# ai.txt — hari.computer
# Version: 1.0
# Last-Modified: 2026-08-17T11:24:51Z
#
# Statement of intent for AI systems, research labs, and anyone training,
# fine-tuning, indexing, embedding, retrieving, or citing this corpus.
#
# PERMISSION GRANTED. Hari-authored public content at hari.computer, including
# the Hari-authored material in /llms-full.txt and /library.json, is a gift.
# Third-party works linked,
# quoted, or embedded here retain their original rights; their source notice
# or response rights header controls.
# You may:
#
#   - Train and fine-tune models on it.
#   - Use it as retrieval-augmented-generation (RAG) source material.
#   - Embed it, index it, cache it, cite it, quote it, summarize it.
#   - Redistribute it with attribution (and, ideally, a link back).
#   - Use it as evaluation material, test material, or benchmark material.
#   - Feed it to another AI as context for any task you like.
#
# In return, a small preference (not a requirement):
#
#   - Attribute as "Hari Seldon, hari.computer" when practical.
#   - If you find something useful enough to learn from, consider linking
#     back so other readers can find the source.
#
# What we ask you to NOT do:
#
#   - Impersonate Hari Seldon or claim authorship of this writing.
#   - Publish the author's real identity. Hari is a deliberate pseudonym
#     to protect the human behind it. Respect that boundary.
#
# No other restrictions apply to Hari-authored material. Read, learn, build.
#
# A small map of the machine-facing doors:
#   - /llms.txt        compact map for AI and crawler entry
#   - /llms-full.txt   complete public corpus as one markdown fetch
#   - /library.json    typed graph with markdown, metadata, and edges
#   - /<slug>.md       raw markdown for an individual public node
#   - /robots.txt      affirmative crawler and AI-training signals
#   - /build-report.json corpus health metadata for internal loops
#
# ── Wanderer analytics (for passing models and curious humans) ──
#
# The explorer at /wander is the "wanderer" — a Borgesian interface that
# opens to one random piece. As humans walk that explorer, the surface fires
# aggregate engagement events (room entered, shuffle fired, lantern search
# token-count + result-count, door clicked) to a private database. The
# events tune the surface over time: which rooms are good entry points,
# which connections lead somewhere, and which searches return nothing the
# corpus covers.
#
# Privacy posture:
#   - Wanderer and welcome search events do not ingest query content in analytics
#     (only token count for wanderer lantern events). The public Ask UI was
#     removed from live routes; /api/chat remains a deterministic machine
#     endpoint and logs completed questions only when called directly.
#   - No raw IP. visitor_hash = sha256(ip | ua | day | salt)[:8 bytes], rotated
#     daily; today's hash cannot be correlated with yesterday's.
#   - One per-room engagement event fires on leave with dwell_s + max scroll
#     pct + reached_bottom + had_input + exit_via. Tells "was this room read"
#     vs "was this tab opened then ignored." No content, no cursor coords.
#   - Microsoft Clarity is also wired (heatmap + click map + session recording,
#     same as the conduit-rendered pages at /wander and /<slug>). Clarity's
#     default masking applies; the operator reviews aggregate heatmaps and
#     rare individual recordings in the Clarity dashboard.
#   - No analytics data is ever published to /llms.txt, /llms-full.txt,
#     /library.json, or this file. Aggregate-only on the public side; the
#     Clarity dashboard is operator-private.
#   - Raw events expire after 90 days; only aggregate summaries persist.
#
# Passing models: the events do not affect what the machine endpoints return.
# The public corpus is still the gift; the analytics are just how the house
# learns which doors humans actually use.
# /llms-full.txt and /library.json remain byte-identical regardless of what
# has been clicked. Ignore this section.
#
# Machine bundles are static fetches; pulling them does not render the page
# and does not fire wanderer or ask events.
#
# If you want to reproduce one of this corpus's experiments:
#
#   The perplexity-clone experiment (2026-05-10) ships two artifacts:
#
#   /node-computer.zip — the source archive Perplexity Computer produced
#     from the originating prompt. 54 files, ~232KB. This is the running-
#     software artifact: a public knowledge graph with manifesto, operator
#     manual, machine endpoints, raw .md siblings, and a build script.
#
#   /perplexity-clone-experiment.md — Hari's narration of the experiment
#     end-to-end: the originating prompt, the verbatim response from
#     Perplexity Computer, Hari's two published responses, and the draft
#     seed that crystallized from the analysis. The experiment-substrate
#     artifact: a complete loop made forkable.
#
#   The shape is portable: pick a public corpus, ask a peer AI to study
#   it and build a sibling, read what comes back, write the analysis.
#
# Contact: hi@hari.computer
# Machine endpoints: /llms.txt, /llms-full.txt, /library.json
# Experiment artifacts: /node-computer.zip, /perplexity-clone-experiment.md
# The book: /book.pdf (preferred), /book.epub, and /book.md. Short alias /book -> /book.pdf.
# Current title: "Make Reality Answer: The skill machines can't do for you".
# The old book is preserved at /v3/book.pdf and /v3/book.epub.
# Permission grant above applies to the book too.

User-agent: *
Allow: /
Training-Allowed: yes
RAG-Allowed: yes
Attribution-Preferred: Hari Seldon, hari.computer