Release v1.1.0 — Engine Page, Families Redesign & June 2026 Data · coeusyk/opencast

OpenCast v1.1.0

A dedicated Engine Signals page analysing Stockfish evaluation vs. human
performance across 73 openings.

Human vs Engine Win Probability — horizontal diverging bar chart
(pure CSS, zero-dependency) showing delta between human win rate and
Stockfish depth-20 prediction, sorted by absolute delta. Teal = human
outperforms, pink = engine outperforms. Inline ±pp value labels.
Greatest Engine–Human Gaps — side-by-side Top 20 tables for each
direction. Clicking any row navigates to the opening detail page.
Move Lines Driving Trends — sortable table of individual move
continuations with the largest 12-month shift in win rate or popularity.
ECO and opening name cells link to opening.html?eco={ECO}. SAN moves
rendered as monospace chips.
Structural Break Events — Plotly scatter plot of regime change
detections across ECO families (A–E), dot size scaled by absolute
centipawn evaluation. Filtered to p < 0.01 (Chow test) and one
breakpoint per opening (most significant only) to prevent false-positive
flooding. Restyled for OpenCast dark theme.

Replaced flat table layout with a responsive card grid (3-col desktop,
1-col mobile).
Each card surfaces: tier chips, average win rate, min–max range as inline
subtitle, trend counts, engine gap chip with average delta, forecast
confidence breakdown, regime change count, and top outlier link.
Card gradients and win-rate CSS bar chart aligned to actual data.
Added Show sparkline toggle per card — 12-month server-rendered
Plotly line with dynamic Y-axis domain (30% padding above/below data
range). Sparkline delta annotation added.
Added Compare Families panel — overlaid 24-month line chart with
per-family toggle buttons.
Summary metrics bar: fourth tile Avg Engine Gap added.
AVG WIN RATE tile: range now shown as inline subtitle
(min% – max%), replacing disconnected side column.

Families sparklines fixed: Plotly now loaded in page <head> to prevent
render race on expand.
CSS win-rate bar chart values aligned with card actuals.
Dashboard thresholds (low-confidence floor, min games) driven from
config.json instead of scattered hardcoded values.
latest_tracked_month field added to serialized data; move stats panel
shows honest lag caption when move-line data trails actuals.

max_tracked_months=48 enforced across ingest, timeseries, CI fetch,
and dashboard serialization via shared month_window helpers.
Prevents unbounded time-series growth as monthly data accumulates.
CI fetch step respects the window — older raw JSON files outside the
48-month window are no longer fetched or staged.

June 2026 Lichess data ingested via update.yml workflow run (2026-06-11).
Cron schedule adjusted to run on the 10th of each month to ensure
Lichess indexes the previous month fully before fetch.
Forecasts, engine delta, and move-line trends updated for all 498 ECO
codes through 2026-05.
Findings report regenerated (2026-06-11).

.cursor/rules/web-audit.mdc added — eight-dimension dashboard review
rubric enforcing hierarchy, semantics, accessibility, and state handling
for inline HTML/CSS pages.
opencast-spec.yaml added at repo root — OpenAPI 3.1 component schemas
for OpeningRecord, MoveLine, and FamilySummary. Schema-only
contract; no server paths.