A practical guide to defining observable performance, building meaningful proficiency levels, and bringing your rubric into Airtable.
Dr. Aurelia O’Neil
10
min read

TL;DR
Start with one observable workplace skill and define what independent performance looks like. Build distinct proficiency levels, specify the evidence and scoring rules, and pilot the rubric with two trainers. Then use Airtable to collect observations and support feedback without confusing a rating with proof of readiness.
Two trainers watch the same person complete a task. One sees someone ready to work independently. The other sees someone who still needs support. Both may have paid close attention. What they have not necessarily shared is a definition of readiness.
That is where I would begin an on-the-job rubric: with the judgment people are already making, and the evidence they need to make it fairly. Before choosing a scale or opening a database, ask what a trainer would need to see, hear, or inspect to explain their decision to the person being assessed.
The approach below follows the sequence in my Build Better Rubrics workshop materials. Start with a skill, define the standard, describe the progression, and test the rubric before bringing it into Airtable. The digital tool comes later, once there is something worth recording.
1. Decide what you are trying to observe
Knowledge, tasks, and skills are related, but they call for different evidence. Someone can explain a procedure without yet being able to carry it out. Someone can finish a task while relying on a trainer to catch the errors.
A knowledge question checks what someone can explain. A checklist can show whether required steps were completed. A skill rubric describes how well someone performs against a defined standard. In practice, you may need all three.
For a first rubric, choose one to three skills that matter to a real piece of work. Write each skill as an action, its object, and the standard it must meet. “Understands documentation” leaves the observer with little to assess. “Records a service request with the required information, category, and next action specified in the team procedure” gives the observer somewhere to look.
Use an approved procedure, work specification, or agreed example to establish the standard. Ask someone who performs the work and someone who trains it to check the statement. If they disagree about what successful work looks like, resolve that disagreement before turning it into a rating scale.
2. Define readiness before the other levels
I find it useful to write the independent-performance level first. Ask: what must be true before someone can do this work without coaching? “Without coaching” should still allow the procedures, reference materials, tools, and approved accommodations normally available in the role.
The workshop uses four levels:
1 · Beginner: practices with step-by-step guidance and checks throughout the task.
2 · Developing: completes routine work with limited prompting and a trainer’s review of the finished result.
3 · Proficient: meets the defined standard independently within the agreed scope, checks the result, and handles or escalates problems appropriately.
4 · Master: sustains proficient performance and demonstrates an agreed extension, such as handling more complex cases or coaching others.
These labels are a starting structure, not a universal classification of people. Decide which changes matter for the skill: the support required, consistency across opportunities, complexity of the work, or ability to help someone else. Do not add teaching as a requirement for every job merely because it appears at the top of a template. If coaching is a separate responsibility, it may deserve its own skill row.
“Sometimes” and “usually” need an agreed meaning. “Every time” cannot be established from a single observation. Define the number and range of observations needed for a readiness decision with the people accountable for the work. More consequential tasks require an evidence plan suited to their risk.
3. Turn the standard into visible evidence
For each skill, list three to five actions or results a trainer can observe. Keep the list focused enough to use while work is happening. The behaviors direct the trainer’s attention; the level descriptors explain how that evidence supports a rating.
Here is a fictional, low-risk example you can adapt. Its requirements are illustrative, not a substitute for your own procedure.
Skill: Record a service request so another team member can take the correct next action, following the team’s documentation procedure.
What the trainer looks for:
Captures each required detail from the request.
Selects a category supported by the information provided.
Records the next action and responsible team.
Checks the record for missing or contradictory information before handing it over.
Write Level 3 first, then work outward:
3 · Proficient: Records all required details, selects the supported category, and documents the next action without prompts. Checks the record and resolves missing information or escalates it through the agreed route before handoff.
2 · Developing: Records a routine request and documents its next action. May need a prompt to identify missing information; corrects omissions identified during the trainer’s review before handoff.
1 · Beginner: Records the request with step-by-step guidance. Needs the trainer to identify required details, guide category selection, and check the next action before handoff.
4 · Master: Meets the proficient standard on the agreed range of complex requests, explains category decisions to a colleague, and helps the colleague identify and correct documentation gaps.
Notice what stays steady: the same work and standard appear throughout. The support and scope change. A descriptor such as “produces excellent documentation” would force trainers to supply their own definition of excellent.
4. Write the decision rules beside the rubric
A useful rubric also explains what to do when the evidence is incomplete. Include “Not observed / insufficient evidence” as a separate response. An unobserved skill is not a beginner rating, and a blank field should not silently become a zero.
For this starter approach, select the highest level whose required conditions are supported by the observation. If performance falls between levels, record the lower supported level and explain the gap. If the task did not provide enough opportunity to assess the skill, record insufficient evidence instead of guessing.
Record what happened, how much support was provided, and the context. “Trainer prompted the learner to add the responsible team before handoff” is more useful than “needs confidence.” Share the criteria before the observation and use the resulting notes to agree on the next practice opportunity.
Where safety or critical quality requirements apply, identify them explicitly and state how a missed requirement changes the decision. Have the responsible subject-matter expert approve those rules. Do not let an average across unrelated skills conceal a missed critical requirement or automatically authorize independent work.
5. Check the rubric, then test it with people
The workshop’s eight-point check is a practical review aid. Ask a colleague whether each row has:
One clearly defined skill.
Evidence that can be observed or inspected.
A meaningful progression between levels.
Differences specific enough to distinguish adjacent levels.
Consistent comparisons of the same work.
Explicit expectations instead of vague quality labels.
Rules for critical requirements, where relevant.
Clear scoring instructions, including insufficient evidence.
The workbook suggests seven or eight positive checks as readiness to pilot, five or six as a signal to revise, and four or fewer as a reason to rebuild the levels. Treat these as workshop heuristics, not proof that a rubric is valid. A missing critical requirement still needs attention whatever the total.
Next, ask two trainers to rate the same sample independently. Compare the evidence they used before comparing their numbers. Where did their judgments diverge? Did one infer something the other could not observe? Did a descriptor combine two different standards?
Revise those passages and try again with another sample. Include the people being assessed in the pilot: can they understand why a rating was given and what practice would help? Keep a version number and review date so later observations remain connected to the rubric actually used.
6. Build a small, usable version in Airtable
For a pilot with up to three skills, the workshop uses two tables: a Rubric Library for the definitions and Observations for the evidence and ratings. A form collects observations; a grid view and an interface help people review them.
You can explore the OJT Rubric example in Airtable. The shared base includes Rubric Library and Observations tables and a Copy base option. Use a separate copy for practice. Keep real employee information out of public demonstration bases.
Create the reference table
Create a base named Skills Rubric and a table named Rubric Library. Use a single-line text primary field called Skill. Add long-text fields for Skill Statement, Observable Behaviors, and each of the four level descriptors. Add Rubric Version and a review date. Enter one record per skill.
This is your maintained reference. In the simple setup below, copying descriptors into another table does not create a live connection. If a descriptor changes, you must deliberately update the form options too, while preserving the meaning of earlier observations.
Create the observation table
Add an Observations table with these fields:
Observation ID: an autonumber primary field.
Employee and Trainer: single-line text for a fictional pilot; use controlled identifiers or linked records when scaling.
Date Observed: a date field.
Rubric Version: the version used for this observation.
Task Context and Trainer Notes: long-text fields for the work attempted, prompts given, evidence, and next practice step.
One rating field per skill: a single-select field with the four complete descriptors, each beginning with its level number, plus “Not observed / insufficient evidence.”
For example, call the first rating field Documentation Rating. Paste its observable behaviors into the field description so trainers know what to attend to. Keeping the full descriptor in the choice helps prevent the number from becoming a substitute for reading the evidence.
Build the trainer form
Use Airtable’s Forms area to create a form with Observations as its source table, and name it On-the-Job Rubric. Airtable recommends this form builder for new forms; older bases may still show legacy form views. Add identity, date, version, context, ratings, and notes. Put each skill in a clearly labeled group, with behaviors visible as help text. Preview the form before sharing it.
Give trainers a short instruction: observe the work, read every descriptor, select the supported level or insufficient evidence, and record the evidence and support provided. Require the identifying fields and ratings, while keeping the insufficient-evidence option available. Test the form on the device trainers will actually use. See Airtable’s form-builder guide for current controls and plan-dependent options.
Make a report without losing the meaning
Create a grid view of Observations called Skills Report. Group by Employee and sort by Date Observed, newest first. A view reorganizes the same records; it does not create another table.
Add a formula field called Documentation Level. With rating options beginning 1, 2, 3, or 4, this formula returns a number only for those four prefixes:
Blank and insufficient-evidence responses remain blank. Repeat for other skills using their exact field names. Test all four choices, an unobserved choice, and a blank record before relying on the report. The syntax follows Airtable’s formula reference.
Under Interfaces, create a Dashboard connected to Observations. Start with an observation count and a chart showing the distribution of ratings for one skill. Provide a detail view containing dates, context, and notes so a reviewer can examine the evidence behind a pattern. Airtable’s Dashboard guide explains the available layouts and controls.
Label summaries carefully. A count measures observations, not people qualified. A chart containing every historical rating does not show each employee’s current proficiency. Keep skills separate and inspect recent evidence rather than treating an average of ordered levels as a precise measure of capability.
Finally, submit fictional test observations, check that they reach the correct fields, and verify the formula outputs and dashboard filters. Set access deliberately before collecting real data. Give trainers and managers only the access their work requires, and check what a form or interface recipient can actually see. Do not publish employee records through a public share link.
Let the next conversation improve the rubric
A small pilot can reveal more than a large collection of untested criteria. Start with one skill, one task, and a pair of trainers willing to compare their judgments. Watch where they hesitate. Ask the learner whether the feedback gives them something concrete to practice.
As the system grows, you may want one observation record per employee, skill, and occasion, linked to a versioned rubric library. That structure is easier to extend than adding a new column for every skill. The small two-table design is a useful place to learn what the workflow needs before making it more elaborate.
The question to carry through the whole process is straightforward: could another person follow this evidence and understand the decision? If the answer is uncertain, return to the description of the work. That is where the rubric earns its usefulness, long before the first chart appears.