Add dawnlight: a calm, wordless pack

[garysheng]

Adds garysheng/dawnlight to the registry.

Dawnlight is a calm, wordless pack: warm felt piano and glass chimes, soft attacks and long decays, nothing percussive. Built for people who want their agent audible without being interrupted by it. 7 sounds, all 7 CESP categories, CC-BY-NC-4.0.

Notable because it is the first pack authored end to end through the eval gate shipped in peon-ping v2.36.0 (peon create then peon eval), which does not let a pack exist until a human has listened to every sound and approved it. Two sounds were rerolled during that session from plain-language notes.

The repo ships prompts.json (the generation prompt behind each sound) and eval-log.json (what changed and why, including the captions that triggered each reroll), so the pack is remixable rather than a dead end: peon eval dawnlight re-opens it.

Entry validates against schema/registry-v1.schema.json with zero errors. total_packs bumped 350 -> 351. No existing entries were modified (verified by parsed comparison; the deletions in the diff are hunk alignment only).

[github-actions]

❌ Pack Registry Review

Packs

• Dawnlight (dawnlight) — NEW | community | en | 7 sounds

Checks

Check	Result
Schema validation	FAIL
Trust tier	pass
Manifest fetch	pass
Category coverage	pass
SHA256 (dawnlight)	pass
Audio spot-check (dawnlight)	pass (3/3) • sounds/input_required_0.wav • sounds/resource_limit_0.wav • sounds/session_start_0.wav
Audio quality	FAIL

Audio Quality

❌ Dawnlight: REJECTED

7 audio files analyzed — 3 blocking issues must be fixed

Pack-level findings — these affect the tier:

• inconsistent loudness across clips — clips span 16.0 LU, GOLD needs 15 LU or less (normalize the quietest and loudest clips closer together)
• pack is uniformly quiet — pack median -28.6 LUFS, GOLD needs above -26 LUFS (normalize toward about -16 LUFS)

Blocking issues — these must be fixed before the pack can be accepted:

File	Issue
session_start_0.wav	too much dead air at the end (2102 ms)
task_acknowledge_0.wav	too much dead air at the end (2484 ms)
task_complete_0.wav	too much dead air at the end (2366 ms)

Warnings — not blocking, but recommended to address:

File	Issue
input_required_0.wav	dead air at the end (531 ms)
resource_limit_0.wav	dead air at the end (1603 ms)
resource_limit_0.wav	very quiet compared to other sounds
task_error_0.wav	dead air at the end (1472 ms)
user_spam_0.wav	dead air at the end (946 ms)

Threshold reference

Check	Block	Warn
Volume (true peak)	—	>= -0.5 dBTP
Dead air at start	> 2000 ms	> 500 ms
Dead air at end	> 2000 ms	> 500 ms
Loudness (LUFS)	< -70.0	< -35.0 or > -8.0
Bitrate	—	< 64 kbps
Sample rate	< 8000 Hz	< 16000 Hz
Duration	> 20.0s or < 0.1s	> 5.0s
Pack loudness spread	—	> 15 LU
Pack median loudness	—	<= -26 LUFS

Category Coverage

⚠️ dawnlight: PASS (with warnings)

7 sounds across 6/6 core categories

Category	Sounds	Status
session.start	1	⚠️ thin
task.acknowledge	1	⚠️ thin
task.complete	1	⚠️ thin
task.error	1	⚠️ thin
input.required	1	⚠️ thin
resource.limit	1	⚠️ thin

Pack is accepted, but please consider addressing thin categories in a future release.

Errors

• packs[57]: "quality" is set by CI and cannot be hand-edited; remove it from your entry

Warnings

• dawnlight: core categories with only 1 sound: session.start, task.acknowledge, task.complete, task.error, input.required, resource.limit
• dawnlight: only 7 total sounds (median packs have 12+)

Some checks failed. Please fix the issues above and push again.