HN Debrief

Mea Culpa – Dark Hours

  • AI
  • Open Source
  • Programming
  • Platforms
  • Media

This post is Terry Godier’s follow-up after a messy App Store story unraveled. The original narrative was that Apple had absurdly rejected a legitimate astronomy app as if it were astrology. The new post admits that a web app he launched with Claude closely resembled an existing open source astronomy project called Dark Hours, including a bug, and says he is taking it down and stepping back from using AI this way. What changed the story is that the earlier account left out the key fact that Apple had actually rejected a different app, Asterly, which had started as an astrology app in a category Apple treats as saturated unless a submission is clearly differentiated. Commenters came away believing the astronomy angle was used to launder an ordinary rejection into a much more viral “Apple is incompetent” story.

Treat “AI made me do it” as a governance failure, not a defense. If your team ships AI-assisted code or content, you need provenance checks, basic review, and clear ownership before it goes public.

Discussion mood

Strongly negative. People saw the apology as incomplete and self-serving, mostly because the earlier Apple story omitted crucial facts, the copied app matched an existing project too closely to blame on chance, and the post arrived only after outside scrutiny forced the issue.

Key insights

  1. 01

    The Apple rejection was ordinary, not absurd

    The missing context turns the whole saga inside out. Apple did not ban astrology outright. It applies stricter review to saturated categories like fortune telling and wants new submissions to be meaningfully different. That makes the original complaint look less like platform incompetence and more like an attempt to reframe a routine rejection as a scandal. One commenter added that astrology apps had become a subscription-heavy scam magnet, which helps explain why Apple watches the category closely.

    If a founder comes to you with a platform horror story, ask for the exact rejection text and the full submission history. A lot of outrage disappears once you see the category rules and prior app versions.

      Attribution:
    • Aurornis #1
    • user43928 #1
    • dd8601fn #1
    • runtime_terror #1
  2. 02

    The original post appears to have been revised

    Archived copies show the earlier blog post more strongly implied bafflement that Apple had linked the app to tarot cards at all. The current version now includes language saying the astronomy app began as an astrology app. That edit matters because it suggests the key omission was not just a forgotten detail in conversation with reporters. The public record itself was adjusted after the story started collapsing.

    When a controversy involves changing explanations, pull archived versions before trusting the current writeup. Post-publication edits are often the fastest way to tell whether you are looking at a mistake or a managed narrative.

      Attribution:
    • SamBam #1
  3. 03

    Experienced AI users did not buy the cloning story

    People who use Claude, Codex, and similar tools regularly said the alleged behavior did not match what they see in practice. Their claim was not that models never echo training data. It was that reproducing a project that closely, while keeping the same name and bug, is not a normal failure mode from a generic “build me an astronomy app” prompt. Some also said these tools often add attribution when they lean heavily on a public repo, or they diverge enough that a bug-for-bug copy is especially suspicious.

    Do not rely on vague model behavior claims in an incident review. Ask for prompts, session logs, and commits. If someone cannot produce them, assume your process for AI-assisted development is not auditable enough.

      Attribution:
    • supriyo-biswas #1
    • vikramkr #1
    • arcfour #1
    • mft_ #1
  4. 04

    This is really a provenance and accountability problem

    The sharpest framing was broader than this one apology. AI-assisted development creates a provenance gap for people who cannot inspect what they ship. That is dangerous with apps, but the same issue scales to code review, agents, and internal automation. Once teams start saying the system made the decision, accountability evaporates even though the organization still chose to deploy it.

    If your team uses AI agents, add provenance to your shipping checklist. Keep prompts, source references, diffs, and human sign-off, especially for anything customer-facing or legally sensitive.

      Attribution:
    • bartread #1
    • MrDrone #1
    • padjo #1
  5. 05

    AI disclosure fights are becoming impossible to interpret

    A useful side thread pointed out that “used AI” is too blunt to tell readers anything. A game or app might use AI for translation, bug hunting, boilerplate UI, or a core user-facing asset, and those are not the same risk. That matters here because blanket denials and blanket labels both hide the real question, which is where AI touched the product and whether a human actually validated the result.

    Write internal and external disclosure rules that name the layer of AI use. “AI-assisted translations with native-speaker QA” tells users far more than a binary yes or no badge.

      Attribution:
    • avaer #1
    • teamonkey #1
    • dghlsakjg #1
    • bsder #1
  6. 06

    Gruber took the hit faster than the source did

    Several comments singled out the retraction itself as the most credible part of the affair. Gruber explicitly said he had been misled, owned publishing the piece, and retracted it. That response stood in contrast to Godier’s narrower apology, which focused on careless AI use while leaving many readers unconvinced about the earlier omissions. The comparison sharpened the feeling that the post was triage, not full accountability.

    If you publish based on a source’s grievance, document what you verified and what you are taking on trust. When the source story fails, your credibility depends on how cleanly and quickly you correct the record.

      Attribution:
    • ChrisMarshallNY #1
    • sersi #1
    • simonw #1
    • kstrauser #1

Against the grain

  1. 01

    Model regurgitation is still a real risk

    A minority view held that people are too eager to dismiss the possibility that an LLM reproduced more of a prior project than expected. Ongoing copyright suits, model guardrails around style imitation, and experience with recycled project names all point to a real failure mode where training data leaks into outputs more directly than vendors admit. That does not excuse shipping a clone, but it does mean the tool risk should not be waved away just because the source is untrustworthy.

    If you let coding models generate substantial product code, test for similarity against known repos and check naming collisions before launch. Tooling risk and operator misconduct can both be present at once.

      Attribution:
    • CircuitSeuss #1
    • K0balt #1
    • vintagedave #1
    • viccis #1
  2. 02

    The earlier RSS app may not prove real craftsmanship

    Some people pushed back on the idea that this was a shocking fall from a previously strong reputation. They said Current, the RSS reader that drove a lot of goodwill, already looked and behaved like an AI-built app, with buggy shortcuts and generic patterns beneath a compelling essay and polished branding. That does not prove misconduct, but it weakens the narrative that a careful craftsperson suddenly snapped.

    Do not confuse a strong essay, launch story, or visual taste with engineering quality. If you are betting on a builder, inspect the shipped product and support burden, not just the discourse around it.

      Attribution:
    • huebnerob #1
    • 1123581321 #1
    • terabytest #1
    • latexr #1
  3. 03

    Forking openly would have been acceptable

    A few commenters argued that the underlying act of building on an open source astronomy app was not the real problem. The failure was pretending it was original while publishing under a near-identical name and stripping the social norms that come with a visible fork. In that framing, AI was a distraction from a straightforward open source etiquette failure.

    If you want to commercialize or extend an open source product, do it in the open. Fork it, keep attribution clear, and make your changes legible enough that users know what is original and what is inherited.

      Attribution:
    • scotty79 #1
    • overtone1000 #1

In plain english

App Store ↩
Apple’s marketplace and review system for apps on iPhone, iPad, and related devices.
Claude ↩
A family of AI language models from Anthropic used here to generate code and text.
Codex ↩
An AI coding model and tool from OpenAI for generating and editing software.
LLM ↩
Large language model, a type of AI system that generates and edits text by predicting likely next words from patterns in training data.
open source ↩
Software whose source code is publicly available for others to inspect, use, modify, and often redistribute under a license.
provenance ↩
The record of where code, content, or data came from and how it was created or changed.
repo ↩
Short for repository, a stored collection of source code and related project files, usually managed with version control.
RSS ↩
Really Simple Syndication, a feed format used to subscribe to updates from websites and blogs.
vibe coding ↩
A slang term for building software mainly by prompting AI tools and accepting their output with limited manual understanding or review.

Reference links

Primary story and retraction

Evidence and archives

Related reading on AI copying and copyright

Background on dark adaptation

Referenced products and prior coverage