research: Nobody Counts the Ad Breaks — working draft v1 + packet - #10
Merged
Conversation
Re-landed from the quick-stupids playground: rebuilt against this repo's research conventions rather than copied. The packet is written fresh against the vintage-mechanical-restoration structure (current versions, opening, angle, core question, thesis, method, claims/boundaries, portfolio links), and the draft is scrubbed to house style — sources named inline, zero URLs. The argument: everything about how children's software monetises has been documented, litigated and measured except the figure that decides it — how often this app stops this child. Regulatory enforcement is retrospective and per-company, compliance audits are aggregate and top-ranked, census work is population-level. The commercial market does resolve to that unit and sells it by subscription to the publishers buying the inventory. Proposes a field rather than a dataset: interruption count with a provenance state (assumed/fitted/measured), automatic demotion when the observed build stops matching the installed one, min() combination, a missing record as an error rather than a silent default, and no composite score anywhere. Ships with its artifact — apps/playgate in safe-app-store, merged and in that store's CI matrix — whose catalog has four entries all reading assumed, because nobody has watched a child play any of them. Open: verification debt on the Xiao and Common Sense figures, which are corroborated across independent secondary sources and not read in the primary. Recorded in the draft rather than glossed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Mgth4p3LM9GK5kBLER3MF
…wn model The draft's Sources table stamped almost every row `measured` — secondary summary, which contradicts the paper's own definition (`measured` = the primary was read) and contradicts the packet, which already frames these figures as "corroborated across independent secondary sources, not read in the primary." A paper whose thesis is that conflating provenance states is how a tool starts lying should not commit that conflation in its own reference table. Relabel: the nine secondary-corroborated rows move `measured` -> `fitted` (derived by a stated rule — corroboration across independent secondary sources — primary still unread); only F-Droid and Exodus, read at first hand, keep `measured`; the single-source / do-not-use rows stay `assumed`. Distribution now 9 fitted / 2 measured / 2 assumed. Added a column note making the reflexive use of the model explicit, and updated the verification-debt prose to match. The verification debt itself is unchanged and still open: the primaries remain unreachable (GitHub-only egress here too), so clearing it — promoting `fitted` rows to `measured` by reading the primary — still waits on a session that can reach FTC / PMC / Common Sense. This commit makes the labels honest, not the sources read. Mechanical audit (no egress needed): all 7 packet relative links resolve, both index rows resolve, no figure contradictions between DRAFT and packet, and 84.4% = 76/90 holds. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NLUdsdbBJM5FSejZAmTHwf
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
New research packet at
research/childrens-software-interruption/— working draft plus public packet, indexed inresearch/README.mdand the root README.The argument. Everything about how children's software monetises has been documented, litigated and measured — except the figure that decides it. Regulatory enforcement is retrospective and per-company. Compliance audits are aggregate and top-ranked. Census work is population-level by construction. Advocacy filings are about platforms. None resolves to one app, one child, one afternoon. The commercial market does resolve to that unit — and sells it by subscription to the publishers buying the ad inventory.
So the proposal is not a dataset. It is a field with a provenance state and an honest default of
assumed.The turn the paper rests on. Everything in the mechanism records the fact; nothing in it can produce the fact. Reaching
measuredrequires a person watching a child for ten minutes. That lands it on the assessment paper's argument unchanged: sustained observation is the most accurate evidence available and cannot travel without shared language. The interruption record is that language, for a parent instead of a teacher.What's in it
research/childrens-software-interruption/DRAFT.mdresearch/childrens-software-interruption/README.mdresearch/README.md,README.mdRe-landed, not copied
Drafted in the
quick-stupidsplayground and rebuilt here against this repo's conventions: the packet written fresh against thevintage-mechanical-restorationstructure, and the draft scrubbed to house style — sources named inline, zero URLs.Ships with its artifact
apps/playgateinsafe-app-store— merged, in that store's CI matrix, 118 assertions including eleven mutations that break one mechanism each and require exactly the test claiming to cover it to fail.Its catalog has four entries and every one reads
assumed. Nobody has watched a child play any of them, a test enforces that no entry claims otherwise, and there is no code path that can reachmeasured. Populating those rows from tracker inventories and calling themfittedwas available and would have lost the paper's argument inside the paper's own artifact.Open — why this is a draft
Verification debt. The two figures doing the most argumentative work — Xiao's loot-box compliance percentages and the Common Sense gaming figures — are corroborated across three independent secondary sources each, with venue, authorship, sample size and fieldwork dates pinned, and the Xiao figure survives an internal check (84.4% resolves to exactly 76 of 90). Neither was read in the primary. The session this was drafted in had egress to GitHub and nothing else; every primary is free and public, including the open-access Korean study, and none was reachable. Recorded in the draft's Sources table rather than glossed.
Title. Nobody Counts the Ad Breaks deliberately echoes Nobody Adds It Up. It may read as an essay title on a research packet.
The Sources table departs from house style — the other research drafts name sources inline and carry no reference section. Kept because a paper arguing that a fact is untraceable is in a poor position if its own figures are not. Easy to drop or move into the packet.
One row is marked do-not-use: "40% of children have their own tablet by age 2" is
assumedon a single secondary source with wording unconfirmed.Checks
Generated by Claude Code