Every number fastpace produces. In one place. Plain English.

Six grades on one report card: how much of the codebase is documented, how fresh that doc is, do you share vocabulary, are decisions written down, are patterns visible, are lessons captured. Higher = AI has more to work with.

What level your codebase is on. L1 = just code. L5 = code + architecture + PRDs + execution plans + ADRs + extracted lessons. AI quality compounds as you climb the ladder.

Of the architectural decisions your team actually made, what % did you write down in a numbered ADR? Low = the next engineer rediscovers what you decided.

How old is your fastpace/context/ directory? Months since the last refresh. Old context = AI working off stale assumptions.

Of every AI-touched commit you shipped, what % has the complete evidence chain — audit-log entry + signed manifest + provenance trailer in the commit message? 100% = your auditor never asks "why did the AI do that?" without a receipt to show them.

Did anyone tamper with the audit log? fastpace verify walks every link in the hash chain. Pass = your forensic trail is intact. Fail = something changed history.

What % of your auditor's checklist for NIST AI RMF / ISO 42001 / EU AI Act / SOC 2 fastpace evidence already covers. The number that drives the Enterprise outcome ladder pricing — you pay less per seat as readiness rises.

Every non-human identity (AI agent, hook, MCP server) and whether it stayed inside its declared scope. Violations = an agent did something it wasn't supposed to.

How often AI commits got reverted or corrected. Lower = AI is fitting your codebase; higher = AI is producing code your team has to rewrite.

"$1 spent on AI calls returned $X in saved engineering hours." Computed from manifest token counts × model pricing × your blended dev-hour rate. <2× = you're probably losing money; 4×+ = AI is paying for itself many times over.

Who is spending what on which model this month. Same view your invoice will surface — but sliced by repo and dev so you can spot the runaway agent.

Naive linear projection of next-month spend from the last four weeks. Surfaces the $20K surprise before it lands on the invoice.

Of your monthly token cap, how much have you used? Warns at 70%, alerts at 100%. Today this is advisory (the cost dashboard surfaces warnings); real enforcement waits on the model gateway.

"Anthropic just raised Opus prices 12% — at your current usage that's +$340 next month." Diffs the bundled pricing table against your local snapshot and multiplies by usage.

How often the team ships, broken out AI-vs-human. The AI side should be ≥ the human side, sustainably.

Time from first commit on a feature to merged to main. Lower = AI is shortening cycles, not lengthening them.

What % of AI-assisted releases broke production. CFR for AI commits should not exceed CFR for human-only commits.

When something does break, how fast did the team fix it? Sliced by whether the broken commit was AI-touched.

Commits with a fastpace manifest trailer that landed in the last week, per developer. The shippability of each dev as a signal — but easy to game, so don't outcome-bind it.

"How often does this team tag a release?" Average days between recent tags. Surfaces the overdue signal — "your team typically tags every 21 days; you're 7 days late on the next tag." Triggers a release-train conversation without a meeting.

"What's sitting in main right now, waiting to be released?" Each commit risk-ranked (lines changed × age × AI-attribution). The release-train view a VP Eng would otherwise piece together from `git log` + Slack threads.

"How many production deploys did we record this month?" Each is a signed attestation written to the audit chain — the auditor-grade answer to "who shipped this code, when?". Different from "did the tag exist" — that's git; this is the signed deploy event.

"If one person quits tomorrow, would part of the codebase go dark?" Smallest set of authors covering 80% of recent commits per area. 1 = single point of failure. Healthy = 3+.

"How much of this code did AI write vs humans?" Per top-level directory. Reveals where AI is doing the heavy lifting and where it's been kept out.

"If our top AI vendor doubled their prices tomorrow, how exposed are we?" 0 = perfectly multi-vendor. 100 = single vendor. Above 60 = real concentration risk.

"How many files did AI touch that have zero test coverage?" Intersects fastpace's manifests with coverage/lcov.info. The AI's untested footprint.

"How far has each repo's .claude/ wandered from the org baseline?" Counterpart to fastpace sync — measures whether everyone is actually on the baseline you publish.

"Which prompt-injection tricks worked against our live agents and hooks this week?" The number a CISO defends to their board.

How many times a guardrail caught a real risk before it reached the codebase. Sliced by hook. High = guardrails are working; spike on a specific hook = a new policy gap or a regressed agent.

Open PRs ranked by review-attention ROI (size × age × AI-touched label). Surfaces the "we forgot about this PR" failure mode.

Time from fastpace welcome on a fresh checkout to first merged PR. Senior-dev hours saved per hire. Onboarding tax is a real org cost; this number makes it measurable.

When someone leaves, how complete is the auto-generated knowledge bundle for them? Counts commits, manifests, ADRs, decisions, learnings, ownership %.

A private nudge from your past self. The fraction of your AI sessions in the past 14 days that happened between 22:00 and 06:00 local time. Never sent to the org dashboard. Never in a contractual promise. Opt-in only, default OFF.

Hook-blocks resolved + recent shippers + new learnings — the appreciation moments derived from data. Cultural, not contractual.