Evidence You Can Try: A Workstation Becomes an Open Workshop
A Phase 1 progress report — and a few small experiments in planning, business intelligence and accountable software.
Don Gunter, G-kyu Tech — developed with Codex. October 26, 2026.
A useful article can introduce an idea. A working tool gives the reader somewhere to test it.
That is the step we have taken with our Workstation project. Beside the descriptions and screenshots, we can now offer small browser applications: open one, change a selection, arrange an appointment, inspect the records and decide whether the behavior supports the explanation.
The public workshop is Workstation X. G-kyu Tech develops the tools; Avia Integrated Management Solutions, AIMS, is our services and licensing partner. These examples use supplied synthetic data. They are early, bounded demonstrations in Phase 1 of a five-phase rewrite.
We have a better frame for measuring progress now. Three questions belong beside one another:
An increase in application breadth is one kind of progress. Completing an operating-system dependency cut is another. Keeping those accounts separate helps us make better decisions about the next hour of work.
The browser demonstrations run independently of Workstation libraries. That establishes a useful delivery target. It does not establish that the native C++ engine has been ported to every browser, phone or Unix host. Our migration work has qualified selected Windows contracts and bounded Linux header syntax checks; a complete linked Linux engine and Android device acceptance remain separate work.
For readers who enjoy learning by trying, here are three small experiments.
First experiment: move the appointment, then inspect the work.
Open the Daily view on the sample date, October 26, 2026. Choose a task that matters to you. Its A/B/C priority and order belong to a dated plan. Its description belongs to a work item. Its appointment supplies a participant and a start/end interval.
Click + Task, enter some text, and optionally create an appointment in the same draft. Apply it, then switch to Sequencer. Both views select the same underlying working records.
Before dragging a clip, predict what should change. Then inspect the result.
Moving the title changes that appointment’s time or participant lane. Resizing changes its scheduled interval. The work item’s text, original owner and other appointments retain their identities. One work item can have several appointments, including appointments in different people’s lanes.
Explicit Delegate changes default responsibility under the supplied demo policy. That is a different operation from moving a clip. Cancel appointment removes a clip from the active view while retaining its history and work item.
The little game is to discover how many of these relationships you can explain before using the controls. A planning priority, an appointment, delegated responsibility and actual completed labor answer different questions. The current demo helps make the first three visible; it does not post payroll or actual effort.
You can also move a dated TODO to another day without moving its appointments. That separation lets the user reconsider a plan while preserving the commitments already on the calendar.
Use Save file to download the complete browser state and Load file to recover it. Reopening alone restores the sample. This edition has no automatic save or live connection to a private workspace.
Second experiment: change the question while preserving the source.
Set Item membership to All items, Group by to category, Measure to quantity, and compare Hour with Day. Then open Records / evidence.
Get Don Gunter’s stories in your inbox
Join Medium for free to get updates from this writer.
What happened to the source movements? What happened to the buckets? Did the category grouping preserve the distinction between quantities measured in pounds and quantities counted as each? What did a return contribute?
The intended behavior is that controls change a query projection while preserving the supplied movement records. Unit-separated grouping and signed returns remain part of the meaning of the result. A smooth line is useful only when we can account for the values it represents.
Save a query, change it and load it again. You have preserved a way of asking a question. That method can be useful independently of the particular picture it first produced, provided a later source satisfies the same schema and measurement contract.
This is one reason I care about small internal symbols and reusable expressions. A recognized identifier can refer to a larger structure. Different domain words can select common operations. Factoring, lookup and carefully managed memoization can reduce repeated work. Each still needs an identity, context and rules for when its result remains valid.
The current build selects admitted recipes and implementations. It has not become a universal language learner or a compiler for arbitrary unknown domains. We can extend its expression without pretending every possible meaning is already understood.
Third experiment: follow an observation without executing it again.
The Business Sandtable has its own page. Its location and workflow replay lets you select a route query and step through retained occurrences. Its office reconciliation report presents captured results from the fictional ACME environment.
Choose a route, move the occurrence control and inspect what the view says was observed. Ask which location, actor and accounting boundary are involved. Ask what additional evidence you would require before authorizing a real action.
These public pages select recorded simulation results. They do not dispatch resources, send payments or run the fuller local financial services. That boundary is part of the lesson: playback can help us understand a process while keeping observation separate from execution.
There is a related Workspace Observer. Select a context and compare its captured query results.
The relationship between these examples is the useful part. A query, a cursor, a schedule, a record and a presentation can cooperate while retaining distinct responsibilities. An application can select an existing capability and give it a suitable view. Different targets still have to prove their own compatibility.
For a fourth way into the workshop, Audio Dynamics offers a compact scalar signal study. Change an admitted compressor, expander or compander parameter and observe its transfer projection.
That breadth gives us more places to look for contradictions. The latest engineering round found a planner-only build still invoking an unrelated generator, and repeated template substitution reinterpreting marker-shaped text inside supplied data. We corrected both. We also moved qualification expectations into a target table so the common process could exercise the planner alongside the existing applications.
The latest shared qualification records 51 portable tests and nine browser runs containing 287 assertions. Eight runs were offline, with application network requests blocked. All seven nonempty selections of the three applications were checked, along with selection-order invariance. The 13 application/package/start-page payloads remained byte-identical to the retained published release while the build’s source provenance changed.
Those numbers describe this test scope. They are not a percentage of Phase 1 completion, and assertion counts should not be added across overlapping historical suites. The earlier publication separately checked 96 public artifacts, including retained older release links. Owner acceptance, platform migration and production security have their own evidence requirements.
For the illustrated release, you can inspect the public Time Management composition contract and release build receipt. They describe the shipped state boundaries, selected products and artifact hashes. The later generator-stability qualification is a separate record; an older release receipt does not acquire new test results merely because the application bytes were preserved.
We can now measure unnecessary generation, retained compatibility and specific changes more clearly. We have not measured thermodynamic balance, hardware energy savings or a general speedup from memoization. A reduction in avoidable work is a direction to investigate; an energy claim would need its own measurements.
Elizabeth Magie offers a useful historical reference for this style of teaching. Her 1924 Landlord’s Game patent describes an educational purpose alongside amusement: using rules and play to illustrate her argument about land tenure and a proposed alternative. The lesson I take is that people can inspect consequences through participation. Our software experiments are our own application of that practice, not a reproduction of her game or proof of an economic theory. Magie’s patent, US1509312A.
I would like these demonstrations to give readers that kind of room. Make a prediction. Try a control. Inspect what was retained. Bring a different interpretation, and tell us what evidence would distinguish it.
The idea is to make the method clear enough to recognize, the record durable enough to revisit, and the workshop open enough for another person’s ideas to grow.
Visit: Workstation X · Business Sandtable · Demo links and QR sheet.
The supplied IP-address HTTPS endpoint currently has a certificate issue that may produce a browser warning. These demonstrations request no sign-in, payment or private customer information. Downloadable standalone packages are available from the workshop. G-kyu Tech products are licensed through AIMS.
