📊
PRODUCT
Harness Score
Establishes the baseline, evaluates frozen evidence, exposes confidence, and verifies improvement.
THE KNOWLEDGE-SYSTEM ENGINEERING SUITE
Know what your
harness can prove.
Score the system you have. Understand the evidence behind it. Improve it safely. Verify what changed.
Executable alpha · three Agent Skills · read-only review · Forge plan-only
Scoremeasure
Atlasunderstand
Forgeevolve
Foundationdefine
Traceevidence
Wiresconnect
THE PROBLEM
A folder can look complete while the system behind it is drifting.
Knowledge harnesses emerge from instructions, agents, tools, permissions, memory, generated projections, runtime state, and years of accumulated change.
File presence is not architecture. Documentation is not mechanism. Activity is not health.
Without evidence and dependency context, a confident score is theater—and a simple cleanup can become a migration incident.
THE OPERATING LOOP
Score first. Evidence underneath.
Harness Score is the first user experience. Foundation, Trace, and Wires quietly establish the admissible evidence before any number appears.
01ScoreBaseline the existing harness
→
02UnderstandNavigate evidence in Atlas
→
03ImproveRehearse change in Forge
→
04RescoreVerify measurable change
LIVE ALPHA
harness review TARGET --output RUN_DIR produces a verified offline report. See the real synthetic run ↗CORE DISTINCTIONScore is first in the experience. Evidence is first in the computation.
SIX COORDINATED LAYERS
One system. Clear authority.
🗺️
PRODUCT
Harness Atlas
Turns architecture, vitality, evidence, drift, and risk into navigable human views.
🔥
PRODUCT
Harness Forge
Creates baseline-bound impact and rollback plans. Mutation is deliberately unavailable in the alpha.
🧱
ENGINE
Harness Foundation
Defines schemas, standards, invariants, authority rules, and compatibility contracts.
🔎
ENGINE
Harness Trace
Discovers artifacts, binds evidence, models vitality, compares history, and reports blind spots.
⚡
ENGINE
Harness Wires
Resolves dependencies, lineage, data flow, coupling, and change-propagation paths.
TRUST BEFORE THEATER
Unknown is a valid answer.
01
Read-only by default
Target content is untrusted data. Discovery never executes it.
02
Coverage is evidence
Blind spots, budgets, and partial status remain visible.
03
Dimensions stay separate
Architecture, mechanism, vitality, and proof are not interchangeable.
04
Change is governed
The alpha stops at a baseline-bound plan and numbered approval record. It cannot mutate.
BUILD WITH US
Help make harness claims testable.
Install the three skills, run the synthetic fixture, and challenge the evidence: false-positive reports, adversarial fixtures, accessibility feedback, and clean-install witnesses are the highest-value contributions.