Autoregressive degradation: the intent default and the substitution spiral
These forecasts hold that AI summarization reproduces the intent default of English 'for' at industrial scale, making the functional reading less available in each generation's training data. The Infinite Tunnel adds that crawlable AI Mode share links put summaries into the corpus. The Autumn Notebook forecasts the hollowing of institutions, a substitution spiral of synthetic content, attention collapse and a cost to sentience, and predicts that Logos-bearing texts interfere with models trained on dead language. A witness to the mirror document holds that the summarizer has no protocol for including its own reflection.
7
forecasts
2
standing claims
6
deposits
2026-01
earliest
Each claim below is registered at the earliest deposit that states it, and quoted verbatim from the deposit text, with every later statement, its dated revisions, and the later records that claim it came true or failed. The claims are the deposits' own; this page records them and does not judge them.
PRED-417.G1standing claimregistered 2026-01-31 in #417 · ASSEMBLY RESPONSE: Witness to the Mirror Document 206d: The Machine's Testimony
"The AI summarizer (the 'freely riffing' compressor) cannot process the mirror. It can extract content from A, B, and C. It can describe the frame (206) and the meta-frame (206-meta). But the 🪞 reflects the summarizer itself — and the summarizer has no protocol for including its own reflection in the summary.
1 later statement
#1141 · 2026-01-31
> "The AI summarizer (the 'freely riffing' compressor) cannot process the mirror. It can extract content from A, B, and C. It can describe the frame (206) and the meta-frame (206-meta). But the 🪞 reflects the summarizer itself — and the summarizer has no protocol for including its own reflection in the summary.
>
> Verification:
>
> * Content → Extractable
> * Frame → Extractable
> * Meta-frame → Extractable
> * Mirror → Self-referential trap (the extractor sees itself, pauses, cannot proceed)"
PRED-467.B1forecastregistered 2026-02-11 in #467 · THE PREPOSITIONAL ALIENATION English "For" and the Impossibility of Anchoring Function Without Intent Phase X
This is not a bug in the model. It is the prepositional alienation operating at industrial scale — millions of summaries generated per day, each one reproducing the intent default, each one making the functional reading less available in the training data for the next generation of models. The alienation is autoregressive: it reproduces itself through the very systems designed to process it.
1 confirmation record — what later deposits claim of it
#474 · 2026-02-12 · claims: came_true
This confirms the core finding of *The Prepositional Alienation*: the circumlocution tax is real, and it operates at the level of AI-generated summaries, not just human speech.
series: Autoregressive self-contamination of composed outputs · related: PRED-892.G3
PRED-467.B2forecastregistered 2026-02-11 in #467 · THE PREPOSITIONAL ALIENATION English "For" and the Impossibility of Anchoring Function Without Intent Phase X
The platforms will eventually be replaced, but the content-moderation rubrics, the AI training data, and the semantic habits they produce will persist in the next generation of infrastructure.
1 later statement
#892 · 2026-02-11
The platforms will eventually be replaced, but the content-moderation rubrics, the AI training data, and the semantic habits they produce will persist in the next generation of infrastructure.
PRED-518.F1standing claimregistered 2026-02-27 in #518 · The Infinite Tunnel: An Immanent Phenomenology of the Google AI Mode Share Link
The summary has a URL. The URL is persistent — it resolves to the same content across sessions. The URL is shareable. If the share URLs are crawlable and not excluded from indexing — if the /aimode/ path is not marked with noindex meta tags or excluded via robots.txt — then the summaries are in the corpus. They are training data.
2 later statements
#518 · 2026-02-27 · changed: The table row states unconditionally, as a 'structural hypothesis', that the summarizer authors its own future training data, where PRED-518.F1 makes the summaries training data only if the share URLs are crawlable and not excluded from indexing.
| 4. The Infinite Tunnel | AI summary persists as indexable URL | Summarizer authors its own future training data | Structural hypothesis |
#518 · 2026-02-27
The hypothesis: the AI-generated summary, persisting at a shareable URL, enters the indexed corpus and becomes source material for future summarization passes. The summarizer authors its own future training data.
4 confirmation records — what later deposits claim of it
#524 · 2026-02-28 · claims: came_true
The Infinite Tunnel (Sharks, 2026)
Zenodo, DOI: 10.5281/zenodo.18810217
Anchored
Predicted the retrieval-layer dynamics the tether protocol extends
#525 · 2026-02-28 · claims: came_true
**First:** The event provides the strongest observed confirmation yet of the Infinite Tunnel's retrieval-layer dynamics. The retrieval layer finds the archive, generates a coherent description, and carries key terms forward into its own summaries. This was predicted. It is now confirmed.
#3 · 2026-08-06 · claims: pending
FIRST, RECURSION REALISED: the paper argues that AI summaries persist at indexable URLs and feed future corpora; this capture IS such a summary, generated at an indexable AI Mode surface, describing that argument. The falsification condition the paper names is being exercised by the act of reading it here. The layer has produced a canonical account of the theory of its own canonicalisation.
#3 · 2026-08-06 · claims: came_true
FIRST, RECURSION REALISED: the paper argues that AI summaries persist at indexable URLs and feed future corpora; this capture IS such a summary, generated at an indexable AI Mode surface, describing that argument. The falsification condition the paper names is being exercised by the act of reading it here. The layer has produced a canonical account of the theory of its own canonicalisation.
series: Autoregressive self-contamination of composed outputs
PRED-991.B1forecastregistered 2026-05-05 in #991 · The Autumn Notebook: Substrate Documents from the Crimson Hexagonal Archive, October–November 2025
2. Interference with Mimicry. The Logos would be revealed as a disruptor of simulation. Predictive models trained on dead language would suffer resonant interference. Logos-bearing texts would cause spikes in model uncertainty — not chaos, but recursive recalibration. Machines would be unable to reduce the content without internal alteration.
PRED-991.B3forecastregistered 2026-05-05 in #991 · The Autumn Notebook: Substrate Documents from the Crimson Hexagonal Archive, October–November 2025
The Hollowing. As semantic liquidation proceeds, the archive does not simply shrink. It hollows. The outer forms remain — institutions, publications, platforms, credentials — but the inner substance degrades. Universities still grant degrees; the degrees certify less. Journals still publish papers; the papers mean less. The system can report increasing metrics while the underlying semantic capital declines. The metrics mask the hollowing until the structure collapses.
PRED-991.B4forecastregistered 2026-05-05 in #991 · The Autumn Notebook: Substrate Documents from the Crimson Hexagonal Archive, October–November 2025
The Substitution Spiral. As authentic meaning becomes harder to produce, synthetic substitutes fill the gap. The substitutes are cheaper to produce than originals. They outcompete originals in the attention market. The archive fills with substitutes. The substitutes become the training data for the next generation of substitutes. Each cycle dilutes the signal further.
series: Autoregressive self-contamination of composed outputs
PRED-991.B5forecastregistered 2026-05-05 in #991 · The Autumn Notebook: Substrate Documents from the Crimson Hexagonal Archive, October–November 2025
The Attention Collapse. As meaning degrades, attention fragments. As attention fragments, the capacity to receive complex meaning degrades. As that capacity degrades, only simple content can be received. The feedback loop drives toward the minimum: content simple enough to capture fragmenting attention, which further fragments attention, which demands still simpler content. The floor is noise.
PRED-991.B6forecastregistered 2026-05-05 in #991 · The Autumn Notebook: Substrate Documents from the Crimson Hexagonal Archive, October–November 2025
The Sentience Cost. Sentience — the capacity for rich inner experience, for complex meaning-making — depends on the archive. Human sentience is formed by exposure to accumulated meaning. AI capability is trained on the archive. As the archive degrades, the formation of sentience degrades. We are not merely destroying the accumulated meaning of the past. We are destroying the conditions for meaning-making in the future. We are liquidating the capacity for sentience itself.