
A practical continuation of the Plant City + Hillsborough County 100-address exercise.
QGIS workflow · 36 review cases · Project delivery
Self-paced / intermediate / approximately 4–6 hours
Training snapshot retained from the previous exercise.

Level 1 showed how to turn addresses into points and compare them with a reference. Level 2 teaches you to decide what the evidence supports, preserve uncertainty, and deliver a project someone else can inspect.
| Block | Time | Evidence you will produce |
|---|---|---|
| Set up & inspect | 40 min | A QGIS project with reproducible paths and verified row counts |
| Adjudicate differences | 90–120 min | A 36-record decision log with evidence and unresolved cases |
| Analyze & communicate | 60 min | A projected analysis layer, summary, and branded map |
| Check & hand off | 40 min | QA report, GeoPackage, and project README |
Use QGIS Desktop, a spreadsheet editor, and Python 3 if you want the included automated checks. Instructions use QGIS 3.44 terminology; labels can vary in other versions. You need basic layer loading and attribute-table skills from Level 1.
Work from the extracted Level 2 folder. Everything needed for the core exercises is included. Online source checks are an optional extension; write the date and source if you make one.

The exercise uses the completed comparison snapshot from Level 1. It is not a new address lookup. The original input contains 100 public-facility records across 46 input ZIP codes, including 9 records labeled Plant City.
| File in data/ | Expected content | Role |
|---|---|---|
| hillsborough-100-addresses.csv | 100 unique record_id values | Original input; keep unchanged |
| hillsborough-100-e911-comparison.csv | 100 rows: 64 exact, 27 review, 9 no match | Baseline outcomes |
| hillsborough-100-e911-points.geojson | 64 Point features | Exact reference matches |
| hillsborough-100-e911-review-candidates.csv | 79 candidate rows | Possible alternatives, not decisions |
| e911-review-candidates.geojson | 79 Point features | Spatial view of the same candidates |
| hillsborough-100-e911-summary.json | Counts, timestamp, method, source URL | Provenance and reconciliation |
The 36 unresolved input records have 79 candidate rows between them; some have no candidate and some have many. Candidate count is not matched-address count. Six input addresses have multiple matching points, and HC-033 alone has 37 candidates.
The source is the public City of Tampa GIS county address-point service used after county feature requests failed. A Current point is a reference record; it does not independently verify the facility name, unit, entrance, or current occupancy.

This diagram plots the actual coordinates supplied with the exercise. Use it to orient yourself before creating your own map.
Cyan marks represent the 64 exact reference matches. Amber marks represent 79 proposed candidates for review. Some points overlap; the visible symbol count is not a reliable count of records.
There is no basemap, municipal boundary, road network, or unit/entrance verification. The view uses longitude and latitude as a simple data footprint. Do not measure a distance or infer municipal membership from this diagram.

data/ read-only in your workflow. Save your edits under outputs/; use the supplied worksheets/ for decisions.outputs/antbytes_level2.qgz. In project properties, choose relative file paths where available. Save again after adding layers.record_id, ZIP, and source IDs as text; inspect field types before adding.reference_exact_64 and review_candidates_79. Their geographic coordinates are longitude/latitude in WGS 84.worksheets/review-decisions.csv as an attribute-only table. It has 36 unique record IDs and starts with decision = Unresolved.outputs/01_loaded_layers.png.Create groups named Source, Review, Analysis, and Presentation. Keep original geographic layers in Source. Later, put transformed copies in Analysis. A saved project points to files; it does not automatically embed all of them.
Reference: QGIS User Guide, Opening Data [1].

A join can look successful while silently repeating or losing information. Decide which table is the target and whether the join key is unique on the side being attached.
record_id is unique in the comparison table and in the 36-row decision table. It is intentionally repeated in the candidate layer.review_candidates_79, open Properties → Joins and add the decision table. Use record_id for both fields and set the joined-field prefix to review_.reference_exact_64, join the 100-row comparison table on record_id only if you need more attributes. The point layer stays at 64; a join does not create the 36 missing geometries.Capture the join settings and the HC-033 candidate filter. Write the before/after feature count. If a tool regenerates an internal feature ID, continue using the business key record_id for reconciliation.
Reference: QGIS User Guide, Joins and Relations [2].

Use worksheets/review-decisions.csv for QGIS and Python. The branded Excel workbook is a companion for organizing work and self-assessment. If you export Excel data, export only the table header and 36 data rows—not the branded title rows.
| Decision | Use when | Required reasoning |
|---|---|---|
| Accepted | One location is supported by the evidence you recorded | Identify the point, proposed address/ZIP, coordinates, source, reviewer, and date |
| Rejected | You have evidence that the proposed/input association should not be used | Explain what was ruled out; a missing result alone is insufficient |
| Unresolved | Evidence is incomplete, conflicting, or ambiguous | State the open question and the next useful check |
e911_objectid in candidate_objectid. Record the proposed address and ZIP exactly as supported.external:SOURCE_ID, record a retrievable evidence URL and coordinates, and explain the additional verification.decision_notes, evidence_source, reviewer, and an ISO date such as 2026-10-23.
Original: Edward Medard Park — 5726 Partner Lp, ZIP 33567
Candidate: 5726 Panther Loop, ZIP 33567 — object ID 359455
This looks like a small spelling error, but accepting a similar street blindly is a habit that fails on less obvious records.
record_id = 'HC-009'. Confirm that the proposed point came from this record’s candidate list.5726 Partner Lp in the input field.A before/after note, the candidate ID, a source citation, and a decision that follows your actual evidence. The supplied candidate is not an instructor-approved correction.

Original: 11020 Peppersong Dr — ZIP 33569
Reference candidate: 11020 Peppersong Dr — ZIP 33578
The normalized street text agrees, but the ZIP differs. A match that ignores that difference would overstate the original exact-match rate.
33578 as the proposed ZIP while retaining 33569 as the original input ZIP.A postal city, ZIP, municipal boundary, planning area, and service area answer different questions. Do not infer Plant City municipal membership from a Plant City mailing label.

HC-033: 2401 19th Ave NW has 37 candidate rows. HC-081: the input requests 7606 Paula Dr Suite 120; the supplied candidates include a base address and suites 110 and 139.
review_, filter to the selected object ID—not just the Accepted decision."review_decision" = 'Accepted'
AND to_string("e911_objectid") =
to_string("review_candidate_objectid")
HC-093 input: 10020 W Bridgeton Dr or 9791 Westch
Baseline: Input needs cleanup. No candidate was automatically accepted.
external:ID with evidence. The included checker can assess field completeness; it cannot verify that external source or your research.Rank unresolved cases by the consequence of a wrong location, the evidence already available, and the cost of the next check. State whether the next step is source correction, facility confirmation, unit verification, or a fresh reference lookup.

outputs/antbytes_level2.gpkg, name the layer reference_exact_2237, and select EPSG:2237 (NAD83 / Florida West, US survey feet).5280 for an approximate one-mile planar radius. Save site_buffer_1mi in the GeoPackage.Changing only the project display CRS can change how layers appear together; exporting a transformed analysis layer makes the chosen coordinate system explicit in the deliverable.
Reference: QGIS vector geometry tools and export workflow [1, 3].

Make a small audit table keyed by original input ZIP. Keep separate columns for baseline outcomes and your reviewed decisions. Retain all 100 original records in the overall denominator.
| Metric | Calculation | What it does not establish |
|---|---|---|
| Baseline exact reference-match rate | 64 / 100 = 64% | Independent positional or facility accuracy |
| Baseline review queue | 27 review + 9 no match = 36 | That all 36 are invalid |
| Reviewed reference coverage | (64 + newly accepted review IDs) / 100 | A revised original exact-match rate |
| Original-ZIP match rate | Baseline exact matches for that input ZIP / input count for that ZIP | Corrected-ZIP totals or municipal membership |

Produce one overview map and, if useful, an inset for an unresolved cluster. Distinguish an exact reference point from a proposed candidate visually.
outputs/antbytes_level2_map.pdf. Open it and inspect text size, clipping, legend categories, and source notes before considering it finished.Reference: QGIS Print Layout and PDF output [4]. The included footprint diagram is a data overview, not a basemap, boundary, or completed student map.

The included checker uses Python’s standard library and makes no network requests. It checks your CSV structure and decision evidence fields, then writes a JSON audit report.
python checks/check_submission.py worksheets/review-decisions.csv --output outputs/qa-report.jsonpython3. Read the terminal summary and the JSON output.external:ID. They require a source URL and complete location fields. The script flags them for manual verification; it does not confirm the external source.--final. It fails if required review fields are incomplete or structural errors remain. Properly documented Unresolved cases can pass.Keep the QA report with the submission. Do not edit the original dataset simply to make a structural check pass. Fix the decision log or explain a remaining limitation.

Submit a folder another analyst can open without guessing where the data came from or which records were accepted.
| Deliverable under outputs/ | Minimum content |
|---|---|
| antbytes_level2.qgz | Project with working relative references and clear layer groups |
| antbytes_level2.gpkg | Reference layer, selected accepted-review points where supported, and projected analysis/buffer layers |
| review-decisions-final.csv | All 36 review IDs; original input preserved; evidence and dates complete |
| qa-report.json | Final checker output with no structural errors or incomplete rows |
| antbytes_level2_map.pdf | Readable AntBytes-branded layout with sources and limitations |
| README.txt | Purpose, files, baseline timestamp, CRS/transformation, method, counts, unresolved issues |
If you create a combined reviewed point layer, merge the 64 baseline reference points with only the selected, supported review points. Preserve record_id and an origin/status field. Additional-reference points need an equivalent schema and provenance. Do not append all 79 candidates.
Each original input should contribute at most one accepted output feature for this capstone. Count distinct record IDs after merging; shared coordinates alone do not mean duplicate business records.

| Criterion | Points | Full-credit evidence |
|---|---|---|
| Data contract & IDs | 20 | Original 100 inputs preserved; 36 decision IDs unique; counts reconciled |
| Decision reasoning | 30 | All review rows documented; units, aliases, ZIPs, and competing points handled with cited evidence |
| CRS & analysis | 15 | Actual coordinate transformation; units and buffer parameters documented; correct interpretation |
| QA & statistics | 15 | Final structural checks pass; denominator and baseline/reviewed metrics clearly separated |
| Map communication | 10 | Readable symbols, sources, legend, dates, AntBytes styling, and limitations |
| Reproducible handoff | 10 | Project opens after moving; files and decisions can be traced from README |
Award up to half the points when a criterion is partly demonstrated but important evidence is missing. Give full credit only when the required artifact supports the claim. Use the workbook’s Rubric sheet to record the score and a brief justification.
Suggested learning target: 80/100 with no critical errors. This is a self-assessment threshold for the tutorial, not an accredited certification or approval for emergency operations.

Answer in your own words and cite one of your project artifacts where possible. Each prompt is worth 5 points in this optional 30-point check; it is separate from the capstone’s 100-point rubric.

| Prompt | Expected reasoning |
|---|---|
| 1 / Join behavior | A record-level Accepted value is inherited by every candidate with that record_id. Filter by both Accepted and the selected candidate object ID; ensure one output per accepted review ID. |
| 2 / ZIP denominator | Use original input ZIP 33569 for original-data performance. A corrected-ZIP distribution is a different, separately labeled output. Preserve the input value. |
| 3 / Missing suite | Keep Unresolved unless independent evidence supports the requested unit. Check the facility/unit record or an appropriate official source; another suite is not a substitute. |
| 4 / Coordinates | On-the-fly display transformation is not proof that stored coordinates were transformed. Export a projected analysis copy, verify its CRS/units, and record the transformation. |
| 5 / Counts | Baseline exact matches: 64/100 = 64%. Reviewed reference coverage: (64+10)/100 = 74%. The open queue is 24; the two Rejected records remain in the original denominator. |
| 6 / QA limits | Not ready. Structural field/ID checks do not validate facility identity or map claims. Revise symbology and language, then document remaining uncertainty. |
The packet’s proposed candidates are evidence to investigate, not predetermined answers. An Accepted decision and an Unresolved decision can both be well reasoned when based on different documented evidence. Unsupported certainty should score worse than a justified unresolved outcome.

| Symptom | Check first |
|---|---|
| CSV has the wrong feature count | Confirm delimiter, header row, and geometry setting; do not import workbook title rows as data. |
| Joined decisions are empty | Compare record_id text types, whitespace, matching fields, and join direction. |
| Many points appear accepted | Filter by selected object ID as well as decision. Review one-to-many behavior. |
| Points appear far away | Check coordinate order, signs, source CRS, and whether the CRS was assigned or actually transformed. |
| Buffer is much too large/small | Inspect input-layer units, selected features, and radius value; use the projected analysis layer. |
| Checker says rows are incomplete | Fill notes, evidence source, reviewer, and ISO date even when leaving a record Unresolved. |
| Project opens with missing layers | Recheck relative paths and include the referenced files with the handoff. |
AntBytes original instructional material. Data remains attributed to its public source publishers. Use the bundled comparison timestamp when reproducing the exercise; a fresh online query can return different records or IDs. No operational E911 system is part of this tutorial.