Repository navigation
feat: simplify profiling results and model detail pages #25
Description
Overview
Simplify the profiling dashboard, open model details in separate tabs, and show a focused set of preserved profiling results. User approved this plan with “go”.
Plan
Simplify profiling results navigation and presentation
Type: feature
Difficulty: medium
Consequence: judge
@PyAutoPulse
Implement the requested dashboard presentation changes without deleting captured results or changing measurement producers. Use the workspace development route. Plan approval pending.
Implementation plan awaiting approval
- Simplify the landing board and standardize all requested display names.
- Open each model in a dedicated browser tab with a stable, reloadable URL.
- Put Instrument, Device and Configuration selectors together; show Configuration details before measurement panels. Retain the lower measurement navigation and remove duplicate buttons and browse steps.
- Restrict normal dashboard choices to float64 and 1500 source pixels where source pixels apply, choosing explicit default variants for remaining settings. Preserve all raw captures. Missing matching results get an honest empty state, not a silent nonmatching fallback.
- Use compact bars sharing an appropriate timing scale, with the matching total likelihood as denominator where recorded. Keep incompatible units, batch semantics, hardware and configurations separate; never invent totals or join unrelated runs.
- Remove the specified redundant prose and sections, and split result links from script links.
- Verify rendering, filtering, navigation and bar scaling with Python and Chromium checks; regenerate the board from retained evidence.
Tier: judge — merge mode: human /prm
Detailed implementation
Primary repo: PyAutoPulse. Classification: workspace/organ presentation (the Feature CLI classified this unknown organ as library; this task changes no scientific library or public API).
Suggested branch: feature/profiling-results-ui.
Branch survey: canonical PyAutoPulse is clean on main, only local branch main; conflict helper returned 0, no active claim. Mind clean on main before drafting.
pulse/campaigns.py::render_htmland markdown: remove Fix Profiling Systematically and its check-in review line; rename evidence heading and remove its introduction. Retain underlying campaign state.pulse/board.py::render_html: remove legacy capture disclosure; update navigation label. Reuse Brain shared theme without changing shared consumers.pulse/setup_browser.py::render,pulse/setup_browser.jsand browser CSS: central display-name mapping; model links with new-tab targets and URL state rendering a model detail page. Preserve setup identifiers and same-commit shard verification. Keep instrument control dimensions, arrange Device and Configuration beside it responsively.- Replace axis-dependent repeated browsing with direct configuration selection and measurement panels. Move configuration details first; remove overview cards, qualification/method and shared-findings sections; split evidence/script disclosures. Hide only requested presentation, preserving provenance in captured data and linked results.
- Add explicit display eligibility/default selection for precision and source size based on recorded metadata; do not assume unknown precision is float64. Analytic models have no source-pixel restriction. Retain models/devices with an explicit missing-results state where no eligible evidence exists.
- In
panels, replace per-method singleton normalization with compatible common scales. Breakdown uses a recorded matching full-likelihood total when available; otherwise label an honest shared component scale. Runtime uses a common maximum only within comparable measurement semantics. Reduce row spacing and bar height, remove repeated captions and missing-statistic filler. - Update
tests/test_campaigns.py,test_board.py,test_setup_browser.pyandbrowser_setup.cjs: new-tab/reload navigation, selector behavior, default filtering, retained archived data, missing matches, bar proportions/units and responsive overflow. Run required Ruff, Python tests, Chromium checks and offline contract validation. Regenerate published artifacts via the owning CLI without new profiling runs.
Planning evidence
Heart door: STALE, exit 1; development planning permitted. Memory consultation located the completed setup browser/redesign work. Inspection confirms singleton grouping in panels explains full-width bars. No source files edited before plan approval; issue/worktree setup follows approval through start-dev.
Original Prompt
Original request
-
Remove the "Fix Profiling Systematically thing and the Last check-in: 2026-10-04 · review date" underneath it.
-
Remove lens · capture, qualification and legacy diagnostics which feels legacy.
-
In profiling evidence remove "Choose a project, dataset and model. Qualification belongs to each recorded setup."
-
Rename Profiling evidence to Profiling results.
-
rename AutoLens to PyAutoLens
-
I want us to adopt standard captilization and case for all things so:
delaunay matern -> Delaunay Matern
Delaunay natural neighbor -> DelaunayNN
knn -> KNN
mge mass -> MGE Mass
remove "pixelized"
sersic -> Sersic
Inside a thign remove "Browse recorded runs by measurement and device. Runs can have different settings and revisions; they are not a combined benchmark."
I think we should remove the 4 buttons Runtime / Breeakdown / Compilation / Memory, they duplicate the tabs below and the
tabs are more natural to navigate.
Top row should be instruemtn with drop down (same as it is now sizing is good) but then to the rigth should be Device and Confiruaiton (remove / evidence run.)
There are too many things under Configuration that currently I find it hard to navigate. I think for now we should just do
float64 for all devices with 1500 source pixels (and try choose the other most default options for other things)
and we build up. dfont lose other results but avoid them meeting dashboard for now.
Under breakdown this stuff is repetitive or redundant so remove:
Statistic not recorded · repetitions not recorded
Component observations · a100 · gpu · float64 · euclid-ral-gpu-2 · s
Also remove this:
Instrumented component costs are not the full compiled likelihood time. Bars compare values only within this setup and unit.
Component observations · a100 · gpu · float64 · euclid-ral-gpu-2 · s
The breakdown bars are good but they are not relative to the total time so they are all the same size (e.g. full)]
and they should be samller so I can get most on my screen without issue.
I dont think we need these "Browse Likelihood Runs" bottons, they are an unecesary click, all info should be after
we click the drop down but with the extra text remove and smaller bars they will be more visible on one poage.
Agian for likelihood run time all bars are full so we are not seeing sizes relative to some main run?
Cofiguration details, these details are important for all profiling inspection so should be the top option befpore likelihood runt ime.
Remove Qualification and measurement method.
Remove shared component and method findings
Split "Profiling evidence and scripts" in "Profiling Results" and "Profiling scripts".
I think when we click "Delaunay" (or any button under imaging or another thing) it should open a new tab with all the
above as a new page, theres a lot of info and we will ultimately want to compare run times across settings so want
new tabs in my browser anyway!
Activity
Jammy2211 commentedon Oct 7, 2026
Implementation and local validation are complete on
feature/profiling-results-ui, at/home/jammy/Code/PyAutoLabs/tmp/worktrees/profiling-results-ui/PyAutoPulse. Source changes remain uncommitted because the current Heart shipping gate is RED.193 Python tests, Chromium interactions/responsive checks, Ruff lint/format, offline contract validation, and
git diff --checkall passed. Full captured Delaunay/A100 preview verified with exact-commit shards. No raw results, captures or profiling producers changed. No independent review is claimed; this human-approved task does not require the autonomous review gate.The prepared PR body is
../pr-body.md; logs are../pulse-tests.log,../browser.log,../offline.log,../ruff.logand../preview.log. Screenshots aretmp/browser/delaunay-a100.png,model.pngandlanding.pngin the worktree.Heart RED reasons (verbatim):
Next step: obtain the live human development-only shipping override required by Brain AUTONOMY.md, record it in all four required sinks, then commit/push/open the prepared PR. Merge remains a separate human action.
Jammy2211 commentedon Oct 7, 2026
PR opened
868d929, branchfeature/profiling-results-ui.User authorization: “I authorize, maybe we should get those repos up to date first?” Verified PyAutoFit, PyAutoArray, PyAutoGalaxy and PyAutoLens clean on main and equal to fetched origin/main. No pulls were necessary. Refreshed Heart cleared all four previous RED reasons before source commit/push/PR; no RED override was exercised.
Current Heart is YELLOW:
manifest drift: workspace checkouts (manifest ↔ disk) — 2 mismatch(es) vs PyAutoMind/repos.yaml. Stale evidence:release validation incomplete: no rehearsal for current source.Next step: human
/prmafter all required checks pass. No merge or deployment performed.Jammy2211 commentedon Oct 7, 2026
CI reload-race repair — 2026-10-07
User requested “fix the bug” after
/prmstopped on PR #26's Chromium failure. Resumed the existing approved task/branch; no scope expansion or second issue.Root cause: panel opening was immediate, but URL persistence waited for the asynchronous native
toggleevent. A reload before event delivery restored Runtime. A same-task activation regression reproduced the failure deterministically (open: true,axis: runtimeinstead ofbreakdown). The fix persists selection synchronously from summary click activation, including native keyboard activation; closing retains native behavior. No sleeps or weakened assertions.Commit:
f3a34c5, pushed to PR #26.Validation: 193 Python tests, Chromium including deterministic activation/reload and keyboard checks, Ruff lint/format, offline contract and whitespace checks PASS. Dashboard regenerated. Heart remains YELLOW for the previously recorded manifest mismatch; no new RED reason. Logs: task bundle
bug-repro.log,bug-browser.log,bug-pytest.log,bug-offline.log,bug-readiness.json.Lesson: native details toggle events are queued; persist route state during activation when reload/share correctness requires synchronous state. GitHub CI on the new head is the remaining merge gate. Merge still requires
/prm.Jammy2211 commentedon Oct 7, 2026
Profiling Results UI — completed
Simplified the profiling dashboard, standardized the requested model labels, removed legacy/redundant controls and prose, and split results from script links. Model links open dedicated browser tabs; Instrument, Device and Configuration selectors share a row with Configuration details first. Normal choices are curated float64 / 1500-source-pixel defaults where applicable; captures and archive links remain intact. Compact timing bars use matching totals or a labeled shared maximum without joining unrelated setups or units.
Fixed the CI-discovered reload race by persisting panel selection synchronously on activation rather than waiting for queued toggle events. The deterministic regression failed before the patch and passes with it; keyboard activation and reload behavior remain covered.
Validation: 193 Python tests, Chromium responsive/interaction checks, Ruff lint/format, offline contract validation and git diff checks passed. Exact final head f3a34c5 had two successful Actions runs: lint 37629490960 (all steps including Chromium and link checks passed), Dashboard Refresh 37629490950. Mergeability CLEAN; human invoked /prm. Git ancestry proves the sole claimed branch merged.
Heart's previous library-behind RED reasons cleared after refresh; the four libraries were already clean/current. Shipping proceeded under the recorded authorization with the existing manifest-mismatch YELLOW. No release or RED override was exercised.
Local logs and rendered previews preserved outside the removed task bundle at /home/jammy/Code/PyAutoLabs/tmp/profiling-results-ui-evidence/ (preview/browser/ contains screenshots). No scientific data products were created or removed. Post-merge publication is handled by the normal Pages workflow.