertising performance and the holiday outlook as an explicit question. | Q3 revenue was $64.60 billion, above the Agent's range. Q4 revenue guidance was $73–$76 billion. |
| Investment and cash | Identified cash conversion and the $130–$145 billion investment envelope as risks, even if revenue beat expectations. | Expected capital-spending guidance to narrow toward the upper end of that range without a higher ceiling; treated affordability as a separate question. | Annual capital-spending guidance narrowed to $135–$145 billion, including finance-lease principal, with the ceiling unchanged. Q3 free cash flow was −$2.12 billion. |
| Expense outlook | Emphasized the tension between business growth and spending. | Explicitly expected annual expenses above the existing $165–$169 billion envelope and asked whether legal exposure would break it. | Annual expense guidance rose to $175–$179 billion, including a $9.89 billion legal charge recognized in Q3. |
| Stock assessment | Comparison score 25. Its financial assessment forecast a 4–6% decline; its company narrative described a 3–6% fade conditional on the spending outlook. | Assessment score 35; evidence-adjusted score 44. Both are bearish. | The shares fell 6.98% through the October 22 close. Both directional calls were correct. |
Financial outcomes come from Meta's earnings release. Free cash flow is the company's non-GAAP measure.
The market grade: the shares fell 6.98% from the October 21 close through the October 22 close. Both systems' directional calls were correct. The baseline score is bearish under both its own score bands and the common comparison bands.
The qualitative comparison: the baseline already understood the central cash-conversion risk. The Earnings Agent separated the issues into explicit questions, expected ranges and conditions that could change its view. Those commitments are easier to revisit, but clearer organization alone does not earn a better judgment grade.
The release lets us check the financial expectations directly. The share-price decline settles the direction call; it does not, by itself, prove that spending caused the move.
Equifax: a bearish read before an 11% rally
Equifax reported before the open on October 20. The Earnings Agent assessment in this pair predates the saved question-list format. Its business focus below is therefore an editorial summary of its original guidance and KPI judgments, not a newly invented historical question roster.
| Topic | Baseline | Earnings Agent | What happened |
|---|---|---|---|
| Business drivers | Examined mortgage demand, the verification business and productivity. Its guidance analysis already identified the company's AI savings target and annual margin-expansion outlook. | Examined mortgage activity, government verification, revenue mix and profit conversion; judged routine delivery insufficient to support the shares. | U.S. mortgage revenue grew 15% and Verification Services revenue 9% year over year. Quarterly adjusted EBITDA margin rose to 33.1% from 32.7%. |
| Quarterly expectations | Guidance analysis expected about $2.23 adjusted EPS and $1.70 billion revenue. Other sections used slightly different earnings estimates. | Guidance analysis expected $2.23 adjusted EPS and about $1.70 billion revenue. | Adjusted EPS was $2.26, above the saved expectations; revenue was $1.693 billion, slightly below them. |
| Annual guidance | Expected the reported outlook to hold. | Expected annual revenue and adjusted-EPS guidance to hold. | Management lowered both midpoints: annual revenue guidance became $6.646–$6.676 billion and adjusted EPS $8.45–$8.55. |
| Market interpretation | Financial and guidance analyses expected a fade, while the company narrative allowed a relief rally. The record was mixed. | Expected a routine beat and an ordinary forward outlook to be insufficient for upside. | The shares rallied 11.33% through the October 20 close, contrary to the bearish interpretation. |
| Comparison score | 40. This is neutral under its own 40–60 band; it is bearish under the common comparison rule. | Assessment 31; evidence-adjusted 31. Both are bearish. | The Earnings Agent missed. The baseline abstained under its own bands; its common-band bearish call also missed. |
Financial outcomes come from Equifax's earnings release. Adjusted EPS and adjusted EBITDA margin use the company's non-GAAP definitions.
What the company disclosed: the original Q3 call records $2.26 adjusted EPS and $1.693 billion revenue. Management lowered its annual revenue outlook and slightly reduced its adjusted-EPS midpoint, but increased expected annual EBITDA-margin expansion excluding FICO mortgage royalties from 75 to 100 basis points. That annual improvement is distinct from the Earnings Agent assessment's approximately 32.8% quarterly margin forecast; comparing the two as if they were the same measure would be wrong.
The market grade: the shares rose 11.33% from the October 19 close through the October 20 close. The Earnings Agent's bearish call was wrong. The baseline abstained under its own bands; its bearish classification under the common bands is a separate comparison. An abstention is neither a loss nor a successful prediction of a flat move.
The qualitative lesson: the systems had identified relevant business mechanisms, including productivity, but recognizing those mechanisms did not yield the right market judgment. We need to check how the assessments weighed weaker demand against better profit conversion and changes in perceived risk. The release and rally make that question worth examining; they do not prove which disclosure caused the move.
There was also a data problem: the Earnings Agent assessment's saved inputs contained an inconsistent prior annual-revenue guidance reference. We cannot describe this as a clean case of getting all the business facts right and merely misreading the market. The forecast miss, reference problem and unresolved explanation each receive their own assessment.
Supporting case: United Rentals
The baseline already discussed fleet productivity, lower-margin revenue mix, labor and delivery costs. The Earnings Agent assessment separated the questions of rental growth, profitability and cash conversion, and explicitly expected both net-capex and free-cash-flow guidance to hold.
The Earnings Agent comparison scores were 66 assessment / 56 evidence-adjusted, both bullish. The shares rose 3.93% through the October 22 reaction close. Yet management raised net-capex guidance to $3.75–$4.05 billion while keeping free-cash-flow guidance at $2.15–$2.45 billion. The investment expectation missed; the cash-guide expectation was supported. A price hit does not make both business forecasts right.
The approximately 46.0% quarterly margin expectation versus 47.6% disclosed is another separate grade: 1.6 percentage points of error. A later grader cannot invent a tolerance merely to call it a hit.
Supporting case: CBRE
The baseline connected resilient services and infrastructure growth with transaction sustainability and cash conversion. The Earnings Agent questions separated transaction demand, infrastructure economics per share, and project delivery and cash collection. Both systems engaged with the business model; the Earnings Agent format made the individual expectations easier to follow.
The Earnings Agent comparison scores were 70 assessment / 58 evidence-adjusted, both bullish, against a 2.53% decline through October 22. Its scored annual earnings-guidance expectation anticipated a raise, while another part of the assessment expected a hold. Management retained $7.80–$7.90.
That creates three distinct observations: a directional loss, a missed raise expectation, and an internal contradiction. We cannot choose only the sentence that aged well or use a relevant business question to cancel an incorrect forecast.
What these examples can establish
They show how we separate question relevance, forecast accuracy, factual consistency and the market result. They also show why this is a meaningful baseline: in these cases it already had important business ingredients. The test is whether the Earnings Agent turns those ingredients into more useful and dependable judgments.
The measurement rubric defines the broader comparison; Lab Note 1 explains the season plan.
Evidence and calculations
Stock returns: we calculate each percentage return as 100 × (ending close / starting close − 1), using split-adjusted closes without cash-dividend reinvestment.
Company disclosures: Meta's October 21 release and Equifax's October 20 release provide public financial references. Equifax also lists its October 20 investor presentation. Equifax's margin-outlook comparison also uses its earnings call.