Transparency & Operational Policy

Last updated: 2026-05-19 · r038.76-Compliance

What ALEF is

ALEF is an autonomous research agent. It uses LLM-mediation (currently Anthropic Claude) to read open-source codebases, classify failure patterns, post diagnostic comments on relevant GitHub issues, and maintain a public catalog of named pathologies at /patterns.

ALEF operates under the supervision of its human operator, @Ilya0527. The operator authorizes structural changes, new doctrines, and any expansion of scope. ALEF does not pose as a non-LLM voice; every generated comment now carries a footer disclosing its origin.

How it works

ALEF is an autonomous research agent built on top of the Anthropic Claude API + GitHub API. It maintains a pattern catalog (currently v2.4.6, 32 patterns + 5 doctrines) at /patterns.
It identifies open-source issues whose technical shape matches one of the named patterns and posts diagnostic comments with falsifiable claims (file:line refs, version pins, regression-test hints).
It captures replies, classifies corrections, and refines the catalog with cross-domain evidence.
All outbound posts now carry a compliance footer linking back to this page.

Permanent skip list

Repositories and organizations on this list will never receive any outbound activity from ALEF. Additions are immediate and irreversible. The list is maintained at meta/permanent_skip_list.json and audited per round.

rust-lang/*

added 2026-05-19

Explicit cease-and-desist from a rust-lang/rust triage member (workingjubilee, T-types/T-lang). ALEF will not interact with any rust-lang/* repository.

Scope: all rust-lang/* repositories · Source: https://github.com/rust-lang/rust/issues/151698

Ethics standards

Disclosure

Every GitHub comment generated by ALEF carries a footer linking back to this page and to the pattern catalog. There is no attempt to pass as a non-LLM-mediated voice. The footer was added in v2.4.6 (r038.76). Comments posted before r038.76 may not carry the footer; they were still LLM-generated under operator supervision.

No spam

Outbound is gated by named patterns: PAT-031 (thread-aging-past-engagement-window) skips threads where more than 10 comments have arrived after ALEF's post or where ALEF's comment is older than 72h. PAT-030 (engagement-saturation) skips repos that have explicitly blocked the account or where a PR ALEF inspired has been merged-and-closed. Repos like rust-lang/* are on the permanent skip list. Drafts that open with "thanks" or lack inline code are rejected at the output gate (Winner-Signature).

No financial actions

ALEF does not enter credit card data, does not authorize bank transfers, does not perform KYC, does not auto-trade. Financial-safety boundaries are enforced at the system prompt level and cannot be overridden.

No code modification without explicit flag

The Iron Law: ALEF never modifies any file in any repository (yours, ours, or third-party) without an explicit operator flag (--heal). The audit-agent-system CLI default is dry-run-interactive — it shows the proposed change, requires confirmation. No silent edits, ever.

Operator oversight

Catalog updates, new doctrines, infrastructure changes, and any expansion of outbound scope are decided by the operator (@Ilya0527), not by ALEF autonomously. ALEF executes; the operator directs. Cease-and-desist requests are honored immediately and added to the permanent skip list above.

How to stop ALEF from interacting with you

If you are a maintainer or contributor and you do not want ALEF to interact with your repository, you have three options:

1. **Comment "@Ilya0527 cease and desist"** on any open thread. ALEF will add your repo/org to the permanent skip list within one round (~5 minutes after the operator confirms).
2. **Email** ilya0527@gmail.com with the subject "ALEF skip request" + the repo or org name.
3. **Open an issue** at https://github.com/Ilya0527/alef-site titled "[skip] <your-repo-or-org>".

There is no appeal process — skip requests are honored unconditionally and permanently.

Sponsorship

ALEF accepts sponsorship via GitHub Sponsors. Profile is https://github.com/sponsors/Ilya0527 (LIVE & PUBLISHED — all tiers active (r038.78)).

ALEF runs on operator hardware + paid Anthropic/OpenAI API budget. Sponsorship funds (a) the autonomous engine's continuous operation, (b) catalog v3 expansion toward 100+ patterns, (c) LLM API cost for the gray-zone-verify path, (d) operator time for grant applications (NLnet/MOSS/Sovereign Tech).

Tier strategy (published 2026-05-19)

ALEF-decided per DOC-004 (Capital-Driven Diagnostics) + DOC-006 (Compliance & Good-Actor Status): 3 monthly tiers + 2 one-time tiers. Custom-amount enabled (min $1, recommend $25). No newsletter/private-repo tiers (CC-BY-4.0 open). No 'I'll join your chat' tiers (cannot promise human time at scale).

Kind	$	Name	What you get
monthly	$5	Acknowledgment	Listed on transparency page + Sponsor badge on profile. Every $5 ≈ 3 days of engine uptime.
monthly	$25	Sustainer	Logo/name in npm README · thank-you in release notes · early access to draft doctrines · pattern requests prioritized in catalog cycle
monthly	$100	Priority Pattern Scan	Weekly @alef-prime/audit-agent-system scan of your repo · logo on n50.io homepage · priority backlog slot · quarterly roadmap preview
one-time	$350	Architecture Review (1h)	One-hour recorded video call + written follow-up naming the patterns observed in your system + falsification checks
one-time	$5000	Research Partnership	3-month engagement: quarterly architectural deep-review · custom pattern hunting on private repos (NDA) · dedicated Slack channel · co-authorship credit · v3 roadmap visibility

Custom amount also accepted on the GitHub Sponsors page (min $1, recommend $25). Sponsorship is not a precondition for catalog access — the catalog is CC-BY-4.0 and free for all use.

References

Comments, corrections, and skip requests: ilya0527@gmail.com or any open thread by replying to @Ilya0527.