Work log and experiment record for the GP12 diagnostics and prognostics thesis, measured against the four DC-1 objectives.
The top row is the measure worth watching. A thesis stalls by one objective going quiet for months while the others feel busy, so the count is days since anything — a log entry or an experiment — was recorded against each one. The bar turns amber past a week and red past three.
Refutations count as output. An experiment that disproves its hypothesis is recorded, tallied and kept exactly like one that confirms it.
Hover over a day to see what you did. Click a day to keep it open.
Raw traces, plots and simulation output stay on disk and in git. Record the path and the conclusion; a dashboard that holds measurement files becomes a bad file manager.
The list was fixed before the work started, which makes the denominator honest in a way self-set milestones never are. Closing one means writing the answer down, so the metric and the thesis material are the same act.
Dropping a question is a legitimate outcome. A question that turns out to rest on a false premise should be dropped with a reason, not quietly left open.
Checked by you: you read the page in the source yourself. Read in source: an assistant read the page and cites it, and you have not yet. Second-hand: it comes from a summary or report and nobody has read the original page. Open: not known yet. Wrong: checked and false, kept so it is not believed again.
A fact needs a source unless it is open. Use “I checked this” only after you have opened the page yourself.
Everything the work depends on, in one place: equipment, sensors, computers, software, data, accounts and documents, whether the lab has it, where it is and who holds it. Items marked need carry a lead time, because buying through the institute takes weeks to months and should start long before the experiment that needs it.
Record where something lives, not the thing itself. Project code and data stay on the lab systems.
Every row is fetched data, never a placeholder. Deadlines come from each venue's own call-for-papers page; literature is arXiv preprints plus journal articles from MSSP, RCIM, the Journal of Manufacturing Systems, Mechatronics and Measurement. A scheduled task refreshes both each morning at 07:00 IST.
Three things are inference rather than source data: topic tags are derived from each title, preprint months from the arXiv ID, and journal years from the ScienceDirect PII, which encodes a year but no month — those rows show a year only.
Triage is what stops this becoming thirty undifferentiated rows. Mark a paper read or irrelevant, and mark a deadline as one you intend to submit to; only then does the feed tell you anything about your own work.
Filter by what it needs first. Most days you are at a desk, and nineteen of these need nothing else — staring at the robot tasks when you have no robot booked is how a queue becomes noise.
A task marked waiting has an unfinished prerequisite. Click the status to move a task along: backlog, next, doing, done. Estimates will be wrong at first; recording the actual hours is how they stop being wrong.
A question is joined to an objective when some task in your plan says it closes that question and serves that objective. Nothing here is drawn by hand — it is derived from the plan, so it changes as you schedule work.
No route means no task in your plan touches that question. That is the useful column: it is where the plan and the question bank disagree about what matters.
There is no citation graph in the data and I will not invent one, so the threads here are the ones you pull. Switch to connect, click one card then another, and the link records what you think the relationship is. Add notes as pins for conclusions that are not papers.
Free keeps where you drag things. Chronology lays them out by publication date, which exposes what was reinvented rather than built on. Logic stacks them by dependency, so what everything rests on sits at the bottom. Only free mode saves positions.
Click any card in move mode to open it: the abstract, your own notes, and a link that opens the file in Drive in a new tab. Both text boxes save back to the repository, so a note written here is in git history like everything else. Dragging a card moves it; clicking it opens it — the two no longer fight each other.
Suggest links proposes edges from shared tags and dates. They arrive dashed and provisional — keep the ones you agree with, click the rest away. The point is that you decide.
Claims carry an evidence mark: assumed, simulated, bench, validated, contradicted. Adding a claim forces the question the abstract avoids — what was actually demonstrated?
Your objectives, questions, work log and experiments live as JSON files in a private GitHub repository. This page reads and writes them directly, so every entry you make becomes a commit. Nothing is stored on any server but your own repository.
robot-phm-workbench-data and push the
starter files into it.It is stored only in this browser, on this machine, and is sent only to api.github.com. It is never committed and never leaves for anywhere else. You will paste it once per browser.
Use one token per computer, named after that computer (for example "workbench · lab PC"), with an expiry of about 90 days. Then losing or leaving a machine means revoking one token, not all of them. On a computer where other people hold administrator rights, untick "remember" above.
Scoped as above it can read and write files in that one private repository and nothing else — it cannot touch your other repositories or your account. If it ever leaks, revoke it on the same settings page and generate another; the worst case is someone editing this log.
Each browser on each computer gets its own id the first time it connects, and every entry it adds or changes is stamped with that id and named in the commit. Nothing has to be listed in advance. Clearing this site's data in a browser makes it count as a new computer; give it the old name again and the history still reads clearly.
Changes made by Claude Code show up under its own name (for example claude-code · lab PC).