DYSARTISANIDEAS, FORGED INTO FORM.Search the episode library ↗

THE INNOVATION FORGE / Forgeworks Grimoire

Compare activity with a Slate Engagement Insights Engine

Compare lifetime, milestone and recent activity in Slate, then test the definitions, units and import steps behind an Engagement Insights Engine.

The Aetherbound card from the Codex Practitioners Tarot collection.
The Aetherbound asks what the visible signals leave unseen.

THE IDEA, IN A MINUTE

A total needs a time window.

The second Forgeworks Grimoire training builds event and website activity measures in Slate. Separate time windows, readable subscores and an export/import workflow make the calculation inspectable before anyone acts on its result.

01 / THE RECORDING

Watch the episode.
Or find your moment.

Give the total a second view

An engagement total carries history with it. Someone who attended events months ago may still have a substantial score even when little recent activity is visible. A new inquiry can have a modest total while actively exploring this week. Read either number alone and the timing disappears.

The second Forgeworks Grimoire training puts several views beside one another: total engagement, engagement as of a milestone, recent engagement and behavior-specific subscores. The Aetherbound guides the episode, with Ghosted Greg as its fictional student persona. The practical question is how to make changes in recorded activity visible without pretending a score explains the person behind them.

In the recording · 05:21 ↗

Define the behavior before assigning points

The campus-visit example begins with a person-based query and Configurable Joins. Form responses supply attendance information; the related form supplies event characteristics. The demonstration narrows the count to attended events, excludes online events and selects an example template. Those choices define what the number actually means.

Scope matters. An institution may record responses against people, applications or both. Counting only person-scoped responses will miss qualifying application-scoped activity. Before adding the results, check that the two paths do not count the same occurrence twice. Decide whether two events on one day represent two activities or one visit.

The build also explores distance bands and a cap on points from this behavior. Treat the weights as demonstration choices. Travel depends on access, time and resources, so a campus visit cannot stand in for commitment. Keep missing addresses distinct from measured distances and inspect virtual or other accessible ways of engaging before interpreting a low visit count.

In the recording · 08:00 ↗

Make each date comparison explicit

For the milestone view, the episode compares the event's start time with application submission. It also explores a fallback date derived from an entry term when the submission date is absent. That fallback changes the meaning of the result: it is activity through a chosen deadline rather than activity through an actual submission.

Write that distinction into the field description. Name the timestamp being compared, the cutoff and whether equality is included. Test an event before the boundary, exactly on it and just after it. Test a missing milestone and a missing entry term separately, rather than allowing either to become a plausible-looking date.

The recent view uses the current date minus 30 days. Keep its behavior definition aligned with the lifetime view if you intend to compare them. A trailing-month score is a smaller window, so a smaller number is expected. To describe an actual decline, examine comparable earlier windows or another documented baseline. A lifetime total and one recent value alone do not establish a rate of change.

In the recording · 11:50 ↗

Keep the raw measure beside the score

The website example sums ping duration for a selected URL and discusses turning that duration into bands. The narration identifies the stored duration as seconds, then talks about thresholds in minutes. That is a useful point to pause and verify the actual formula before reusing it: normalize the units before applying any threshold.

Keep raw seconds, the normalized value and the awarded points visible in a test query. Try values immediately below, at and above each boundary. Check whether the URL filter matches the intended page and whether repeated activity is being summed or counted distinctly. These choices answer different questions.

A tracked visit is evidence of recorded activity, not proof of comprehension, enthusiasm or a particular personal concern. The episode also flags source hygiene elsewhere: distinguish direct portal logins from impersonations and identify the sender when measuring incoming messages. A technically valid count can still count the wrong behavior.

In the recording · 25:20 ↗

Test the route back into Slate

The demonstration exports query results to Slate's SFTP location and uses a source format to bring calculated values back into fields. David recommends update-only behavior for existing records and failure notifications. A later calculation can combine the behavior fields into a total.

Check the matching key, field mappings and update behavior on fictional records first. If totals depend on earlier imports, allow those imports to finish before calculating the total. The training recommends staggered delivery windows rather than treating requested priority as an ordering guarantee.

Filename patterns deserve the same care as formulas. A wildcard for an engagement-behavior file can also match a similarly named totals file. Give each import a pattern that selects only its intended files, then test that an unrelated file is rejected. Record when the last successful refresh occurred so an old value is not mistaken for today's observation.

In the recording · 35:44 ↗

Finish with a question the data can answer

The closing Forge Trial proposes comparing total and recent engagement in a small pilot. Start with the fictional cases below and explain each result in ordinary language. Separate the observed behavior, the calculation and the questions still unanswered.

Use the comparison to review data quality and consider useful support. A quiet window cannot tell you why someone stopped appearing in the logs. Respect expressed communication preferences, seek context and keep these measures out of decisions about admission, aid or access. The first useful outcome is a number whose limits the team understands.

In the recording · 1:01:18 ↗

03 / A SMALL EXPERIMENT

Test three engagement windows

Use these fictional cases to inspect definitions and boundaries. Write the expected result before checking a query.

One total, two histories

Two fictional records each have 40 lifetime points under the same rules. One earned 20 in the last 30 days; the other earned none. What can this comparison establish?

Compare your reasoning

It distinguishes recorded activity within the recent window. It does not explain motivation or establish a decline without a comparable prior window. Check source coverage and refresh time.

Revisit this moment · 31:53 ↗
An event at the cutoff

A fictional application was submitted at noon. Qualifying events began at 11:59, noon and 12:01. Which enter an as-of-submission count? What changes if submission is missing?

Compare your reasoning

Under an inclusive cutoff, the first two qualify. Verify the actual operator and timestamp precision. A missing submission needs a documented fallback or missing result, not an invented milestone.

Revisit this moment · 11:50 ↗
Correct input, wrong interpretation

A fictional ping lasts 300 seconds. Your scoring notes say five minutes. Separately, an import pattern for behavior files also matches a totals filename. What needs testing?

Compare your reasoning

Confirm the conversion and exact five-minute boundary before awarding points. Narrow or rename the file pattern and verify that only the intended behavior file imports.

Revisit this moment · 27:06 ↗
Build a test card to take with you

Notes stay in this tab. Reloading clears them.

04 / INSIDE THE STUDIO

The seams are part
of the story.

Behind the episode

The live build leaves its provisional choices visible. At 41:19, David uses a subscore for the engagement-score mapping and says a specific calculation may need its own export. At 44:13, he warns that a wildcard can ingest both behavior and totals files when their names share a prefix. These moments make useful review checks: finish the total calculation and test file selection before treating the demonstration as a working pipeline.

Revisit the studio moment · 41:19 ↗

Engagement Insights Engine

Song 02 of The Forgeworks Grimoire, published separately as the outro for Episode Two.

Watch the standalone lyric video

Art from the workshop

The Aetherbound guides this training's exploration of movement across distance and behavior.

Open the artwork ↗

KEEP THE IDEA MOVING

Where would you take it next?

CONTINUE EXPLORING