Monday’s map is only as good as Friday’s packet. If the CRS, the capture date, and the hole in the south line are not written down, the next person will invent them.
Why this matters
Most mapping failures are not in the algorithm. They are in the interval between the field and the office: the wrong mission file, last week’s dictionary, a mosaic named final_final2, a rover that was still on the phone’s location service, a product sent without saying the east lots were not flown.
Grok is excellent at blank forms. It is reckless at filling them. This lesson is the packet you should have been writing already — flight or collection log, data dictionary, processing note, and a one-page “what this map answers.” The capstone will attach a comms SOP and an analysis brief to the same job. Today you make the spine.
Teach
A field-to-office packet has four sheets. They can live in one document. They cannot live in someone’s memory.
1. Collection log
- Job name and product promised (ortho, elevation, rover points, scout photos, as-built).
- Who collected, who was PIC or lead, who processed.
- Date, local start/stop, site in plain language.
- Platform / instrument (aircraft + sensor, rover + rod height, phone — if it was a phone, say phone).
- Planned CRS and the CRS actually on the collector. If you do not know, write
UNKNOWN — do not deliver.
- Control: GCPs yes/no, which set, check shot against a known point (pass/fail + residual if you have it).
- Weather or field conditions that affect the product (low sun, standing water, dust, canopy).
- What was not collected.
2. Data dictionary (if you collected features)
- Feature class or layer name.
- Field names, types, domains (the allowed values).
- Required vs optional.
- What “done” means for a point (photo required? two offsets?).
- Version of the dictionary. Last week’s
status domain is how you get three spellings of “complete.”
Grok can draft a dictionary from your list of fields. It must not add cute fields you will never collect. Extra fields are how crews skip the ones that matter.
3. Processing note
- Software and version.
- Inputs (raw names, not
that folder).
- Outputs and their names.
Site_YYYYMMDD_ortho_v1 beats final.
- Ground sampling distance you computed, or
not computed.
- Whether you have a DEM / DSM, and which.
- Known artifacts: seam, hole, moving truck, reflective roof, standing water.
- What you did not do (no color-balancing, no hydro-flattening, no official flood overlay).
4. Map contract — what it answers, what it does not
Two short lists. “This product can show the wet signature on the south line as of Thursday.” “This product cannot set a finished floor, a spray buffer, a property line, or a conductor clearance.” If you skip this sheet, someone else will use the pretty mosaic as a survey.
Versioning. Do not overwrite. v1 is the first deliverable. A reflight is a new date. A new process is a new version. The packet says which file is current.
Empty cells stay empty. The prompt you will reuse:
Draft the packet structure. Fill only facts I typed.
Every other cell is NEED or NOT COLLECTED.
Do not guess a CRS, an EPSG, a GSD, an acreage, a county, or a software version.
Do not write "assumed NAD83" or "typical 1.5 in/px."
What the office must not assume. This line is the whole point of after-action. “South lots not in the mosaic.” “Rover was WGS84 geographic; do not drop into the state-plane feature class without a transform we name.” “No GCPs; relative only.” If you are ashamed to write it, the superintendent still needs it.
Ag and utilities. Spray or scouting packets add the field ID the producer actually uses, the growth stage you observed (not Grok’s crop ID), and whether a labeled buffer was your observation or a legal requirement you have not checked. Corridor packets add milepost or structure IDs you recorded, not IDs the model inferred from a thumbnail.
Students and shops. If your “office” is a classroom server, the packet still exists. The next lab cannot start from ortho.tif on the desktop with no date.
Worked example
Thursday flight, same development. Friday morning the GIS lead opens a thread named ops packet — prairie edge thu.
They paste: two-person crew, names as functions only, 10:12–11:04 local, tablet mission prairie_edge_v3, raw on the card DCIM/100MEDIA, intended CRS on the processor (they type the one they use; they do not ask Grok which to pick), no GCPs this flight, cattle delayed the south line, last three lots not flown, processing in the photogrammetry package they name and version, output PrairieEdge_20260409_ortho_v1, GSD they copy from the report, visible seam on the gravel spur, standing water in the slough.
They tell Grok to build the four sheets and leave holes as NEED.
Grok adds “EPSG:26914” because it likes South Dakota examples, and “1.2 cm GSD” because that is a brochure number. The lead deletes both. The processing report already had a GSD. The EPSG is theirs to confirm in the project, not the model’s to gift. Grok also wrote “site is suitable for grading design.” The lead replaces that with the map contract: not a grading surface; no DEM delivered in v1.
The packet that leaves the shop is one page plus the dictionary they already had (v3, unchanged). The superintendent gets the map contract, not the raw log. The log stays with the files.
Your turn
Open grok.com. New thread named ops packet — [job]. Use a collection you have run, or the next one on the calendar. If you have a processing report or a collector screenshot, attach a redacted crop. Strip client names if you need to.
Exercise d-geo-l03-e1
| Field | Value |
|---|
| whereToDo | grok.com |
| time | 22 min |
Copy-ready prompt
I am in Grok Mastery (Russell Tech), Pathway D, Geospatial, lesson 3 — ops documentation.
Interview me with exactly eight short questions, one at a time:
1) Product promised
2) Who collected / who is processing
3) Date and roughly how long
4) Instrument or platform
5) CRS I actually used, or UNKNOWN
6) Control: GCP / check shot / none
7) What we did not collect
8) Software I process in, or "not processed yet"
Then write an Ops Packet with four sheets:
- Collection log
- Data dictionary (only if I collect features; otherwise write "n/a — imagery/raster product" and do not invent fields)
- Processing note
- Map contract (answers / does not answer)
Use NEED or NOT COLLECTED everywhere I did not give a fact.
No guessed EPSG, GSD, acres, county, or "typical" accuracy.
End with:
- Three things the office must not assume
- Session stamp: today's date and "Ops packet written by Grok in this thread."
Artifact to paste
- The Ops Packet (four sheets, office-must-not-assume, stamp).
- Your eight answers.
- One line from you: the cell you refused to let Grok fill, and the true value or UNKNOWN.
Rubric
- Pass: Packet is Grok-authored and stamped. Empty facts are NEED/UNKNOWN/NOT COLLECTED, not guesses. Map contract forbids at least one misuse (survey, flood, clearance, spray, grading). Job is real. CRS is what you said, not a volunteer EPSG.
- Fail: A template with fake numbers. You typed the packet offline. “Accuracy ±2 cm” with no control. Dictionary full of fields you do not collect. No map contract.
Common failure
The common failure is naming the file final. The next flight overwrites it. The packet points at a ghost. Use a date and a version. Boring names are how shops stay honest.
The other failure is letting Grok pick the CRS because you did not want to look. Wrong CRS is a silent shift. Write UNKNOWN and go look. Delivering UNKNOWN is better than delivering the wrong zone with confidence.
What’s next
You have comms, analysis rails, and a packet spine. Capstone: one field SOP and one analysis brief for the same real job — tied to this packet so Monday can run it.
