Part 6 — Sample Methods Paragraph (intentionally weak)

Part 6 — Sample Methods Paragraph (intentionally weak)#

The paragraph below is drawn from a real-style undergraduate report on a multi-disciplinary geophysical survey. It is intentionally weak in several ways — vague, missing parameters, missing software versions, missing data provenance. Your job in Part 6 is to design a “Methods Reviewer” AI agent that catches these weaknesses automatically.

This paragraph is your test case for the agent. Do not fix the paragraph yourself — feed it to your agent and evaluate whether the agent’s critique would be useful to a peer reviewer.

Methods (excerpt from a hypothetical student report)#

We collected seismic refraction and reflection data along a survey line crossing the eastern flank of a Cascade volcano. The geophone array was deployed at the surface and we shot the line with a seismic source. The data were processed using standard methods in Python. First arrivals were picked from the shot gathers, and we fit lines on a T–x plot to derive velocities. We also picked reflection events and applied NMO correction to remove the hyperbolic moveout. The velocities were then used to estimate the depth to the bedrock interface, which we found to be at a few tens of meters.

Gravity data were collected at the same stations as the seismic survey, using a gravimeter. We applied the standard Bouguer correction, terrain correction, and latitude correction to obtain a Bouguer anomaly map. The gravity data showed a low anomaly over the western half of the survey, which we attributed to a low-density body. Magnetic data were collected with a magnetometer and processed in the usual way. We then interpreted all three datasets jointly to constrain the geometry of the subsurface magmatic system.

Why this paragraph is a useful test case#

A real reviewer reading this paragraph would have questions:

  • Which seismic source was used (sledgehammer, weight drop, explosive, vibroseis)?

  • What was the geophone spacing? Shot interval? Total line length?

  • Which Python package handled the processing? With what version? Was the inversion linear, MASW, or tomographic?

  • What filter was applied before picking? Corner frequencies?

  • What is “a few tens of meters” quantitatively? With what uncertainty?

  • What density was assumed for the Bouguer correction?

  • Where was the magnetic survey calibrated? Diurnal correction applied?

  • Where are the data stored? DOI, URL, repository?

  • Which coordinate reference system?

These are exactly the kinds of issues a well-designed methods reviewer agent should flag. If your v0 agent only flags one or two of them, that’s normal — iterating the agent until v1 catches more is itself a learning artifact, and is what you will do in Week 8.