Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.terrapinfinance.com/llms.txt

Use this file to discover all available pages before exploring further.

Data Tables (/munis/data-tables) is the workspace for comparable metric extraction across a watchlist. You write one prompt describing the metrics you want, and the platform runs it against every entity, normalising the results into a sortable, filterable matrix.
Data Tables is still under active testing and not yet generally available. It’s reachable at /munis/data-tables but is not advertised in the main sidebar, and behaviour may change without notice while we finalise it. For early access or feedback, email team@terrapinfinance.com.
Data Tables replaces the old Metrics and Analysis pages. The previous workflow (define each metric template up front, submit them one by one via Watchlists, then view in Analysis) has been collapsed into a single prompt-driven surface. If you used Metrics/Analysis before, your data is still extractable here; the path is just shorter.

​
When to use Data Tables vs other surfaces

You want…Use
A long, prose write-up for one obligor (tear sheet, memo section)Deep Dive. One session, one answer.
The same write-up for every obligor in a watchlistA Prompt Template assignment. One event per obligor.
A comparable matrix of metrics across every obligor (e.g. revenue, expenses, DSCR by fiscal year)Data Tables. One matrix, sortable rows × columns.
The difference from a Prompt Template assignment is the shape of the output: an assignment gives you N independent write-ups; Data Tables gives you one matrix where every cell is comparable across rows.

​
The workspace

There’s one workspace per prompt. The sidebar lists every prompt you’ve authored, plus any saved views nested under it. The main area has three sections:
  • Prompt panel: the prompt body. Editable inline.
  • Matrix: rows (typically entities) × columns (typically fiscal years or another comparable dimension). Each cell shows the extracted value and links back to the source page.
  • Filters and view controls: sort, hide rows, switch orientation (trend by entity vs. trend by metric), show or hide blanked cells.

​
Sandbox vs production

Data Tables has a built-in two-step extraction flow:
1

Preview on a sandbox

Click Preview. The platform seeds a sandbox with up to 3 CUSIPs from the watchlist and runs the prompt against them. Use the sandbox cells to check the prompt is extracting what you expect.
2

Tune the prompt

Adjust the prompt body until the sandbox cells look right. Sandbox runs are cheap and don’t touch the full watchlist.
3

Apply to production

Once the prompt is dialled in, apply it to the full watchlist. Extraction runs across every entity and the matrix populates as cells come back.
You don’t have to think about the sandbox watchlist explicitly. It’s seeded automatically from whichever watchlist the prompt is associated with.

​
Saved views

A view is a bookmark on top of a prompt: a saved combination of filters, sort, and orientation. Views are nested under their parent prompt in the sidebar so you can stack several lenses on the same data (e.g. “DSCR descending”, “missing audits only”, “FY2024 cohort”).

​
Tips

  • Author against a small sample first. Use the sandbox to nail down output shape and disambiguation rules before paying for a full extraction.
  • Be explicit about units and time bases. “Revenue in USD millions” and “Fiscal year ending June 30” beat “operating revenue” for sortable comparability.
  • Pin sector idiosyncrasies in the prompt. For example, exclude entrance fees from DSCR for charter schools, or specify GASB vs FASB.
See Prompt Templates → Writing better prompts for more.