Where the base is going

Plans are the index's design documents: what is being built into the base every ix VM boots from, and what alternatives were considered, which git log does not capture.

Map

Every Plan scores itself 1-10 in frontmatter along the axes defined below, and every score carries a one-clause why. Pick any two axes; click a dot to read the Plan.

x AmbitionImpactEffortRiskMaturityLeverageTaste y AmbitionImpactEffortRiskMaturityLeverageTaste Expand

+ −

Index

• User-built router VM for fleet networking policy 0001 Draft
• Third-party integrations for partner-mediated services 0002 Draft
• MCP capabilities as composable CLIs 0003 Draft
• OpenTelemetry ingestion bus 0004 Draft
• Hosted credentials for external tool services 0006 Draft
• Forkable dev VM environments 0007 Accepted
• Module-typed builders 0008 Accepted
• Cross-compiled Darwin packages from Linux 0009 Load-bearing
• Fork patch-series machinery vs the state of the art 0010 Load-bearing
• Structured commit history from the Nix git fetcher 0011 Load-bearing
• One PR landing loop 0012 Sketch
• The kernel job contract 0013 Sketch
• Guards as typed policy 0014 Sketch
• Retire the Mac from CI entirely 0015 Sketch
• Every red check carries its evidence 0016 Sketch
• One delegation contract for agents 0017 Sketch
• Store and cache as a managed lifecycle 0018 Sketch
• Search results you can trust 0019 Sketch
• Agents should feel the cost of being stuck 0020 Input wanted RFC
• The filesystem as a queryable database 0021 Input wanted RFC
• Property-based UI testing with Bombadil 0022 Input wanted RFC
• Evals as the gate for behavioral change 0023 Input wanted RFC
• Version-tracking forks, base the series on the source nixpkgs already pins 0024 Input wanted RFC
• Let agents gossip, then fix what they gossip about 0025 Input wanted RFC
• A shared brain, memories that strengthen and decay 0026 Input wanted RFC
• Session provenance: agent history uploads keyed by session id 0027 Draft
• Cross-agent memory: share the fact files, keep the format 0028 Input wanted RFC
• Blog on ix.dev: fold the margins site into index 0029 Input wanted RFC
• Global knowledge graph: one ontology for people, biases, facts, and posts 0030 Sketch RFC
• Recommended next actions: a multi-select menu when the work lands 0031 Input wanted RFC
• Web-search eval: Claude built-in WebSearch vs Exa 0032 Input wanted RFC
• Agent truthfulness: decomposed claims, per-claim provenance, and an audit eval 0033 Input wanted RFC
• Proactive improvement: one controlled pipeline for drive-by fixes 0034 Input wanted RFC
• Endorsement trust lens: judge ideas by local PageRank over people we trusted 0035 Input wanted RFC
• Evals for tmux agency: will an agent drive a TUI via tmux when no proper tool is present? 0036 Sketch RFC
• Datalog as a generalization of Nix: cached demand-driven rules over a content-addressed memo table 0037 Sketch RFC
• One memory, queryable across the whole world 0039 Input wanted RFC
• The same roadblock never costs the same time twice 0040 Input wanted RFC
• Provenance lookback: from an installed thing to the line that put it there 0041 Draft
• VM templates, the systemd unit model 0042 Draft

Dimensions

Ambition Incremental → Moonshot

Distance from current practice, independent of payoff: rebuilding the Rust build on experimental Nix is a 9 even if it never lands; typing an existing builder is a 4 even though its impact is real.

Impact Marginal → Transformative

Size of the win if it lands, regardless of novelty or compounding: a boring change that saves everyone an hour a day is high impact with low ambition and low leverage.

Effort Weekend → Quarter

The time price: wall-clock and complexity to implement fully, not just to prototype.

Risk Safe → Hairy

The downside price: what breaks, or what the repo is stuck carrying, if the bet turns out wrong. Blast radius times how hard it is to back out.

Maturity Rough → Battle-tested

Not a judgment of the idea: the present state of its implementation, from unbuilt to daily-driven and hardened. The one axis that moves on its own as work lands. Orthogonal to API stability.

Leverage One-off → Flywheel

The shape of the win over time, not its size: does each use make the next cheaper? A contract every future integration inherits scores high before much impact has arrived; a big one-time saving scores low however large.

Taste Mechanical → Judgment

The judgment price: how much of the outcome depends on a human knowing what good looks like. Low means the path is clear and an advanced model can execute it; high means taste moves the result more than execution does.

Process

Plans are authored in the index repo as .svx files; the status ladder, numbering rules, and template live there. PR review is the discussion.