Canonical launch evidence#

Noema has one machine-readable source for launch-facing quantitative material: launch_evidence.json. Think of it as the receipt behind the public result: it records the measurements, calculation, limits, and source-file hashes used by the README, website, figures, tables, and interactive demo. README claims, the documentation landing page, the public webpage, experiment figures, result tables, and video overlays must derive their numbers from that projection instead of copying values from tutorial prose.

The current projection selects the synthetic QPSK I/Q-calibration demonstration. It contains all 63 per-run integer error counts and identities, their deterministic seven-cell aggregation, the headline selection rule, figure/table/video data contracts, and the warning and limitations inherited from the retained benchmark manifest. The selected CSV bytes and the embedded semantic observation grid are both frozen by digest. Because the observations are embedded, the arithmetic remains verifiable when the private authoring inputs are absent from a public-source export.

Scientific status and distribution clearance are deliberately independent:

  • scientific_status remains completed_experimental_benchmark, warning, and publication_ready: false. Completing a rights review cannot upgrade those claims.

  • distribution_clearance.status is pending_final_review for the future stable v0.2.0 release and paper bundle. The public repository is a development preview, not a blocked or cleared stable release. This field preserves the private authoring state; it is not a scientific-quality claim.

  • Observed minima and maxima are display ranges across three paired held-out seeds. They are not confidence intervals or population guarantees.

  • The calibrated I/Q oracle is a diagnostic reference with calibration knowledge, not a deployable same-information competitor.

Verify the public projection#

The public repository verifies the self-contained projection and all generated launch assets:

uv run --frozen python tools/generate_qpsk_iq_calibration_demo_assets.py --check
uv run --frozen python tools/generate_launch_evidence.py --verify
uv run --frozen python tools/generate_launch_assets.py --check

--verify validates the standalone JSON’s schema, self-hash, frozen observation-grid digest, coordinate closure, arithmetic, embedded distribution state, and consumer contract. Regeneration of the projection remains in the private authoring repository because it also consumes private release-control inputs; the public repository contains the frozen benchmark CSV and manifest needed to inspect the demonstration itself.

The strict contract is schemas/launch_evidence.schema.json, and the deterministic producer is tools/generate_launch_evidence.py. The canonical F3 figure and T0 table have generated paths in the projection. The full F0–F5 and T0–T2 launch set is rendered by tools/generate_launch_assets.py and closed by a digest manifest. The presentation contract also records the stable identity and external URLs of the learned QPSK I/Q calibration walkthrough; the recording itself remains outside the source repository.