Roadmap

What has to be true before a release

Open items, with the ones the audit closed marked as done. Nothing here is scheduled, and the order is a grouping, not a commitment.

Closed by the audit

18 September 2026, 20:12:15 UTC
✓Process-kill recovery for a bounded import — 8,192 records, verified raw hash and single input charge.Physical power loss and live-feed reconnects remain open.
✓An end-to-end runtime chain measurement exists as an engineering baseline.A controlled release comparison is still open.
✓A read-only doctor and deterministic health assessment, in gco-ops and through the CLI.
✓Native tick and book families select genuinely different codec profiles from the header, not from the extension.
✓Documented counting rules for the input budget, and a runtime manual.
✓gqodb-codec and gco-ops now carry source and tests — 30 packages, no design-only pair.

Release and installation

□A clean source commit and release bundle with current docs, lockfile and full release checks.
□A complete suite install from an empty prefix, followed by real installed import, capture, query and recovery.
□Storage and platform qualification: cold disk, full disk, and physical power loss where it is required.
□Licence, dependency and redistribution rights, package name and distribution channel.

Measurement

□A controlled chain benchmark with equal workload, durability, CPU and RAM budget and storage cost for the reference systems.
□A separate assessment of the current group-commit and metadata work — historical per-record figures must not be swapped for it without a new run.
□Long-running load with multiple consumers, bursts, source reconnects, receipt ambiguity and producer deduplication.
□The kdb+/q comparison, once a usable runtime and the right licence exist.
⊘Time-contract overhead against real ticks and L10 — base vs. times vs. times plus clock quality.Deferred on request since 8 September and still deferred; physical clock and provider qualification stays with it.

Function

□Persistent query indices, the remaining domain adapters, and retention and checkpoint policy where they are still missing.
□Feature-side trade corrections and cancels, candles, multi-source joins and a persistent feature store.
□Continuous source integration for news, fundamentals and reference, including calendars, DST and corporate actions.
□Model bundles with preprocessing, calibration and feature order, plus runtime and inference RAM measurement.

Claims and commerce

□Commercial figures, interview status, publication status and licence claims confirmed separately before any of them appears on a public page.
□A stated scope for what a chain benchmark would support — it covers the tested scenarios and chosen baselines, and market-wide superiority does not follow from it.