01
Sources
evidencepartial
System / public projection
The sum of what you save.
Sources become signals. Stable owners change only through named loops, versioned workflows, exact authority, and proof.
Goals
6
measurable promises
Workflows
14/16
ready command contracts
Owners
192
stable write targets
Reviews
3295/3376
final source decisions
OverviewSystem / change view
00 / living architecture
One graph, three kinds of truth
Evidence, operating objects, and proof stay distinct so retrieval never becomes accidental authority.
02
Signals
evidenceready
03
Stable objects
knowledgeready
04
Goals
controlready
05
Loops
controlpartial
06
Harness
executionpartial
07
Proof
proofready
08
API · skill · UI
interfaceready
sourcessignalsexpand + extract
signalsobjectsbind + challenge
objectsgoalsinform
goalsloopsgovern
loopsharnessassemble
harnessproofexecute + verify
proofobjectswrite back
objectssurfacesproject
Wikibot / governed agent
One projection-bound plan powers this API and the private Daily Brief producer. Public reads, private proposals, and no mutation before exact approval.
read / publicpropose / privatemutate / none
01
evidence-reviewReview the next evidence cohort
54 evidence clusters are queued and 27 are partially reviewed. Capture is not learning until a receipt names the stable owner and disposition.
source.absorb02
coverage-recoveryRecover learning from agent work
1,842 agent evidence traces are replayed and 28,019 remain pending. Use the private task library to choose a bounded project or month; indexing alone is not learning.
source.absorb03
workflow-repairChoose one workflow gap to close
voice.realtime is partial. Each must pass its own completion proof; this brief does not authorize a build.
work.quality-loop04
source-repairStrengthen a source adapter
13 of 15 public source adapters declare a gap. Improve evidence completeness before increasing ingestion volume.
source.absorbEvidence graph
Exact revisions, artifacts, completeness, and cross-platform correlations.
Operating graph
The small set of instructions, workflows, skills, rules, projects, and beliefs that agents use.
Proof graph
Plans, runs, tests, decisions, failures, and writebacks that justify current state.
product program
The applications this system is here to realize
Discord instructions resolve into five durable product owners. The status is literal; the next slice is the smallest proof-bearing build.
fieldworkactive
Fieldwork Installable design IDE and agent-connected visual inspection environment.
proofexternal
Implementation is active in a separate Codex task; this checkout owns the product contract, not that task's worktree.
next proof-bearing slice
Continue implementation in the dedicated Fieldwork Codex task; reconcile proof here only after that task hands off a verified artifact.
daily-briefimplemented
Daily Brief Installed local review and execution surface for ranked proposals, exact decisions, confined verification, and commit or PR delivery.
proofpartial
24 control-plane/API tests pass; real brain-health, portable-memory, source-review, and private agent-session proposals were stored; exact approval and dispatch are separate; confined Git apply/verify, failure recovery, local-commit and PR delivery fixtures, notification privacy, and LaunchAgent security pass; the installed loopback daemon is healthy; and the SwiftUI client builds, packages, signs, launches, and shares the queue contract. A live local fixture produced and replayed one verified commit receipt, but one real Kevin-approved PR receipt remains open.
next proof-bearing slice
Complete one real Kevin-approved proposal-to-verified-PR receipt; then add authenticated remote producers without weakening local approval authority.
local-searchshipped
Local Search CLI Agent-callable local-browser search with structured output and cleanup.
proofpassing
Published v0.1.4; 31 tests pass, clippy is clean with warnings denied, and lsearch reports 0.1.4.
next proof-bearing slice
Add an end-to-end live browser smoke receipt to the wiki's source loop while preserving the already-passing release contract.
agent-broomavailable
Agent Broom Standalone cleanup and process-hygiene CLI for agents.
proofpassing
CLI and vendored skill are present; a fresh doctor run at ea8dffb6 exited 0 in dry-run mode and reported listeners, browsers, artifacts, and candidate orphans without mutation.
next proof-bearing slice
Add automated empty-nested-repo, protected-process, and mutation-mode regression fixtures while keeping audit and doctor dry-run by default.