Coder/deepseek-harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(desktop): fold Rubrics/Growth/Runtimes into one Evals nav entry (3 tabs + shared rubric selector)

Browse Source
This commit is contained in:
ZiyaZhang
2026-07-19 23:31:47 -07:00
parent 211ffb4b7c
commit 0207b5edd2
14 changed files with 781 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

190
examples/desktop/src/renderer/index.html
View File

@@ -104,35 +104,31 @@
</button>
</div>
<div class="nav-group" data-nav-group="runtime" data-item-count="6">
<div class="nav-group" data-nav-group="runtime" data-item-count="4">
<div class="nav-group-header">runtime</div>
<!-- lane-rubrics slot — live as of NAV delta merge (feat/lane-rubrics-annotation
@ 70293ba already merged as a13e020; this merge is the flip point per team-
lead rule "hub+rubrics flip in NAV delta since neither NAV nor RUB touched
index.html on the merge path"). data-fixture-tier stays because the 28-task
rubric catalog is still fixture-backed until the annotation-store producer
wires up; the "· demo" chip decorateNav() paints from it stays truthful. -->
<button class="tab-btn nav-item" data-tab="rubrics" data-fixture-tier="true"
title="Rubrics — catalog-first scoring rubric library, 28-task groups (fixture-tier)">
<!-- Evals entry (lane-evals-merge, 2026-07-19). Three former nav
items — Rubrics, Growth, Runtimes — are collapsed into a
single Evals door because they already share one rubric
scoring event log (rubric-fusion-model.js) and mean much
less on their own than they do together. Inner surfaces are
unchanged; the pane just grows a top tab strip that mounts
each existing controller under [ Rubrics | Growth | Runtime ].
data-fixture-tier stays: the fusion event log is still
fixture-backed (rubric-fusion-seed.js) until the annotation
store and rubric library seams wire up. -->
<button class="tab-btn nav-item" data-tab="evals" data-fixture-tier="true"
title="Evals — three views over the shared rubric scoring event log (Rubrics catalog · Growth timeline · Runtime rollout grid).">
<svg viewBox="0 0 20 20" width="16" height="16" aria-hidden="true"><path fill="none" stroke="currentColor" stroke-width="1.6" stroke-linecap="round" stroke-linejoin="round" d="M4 5h12M4 10h12M4 15h8M14 14l2 2 3-3"/></svg>
<span>Rubrics</span>
<span>Evals</span>
</button>
<button class="tab-btn nav-item" data-tab="plugins" title="Plugins">
<svg viewBox="0 0 20 20" width="16" height="16" aria-hidden="true"><path fill="none" stroke="currentColor" stroke-width="1.6" stroke-linejoin="round" d="M7 3v3H4v4a2 2 0 0 0 2 2h1v4h3v-4h2v4h3v-4a2 2 0 0 0 2-2V6h-3V3h-3v3h-2V3z"/></svg>
<span>Plugins</span>
</button>
<button class="tab-btn nav-item" data-tab="runtimes" title="Local runtime profiles + isolated daemons">
<svg viewBox="0 0 20 20" width="16" height="16" aria-hidden="true"><path fill="none" stroke="currentColor" stroke-width="1.6" stroke-linejoin="round" d="M3 5h14v4H3zM3 11h14v4H3zM6 7h.01M6 13h.01M9 7h.01M9 13h.01"/></svg>
<span>Runtimes</span>
</button>
<button class="tab-btn nav-item" data-tab="mission" title="Missions">
<svg viewBox="0 0 20 20" width="16" height="16" aria-hidden="true"><circle cx="10" cy="10" r="6.5" fill="none" stroke="currentColor" stroke-width="1.6"/><circle cx="10" cy="10" r="2" fill="currentColor"/><path stroke="currentColor" stroke-width="1.6" stroke-linecap="round" d="M10 1.5V4M10 16v2.5M1.5 10H4M16 10h2.5"/></svg>
<span>Missions</span>
</button>
<button class="tab-btn nav-item" data-tab="growth" title="Growth (self-evolution log)">
<svg viewBox="0 0 20 20" width="16" height="16" aria-hidden="true"><path fill="none" stroke="currentColor" stroke-width="1.6" stroke-linecap="round" stroke-linejoin="round" d="M3 16.5h14M4.5 16V12M8 16V8.5M12 16V5M15.5 16V10M4.5 12l3.5-3.5L12 5l3.5 5"/></svg>
<span>Growth</span>
</button>
<button class="tab-btn nav-item" data-tab="prs" title="Pull Requests">
<svg viewBox="0 0 20 20" width="16" height="16" aria-hidden="true"><circle cx="5" cy="4" r="1.8" fill="none" stroke="currentColor" stroke-width="1.5"/><circle cx="5" cy="16" r="1.8" fill="none" stroke="currentColor" stroke-width="1.5"/><circle cx="15" cy="16" r="1.8" fill="none" stroke="currentColor" stroke-width="1.5"/><path fill="none" stroke="currentColor" stroke-width="1.5" d="M5 5.8v8.4M5 6.5c0 4 5 4 5 6.5V6M15 14.2V9l-2 2M15 9l2 2"/></svg>
<span>PRs</span>
@@ -237,35 +233,34 @@
</div>
</div>
<!-- Runtimes sidebar — page owns its own body; the sidebar just
orients + provides a refresh button. See runtimes-page.js. -->
<div class="tab-panel" data-tab-panel="runtimes" hidden>
<!-- Evals sidebar (lane-evals-merge, 2026-07-19). Rubrics / Growth /
Runtimes were three siblings until 2026-07-19; each had its own
sidebar panel. They're now one door because the three views
project the same rubric scoring event log. This single panel
orients the researcher to the three tabs (top of the pane) and
preserves both the rubrics-catalog Refresh button and the
Export annotations action from the old Rubrics sidebar so no
deeplink from Chat header / Recent rows loses its target. The
Runtimes refresh button is preserved under the same id so
runtimes-page.js click handlers keep working. -->
<div class="tab-panel" data-tab-panel="evals" hidden>
<div class="sidebar-section-head">
<span class="section-label">Runtimes</span>
<button id="runtimes-refresh" class="icon-btn ghost" title="Reread profile list + status" aria-label="Refresh">
<span class="section-label">Evals</span>
<button id="rubrics-refresh" class="icon-btn ghost" title="Reload the rubric catalog + fusion store" aria-label="Refresh">
<svg viewBox="0 0 20 20" width="14" height="14" aria-hidden="true"><path fill="none" stroke="currentColor" stroke-width="1.6" stroke-linecap="round" stroke-linejoin="round" d="M15.5 4.5v4h-4M4.5 15.5v-4h4M5.2 7.6a6 6 0 0 1 10 .4M14.8 12.4a6 6 0 0 1-10-.4"/></svg>
</button>
</div>
<div class="runtimes-sidebar-note muted">
Local profiles, adapters, and any isolated daemon currently up. Composed
locally until <code>runtime/list</code> lands on the wire (gap G8).
</div>
</div>
<!-- Rubrics sidebar (task #188) — orients the user to the catalog. -->
<div class="tab-panel" data-tab-panel="rubrics" hidden>
<div class="sidebar-section-head">
<span class="section-label">Rubrics</span>
<button id="rubrics-refresh" class="icon-btn ghost" title="Reload the catalog" aria-label="Refresh">
<svg viewBox="0 0 20 20" width="14" height="14" aria-hidden="true"><path fill="none" stroke="currentColor" stroke-width="1.6" stroke-linecap="round" stroke-linejoin="round" d="M15.5 4.5v4h-4M4.5 15.5v-4h4M5.2 7.6a6 6 0 0 1 10 .4M14.8 12.4a6 6 0 0 1-10-.4"/></svg>
</button>
</div>
<div class="rubrics-sidebar-note muted">
Evaluator library — one SKILL.md file per rubric, grouped by task category.
Every rubric attaches to Bench for scoring runs; "Create from scratch"
lives at the bottom of the page as a fallback CTA.
<div class="evals-sidebar-note muted">
One rubric scoring event log, three views. Rubrics = catalog, Growth =
timeline over sessions, Runtime = per-rubric rollout grid. Pick a rubric
from the top selector to keep them in sync.
</div>
<div class="sidebar-foot">
<!-- Runtimes refresh handler still bound by id; keeps the sidebar
button contract stable for runtimes-page.js's click handler
and any deeplink that targeted it. -->
<button id="runtimes-refresh" class="ghost small" title="Reread profile list + status (Runtime tab).">Reload runtimes</button>
<button id="growth-refresh" class="ghost small" title="Reread the growth log (Growth tab).">Reload growth</button>
<button id="rubrics-annotate-open" class="ghost small" title="Open the export drawer for annotated sessions">Export annotations</button>
</div>
</div>
@@ -281,20 +276,11 @@
</div>
</div>
<!-- Growth sidebar (kept minimal — this page is a self-contained page,
the sidebar just orients the user to what it's for). -->
<div class="tab-panel" data-tab-panel="growth" hidden>
<div class="sidebar-section-head">
<span class="section-label">Growth</span>
<button id="growth-refresh" class="icon-btn ghost" title="Reread the growth log" aria-label="Refresh">
<svg viewBox="0 0 20 20" width="14" height="14" aria-hidden="true"><path fill="none" stroke="currentColor" stroke-width="1.6" stroke-linecap="round" stroke-linejoin="round" d="M15.5 4.5v4h-4M4.5 15.5v-4h4M5.2 7.6a6 6 0 0 1 10 .4M14.8 12.4a6 6 0 0 1-10-.4"/></svg>
</button>
</div>
<div class="growth-sidebar-note muted">
Every entry links to a real event — plugin install, overlay apply, vibe
session. This is an audit trail, not a persona.
</div>
</div>
<!-- Growth sidebar folded into the Evals sidebar above (2026-07-19,
lane-evals-merge). The growth-refresh button lives in the new
Evals sidebar-foot so growth-v2.js listeners keep binding by id.
If a future lane splits Growth back out, restore this block
with the original heading + refresh icon. -->
</aside>
<main class="main">
<!-- Chat pane -->
@@ -937,13 +923,67 @@
</div>
</section>
<!-- Rubrics pane (task #188) — catalog-first evaluator library.
<!-- Evals pane (lane-evals-merge, 2026-07-19). Wraps three sub-panes —
Rubrics catalog, Growth timeline, Runtime rollout grid — that
share the same rubric scoring event log (rubric-fusion-model.js).
The outer .pane is what switchTo('evals') toggles; only one of
the three inner sub-panes shows at a time based on the
data-evals-tab strip below. The inner sub-panes keep their
original data-pane values because rubrics-page.js / growth-v2.js
/ runtimes-page.js all query them by that selector — the
renderer.js switchTo() loop is updated to iterate direct main
children only so nested data-pane doesn't fight the outer
hidden flag. Shared selector (rubric picker) sits above the
tab strip so choosing a rubric syncs all three views. -->
<section class="pane" data-pane="evals" hidden>
<header class="header evals-header">
<div class="header-lead">
<div class="page-title">
Evals
<span class="demo-tier-chip" title="The rubric scoring event log is still fixture-backed (rubric-fusion-seed.js). User-authored rubrics and real annotation results land as the G1 library seam and the annotation-store producer wire up.">demo · G1 pending</span>
</div>
<div class="page-sub muted">Rubric library, evolution over time, and per-rubric rollout grid — three views over one scoring event log.</div>
</div>
<!-- Shared rubric selector. Populated on show() by
renderer.js/evals-shared-selector from window.__dshRubricFusion.
Empty by default so first paint is stable and the "all rubrics"
scope is honest. Emits `dsh:evals-rubric-change` when the user
picks — each inner page listens and re-scopes its view. -->
<div class="evals-shared-selector" role="group" aria-label="Shared rubric selector">
<label class="evals-shared-label muted small" for="evals-shared-rubric">Rubric focus</label>
<select id="evals-shared-rubric" class="evals-shared-select"
title="Filter Rubrics catalog · Growth curves · Runtime grid to one rubric. 'All rubrics' keeps the default per-page view.">
<option value="">All rubrics</option>
</select>
</div>
</header>
<!-- Inner tab strip. Only one .evals-tab-pane is visible at a time;
visibility is driven by [data-evals-active] on this .pane
(CSS in style.css matches the attribute so no JS querySelector
per pane is needed). Rubrics is the default because it's the
catalog entry point new researchers see first. -->
<div class="evals-tab-strip" role="tablist" aria-label="Evals view">
<button type="button" class="evals-tab active" data-evals-tab="rubrics"
role="tab" aria-selected="true"
title="Catalog of scoring rubrics grouped by task class.">Rubrics</button>
<button type="button" class="evals-tab" data-evals-tab="growth"
role="tab" aria-selected="false"
title="Score curves over harness versions / models / data mixes.">Growth</button>
<button type="button" class="evals-tab" data-evals-tab="runtime"
role="tab" aria-selected="false"
title="Rollout × rubric-dim red/green grid for the current runtime.">Runtime</button>
</div>
<div class="evals-tab-body">
<!-- Rubrics sub-pane (task #188) — catalog-first evaluator library.
Rendered by src/renderer/rubrics-page.js; the pure model lives at
rubrics-model.js and the fixture blobs are inlined via
rubrics-seed.js. The detail drawer sits fixed to the right; the
annotation flow (task #191) lives in annotation-panel.js and is
reachable from Chat header + Recent list rows. -->
<section class="pane" data-pane="rubrics" hidden>
reachable from Chat header + Recent list rows. Kept as
data-pane="rubrics" so page controllers that query it by that
selector still bind; the class no longer includes 'pane' because
only the outer wrapper is the main-child pane. -->
<section class="evals-tab-pane" data-pane="rubrics" data-evals-tab-pane="rubrics">
<header class="header">
<div class="header-lead">
<div class="page-title">
@@ -976,10 +1016,15 @@
</aside>
</section>
<!-- Growth pane. Rendered by src/renderer/growth-v2.js — the
compact-window "harness evolution log" for researchers. See the
header of growth-v2.js for pane structure + data source. -->
<section class="pane" data-pane="growth" hidden>
<!-- Growth sub-pane (2026-07-19, lane-evals-merge). Rendered by
src/renderer/growth-v2.js — the compact-window "harness evolution
log" for researchers. Section keeps data-pane="growth" so
growth-v2.js's `document.querySelector('.pane[data-pane="growth"]')`
still binds; the .pane class is preserved for the same reason
(growth-v2.js scopes CSS on `.pane[data-pane="growth"]`). The
outer .pane[data-pane="evals"] toggles overall visibility; this
section shows only when data-evals-active="growth" via CSS. -->
<section class="pane evals-tab-pane" data-pane="growth" data-evals-tab-pane="growth" hidden>
<header class="header">
<div class="header-lead">
<div class="page-title">Growth</div>
@@ -1020,14 +1065,17 @@
</section>
</section>
<!-- Runtimes pane (task #189 / IA §3 Runtimes one-pager). Rendered by
src/renderer/runtimes-page.js on switchTo('runtimes'). Body has
two sections: the profile list (one row per registered profile,
expandable into a capability matrix) and the isolated-daemon
section (any Playground scratch daemon currently up). The
legend chip up top names the wire seam this composes locally
until `runtime/list` (G8) lands. -->
<section class="pane" data-pane="runtimes" hidden id="runtimes-pane">
<!-- Runtimes sub-pane (task #189 / IA §3 Runtimes one-pager), now
the Runtime tab inside the Evals pane. Rendered by
src/renderer/runtimes-page.js on switchTo('evals') + Runtime tab
active. Body has two sections: the profile list (one row per
registered profile, expandable into a capability matrix) and
the isolated-daemon section (any Playground scratch daemon
currently up). Section keeps data-pane="runtimes" + id so
runtimes-page.js's `document.getElementById('runtimes-pane')`
still finds it. .pane class preserved so the CSS geometry
(flex column + min-height) still applies. -->
<section class="pane evals-tab-pane" data-pane="runtimes" data-evals-tab-pane="runtime" hidden id="runtimes-pane">
<header class="header">
<div class="header-lead">
<div class="page-title">Runtimes</div>
@@ -1044,6 +1092,8 @@
<div class="runtimes-isolated" data-runtimes-isolated></div>
</section>
</section>
</div><!-- /.evals-tab-body -->
</section><!-- /.pane[data-pane="evals"] -->
<!-- Settings pane (task #193). Two sections: Model pricing (editable
table seeded from price-table.js DEFAULT_PRICE_TABLE, edits