Repository navigation
Conversation
alehuo
changed the title
alehuo
force-pushed
the
feature/jev
branch
2 times, most recently
from
October 9, 2026 05:36
fb04384 to
a0dd5ca
Compare
The confirmation dialog is an MUI Dialog. Callers pass a semantic confirmColor (error, warning, primary) instead of a Tailwind button variant, and the icon is optional. LLM and account settings use MUI components and share their tab definitions. A ToggleSwitch wrapper around the MUI Switch replaces the hand-built switches in settings and the provider configuration. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Job cards and the job summary now show "Run #N" (per model, within the project), the start time, and the LLM, evaluation mode, prompting and screening mode as labelled fields with tooltips, replacing the ZS/FS/PC badges. Breadcrumbs use the same name, and the project page lists the newest task first. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The paper card, list header, rows and pagination use MUI components. Header and rows share one column template so long titles truncate instead of widening the page, and the list and side panel stack on narrow screens. MUI Pagination replaces react-paginate, which is removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Projects are rows in one card: the whole row opens the project, and a second line shows the project type, criteria counts and last update. Chips filter the list by type (all, literature reviews, GitHub repositories). Deleting a project now asks for confirmation. Loading and empty states use MUI, and the unused ellipsis dropdown is removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The floating "View logs" pill is replaced by an app bar button with a count badge that opens a bottom drawer, shown and collected only when VITE_APP_ENV is dev. Losing the event stream now shows a warning snackbar, only after an actual error. The page no longer needs bottom padding for the pill. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The footer sits at the bottom of the viewport, aligned with the page content, with a top border and muted MUI links that wrap and centre on narrow screens. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MUI 9's Typography only maps palette names and textPrimary, textSecondary and textDisabled through its color prop; the dotted "text.secondary" form is ignored, so captions and labels rendered in the primary text colour. Use the camel-case values. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The inputs had no value until the schema defaults were filled in, which made React warn about switching from uncontrolled to controlled. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pages inside a project, and the new project page, show their title as "Projects › <name>", with Projects linking back to the list. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A project with no papers opens on one task: uploading them. A large drop zone (or a file button on phones), the CSV columns the server requires, a template download and a progress stepper replace the empty tasks column, the disabled export and the disabled task form. The page waits until the papers are known before choosing a layout, so it no longer flashes one and swaps to the other. "Show evaluation results" no longer appears for a project with no papers. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A project with papers but no tasks shows a "ready to screen" screen: the papers that are in, the progress stepper, and a task form with the method, provider and a searchable model up front. The rarely changed options are summarised as chips behind "Customize", and a summary panel describes what will run, with its token estimate and the start button. The panel stays pinned to the bottom on phones. Once a project has tasks they use the full width, and "New task" opens the same form in a dialog. The provider and model of the last task are preselected next time, and few-shot now needs a model before its modal opens. The old sidebar parts, the Headless UI dropdown and the old drop area are removed. The job e2e test no longer clicks zero-shot, which is the default and now folded away. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Only a 401 from the server now logs the user out. A request that failed on the way, such as one the browser cancelled during a page refresh, says nothing about the session, and redirecting to login interrupted the refresh. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Use the same names for the screening modes in the task form buttons, their tooltip and warning, the option chips and the task labels. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The warning now explains that tasks with the provider can't run, and links to settings with a button. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add PaperList, which puts the toolbar, sticky column headings, rows and pagination footer on one outlined surface, and use it on the papers and job pages. Rows show the inclusion probability as a percentage with a bar, a status such as "Not screened" or "Error" when there is none, and the user's decision as a chip. The footer shows which items are on the page. Replace CriteriaList on both pages with CriteriaPanel, which numbers the criteria IC1, EC1 and so on, and use MUI toggle buttons and switches for the pages' filters. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move the dialog from Headless UI to MUI, with the model suggestions, the paper and the criteria in columns that scroll on their own. Model cards show the decision as a chip, the probability as a bar and the Likert answer as a seven-step meter, with the provider and task setup underneath. The dialog body remounts for each paper itself, so the page no longer keys it. Keyboard shortcuts ignore held-down keys and browser shortcuts such as Ctrl+E, and Escape is left to the dialog. CriteriaList has no users left and is removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move the dialog to MUI with three steps: inclusion examples, exclusion examples and a review. Each step lists the labelled papers as a checklist with a select-all box and their inclusion probability. The task form puts the screening mode in the dialog's URL and exposes its model config, so a refreshed dialog keeps the mode and waits for the form to restore the provider and model instead of crashing. Segmented buttons now set aria-pressed. New end-to-end tests cover starting a few-shot task, refreshing the dialog, starting without a model, remembered examples, select all, and a project with no labelled papers. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move the staggered rise-in helper out of the project folder so every page can use it, and add a fade-in for blocks that appear once loaded, e.g. when content replaces its skeleton. Both stop for users who prefer reduced motion. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add placeholders shaped like the tabs, breadcrumbs, job cards, criteria panel, settings sections and text cards. The page header can show a placeholder title, and a tab can show a placeholder count. The paper list shows skeleton rows on its first load and keeps its rows under a progress bar while it reloads. Rows fade in when they arrive and on each new page. The list and the criteria panel accept sx, so pages can stagger them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The project, papers, job and task result pages show skeletons of their layout while they load, instead of nothing or a single bar. Their blocks rise in one after another when the content arrives. Stepping to another paper's result shows skeleton cards and animates the new result in. A reload of the projects no longer replaces a loaded page with a skeleton. The project tabs stay put while the content below them fades. react-loading-skeleton has no users left and is removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The projects list shows placeholder filter chips and rows only on its first load, so a reload after a delete keeps the list. The results table and the LLM settings get loading states, which they lacked, with skeleton rows and sections. The new project form, account settings and not-found pages rise in like the project page. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The few-shot dialog shows a placeholder while it restores the model selection, and the manual evaluation dialog shows skeleton model cards until the model suggestions load, then fades them in. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The project list now includes each project's job_count, counted by a subquery in the same query, and the single-project endpoint returns it too. The project page shows one skeleton card per task, up to ten, so the page keeps its height when the tasks arrive. Before the projects load it shows three. Regenerate the OpenAPI spec and typed client for the new field. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replace the Tailwind page with MUI cards: a header with the version and links to the source code, API docs and OpenAPI spec; the publications with their DOIs; and cards for supported models, open source and funding. The description now names every kind of screening the tool does, and external links open safely in a new tab. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The job task list, the task result and the evaluation dialog's model results ignored stale answers with a cancelled flag, but let the request run on. They now abort the request when the job, task or paper changes or the view closes, using async/await. The job task service functions take an optional abort signal. The evaluation dialog shows an error when the model results fail to load, instead of its skeleton forever. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Components take the ScreeningTarget enum rather than an isGithubScreening flag, and the task form reads it from the project it already gets. Text that differs by target comes from tables keyed by the enum, so a new target won't compile until it has its own wording. A shared helper gives the item names, e.g. "paper" and "papers", and the DOI-or-repository link that three components each built themselves. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Contributor
|
The whole "Customize" button is missing for Jev so it cannot do Full text screening, one call per criterion or few-shot. I don't know if Jev can do few-shot anyways, but I think full text and per criterion are doable. Other than that, it ran well and got results fast and cheap. |
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR implements paper screening using Jev. ProjectPage component was refactored to accommodate for the different screening architecture of Jev.
The UI of AISysRev is updated to follow Material Design 2 guidelines. It uses MUI components under the hood to easen up the maintenance burden of the app.