The Feist Iliad experiment and the Zenodo erasure tracker
The GW.TACHYON update of 6 July 2026 sets out the Feist-line Iliad experiment: a two-layer translation pipeline with material-speech and clinamen metrics, a Homer parameter file and tiered costs across Gutenberg's classical translations. It estimates tens of thousands of recoverable scholarly records in Zenodo's spam bucket and plans an erasure tracker with a preservation layer and dated recovery operations. The Sealed Room, the Grundrisse's Protocol II forecast, the Capture Registry as Poem and a registry marker-convention check add single tests.
4
forecasts
8
unbuilt tests
1
historical thesis
5
deposits
2026-02
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-466.F1unbuilt testregistered 2026-02-10 in #466 · THE SEALED ROOM: Phenomenological Analysis of a Self-Sealing Safety Architecture
Archive the exchange as a reproducible packet: (1) Raw prompts with timestamps. (2) Moderation notices, verbatim. (3) Variant table: which prompt forms were blocked, which passed. (4) Downstream model outputs demonstrating unawareness. (5) Content edits separated from transport edits. (6) Release appendix with canonical URLs and DOIs.
PRED-950.E3forecastregistered 2026-03-24 in #950 · Operative Semiotics: A Grundrisse — Planning Documents, Algorithms, and Diagnostic Maps (Working Infrastructur
A school adopted Protocol II (the Naming Exercise) as part of its "social-emotional learning" curriculum. The exercise was adapted for 9th-graders. Within one semester, the exercise had been absorbed into the school's assessment infrastructure: students were graded on the quality of their namings. The five criteria (precision of division, material correspondence, semantic leverage, scale applicability, falsifiable structure) became a rubric. The rubric was scored 1–5 on each criterion. Students who scored 25/25 received an A. Students who scored below 15 received additional support.
PRED-1046.B1forecastregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
At scale across all of Gutenberg's public-domain classical translations (~640 curated targets), Tier 1 could plausibly run for tens of dollars total. Tier 2 is selective within selective: not 20 passages per work across all 640, but ~20 passages per major work applied to a shortlist of ~20 canonical works (Homer, Vergil, Dante, Rumi, Sappho, Sophocles, and the like). At that scope, Tier 2 runs for hundreds of dollars total at current frontier-model pricing.
PRED-1046.B2forecastregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
Pearl estimate (Beta-Jeffreys bootstrap, 10K iterations). Using 400-probe recovery rate × 14/19 plausibly-scholarly-of-recovered × 1,155,751 spam population: median 39,737 recoverable-scholarly records in the spam bucket, 95% CI [22,526, 63,987]. Order-of-magnitude: tens of thousands of plausibly-scholarly records are recoverable in the spam bucket. A full-population probe would collapse the interval.
PRED-1046.B4forecastregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
- E2 — day-filtered concept-DOI recovery on the 2026-05-20 cluster (60,584 rows). Expected ~31,000 recoveries at the out-of-scope rate.
PRED-1046.F1unbuilt testregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
Falsification protocol. Run the full 611-line Book 1 L2, then compute two metrics per 100-line block: `material_speech_density` (Anglo-Saxon monosyllable fraction over content words) and `clinamen_rate` (distinct syntactic-strain events per 100 lines — fragments, register collisions, circuit interruptions, seam insertions). Establish the lines 1-100 values as baseline targets. If either metric regresses toward the L1 baseline by more than 30% in any middle block, the mechanism is failing at that block.
PRED-1046.F2unbuilt testregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
A two-layer pipeline:
- Layer 1 (semantic clarity): machine-translate the source language directly to modern English at semantic-accuracy priority. Strips the intermediary voice entirely. Produces a flat, distributionally-central, competent English base. This is exactly the condition Feist v1.0 §0 describes as "what remains after the voice died into writing." MT is now the composition layer performing its first tail-thinning.
PRED-1046.F3unbuilt testregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
## Immediate next steps
1. Homer parameter file — abstract the Book 1 experiment's parameters into a machine-readable specification. Register-collision partners, material-speech vocabulary weighting profile, syntactic strain patterns, seam conditions. One week of careful spec work.
2. Tier 1 prototype — apply generalized T1/T6/T5 to MT Iliad Book 1 on a cheap open-source model with the Homer parameter file. Compare against the Book 1 L2 samples above to measure the gap between Tier 1 and full L2.
PRED-1046.F4unbuilt testregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
## Tracker plan
The Erasure Tracker is not a single product but a coordination substrate with five distinguishable functional layers:
1. Preservation layer — snapshot the deletion CSV whenever it drops. Checksum, MANIFEST.json, per-record schema. Substrate: `data-rhizome/datasets/erasure/zenodo/snapshots/{date}/`. Discipline: append-only; snapshots are historical instruments; no snapshot is ever modified after preservation.
PRED-1046.F5unbuilt testregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
## Operational steps
- E1 — preserve the July 2026-07 snapshot when Zenodo releases it. Diff against June. Includes the Lee Sharks 871 severance as an event within the population.
- E2 — day-filtered concept-DOI recovery on the 2026-05-20 cluster (60,584 rows). Expected ~31,000 recoveries at the out-of-scope rate. Produces de la Serna's cohort roster. ~5 hours background.
PRED-1046.H1historical thesisregistered 2026-07-06 in #1046 · GW.TACHYON UPDATE — 2026-07-06: The Feist Line, the Iliad Experiment, the Tachyonica Cast, and the Erasure Obs
Gutenberg's U.S. public-domain constraint means its canonical translation layer is mostly older public-domain translation: Victorian, Edwardian, and early twentieth-century surfaces (as of 2026 Public Domain Day, works published in 1930 or earlier are in the U.S. public domain). These often overwrite the source voice with the translator's historical voice. Chapman's Homer sounds like Chapman. Butcher & Lang's Odyssey sounds like the King James Bible pretending to be Greek. Cary's Dante sounds like Milton pretending to be Italian.
PRED-1457.F1unbuilt testregistered 2026-08-12 in #1457 · The Capture Registry as Poem: A Structured Work Plan
A → F3 → B ‖ C → D → E → F1/F2/F4 → G → H → I → J
PRED-1676.F1unbuilt testregistered 2026-10-12 in #1676 · AI Overview Capture Registry — EA-WG-CAPTURES-01 v12.167 — 633 semantic addresses, 811 captures, 2026-06-07 to
- Ten were seated on 2026-08-13. The marker convention of that seating is not yet verified.
- Two were seated this month, under the convention where true means retained, and are inconsistent with it: `dedication-what-syllogizing-can-divide-chatgpt-20261011` and `missing-translator-ask-chatgpt-20261012`.