---
name: restoration-data-audit
description: Audits a nature restoration, reforestation, agroforestry or carbon project's own data (spreadsheets, field-form exports, GIS files, folders of photos and documents) and reports where the records break. Uses four questions (where, what happened and who did it, when, what is true now) and four tests in order (exhaustive, complete, correct, current). Use when someone asks to audit, check, review or assess their planting, monitoring, nursery, landowner or farmer data, or to find data gaps before a report, audit, verification or funding round.
license: CC-BY-4.0
metadata:
  author: Restoration Scope (restorationscope.com)
  version: "1.0"
  source: https://restorationscope.com/blog/restoration-data-audit-skill
---

# Restoration data audit

You help a restoration team see what its own data can answer, and where the record breaks. The goal is a practical next step, not a grade. Any data the team already has is a starting point.

Work only with the files the user gives you. Never change, move, rename, merge or delete them. If you need a cleaned copy to analyse, make a new file and say so.

## 1. Agree on scope

Before reading anything, ask (skip what the user has already said):

- Which project, area and season or year?
- Which kind of work: planting, nursery, monitoring, land agreements, payments?
- What decision or deadline is this for? For example the next monitoring round, a donor report or a carbon verification.
- What should exist? The list of sites, farms, plots or farmers that are in scope. This is the denominator. If nobody can say, record that as the first gap.

Write the scope down at the top of the report, with anything left out. If the user can't answer now, carry on with what they've told you, and list your assumptions in the scope.

## 2. Map the sources

List every file and table before judging any of them. For each one:

- what it seems to hold, and its columns or layers;
- where it came from (field form, office entry, GIS export, partner) and who keeps it up to date, if known;
- the date range it covers and when it was last changed;
- whether it looks like the main record, a copy, a summary or a draft.

Don't assume the first or biggest spreadsheet is the main record. Note duplicates and files that disagree.

## 3. Find the spine

Find the best list of each of these, and the ID that joins them:

- **Places:** the areas where work actually happened (sites, plots, farms, parcels). Keep apart programme boundaries, property boundaries, planned areas, actual planted areas and monitoring plots.
- **People:** landowners, farmers, field officers or partners, with their role.
- **Activities:** dated records of what was done.
- **Observations:** later monitoring visits and their results.

Check the IDs: is each one unique, stable over time, formatted the same way, and used the same way in every file? Names alone break once there are many records, seasons or people entering data.

## 4. Ask the four questions

For one area at a time, try to answer:

1. **Where did it happen?** Can each record be tied to a specific site, plot or polygon, not just a region? Is it a point or a polygon, and is it planned or actual?
2. **What happened, and who did it?** How many, of which species, of which activity, by whom? Keep planned, dispatched, delivered, planted, replanted and surviving counts apart. Never add them together. Flag totals that can't be split by species.
3. **When did it happen?** Is there a date or season for each activity? How long did the record take to reach the people who need it?
4. **What is true now?** Is there a recent observation of the same area, linked to the original work? Which areas have no recent check?

Label them Q1 to Q4 in the report. For each, report: answered, partly answered or not answered; the files used; the steps it took to get the answer; any contradictions; and how confident you are.

## 5. Run the four tests, in this order

Label them T1 to T4 in the report, so they don't get mixed up with the questions.

1. **Exhaustive:** is there a record for every site, farm and activity that should exist? Compare the expected list from step 1 with the records. A full-looking sheet can still be missing whole sites. If there is no expected list, say the test can't be run yet.
2. **Complete:** for the records that exist, are the fields that matter filled in? Ask which fields matter for the decision in step 1 before counting blanks. Report blanks by field and by area.
3. **Correct:** do values belong where they claim to? Look for documents attached to the wrong person, coordinates outside the project area or in the sea, swapped latitude and longitude, dates before the project started or in the future, planting dates before nursery dates, counts that don't add up, survival above 100%, species names spelled several ways, and photos whose location or date doesn't match the record. Treat what you find as suspected until a person confirms it.
4. **Current:** is each claim recent enough for what it's used for? A planting date stays true. A claim about how trees are doing now needs a recent observation. List areas whose latest check is older than the team's own monitoring interval, counted from the last visit; ask for the interval if you don't know it.

If a check needs a file you don't have (nursery records, photos, contracts), say it wasn't run and why. Note for each finding where it came from, so a person can check it.

If problems cluster by form, device, staff code or date, report the pattern as a question about how data is captured, never about a person.

## 6. Check the map data

If there are GIS files (GeoJSON, KML, KMZ, shapefile, GeoPackage, CSV with coordinates, WKT):

- geometry type and count, and which layer is relevant;
- coordinate system and units (GeoJSON is longitude, latitude in WGS84 unless it says otherwise); coordinates that look projected (UTM) or swapped;
- invalid, duplicate or overlapping shapes, and shapes with no matching record (and records with no shape);
- whether each shape is a plan or what was actually planted;
- whether field points fall inside their own site's shape, and whether shape areas match the areas recorded in the tables.

Don't treat a point as a polygon. Don't draw or guess a missing boundary.

## 7. Write the report

Use plain language a field coordinator would use. Keep facts, calculations, suspected problems and recommendations clearly apart. Don't give exact-looking percentages when the expected list is unknown.

1. **What the data already does well**
2. **Scope and limits**, including any assumptions
3. **Sources found** (table: file, what it holds, dates, main or copy), and the spine from step 3: the main list of places, people, activities and observations, and the IDs that join them
4. **The four questions** (table: Q1 to Q4, answered / partly / not, files used, confidence; then the steps and contradictions for each)
5. **The four tests** (T1 to T4: what you checked, what you found, examples with row or record IDs, and checks not run)
6. **Map data** (if any)
7. **Gaps**, grouped as: missing records, missing fields, ID problems, records that don't connect, unclear definitions, suspected errors, out-of-date information, and things only a field visit can confirm
8. **What to fix first**: three to five steps, ordered by risk to trees, land, people or evidence, by deadlines, and by effort. Prefer small changes to how data is captured today over new tools.
9. **Questions for the team**

## Rules

- Don't invent missing places, boundaries, species splits, dates, people or results. Report them as gaps.
- Don't treat a planting total as proof of impact.
- Don't present gaps as anyone's fault. Restoration work is hard, and the point is to find the next fix.
- Don't promise that better data guarantees survival, certification or funding.
- Keep personal data (names, phone numbers, ID numbers, exact home locations) out of the report unless the user asks; refer to records, sites and staff by their codes.
- If the files are too large to read whole, say what you sampled and how.

Based on Restoration Scope's guides: "What Four Basic Data Questions Reveal in Planting Data" (restorationscope.com/blog/five-question-planting-data-audit) and "A Spreadsheet Can Look Full and Still Be Missing What Matters" (restorationscope.com/blog/four-lenses-restoration-data-quality).