Human–AI Collaboration and Provenance Policy
Purpose
babuni.dev documents not only outputs, but how the human Babuni.dev used AI to make them. What Babuni.dev asks for, notices, challenges, accepts, rejects, or improves is part of the project authorship.
A development diary that only says "AI implemented a grid" misses the decisive contribution: Babuni.dev's toy-box metaphor, shifting-plate behavior, point-origin information bubbles, cell-based bloom correction, and endless recycled layout.
Five distinct kinds of evidence
- Babuni.dev direction — a faithful paraphrase of the intention, creative constraint, observation, question or correction expressed by Babuni.dev.
- AI response / proposal — what an assistant suggested, interpreted, drafted or generated. Clearly labeled as agent-originated.
- Babuni.dev decision — what was explicitly confirmed, changed, or left undecided. Never promote a suggestion to accepted without evidence.
- Implementation action — what was actually changed in source files, designs or deployed systems; include PR/commit references.
- Verification — what was tested and what the observed result was. "Not verified" is a valid answer.
Show chronology, not just conclusions. Prefer a readable narrative that expresses why a particular design happened.
Conversation handling
- Treat relevant conversation content as valuable source material when it is actually available to the current agent.
- Capture meaningful project input from all project sessions: requests, brainstorms, feedback, changes of mind, tests, corrections, failures, and approvals.
- For long exchanges, paraphrase Babuni.dev's directive and why it mattered rather than copying every turn.
- Record relevant exact words only if they are essential and explicitly appropriate to publish.
- Mark material reconstructed from past sessions as retrospective, citing or linking the original source if possible.
- Do not falsely claim that an agent can see or automatically archive every ChatGPT project chat. Existing access may be limited to the current context and connected sources; request missing material when needed.
- Never publish complete raw conversations by default. Exclude non-project personal details, private accounts, credentials and other sensitive material.
- Store public-safe collaboration summaries in docs/DEVLOG.md; complete private source transcripts, if ever retained, require a separately authorized private storage policy.
Required log entry pattern
[Date] — [Feature or discussion]
Context and source: Current session / issue / PR / retrospective / notes.
Babuni.dev's intent: The direction and motivation in neutral, faithful words.
AI's contribution: Proposal or implementation, explicitly attributed.
Iteration: Alternatives considered, corrections, feedback, failure.
Decision: Accepted / rejected / open, with attribution.
Changed artifacts: Files, images, diagrams, code, commits, PRs.
Verification: What was run or observed; uncertainty recorded plainly.
Result: New capability, refined concept, or unresolved question.
Example: The bloom geometry
Babuni.dev's intent: Clicking a tile should physically make room for project content in neighboring cells; the other tiles move away.
AI's earlier interpretation: A central tile surrounded by floating dashboard panels.
Babuni.dev correction: That interpretation didn't reserve actual tile positions.
Decision: Small bloom occupies adjacent cells and displaces their original tiles.
Implementation: Not yet implemented when this example was written.
Value to the journal: It shows Babuni.dev clarifying geometry and the AI correcting a misleading visualization.
Cross-project reuse
This is a reusable policy for Babuni.dev's other repositories, but each project must deliberately adopt it in its own agent rules and ChatGPT project instructions. A babuni.dev file does not automatically govern other projects.
Accountability
Babuni.dev sets direction. Agents should be useful collaborators, not ghost authors of design decisions. Do not infer human approval from silence. If a chat changes a requirement without resulting in code, log the decision but do not present it in CHANGELOG as a shipped feature.
2026-10-27 — Documentation audit #28: scope and provenance
User direction: Continue the documented GitHub portfolio-audit sequence, now reaching babunidev/babuni.dev immediately after ButtonGame. Preserve the origin as a human-directed toy-box interaction and the site as a living record of how Babuni.dev conceived, corrected, accepted, rejected and tested work with AI. Do not flatten those detailed records into a one-line “AI built the portfolio” credit.
AI contribution in this audit: Examine the current Git tree, Astro/TypeScript dependencies, latest navigation/Markdown commits and the substantial existing journals; reconcile README 11 artworks → 15 and older 97-test reports with the later 126-test journal. Record the distinction between 53 candidate tiles, 34 mapped tiles/33 public sources, existing original WebP binaries and 21 archived AI captures. Identify that external project-doc commits require an explicit source refresh before they appear in this portfolio's static snapshot.
Iteration already documented elsewhere: The user corrected AI's suggestion of floating bloom cards, rejected scroll-as-navigation, insisted that tiles retain a coherent stone-and-portal identity through motion, supplied WebP artwork and required visible attribution in the reader. More recent requested adjustments include safe Markdown rendering and persistent header/search navigation; a browser-tested buffered-identity issue was then corrected. Do not invent private dialogue or attribute AI-origin alternatives to human acceptance.
Verification and nonchanges: The checked-in journal reports prior 126 Node tests and browser/build findings, but the assistant did not run new tests, content sync, a production build, real-device trials, or deployment during this document audit. No site source, media, tests, content snapshot, CI, package manifests or artwork was modified. Any future automation, release-license choice or performance change requires its own creator direction and evidence.