Roadmap to 1.0

World 1.0 is defined by a bar of seven clauses, ratified by the project owner on 2026-07-23. Each clause is a checkable end state, not a work item. The authoritative text lives in the mission charter ("The bar"). This page summarizes it and says where each clause stands.

Current state: 5 of 7 clauses met. The two that remain are the ones that measure World against the status quo. No release date is promised.

#	Clause	State
1	Deterministic kernel	Met
2	Local-first daemon	Met
3	Explicit authority end to end	Met
4	Resident-agent non-inferiority floor	Not yet met
5	The human surface works, with provenance teeth	Not yet met
6	Protocol-native boundary	Met
7	Controlled self-modification, proven once	Met

The clauses

1. Deterministic kernel Met

An immutable, content-addressed world store with an append-only transition log and proven deterministic replay. Replaying any recorded episode reconstructs state bit-for-bit from pure transitions plus recorded effect results. The log-epoch semantics (interpreter version pinning and content-addressed transition functions) were decided and enforced before the log format froze.

2. Local-first daemon Met

ailang-worldd runs on one machine over SQLite, with a CLI and a REST API, and has zero cloud dependencies in the core. Cloud transports may exist only as effect-handler extensions.

3. Explicit authority end to end Met

Every effect goes through the broker with a capability and budget check. Effect results are recorded as replay input. Capsules run with a physical isolation floor beneath the semantic checks. No ambient-authority path exists from an agent to the outside world.

4. Resident-agent non-inferiority floor Not yet met

This is a do-no-harm kill switch, not the value proof. Two reference agents from different providers, Claude Code and codex CLI, each run the same benchmark set in two arms: shell (native tools) and World (MCP transition tools only), with the same model, three or more runs per arm. The floor holds only if both agents show a World pass rate no more than 2 percentage points below the shell pass rate and a median wall-clock overhead of at most 25%. A stability precondition on the shell arm keeps a flaky harness from faking the verdict either way. If the floor fails on eligible agents after honest tuning, World parks, and says so.

Where it stands. Tracked as queue row 93. Its prerequisite, row 134 (the software-engineering tools agents use in the World arm), landed on 2026-10-03, and the floor's grader port, classification and statistics landed the same day (row 93 M1 and M3, no agent run yet). Row 135 widened the ailang-run tool (stdin, argv and per-task capabilities) so all 23 core tasks can run through World; it landed on 2026-10-03 on the v0.52.1 tool binary (#192), with an attended four-task smoke still to run. The ratified order to release is now the rest of the developer CLI (row 138), a toolchain effect for non-AILANG projects (row 140), the floor run (row 93), provenance teeth (row 114) and an interpreter epoch-upgrade path (row 139).

5. The human surface works, with provenance teeth Not yet met

This is the 1.0 value demonstration. One real goal is expressed, becomes proposals, reaches an approval inbox with evidence bundles, and commits. And, measurably: on at least three real "why did this happen" questions, arising from actual operation rather than written for the test, a provenance walk yields the verified answer in five minutes or less each, where the pre-World method was a grep and log archaeology session.

Where it stands. Tracked as queue row 114. The walks so far proved retrieval of answers that had been recorded in advance, and the pre-World baselines were not timed. The clause needs unseeded questions from real operation, with the pre-World method timed alongside. Its corpus is the operation World records now that row 134 has landed.

6. Protocol-native boundary Met

The transition registry is served over MCP, filtered per session by capability, and an A2A agent card is published. No new wire protocols.

7. Controlled self-modification, proven once Met

At least one World behavior change ships as an extension package through World's own propose, verify and commit pipeline, with the self-modification boundaries enforced: the mission-loop machinery, the AILANG compiler and the live daemon are outside self-modification scope.

Beyond 1.0

The standing value evidence beyond the bar is the mission loop that builds this repository migrating onto World, measured on incident classes eliminated and human attention per workstream against the July 2026 markdown-and-scheduler baseline.

Following along

Every mission iteration reports on the project's weekly bookkeeping issue, and the full per-iteration log lives in design_docs/world-mission-log.md.