Launch figures and tables#

This is the generated visual system for Noema’s launch demonstration. Figures F0–F5 and tables T0–T2 derive from the repository-root launch_evidence.json; none is a second source of experimental truth. They are documentation and launch assets only. Selecting any of them for the paper remains a later editorial decision after manuscript review.

Regenerate and verify the complete set with:

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

The machine-readable asset inventory is docs/_static/launch/manifest.json. Every quantitative figure and table carries the experimental disclosure or an adjacent note, uses observed min/max terminology, and preserves the calibrated oracle’s diagnostic-reference role.

Figures required — F0–F5#

F0 · Launch hero#

Noema launch hero stating Make the comparison traceable, with the canonical experimental receiver headline and its non-publication-ready disclosure

The repository and webpage hero. It combines the product thesis with a bounded evidence card.#

F1 · Comparison anatomy#

Five comparison boundaries—condition, pairing, aggregation, metric, and role—feeding a claim that survives the audit

The minimum visual explanation of why the surrounding contract matters more than a percentage.#

F2 · Contract-to-evidence chain#

Noema chain from typed protocol through execution, metrics, retained evidence, and launch projection

The system figure for documentation, talks, and the future launch video.#

F3 · Canonical experiment figure#

Log-scale pre-decoder BER versus SNR for uncompensated QPSK, a calibrated diagnostic oracle, and the learned receiver; the learned curve includes an observed min-max band over paired seeds

The canonical launch result. Its band is the observed minimum and maximum across three paired held-out seeds—not a confidence interval.#

F4 · Break the comparison#

Four contract failures around a broken percentage claim: condition mismatch, seed cherry-picking, metric mismatch, and hidden oracle information

The static companion to the interactive flagship demo.#

F5 · One source, many public outputs#

Launch evidence JSON at the center supplying the README, documentation, web demo, experiment figure, tables, and recorded launch video

The launch-production rule: every displayed number and warning comes from one generated source.#

Tables required — T0–T2#

CSV is the transport format; Markdown is the human-readable projection rendered below.

T0 · Paired BER result summary#

SNR (dB)

Uncompensated mean BER

Calibrated-reference mean BER

Learned mean BER

Learned observed min BER

Learned observed max BER

Reduction vs uncompensated

Relative gap to reference

-2

2.545430e-01

2.133725e-01

2.133560e-01

2.129402e-01

2.136230e-01

16.18%

-0.01%

0

2.124840e-01

1.585229e-01

1.585404e-01

1.582680e-01

1.587467e-01

25.39%

0.01%

2

1.720336e-01

1.039540e-01

1.039718e-01

1.035652e-01

1.042442e-01

39.56%

0.02%

4

1.364164e-01

5.634212e-02

5.636883e-02

5.608177e-02

5.659771e-02

58.68%

0.05%

6

1.073624e-01

2.278264e-02

2.278582e-02

2.270508e-02

2.294254e-02

78.78%

0.01%

8

8.453814e-02

6.036758e-03

6.040891e-03

5.994797e-03

6.095886e-03

92.85%

0.07%

10

6.670666e-02

7.495880e-04

7.514954e-04

7.333755e-04

7.810593e-04

98.87%

0.25%

Observed minima and maxima span three paired held-out seeds; they are not confidence intervals. The calibrated reference has diagnostic calibration knowledge. Completed experimental demonstration—not a publication-ready canonical benchmark.

Open T0 as CSV.

T1 · Experiment contract summary#

Contract field

Declared value

Evidence path

Experiment

Learned QPSK Receiver Calibration

identity.title

Result status

completed

identity.result_status

Methods and roles

Uncompensated QPSK (baseline); Calibrated I/Q oracle (reference); Learned I/Q receiver (candidate)

design.methods

SNR cells (dB)

-2, 0, 2, 4, 6, 8, 10

design.snr_db

Paired seeds

71001, 72001, 73001

design.paired_seeds

Retained runs

63

design.run_count

Compared bits per run

1,048,576

design.bits_per_run

Aggregation

arithmetic_mean_across_paired_seeds

scientific_status.aggregation

Statistical unit

paired held-out seed within a predeclared SNR cell

design.statistical_unit

Uncertainty display

observed_min_max_not_confidence_interval

scientific_status.uncertainty_display

Evidence tier

completed_experimental_benchmark

scientific_status.evidence_level

Publication ready

false

scientific_status.publication_ready

Distribution clearance

pending_final_review

distribution_clearance.status

Reference role

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

scientific_status.reference_role

Distribution clearance and scientific status are independent.

Open T1 as CSV.

T2 · Launch claim guardrails#

Guardrail

Required handling

Evidence path

Primary quantitative wording

At 10 dB, the learned receiver reduced mean pre-decoder BER by 98.87% relative to uncompensated QPSK.

headline.text

Adjacent disclosure

Completed experimental demonstration—not a publication-ready canonical benchmark.

scientific_status.disclosure

Uncertainty language

Call bands observed minima/maxima; never confidence intervals.

scientific_status.uncertainty_display

Reference role

Describe the calibrated oracle as a diagnostic reference with privileged calibration knowledge.

scientific_status.reference_role

Scientific tier

Do not call the selected benchmark publication-ready.

scientific_status.publication_ready

Final release state

The development repository is public; stable v0.2.0 and the paper bundle remain pending final review.

distribution_clearance.status

Numeric source

Derive displayed values from series, comparisons, or headline.

consumer_contract.rules

These guardrails apply to README, documentation, webpage, figures, tables, and video overlays.

Open T2 as CSV.

Launch video#

The recorded learned QPSK I/Q calibration walkthrough shows the clean installation, training-contract export, dataset capture, external training, returned-model validation, and UI comparison against both declared baselines. The reviewed video is hosted externally; no local recording or edit file is committed to the source repository.

launch_evidence.json records the stable video identity and viewer-facing URLs alongside the numerical data paths used by the public README, documentation, figures, tables, and demo.