Space Channel Content Refresh Design

Date: 2026-06-29

Goal

Make the Space Channel phone number useful without depending on live page scraping or live Gemini speech for known content during a call.

The production phone path should:

• answer with static droid audio for known prompts;
• fetch read-only Space Channel data before speaking dynamic content;
• put the dynamic content into slots;
• speak the entire dynamic line through droid_full_buffer;
• pace outbound Telnyx media in 20 ms chunks using the fixed absolute-deadline sender.

Current Implementation

Refresh command:

.venv/bin/python scripts/space_channel_refresh.py --once

Long-running four-hour refresh loop:

.venv/bin/python scripts/space_channel_refresh.py --daemon --interval-hours 4

Cron-style four-hour schedule:

0 */4 * * * cd /Users/davidmar/src/riff && .venv/bin/python scripts/space_channel_refresh.py --once >> logs/space_channel_refresh.log 2>&1

Runtime cache:

data/space_channel/content_snapshot.json

That cache is generated runtime data and is ignored by git. If it is missing or expired, space_content_fetch attempts one refresh during the read-only act state before it speaks.

Data Sources

V1 uses two Space Channel-facing sources:

• Launch Command: https://www.spacechannel.com/data/launches.json, with Launch Library fallback already in space_launches_fetch.
• Dashboard catalog: https://www.spacechannel.com/, parsed for the current Space Channel widget list and subtitles.

The homepage currently exposes widgets such as Launch Command, Daily Briefing, Cosmic Weather, Anomaly Division, Intelligence Feed, ISS Operations, Music, and Solar Operations. The phone briefing uses that catalog so callers can ask what is online, ask about Space Channel news, solar weather, UFO/UAP files, or radio status, and receive a real current dashboard briefing instead of a generic "queued" message.

Phone Flow

Production route:

+15122777311 -> space_channel_network_ops

Important states:

• greeting: static droid WAV.
• suggestions_announce: static droid WAV, regenerated after the capability text changed.
• fetch_launches: calls space_launches_fetch, then enters report_launches_audio.
• report_launches_audio: droid_full_buffer slot-templated launch summary.
• fetch_space_content: calls space_content_fetch, then enters report_space_content_audio.
• report_space_content_audio: droid_full_buffer slot-templated dashboard or topic summary.
• bug_captured_audio: droid_full_buffer acknowledgement of captured bug text.

Terminal states keep entry.line for text-mode transcripts and evals. The phone path records and speaks the preceding play_audio.text, then transitions to the terminal and hangs up at a logical completion point.

Audio Learning Applied

The diagnostic calls showed that prerecorded audio was not sufficient by itself. The phone path became clear only when outbound Telnyx media was sent as:

• 20 ms chunks;
• pre-encoded mu-law/base64 frames;
• absolute-deadline paced sends;
• no synthesis work inside the paced send loop.

For Space Channel V1, static prompts are still pre-rendered, but dynamic content uses full-buffer droid synthesis plus the same paced sender. This avoids the sentence-by-sentence choppiness that appeared when generation and sending were too tightly coupled.

Next Data Connectors

The V1 cache provides current dashboard presence/status and launch data. The next connectors should deepen the summaries without changing the phone flow:

• Daily Briefing article ingestion.
• NOAA/SWPC solar weather values for the solar weather path.
• UFO/UAP document headline ingestion for the intelligence path.
• Radio/now-playing metadata or live stream handoff.

Those can all write into the same topics object in content_snapshot.json; the phone flow only needs space_content_summary.

Verification

Focused test command:

.venv/bin/python -m pytest \
  tests/test_space_channel_content.py \
  tests/test_space_channel_tools.py \
  tests/test_space_channel_flow.py

Live refresh smoke test:

.venv/bin/python scripts/space_channel_refresh.py --once --json

Static menu audio regeneration:

.venv/bin/python -m riff.static_audio generate \
  --flow space_channel_network_ops \
  --state suggestions_announce \
  --force