Context Score

Context Score is one number, 0–100, per domain. It answers a single question: how ready is this domain to give you a good answer? Roll the per-domain scores up and you get Life Readiness — how ready your whole vault is.

A domain is only as good as the context it carries. An empty wealth/ gives you generic answers; a dense, fresh, well-structured one gives you answers grounded in your actual life. The Context Score makes that gap visible — and tells you exactly what to fix next.

Note

The score is computed in the engine, never in the UI. Every shell reads the same number via prevail score <domain> --json. The wire shape is ContextScore.

The breakdown

The score is a weighted blend of six factors. Each is reported in breakdown so a shell can show the sub-scores, not just the total.

Factor	What it measures
coverage	Are the core files present? (state.md, QUICKSTART.md, PROMPTS.md, config.md)
density	Is there real substance — not stub headings?
freshness	How recently was the domain touched? Stale context decays.
structure	Are the document folders (00_current/, 01_prior/, 02_briefs/) and skills organized?
activity	Is there recent log / journal / thread activity?
config_completeness	Is the manifest.json filled in — identity, goals, routing?

Each breakdown.<factor> carries its own score and weight, so the total is reproducible and explainable.

What’s missing

missing is an array of MissingItem — concrete, actionable gaps (“no state.md”, “manifest has no goals”, “nothing in 00_current/”). This is what powers the “fix this next” nudges in every shell, and it’s the path a new user follows after onboarding.

The optional audit

By default the score is deterministic and free — pure file inspection, no LLM call. Add --audit and the engine also asks a model to write a short narrative assessment of readiness, populating assessment and audit_source. Without --audit, both are null.

The audit respects the global guardrails: --budget <tokens> caps the spend and --local-only forces a local engine. See Privacy & local-only.

prevail score wealth --json                 # free, deterministic
prevail score wealth --audit --json         # + LLM narrative
prevail score wealth --audit --local-only --json   # narrative, no network

History

Every computed score is appended to <domain>/_log/score.jsonl — one line per point, append-only. This is the JSONL leg of the storage rule: time series → JSONL. prevail score history <domain> --json returns the series oldest → newest so a shell can draw a trend line.

[ { "ts": 1748000000000, "score": 71 },
  { "ts": 1749220200000, "score": 78 } ]

Life Readiness

prevail score --all --json scores every (non-archived) domain and adds a lifeReadiness roll-up — a single 0–100 number across your whole vault, plus the per-domain scores behind it.

{ "lifeReadiness": 81,
  "domains": [ ContextScore, ContextScore ] }

How shells use it

• Cockpit / desktop — a badge per domain in the sidebar; overall Life Readiness in the header; the missing list as a checklist.
• Onboarding — newly-scaffolded domains start with a low score and a full missing list, which becomes the new user’s to-do path.
• Heartbeat — a scheduled routine can re-score a domain and alert you if it drops below a threshold (stale context). → Heartbeat

All of them call the same --json command. The number is computed once, in the engine.