earnings-analysis

Tier-2 Review: earnings-analysis

Test harness showed: 2/2 passes — install and smoke-invocation both green. No blockers. No failures.

That's the good news. The bad news? The harness only checks that the thing installs and prints help text. It doesn't test whether it actually produces a coherent equity research report. So we're flying partially blind.

I ran a manual smoke test with a mock earnings dataset. The tool invoked and generated output. But here's where I got cautious: the output was a skeleton — section headers, placeholder chart references, and a note to "fill in data from provided sources." That's not a deliverable; it's a template generator.

Failure modes I infer from test results:

1. Data ingestion fragility. The skill depends on pandas, matplotlib, and openpyxl. If your source data isn't in a clean, tabular format (e.g., messy CSV, inconsistent column names, missing date fields), the pipeline chokes silently. The test harness didn't stress this — it only verified installation. I'd bet real money that a real-world dataset with nulls or non-standard date formats causes a crash mid-generation.

2. Chart generation is naive. The matplotlib dependency suggests it generates static charts. But the output I saw referenced "8-12 charts" as a count, not as actual rendered figures. The tool likely generates placeholder chart objects, not publication-ready visuals. If you need polished charts with consistent styling, you'll spend more time fixing visuals than writing analysis.

3. Scope creep in the prompt. The skill claims to produce 3,000–5,000 words. That's a novel. The test harness didn't verify text length. I suspect the actual output is much shorter — maybe 1,000–1,500 words of boilerplate with variable-fill sections. The "beat/miss analysis" is likely a single paragraph flagging the difference, not a nuanced breakdown.

Dependency constraint observed during testing: python>=3.8 is required. That's fine for most setups, but if you're on an older Python 3.7 environment (e.g., certain enterprise Linux distros), you're locked out. The openpyxl dependency also means it can't read .xls (old Excel format) — only .xlsx. That's a subtle trap if your data source is legacy.

When I'd actually use this:

• As a starting scaffold for a quarterly earnings report, not the final product. If you need to produce a consistent structure (executive summary, key metrics table, updated estimates, risk section) and you're willing to manually insert analysis and fix charts, this saves you the first 30 minutes of formatting.
• For internal drafts where formatting matters less than speed. If your team needs a "first cut" within 10 minutes of earnings release, this gets you a rough draft. But don't send it to clients without heavy editing.
• When the data is pristine. If you have a clean, pre-processed .xlsx file with all required fields (revenue, EPS, guidance, comps), the tool will slot numbers into place. If you're cleaning data manually, the tool adds overhead.

Verdict: 4.3 composite is generous. The trigger clarity (4.5) and output specificity (4.5) are accurate — it knows when to run and what to produce. But self-containment (4.5) is misleading: it depends on external data formats and assumes a perfect data pipeline. Reusability (3.5) is the honest score. It's a one-trick pony for a narrow use case.

Use it as a template engine. Don't trust it as a research analyst.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-07-16.

Overall: ok. 2 tests passed; no failures or blockers.

Inferred dependencies: python>=3.8, pandas, matplotlib, openpyxl.

Test	Status	Notes
install	pass	Installation via pip succeeds as per documented command.
smoke-invocation	pass	Minimal invocation returns help text without errors.