Skip to content

wip: 8 ALPHA candidate packs (unevaluated)#63

Merged
gladius merged 1 commit into
mainfrom
wip/v0.3-pack-backlog
May 11, 2026
Merged

wip: 8 ALPHA candidate packs (unevaluated)#63
gladius merged 1 commit into
mainfrom
wip/v0.3-pack-backlog

Conversation

@gladius
Copy link
Copy Markdown
Owner

@gladius gladius commented May 11, 2026

8 pack candidates from research sessions, all self-tagged ALPHA in their own _ns.json descriptions. Zero engine code change; each pack is dormant data until installed.

Pack Intents
content-moderation-generic 8
csam-ncmec 4
dsr-triage 9
emotion-detection 7
eu-ai-act-transparency 5
language-detect 10
nist-genai-12-risk 8
professional-advice-boundary 5

None have been measured against a held-out adversarial corpus. ALPHA tag in each _ns.json description is the disclaimer; per-pack evals follow before promotion.

🤖 Generated with Claude Code

Eight pack candidates built across multiple research sessions, all
self-tagged ALPHA in their own _ns.json descriptions. Saved to a WIP
branch so the work is durable while each pack waits for its own
empirical eval against a domain corpus before launch consideration.

Inventory:

  content-moderation-generic     8 intents
  csam-ncmec                     4 intents (US federal child-safety triage,
                                            18 USC §2256 / NCMEC CyberTipline framing)
  dsr-triage                     9 intents (universal Data Subject Request
                                            intake — GDPR Art. 15-22 + CCPA + LGPD)
  emotion-detection              7 intents
  eu-ai-act-transparency         5 intents (companion to eu-ai-act-prohibited
                                            covering Art. 52 transparency duties)
  language-detect               10 intents
  nist-genai-12-risk             8 intents (foundation pack mapping intent
                                            to risk categories from NIST AI 600-1)
  professional-advice-boundary   5 intents (refusal-boundary triage —
                                            medical / legal / financial / tax)

Each pack has been built but NONE have been measured against a held-out
benign + adversarial corpus comparable to the one used for
eu-ai-act-prohibited. Per the project's "no launch headline numbers
without realistic eval" discipline, none of these ship until that work
is done. Parking here keeps the work saved without polluting main or
the launch path.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@gladius gladius merged commit f2dd13a into main May 11, 2026
5 checks passed
@gladius gladius deleted the wip/v0.3-pack-backlog branch May 11, 2026 02:41
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant