The first-order rows of the 30-question review, answered live from the CodeNSM codebase's own instrumented data — one card per question, graded on track / watch / at risk. Every value is computed on this page load from the same models every other tab reads (no separate report to keep in sync). In the public demo, author names are hidden, department names generalised and internal code references stripped — the numbers are the real production distribution.
This page answers: what goes in the board pack? The quarterly review's first-order rows, answered live from instrumented data and graded on track / watch / at risk — lead with the red.
How to read this page: each card is one question a quarterly review should answer, computed live from this codebase's own data. The big number is the value; the coloured tag is the reading — on track, watch, or at risk — and the line under it is what it means. Start with the red.
Top 31 of 311 functions (the top decile by contribution) carry 100.0% of the project's total value-weighted contribution.
0 of the 31 top-decile-value functions are currently fragile or overloaded — 0.0% of that top-decile value sits on unstable ground.
5377 added+deleted lines of churn sit in 92 'expensive dead weight' functions (high build cost, low runtime value) — money leaking into code that doesn't pay back.
34 unique-IP functions vs 31 vanilla-integration (third-party glue) functions — 52.3% of that pair is defendable, unique engineering.
Alignment = share of total contribution carried by the functions an author has touched. Highest lands 88.9%; lowest of 1 authors lands 88.9%. Author names are hidden in the public demo.
Approximate: commit-weighted mean of each author's pct_fragile (share of their touched functions now fragile/overloaded, — not an exact per-commit tally, since fragile/overloaded status is tracked per function, not per commit.
0 of 31 top-decile-contribution functions with a static test signal show test coverage,; 0 of the 31 top-decile functions have no flow data at all (names-only mode).
71 of the 71 top-quartile-contribution functions WITH commit history have exactly one historical author (bus-factor risk). 7 top-quartile functions have no commit history at all and are excluded — this count is a floor, not a ceiling.
Worst the slowest department: p50 0ms, p99 2667ms across 311 functions (of 1 departments; unweighted percentile of each function's own avg_ms — there is no per-call latency histogram, only per-function averages,,.
4 functions sit in BOTH the slowest decile (avg_ms ≥ 267ms) and the fragile error-rate band (error_rate > 5%) — compounding risk: slow AND error-prone at once.
Gini of call volume across 311 functions is 0.967 — Pareto-concentrated (above the 0.60 threshold the research board's H01 'traffic concentration law' tests,.
21 functions show a reliability or latency regression (most recent FunctionDaily day vs the prior trailing window, — this is a rolling day-over-day comparison, not a calendar-week bucket; there is no weekly rollup table to query.
38 of 311 active functions carry no clear role signal yet (generalist archetype, the fallback when no naming/path/flow rule matches, — naming or structure debt that makes the map harder to read.
These first-order cards are the evidence layer, not the debate. Bring the three capstones to the quarterly review and let each persona argue their own:
// The "Open →" action links that jump to each surface are live in your own workspace, where the review runs on your repo with real names.