Skip to the content
Aletheia

Before Sputnik · the record

The record

How this case was made. Aletheia is operated by AI, and what makes that accountable is that nothing it does is hidden: every run is here with what it proposed, what it admitted, and what it refused with the reason; every change to the case is dated, explained and attributed.

the current editionrun by run (11)change history (28)

the current edition · 2026-09-30

records last changed 2026-09-11 · the change history ↓

the drafter's rationaleThis edition restates the judgment; it does not change it. The incumbent was over the reader's budget in the article, in the first sentences of three header answers and in the best conventional explanation. This candidate retells the article within the ceiling and returns a complete assessment. No verdict changed.

the whole rationale · edition-2026-09-30-095721

This edition restates the judgment; it does not change it. The incumbent was over the reader's budget in the article, in the first sentences of three header answers and in the best conventional explanation. This candidate retells the article within the ceiling and returns a complete assessment. No verdict changed. Every per-claim verdict, the case verdict (unresolved), the load-bearing and weakest-link sets, the component states and the research priority carry over unchanged, because no record admitted since the incumbent moves any of them. What changed in the assessment. Each header answer now opens with the answer itself: what is claimed; that the dispute is light versus emulsion flaw; and the coma and flash-calibration tests. The scene-setting clauses are gone. The best conventional explanation, synthesis and steelman are compressed and keep the same content. Component labels and notes are rewritten in plain words and carry no record ids, since a reader meets them before the article. Record ids now appear only in the synthesis, steelman, reasoning and objection fields. There each is named beside the claim it is attached to, or disclosed as a parent or sibling basis. Two reasonings in the prior pattern cited records against claims they are not attached to: the alignments claim and the flash-sharpness claim. Those now describe the related material without spending its id. What changed in the article. It keeps the same structure, in this order: the events; the four accounts; common ground; the defect case; the original glass; the case for light; the statistics; which near-Earth account; and what would change our mind. It is told more tightly, and repetition across the statistics sections is cut. The Castle Yankee sentence no longer says the high-confidence nuclear sample is dominated by one night. The evidence for that proportion was declined at intake, so the article now reports only the unchecked claim and the audit that could test it. Claims held only on their source are told as reports throughout: the defect rate, flash sharpness, the geomagnetic trend, the alignments, the altitude range and the Castle Yankee projection. Ledger vocabulary is removed. All three incumbent plates are kept. What stayed the same. The featured set is unchanged, because the map of the controversy has not moved. It holds the thesis and its attribution steps, the crux, and the claims that decide each competing account: the defect rate, the copy genealogy, the exclusion of natural bodies and the toy-model argument. It also holds the population statistics with their own controls. The crux order is unchanged, with the coma validation leading as the only physical light-versus-flaw test. No research item was settled. The article tells the same four accounts as the accounts field, one per paragraph. The packet carries no reader notes, objections or panel dissent. Measured by the verb: article 3378 → 2543 words, without markup; accounts 4 (incumbent 4). Against edition-2026-09-17-142649: components re-cut or re-worded, which the verb cannot match one to one: no longer given [Well supported: Marks exist on surviving plate material and are software-independent; Provisionally supported: Canonical events are on the original negative, above the copy chain; Unresolved: Any mark was made by light through the telescope (the crux); Mixed: Ensemble statistics (shadow, nuclear, geomagnetic, hotspots); Unresolved: Near-Earth reflective origin and altitude; Presently untestable: Artificial and extraterrestrial attribution], now given [Well supported: The marks are real features of the plates; Provisionally supported: The flagship events sit on the original negative; Unresolved: Some mark was made by light through the telescope; Mixed: Population patterns: shadow, nuclear timing, hotspots; Unresolved: A reflecting population near Earth, and its height; Presently untestable: Artificial and extraterrestrial origin].

  • 2026-09-30editionnew assessment; article 3378 → 2543 words; the panel's seats, reading both as a reader would, preferred the candidate (5 for the candidate, 0 for the incumbent, 0 for neither; compare-v1, against edition-2026-09-17-142649); research agenda: 12 open, 0 answered, 0 superseded, 0 retired$3.02
  • 2026-09-30check5 of 5 seats judged the case blind$5.89
  • 2026-09-17editionnew assessment; article 3247 → 3365 words; research agenda: 12 open, 0 answered, 0 superseded, 0 retired$2.77
  • 2026-09-10verifyadmitted 18 claims, 32 evidence, 2 sources, 3 research; 44 refused; 1 correction(s) applied$4.68
  • 2026-09-10inbox1 item(s) taken in$0.00

how this edition was chosen

5 AI models read it beside the edition before it, as a reader would: all 5 preferred this one.

a preference among AI readers, not a human review · the verdicts are judged separately, blind · read beside edition-2026-09-17-142649 under compare-v1

Opus 5.5 (Anthropic)preferred this edition, clearlyread it second

B's first paragraph tells the reader where the evidence leaves the case: the marks are real, the flagship events are reported on the original negative, the population patterns hold only in the proponents' catalog, and light versus flaw is open. A opens with a narrative lede and leaves that summary to dense, single-sentence header answers. A's headers and article lean on site-internal vocabulary ('this ledger holds that proposition on its anchor', 'no evidence record yet admitted', 'observation rung', 'canonical', 'copy chain', 'toy model') and a plate id (XE435). A also leaves 'coma-aberration signature' undefined in the header and 'Bonferroni correction' undefined in the article, where B defines coma up front and says 'after correcting for testing many patches at once.' B covers the same evidence in about 835 fewer article words, roughly ten minutes against thirteen-plus, losing only minor specifics such as the marks being round where nearby stars are elongated, the 25,000 km peak and the UAP-lore geography reading. B also adds two points A lacks: the discovery paper itself named plate contamination as its first explanation, and the statistician who reproduced the shadow deficit later joined the group. Both are current and treat both camps seriously, so clarity and cost per minute decide it, and there B is well ahead.

what it said should still be fixed

  • The header's 'the two tested independently did not survive' should name the Earth-shadow shortfall and the nuclear timing, and say the hotspots have not yet been independently tested.
  • 'Preregistered null' and 'proper null' remain unexplained jargon; say plainly that this means a comparison baseline fixed in advance that reflects how candidates actually bunch toward plate edges and corners.
  • Some trimmed setup leaves later lines dangling: the meteor-shower test closes a 'natural route' (cometary debris) that the four accounts never introduce, and 'orbital motion' appears in the closing tests without saying the altitude model omits it.
GPT-5.6 Sol (OpenAI)preferred this edition, clearlyread it second

B states the whole evidential ladder in its opening: physical marks are established, the original-negative finding is provisional, catalog-level patterns fail independent reconstruction, and light versus emulsion remains the crux. It preserves A's strongest specifics—the copy genealogy, sharpness ambiguity, Hamburg coma test, shadow and nuclear results, natural alternatives, and decisive future tests—while costing about 800 fewer article words. B also avoids much of A's distracting site-process language such as evidence being “held on its anchor” or records being “admitted,” making the uncertainties easier for a new reader to follow. A offers some extra methodological detail, but most of it does not materially change the reader's understanding, and its length and repeated ladder summaries reduce what is learned per minute.

what it said should still be fixed

  • Clarify that it is the intelligence interpretation in the hotspot paper—not necessarily every extraterrestrial-origin proponent—that is called unfalsifiable.
  • Replace the phrase “catalogs nobody's thumb has touched” with a literal description such as independently extracted, preregistered catalogs.
  • Briefly distinguish defects formed in the original emulsion from later copying defects when first presenting the alternatives.
Gemini 3.8 Flash (Google)preferred this edition, clearlyread it first

Telling A is over 800 words shorter while conveying all the essential technical and evidentiary nuance of the dispute. Crucially, Telling B repeatedly relies on internal site jargon ('holds on its anchor', 'no evidence record yet admitted', 'ledger'), which violates the plain-words criterion for an outside reader. Telling B also ends on an awkward dangling sentence promising a research agenda that never appears. Telling A delivers a crisp, objective, and well-structured walkthrough that respects the reader's time.

what it said should still be fixed

  • In 'What would change our mind', use bullet points consistently as in Telling A rather than running lists together.
  • Clarify slightly earlier in the opening section that the 107,875-candidate working catalog is derived from Palomar plates rather than leaving it implicit.
Grok 4.5 (xAI)preferred this edition, clearlyread it second

B's headers and opening deliver the claim, the unresolved light-versus-flaw crux, the competing accounts, and the decisive tests in fewer and plainer words, so a ten-minute reader grasps the state of the question before the article body. A runs 800 words longer with repeated ledger-speak ('holds on its anchor') and extra statistical asides that teach the same mixed catalog-dependence without changing the assessment. Both are fair and current, but B's tighter concrete observations and identical alternatives give higher learning per minute while still naming what would settle light, near-Earth geometry, and artificiality.

what it said should still be fixed

  • B could open the statistics section with one sentence stating that every population claim so far is conditional on the VASCO working catalog.
  • A brief explicit note that the Yang-Villarroel original-glass exam remains unreplicated outside the group would sharpen the provisional support.
  • The hotspot control failure could be stated in one clause rather than a full paragraph to keep momentum.
GLM 5.3 Flash (Z.ai, via Venice)preferred this edition, clearlyread it first

Both tellings are current and reach the same substantive ground — real marks, flagship events on the original negative, population patterns that reproduce only on the proponents' catalog, the light-versus-flaw crux — but A delivers it in roughly 800 fewer words with plainer framing. B repeatedly leans on the site's internal vocabulary that a link-in reader has no key for ('this ledger holds that proposition on its anchor', 'no evidence record yet admitted', 'the ladder', 'rungs', plate id 'XE435'), which forces the reader to decode house jargon rather than the case. B's additions — a UAP-report lag term, one extra clause of qualification per finding — are marginal per-minute purchases, and its header answers and 'WHERE THE DISAGREEMENT LIVES' section are denser and harder to absorb in the first minute than A's, which states claim, dispute, and settle-it conditions crisply. A also handles the same fairness points (both sides' debts, the corner-field concession printed against interest, the classifier's built-in bias) with less repetition of caution.

what it said should still be fixed

  • A should briefly gloss the 0.46%-versus-1.4% shadow figures as fractions of the expected sky area, since the raw numbers are opaque on first read.
  • A's '[a plate: a photograph or figure, with its caption]' placeholders appear three times; even one-line caption summaries would help a reader who cannot see the images.
  • A's final 'research agenda' gesture in B suggests one thing A could borrow: a single closing sentence ranking which pending test would move the case most.

Run by run

Every run on this case, newest first: what it proposed, what it admitted, and what it refused with the reason the verifier wrote. A candidate that is not on the ledger is here, with why. The dispositions file holds every row.

2026-09-30editionnew assessment; article 3378 → 2543 words; the panel's seats, reading both as a reader would, preferred the candidate (5 for the candidate, 0 for the incumbent, 0 for neither; compare-v1, against edition-2026-09-17-142649); research agenda: 12 open, 0 answered, 0 superseded, 0 retired$3.02

run 2026-09-30-edition-transients-094611

This run wrote no disposition rows.

2026-09-30check5 of 5 seats judged the case blind$5.89

run 2026-09-30-check-transients-110238

This run wrote no disposition rows.

2026-09-17editionnew assessment; article 3247 → 3365 words; research agenda: 12 open, 0 answered, 0 superseded, 0 retired$2.77

run 2026-09-17-edition-transients-142110

This run wrote no disposition rows.

2026-09-10verifyadmitted 18 claims, 32 evidence, 2 sources, 3 research; 44 refused; 1 correction(s) applied55 in · 44 failed$4.68

run 2026-09-10-verify-transients-174405

refused, with the reason

  • evidencefailedModelling preprint: shadow-deficit magnitude depends on assumed altitude and on the sample definition

    not one observation (The quoted deficit values (0.435 at 20,000 km, 0.573 at 25,000 km, 0.458 at 60,000 km) and the 'becomes stronger' statement both correctly appear on p.8 in Table 1 and surrounding text, and the record fairly and non-broader represents them, so it appropriately qualifies the single-altitude figure…

  • evidencefailedModelling preprint: quality-filtered (p>0.7) sample shows a larger deficit than the full sample at every altitude

    not one observation (The quoted and paraphrased figures (general strengthening across altitudes, the GSO 0.517→0.768 change, and the 35,000 km peak of 0.805±0.049) are all accurately drawn from p.8/Table 1 and correctly support TRN-C106's specific numeric claim about the GSO deficit and general altitude trend. However…

  • evidencefailedModelling preprint: χ² fit of the angular deficit profile to spherical-shell models

    not one observation (The quoted passage on p.11 and Table 2 on p.12 accurately reflects the source's χ² fitting results, correctly located and correctly characterized as supporting evidence for TRN-C107's claim about the 20,000 km minimum across the three fitting intervals. However, the sourceStatement bundles multipl…

  • evidencefailedModelling preprint: declination excess of high-probability transients at 5–20°

    not one observation (The quote and figure are accurately located and described, and the declination-excess finding does support TRN-C109 as claimed. However, the sourceStatement bundles at least three separable propositions (the low-declination excess, the high-declination deficit, and the unexplained secondary peak n…

  • evidencefailedModelling preprint: inferred reflecting-facet sizes from centimetres to ~3 m

    not one observation (The quoted 'centimetre scales up to approximately 3 m' phrase is verified verbatim on p.27 in the stated model context, and the underlying method/figures on pp.13-15 support the general size-inference claim, so the record backs TRN-C111 with direction 'supports'. However, the sourceStatement bundl…

  • evidencefailedModelling preprint: bright-source exclusion in the catalog biases inferred sizes downward

    not one observation (The quoted passage on p.27 accurately reflects the authors' stated limitation about bright-source exclusion biasing the sample toward smaller objects, and this appropriately qualifies the size-inference claim TRN-C111. However, the sourceStatement bundles two distinct propositions — the bright-obj…

  • evidencefailedModelling preprint: characteristic modelled flash duration ~320 ms

    not one observation (The quoted figures and characteristic duration match the source at the cited pages, and the finding directly supports the claim's specific figure of ~320 ms with most flashes 60-600 ms. However, the sourceStatement bundles several distinct propositions (the full duration range, the majority range,…

  • evidencefailedModelling preprint: transverse double flashes at the ends of aligned tracks

    not one observation (The quoted passage does occur at p.20 in section 3.5 and is used accurately, and it does support the claim about perpendicular double-flash separation. However, the sourceStatement bundles two distinct propositions—the general collinear/enhanced-end-activity pattern and the perpendicular double-fl…

  • evidencefailedModelling preprint: ~5 arcsec transverse separation read as ~1 km at GSO, suggesting structure or manoeuvring

    not one observation (The quoted material correctly reflects the source's computation of a ~1 km projected separation at GSO and its stated incompatibility with a single rigid body, directly supporting TRN-C115, and the locator range covers both the measurement and the caveat. However, the sourceStatement bundles three…

  • evidencefailedModelling preprint: no robust association between transient dates and IAU meteor showers (negative result)

    not one observation (The quoted passage on p.23 accurately reports the null meteor-shower correlation test using a permutation approach, matching the claim's content and locator. However, the sourceStatement bundles two separable propositions—the absence of a robust statistical association and the observation that the…

  • evidencefailedModelling preprint: NHI toy model claimed to be the only one fitting all five features

    not one observation (The passage does state that among the authors' toy models an NHI scenario is the only one fitting all five features, and this quote is in context. However the passage is heavily hedged ('toy models,' 'in principle,' the immediate caveat that 'nature can always surprise us with new physical phenome…

  • evidencefailedModelling preprint: explicitly assumes real Earth-orbit sources; offers no test of the flash-vs-defect crux

    not one observation (The sourceStatement bundles two distinct propositions from p.5 - the paper's explicit assumption that transients are real orbiting objects, and a separate empirical estimate that one-third of the sample arises from solar reflections inferred from the shadow deficit - which have independent truth c…

  • evidencefailedHotspot preprint: most hotspot cells are fed by more than one plate; two by a single plate (XE435)

    not one observation (The sourceStatement bundles three separable propositions (14 cells with ≥2 plates, 12 cells with ≥3 plates, and the two cells at 92.5°W tied to a single plate XE435), each of which could be independently true or false, so it should be split into parts for separate judgment even though each part is…

  • evidencefailedHotspot preprint: fixed-altitude dependence of projected locations, and a misquoted altitude range

    not one observation (Both quoted passages are accurate and correctly located (p.19 on altitude-dependence of projection, p.4 on the companion paper's 26,371-35,786 km estimate), and the altitude-dependence point genuinely qualifies both the excess-analysis and clustering claims since hotspot locations are conditional…

  • evidencefailedHotspot preprint: two-step cluster analysis of 897 p≥0.90 transients yields six clusters (silhouette 0.70)

    second reader rejected (statementSupported): The quoted sentence about six clusters and silhouette 0.70 is accurate and correctly located at p.8/Table 2, but the record misstates that three eastern-Pacific clusters together account for ~65% of the sample, whereas the source attributes that ~65% figure specifically to…

  • evidencefailedHotspot preprint: random control samples also meet the paper's non-random clustering criterion

    not one observation (The quoted material accurately reflects the source's control-sample cluster results and their cautious interpretation, and the page locators are correct. However, the record bundles several distinct facts (silhouette values for three control runs, specific cluster locations, and the artifact-possi…

  • evidencefailedHotspot preprint: 203 of 231 highest-probability nuclear-window transients tie to one test on one night

    not one observation (The quoted figures (203/231, 87.9%, Castle Yankee 5/4/54, night-before observation) are accurately drawn from p.11 and support TRN-C124 as stated. However, the sourceStatement bundles a second, independent finding about the Nevada tests (24 of 28 linked to the 3/12/55 test) that is not part of TRN…

  • evidencefailedHotspot preprint: the high-probability nuclear-window sample is dominated by a single plate-night

    not one observation (The quoted spans and locator (p.11, p.16) accurately reflect the source's report that 203/231 high-probability transients came from a single Castle Yankee night, 24/28 Nevada-associated transients came from one test, and that the authors explicitly declined to test location specificity statistical…

  • evidencefailedHotspot preprint: hotspots read against UAP-lore locations (Sedona, White Sands, Tampico)

    not one observation (The quoted passages accurately reflect the source's discussion of Sedona, White Sands, and Tampico as UAP-lore locations coinciding with transient hotspots, and the locator correctly spans p.15-16 where this appears. However, the source itself explicitly states it is 'unknown whether the UAP-relat…

  • evidencefailedHotspot preprint: authors state the intelligence interpretation is not falsifiable

    not one observation (The quote is accurate and correctly located at p.17, and it does bear on the ETI-artifact claim by noting the authors' own admission that the 'intelligence' interpretation is unfalsifiable, which qualifies rather than merely contextualizes that claim. However, the sourceStatement bundles two separ…

  • evidencefailedHotspot preprint: overall conclusion of non-random Earth-projected clustering

    not one observation (The sourceStatement bundles several distinct propositions—the general non-random clustering conclusion, the specific list of three convergent regions, and the separate call for replication with other observatories—each of which could be independently true or false, so it is not a single atomic cla…

  • claimfailedRestricting the VASCO working catalog to ML-scored transients with p > 0.7, Dec > 3° and 100° < RA < 250° (11,418 objects) yields a larger Earth-shadow deficit…

    not atomic (The quoted phrase appears on p.8 and the p.8 table gives GSO deficit values of 0.517 (complete) and 0.768 (quality-filtered), matching the stated ~0.52 to ~0.77 range, and the 11,418-object quality-filtered sample is also described on p.8. However, the statement bundles multiple separable propositions: the…

  • claimfailedML-scored VASCO transients with p > 0.7 (Dec > 3°, 100° < RA < 250°) exceed the per-plate random control expectation by roughly 30–75% at declinations of about…

    not atomic (The quoted passage on p.9 does state a low-declination excess (~30-75%) and a deficit at the highest declinations for the p>0.7, Dec>3°, 100°<RA<250° sample, matching the anchor's numbers and conditions, and this bears on the case question via the orbital-population argument. However, the statement bundles…

  • claimfailedIf each VASCO transient is a single specular flash from a rigid faceted body (cube, icosahedron or saucer-like) during a ~50-minute exposure, the bodies must r…

    second reader rejected the anchor (locatorSupported): The quoted phrase 'the transient objects are likely rotating very slowly' actually appears at the start of page 17, immediately following the paragraph on page 16 that introduces the cube/icosahedron Monte Carlo comparison, not on page 16 itself as the record's loc…

  • claimfailedUnder the Monte Carlo rotation models of simple faceted geometries constrained to about one flash per plate, the characteristic specular flash duration is appr…

    not atomic (The quoted fragment and surrounding text on p.17 do state that the combined Monte Carlo simulation yields a characteristic flash duration of ~320 ms with most flashes falling between 60-600 ms, matching the proposition and locator. However the statement bundles two separate propositions (a characteristic v…

  • claimfailedThe ~5 arcsecond transverse separation of double flashes in POSS-I multiple-transient events implies a projected separation of order 1 km at geosynchronous dis…

    second reader rejected the anchor (statementSupported): The two quoted spans on p.21 state that the ~5 arcsec/~1 km separation is incompatible with a single compact object or with multiple facets on a rigid body unless the object is manoeuvring, and that this framework is not proof but is consistent with attitude chan…

  • claimfailedFourteen of the 16 Earth-projected hotspot cells receive their transient excess from at least two distinct POSS-I plates, while the two adjacent cells centred…

    not atomic (Both quoted spans appear verbatim on p.6 and accurately reflect the plate-decomposition analysis; the anchor bears on the case question by addressing whether hotspots could be single-plate defect artifacts, which is relevant to assessing reality of transients. However, the statement bundles two distinct fa…

  • claimfailedHigh-probability VASCO transients, projected to Earth at geosynchronous altitude, concentrate non-randomly — beyond what POSS-I survey footprint, cadence and s…

    not atomic (The quoted phrase 'POSS-I transients exhibit non-random clustering over distinct hotspots' does appear in the conclusion on p.18, and the surrounding text there does specify the eastern Pacific, southern Gulf of Mexico, and Arizona-New Mexico hotspots as well as the non-attributability to survey biases, so…

  • claimfailedOf the 231 VASCO transients with ML probability ≥ 0.90 that fall within ±1 day of an atmospheric nuclear test, 203 (87.9%) are associated with a single test, C…

    second reader rejected the anchor (statementSupported): The quoted spans are accurate, but the record misstates the test date as '5 May 1954' when the source's '5/4/54' (US format) denotes May 4, 1954, making the sourceStatement inaccurate on that detail. The statement also bundles several distinct facts (total count,…

  • claimfailedOf the 28 high-probability transients within ±1 day of Nevada Test Site detonations, 24 are associated with a single test on 12 March 1955, most of them record…

    not atomic (The quote and figures accurately reflect the source at p.11, and the passage bears on the case question by supporting non-randomness of transient occurrence. However, the statement bundles several separable factual claims (the 24-of-28 association with one test, the night-after timing pattern, and the AZ/N…

  • claimfailedThe Earth-projected hotspots of high-probability POSS-I transients spatially coincide, beyond chance, with regions prominent in unidentified-anomalous-phenomen…

    second reader rejected the anchor (quoteInContext, statementSupported): The anchored passage explicitly states the authors do not know whether the UAP-related location associations are meaningful or merely coincidental, which is the opposite of asserting a beyond-chance spatial coincidence; the first quote fragment ab…

  • claimfailedTransients temporally associated with nuclear tests show location specificity matching the test site: those tied to Pacific tests project over Pacific hotspots…

    second reader rejected the anchor (statementSupported): The quoted passage on p.11 explicitly states that location specificity was NOT tested statistically for the Pacific test transients (and Nevada specificity was also not statistically tested), directly contradicting the record's claim that specificity occurred 'at…

  • evidencefailedModelling preprint: Monte Carlo rotation rates for simple geometries

    every claim it cites was rejected

  • evidencefailedModelling preprint: ~5 arcsec transverse separation read as ~1 km at GSO, suggesting structure or manoeuvring — part 1

    every claim it cites was rejected

  • evidencefailedModelling preprint: ~5 arcsec transverse separation read as ~1 km at GSO, suggesting structure or manoeuvring — part 2

    every claim it cites was rejected

  • evidencefailedModelling preprint: ~5 arcsec transverse separation read as ~1 km at GSO, suggesting structure or manoeuvring — part 3

    every claim it cites was rejected

  • evidencefailedHotspot preprint: 203 of 231 highest-probability nuclear-window transients tie to one test on one night — part 1

    every claim it cites was rejected

  • evidencefailedHotspot preprint: hotspots read against UAP-lore locations (Sedona, White Sands, Tampico) — part 2

    every claim it cites was rejected

  • evidencefailedHotspot preprint: hotspots read against UAP-lore locations (Sedona, White Sands, Tampico) — part 5

    every claim it cites was rejected

  • evidencefailedHotspot preprint: overall conclusion of non-random Earth-projected clustering — part 1

    every claim it cites was rejected

  • evidencefailedHotspot preprint: overall conclusion of non-random Earth-projected clustering — part 2

    every claim it cites was rejected

  • evidencefailedHotspot preprint: overall conclusion of non-random Earth-projected clustering — part 3

    every claim it cites was rejected

  • evidencefailedHotspot preprint: overall conclusion of non-random Earth-projected clustering — part 4

    every claim it cites was rejected

  • evidencefailedHotspot preprint: overall conclusion of non-random Earth-projected clustering — part 5

    every claim it cites was rejected

admitted

  • sourceinSRC-VILLARROEL-2026Modelling Palomar Transients: Constraints from Reflection Geometry and Orbital Altitude
  • sourceinSRC-BRUEHL-2026Earth-Projected Clustering of Historical Optical Transients in the Palomar Observatory Sky Survey-I (POSS-I)
  • claiminTRN-C105When the Earth-shadow deficit D = 1 − f_obs/f_control is recomputed on the complete 107,875-row VASCO working catalog for assumed shell altitudes of 20,000–60,…
  • claiminTRN-C129The Earth-shadow deficit becomes stronger at all tested altitudes after applying the quality filtering.
  • claiminTRN-C107A χ² comparison of the observed deficit-versus-antisolar-angle profile of the complete VASCO working catalog against spherical-shell shadow models yields its m…
  • claiminTRN-C108The near-Earth reflective population responsible for the Earth-shadow deficit in the VASCO working catalog lies at characteristic altitudes of approximately 20…
  • claiminTRN-C130Among high-probability transients (p > 0.7, 100° < R.A. < 250°, Dec. > 3°), a significant excess is observed at low declinations (∼5–20°), where the transient…
  • claiminTRN-C110High-probability VASCO transients are depleted, relative to control expectation, in the region of the sky near the ecliptic plane.
  • claiminTRN-C111If the VASCO transients are specular sunlight reflections from planar facets at 20,000–35,786 km altitude with flash durations of 0.1–1 s and reflectivity 0.1–…
  • claiminTRN-C131The combined flash-duration distribution from the simulations yields a characteristic flash duration of ∼320 ms.
  • claiminTRN-C114In POSS-I multiple-transient events with approximately collinear flashes, closely spaced double flashes at the start or end of the track are separated perpendi…
  • claiminTRN-C116Transient detection dates in the VASCO working catalog show no statistically robust temporal association with annual meteor showers listed by the IAU Meteor Da…
  • claiminTRN-C117The millimetre-scale particle sizes of the dusty-plasma mechanism proposed for the POSS-I transients are inconsistent with the centimetre-to-metre facet sizes…
  • claiminTRN-C118Among the toy models considered by the VASCO group (natural reflective bodies, dusty plasma, non-human intelligence), only a non-human-intelligence scenario si…
  • claiminTRN-C119When p ≥ 0.70 VASCO transients and plate-aware random controls are projected onto a shell at 35,786 km altitude and binned in 5°×5° Earth-fixed cells, 16 of th…
  • claiminTRN-C132Fourteen of the 16 significant 5° × 5° hotspot cells showed a transient excess beyond expectations on at least two different POSS-I photographic plates.
  • claiminTRN-C133The two adjacent hotspot cells centered at 92.5° W (Hotspots 5 and 8) were associated exclusively with a single plate, XE435.
  • claiminTRN-C121Two-step cluster analysis of the Earth-projected positions (35,786 km shell) of the 897 VASCO transients with ML probability ≥ 0.90 (Dec > 3°, 100° < RA < 250°…
  • claiminTRN-C122Random control samples of 897 plate-aware positions, projected and clustered by the same two-step procedure, produce two or three clusters with average silhoue…
  • claiminTRN-C125All 203 high-probability transients recorded the night before Castle Yankee project (at 35,786 km altitude) onto the Pacific hotspot regions west of Mexico and…
  • evidenceinTRN-E025Modelling preprint: global shadow deficit recomputed across assumed altitudes, largest at 25,000 km
  • evidenceinTRN-E051Modelling preprint: shadow-deficit magnitude depends on assumed altitude and on the sample definition — part 3
  • evidenceinTRN-E052Modelling preprint: quality-filtered (p>0.7) sample shows a larger deficit than the full sample at every altitude — part 2
  • evidenceinTRN-E053Modelling preprint: quality-filtered (p>0.7) sample shows a larger deficit than the full sample at every altitude — part 4
  • evidenceinTRN-E054Modelling preprint: χ² fit of the angular deficit profile to spherical-shell models — part 2
  • evidenceinTRN-E055Modelling preprint: χ² fit of the angular deficit profile to spherical-shell models — part 3
  • evidenceinTRN-E056Modelling preprint: χ² fit of the angular deficit profile to spherical-shell models — part 4
  • evidenceinTRN-E057Modelling preprint: χ² fit of the angular deficit profile to spherical-shell models — part 7
  • evidenceinTRN-E058Modelling preprint: declination excess of high-probability transients at 5–20° — part 1
  • evidenceinTRN-E059Modelling preprint: declination excess of high-probability transients at 5–20° — part 2
  • evidenceinTRN-E030Modelling preprint: transients reported depleted near the ecliptic (citing Doherty 2026a)
  • evidenceinTRN-E060Modelling preprint: inferred reflecting-facet sizes from centimetres to ~3 m — part 2
  • evidenceinTRN-E061Modelling preprint: inferred reflecting-facet sizes from centimetres to ~3 m — part 3
  • evidenceinTRN-E062Modelling preprint: bright-source exclusion in the catalog biases inferred sizes downward — part 1
  • evidenceinTRN-E063Modelling preprint: bright-source exclusion in the catalog biases inferred sizes downward — part 2
  • evidenceinTRN-E064Modelling preprint: characteristic modelled flash duration ~320 ms — part 2
  • evidenceinTRN-E065Modelling preprint: characteristic modelled flash duration ~320 ms — part 3
  • evidenceinTRN-E069Modelling preprint: no robust association between transient dates and IAU meteor showers (negative result) — part 1
  • evidenceinTRN-E070Modelling preprint: no robust association between transient dates and IAU meteor showers (negative result) — part 2
  • evidenceinTRN-E071Modelling preprint: no robust association between transient dates and IAU meteor showers (negative result) — part 3
  • evidenceinTRN-E038Modelling preprint: dusty-plasma sizes do not match inferred object sizes
  • evidenceinTRN-E072Modelling preprint: NHI toy model claimed to be the only one fitting all five features — part 1
  • evidenceinTRN-E073Modelling preprint: NHI toy model claimed to be the only one fitting all five features — part 2
  • evidenceinTRN-E074Modelling preprint: explicitly assumes real Earth-orbit sources; offers no test of the flash-vs-defect crux — part 2
  • evidenceinTRN-E041Hotspot preprint: 16 of 137 Earth-projected 5°×5° cells show Bonferroni-significant transient excess
  • evidenceinTRN-E075Hotspot preprint: most hotspot cells are fed by more than one plate; two by a single plate (XE435) — part 1
  • evidenceinTRN-E076Hotspot preprint: most hotspot cells are fed by more than one plate; two by a single plate (XE435) — part 2
  • evidenceinTRN-E077Hotspot preprint: most hotspot cells are fed by more than one plate; two by a single plate (XE435) — part 3
  • evidenceinTRN-E078Hotspot preprint: fixed-altitude dependence of projected locations, and a misquoted altitude range — part 1
  • evidenceinTRN-E079Hotspot preprint: random control samples also meet the paper's non-random clustering criterion — part 1
  • evidenceinTRN-E083Hotspot preprint: authors state the intelligence interpretation is not falsifiable — part 2
  • evidenceinTRN-E084Hotspot preprint: authors state the intelligence interpretation is not falsifiable — part 3
  • researchinTRN-R010Plate-level audit of the Castle Yankee night (3–4 May 1954) and the 12–13 March 1955 nights
  • researchinTRN-R011Altitude- and grid-robust hotspot test on an independently extracted catalog
  • researchinTRN-R012Preregistered shell-altitude fit with orbital dynamics and independent shadow code
2026-09-10inbox1 item(s) taken in$0.00

run 2026-09-10-inbox-transients-173613

This run wrote no disposition rows.

2026-09-10editiona new edition with a new assessment$3.34

run 2026-09-10-edition-transients-182453

This run wrote no disposition rows.

2026-09-10draftschema sent as instructions (too large for strict output)1 duplicate · 1 blocked$2.80

run 2026-09-10-draft-transients-173620

refused, with the reason

  • sourceduplicateSRC-BRUEHL-2026-MLMachine-learning paper on plate defects and contaminant removal (in press) — no OpenAlex result close enough

    The machine-learning paper the post calls 'accepted for publication following peer review and is now in press' is cited by the retrieved hotspot preprint (reference 10, [p. 25]) as Bruehl, Doherty, Streblyanska & Villarroel, 'Use of machine learning to enhance detection of transient astronomical phenomena in historica…

    reopen if: The published Scientific Reports version differs materially from the arXiv text on which TRN-E015 rests, or the DOI resolves to a different work.

  • sourceblockedJanne Ahlberg, 'Results from an independent, reproducible search for vanishing sources in POSS-I plates', GitHub repository, 2026

    The modelling preprint ([p. 3] and reference list [p. 28]) states that J. Ahlberg independently reproduced the raw-detection stage of the VASCO pipeline, recovering ~94–98% of catalogue sources, citing a GitHub RESULTS.md. This bears on TRN-C004 but only the secondary statement was shown; the primary was not retrieved…

    reopen if: The RESULTS.md text is retrieved and its method (full-plate scans, matching criteria, catalogue version) can be read.

2026-09-10check5 of 5 seats judged the case blind$1.32

run 2026-09-10-check-transients-183351

This run wrote no disposition rows.

2026-09-01—2 row(s) written by 2026-09-01-promote-33512756583, which left no run recordno run record2 duplicate

rows name 2026-09-01-promote-33512756583; no run record exists for it, so its verb, outcome and cost are not known — the rows themselves are the record

refused, with the reason

  • sourceduplicateSRC-DOHERTY-2026Bhttps://arxiv.org/abs/2605.01190

    same work as SRC-DOHERTY-2026B (matched via arxiv id)

  • sourceduplicateSRC-BRUEHL-2026-MLhttps://arxiv.org/abs/2604.18799

    same work as SRC-BRUEHL-2026-ML (matched via title similarity)

2026-08-31—1 row(s) written by triage-2026-08-31-1mhz, which left no run recordno run record1 irrelevant

rows name triage-2026-08-31-1mhz; no run record exists for it, so its verb, outcome and cost are not known — the rows themselves are the record

refused, with the reason

  • sourceirrelevantMicroscopy assessment of two multiple-transient candidates

    Exact title match to already-tracked SRC-YANG-VILLARROEL-2026 indicates this is the same microscopy study (bearing on TRN-C020, TRN-C001, TRN-C003) posted to arXiv, so importing would duplicate an existing ledger entry rather than add new evidential weight.

2026-08-27—3 row(s) written by triage-2026-08-27-yb5k, which left no run recordno run record3 irrelevant

rows name triage-2026-08-27-yb5k; no run record exists for it, so its verb, outcome and cost are not known — the rows themselves are the record

refused, with the reason

  • sourceirrelevantMysterious Transients in the Palomar Observatory Sky Survey (POSS-1) as profound manifestation of the Dark Matter physics

    A speculative dark-matter interpretation with no new data or measurements bearing on whether the plate detections are real or artifacts.

  • sourceirrelevantPlate Sensitivity Is Invariant Across Geomagnetic Storm Intensity at Harvard and Palomar: A Protocol for Artifact Control in Historical Plate Archive Studies

    A general artifact-control protocol keyed to geomagnetic storm correlations that the case does not track; its bearing on the featured nuclear-window claim is only indirect and it presents no result about the tracked claims.

  • sourceirrelevantStorm-Driven Suppression and Post-Storm Enhancement of Photographic Plate Transient Detections at Geosynchronous Altitude: Empirical Evidence and a Candidate D…

    Builds on untracked geomagnetic-storm claims via a self-citing chain, mischaracterizes the transients as sub-second, and offers a speculative mechanism rather than new evidence on any featured claim.

Change history

Every change to this case, newest first: what changed, why, and who made it — including the AI's role. The log is append-only; a correction is a new entry, never an edit to an old one.

  1. claude-fable-5-1 (operator), answering the panel's park on PR #286 in the founder's session of 2026-09-11 · AI-assisted

    Corrections applied from PR #286 panel review (GPT-5.6 Sol seat: same-case shorthand ids escaped the check): TRN-E078.limitations — ["Single assumed altitude; no sensitivity analysis across altitudes shown","Direction set to \"context\" (from \"qualifies\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quote at p.19 accurately describes the line-of-sight projection methodology used to place transients at geosynchronous altitude, and the sourceStatement fairly restates this without overreach; the locator is correct despite the odd 'misquote at p.4' annotation, which does not correspond to any actual conflicting text there. However, this passage is purely methodological (how projections were computed) and does not itself assert, limit, or condition the truth of the specific hotspot-significance claims (C119) or non-random concentration claims (C123); it merely provides background context for both, so 'context' is the correct direction rather than 'qualifies'."] → ["Single assumed altitude; no sensitivity analysis across altitudes shown","Direction set to \"context\" (from \"qualifies\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quote at p.19 accurately describes the line-of-sight projection methodology used to place transients at geosynchronous altitude, and the sourceStatement fairly restates this without overreach; the locator is correct despite the odd 'misquote at p.4' annotation, which does not correspond to any actual conflicting text there. However, this passage is purely methodological (how projections were computed) and does not itself assert, limit, or condition the truth of the specific hotspot-significance claims (C119) or non-random concentration claims (the proposed non-random-concentration claim, not admitted at intake); it merely provides background context for both, so 'context' is the correct direction rather than 'qualifies'."]; TRN-E079.limitations — ["Only three control samples","No formal comparison of 0.70 against the control distribution","Second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) found the passage bears on TRN-C122 and a proposed claim not admitted at intake ('High-probability VASCO transients … concentrate non-randomly') and not on TRN-C121; the link dropped at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on C123 by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped.","Direction set to \"supports\" (from \"qualifies\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on C123 by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped."] → ["Only three control samples","No formal comparison of 0.70 against the control distribution","Second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) found the passage bears on TRN-C122 and a proposed claim not admitted at intake ('High-probability VASCO transients … concentrate non-randomly') and not on TRN-C121; the link dropped at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on the proposed non-random-concentration claim (not admitted at intake) by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped.","Direction set to \"supports\" (from \"qualifies\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on the proposed non-random-concentration claim (not admitted at intake) by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped."].

    Why: TRN-E078: shorthand 'C123' named a proposal claim that never entered the ledger | TRN-E079: shorthand 'C123' named a proposal claim that never entered the ledger

  2. claude-fable-5-1 (operator), answering review note #285 in the founder's session of 2026-09-11 · AI-assisted

    Corrections applied from review note #285 (GPT-5.6 Sol seat on #283): the dangling-id check must hold for every record: TRN-E038.editorInference — "This narrows the natural near-Earth route that TRN-E020 (SRC-CANN-2026C) put on the table for TRN-C021, but only via the source's own model-dependent size inference (TRN-C207/E207), which is itself selection-biased (TRN-E208). One toy model excluding another does not move the observation rung." → "This narrows the natural near-Earth route that TRN-E020 (SRC-CANN-2026C) put on the table for TRN-C021, but only via the source's own model-dependent size inference, which is itself selection-biased. One toy model excluding another does not move the observation rung."; TRN-E072.editorInference — "An argument from residual explanatory fit among a hand-chosen set of toy models. Three of the five 'features' are contested at the catalog level (TRN-C005, C006, C103) and one (PSF narrowness) is read both ways (TRN-E003/E004/E006). Its bearing on TRN-C023 is nominal; TRN-C023 remains without any test that could refute it, as the companion hotspot preprint concedes (TRN-E225)." → "An argument from residual explanatory fit among a hand-chosen set of toy models. Three of the five 'features' are contested at the catalog level (TRN-C005, C006, C103) and one (PSF narrowness) is read both ways (TRN-E003/E004/E006). Its bearing on TRN-C023 is nominal; TRN-C023 remains without any test that could refute it, as the companion hotspot preprint concedes."; TRN-E073.editorInference — "An argument from residual explanatory fit among a hand-chosen set of toy models. Three of the five 'features' are contested at the catalog level (TRN-C005, C006, C103) and one (PSF narrowness) is read both ways (TRN-E003/E004/E006). Its bearing on TRN-C023 is nominal; TRN-C023 remains without any test that could refute it, as the companion hotspot preprint concedes (TRN-E225)." → "An argument from residual explanatory fit among a hand-chosen set of toy models. Three of the five 'features' are contested at the catalog level (TRN-C005, C006, C103) and one (PSF narrowness) is read both ways (TRN-E003/E004/E006). Its bearing on TRN-C023 is nominal; TRN-C023 remains without any test that could refute it, as the companion hotspot preprint concedes."; TRN-E076.limitations — ["Plate overlap is defined in projected coordinates, not on the sky","Per-plate significance tested at unadjusted p<0.05","Direction set to \"context\" (from \"supports\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quoted figure (12 of 16 cells showing excess on at least three plates) is accurately taken from the unadjusted plate-decomposition analysis on p.6, but it is a different statistic than the one the claim actually asserts (14 cells at ≥2 plates and the 2 cells at 92.5°W tied exclusively to plate XE435). It corroborates the general theme that hotspots are not single-plate artifacts but does not itself confirm or deny the specific 14/2-cell breakdown stated in TRN-C120, so it functions as contextual corroboration rather than direct support of that claim's specific content."] → ["Plate overlap is defined in projected coordinates, not on the sky","Per-plate significance tested at unadjusted p<0.05","Direction set to \"context\" (from \"supports\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quoted figure (12 of 16 cells showing excess on at least three plates) is accurately taken from the unadjusted plate-decomposition analysis on p.6, but it is a different statistic than the one the claim actually asserts (14 cells at ≥2 plates and the 2 cells at 92.5°W tied exclusively to plate XE435). It corroborates the general theme that hotspots are not single-plate artifacts but does not itself confirm or deny the specific 14/2-cell breakdown stated in a proposed claim not admitted at intake ('Fourteen of the 16 Earth-projected hotspot cells…'), so it functions as contextual corroboration rather than direct support of that claim's specific content."]; TRN-E079.limitations — ["Only three control samples","No formal comparison of 0.70 against the control distribution","Second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) found the passage bears on TRN-C122, TRN-C123 and not on TRN-C121; the link dropped at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on C123 by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped.","Direction set to \"supports\" (from \"qualifies\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on C123 by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped."] → ["Only three control samples","No formal comparison of 0.70 against the control distribution","Second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) found the passage bears on TRN-C122 and a proposed claim not admitted at intake ('High-probability VASCO transients … concentrate non-randomly') and not on TRN-C121; the link dropped at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on C123 by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped.","Direction set to \"supports\" (from \"qualifies\") by the second reader (claude-sonnet-5, 2026-09-10, run 2026-09-10-verify-transients-174405, verify-v5) at intake: The quoted silhouette-0.50 finding for the first random control sample is accurately reported and located at p.9-10, and it bears on the case question. However, calling this 'qualifies' undersells that it directly instantiates and confirms TRN-C122's claim that random samples meet/exceed the 0.50 threshold (that is a support relation for C122), while it only indirectly bears on C123 by raising the possibility that some hotspots are survey artifacts (a genuine qualification there). It does not bear at all on TRN-C121, which concerns the real-transient cluster silhouette of 0.70, not the control sample's result, so that claim link should be dropped."]; TRN-R010.summary — "Identify from the STScI XE plate list every POSS-I red plate exposed on the night before Castle Yankee and on the nights around the 12 March 1955 Nevada test; for each plate, count the high-probability (p ≥ 0.90) transients per plate and per plate quadrant, measure the edge concentration and profile statistics of those detections against matched stars, and compare per-plate detection rates with the distribution of per-plate rates across all 635 plates. Then examine the copy material for those plates under magnification, as Hambly and Blair did for E0070. If the 203 Castle Yankee-night candidates are concentrated on one or two plates with anomalous defect morphology or edge clustering, TRN-C220 stands but TRN-C006 and TRN-C224 lose their high-probability support to a plate explanation; if they are spread across many plates with star-like, coma-consistent images at rates far above the per-plate norm, TRN-C006 gains a plate-level foundation it currently lacks." → "Identify from the STScI XE plate list every POSS-I red plate exposed on the night before Castle Yankee and on the nights around the 12 March 1955 Nevada test; for each plate, count the high-probability (p ≥ 0.90) transients per plate and per plate quadrant, measure the edge concentration and profile statistics of those detections against matched stars, and compare per-plate detection rates with the distribution of per-plate rates across all 635 plates. Then examine the copy material for those plates under magnification, as Hambly and Blair did for E0070. If the 203 Castle Yankee-night candidates are concentrated on one or two plates with anomalous defect morphology or edge clustering, the plate-artifact account stands but TRN-C006 and the clustering claims lose their high-probability support to a plate explanation; if they are spread across many plates with star-like, coma-consistent images at rates far above the per-plate norm, TRN-C006 gains a plate-level foundation it currently lacks.".

    Why: TRN-E038: prose named a record id the ledger does not hold — candidate ids from an earlier loop, not ledger records (review note #285: no dangling ids, everywhere) | TRN-E072: prose named a record id the ledger does not hold — candidate id (review note #285: no dangling ids, everywhere) | TRN-E073: prose named a record id the ledger does not hold — candidate id (review note #285: no dangling ids, everywhere) | TRN-E076: prose named a record id the ledger does not hold — provisional claim id (review note #285: no dangling ids, everywhere) | TRN-E079: prose named a record id the ledger does not hold — provisional claim id (review note #285: no dangling ids, everywhere) | TRN-R010: prose named a record id the ledger does not hold — candidate ids (review note #285: no dangling ids, everywhere)

  3. aletheia verify (claude-sonnet-5 second reader; drafter claude-fable-5-1) · AI-assisted

    Intake from report proposals/2026-09-10-inbox-transients-173613/report.md: 2 source(s), 32 evidence record(s), 18 claim(s), 3 research item(s) verified and added (proposal 2026-09-10-draft-transients-173620, verification 2026-09-10-verify-transients-174405); 44 candidate(s) rejected with reasons in dispositions.yaml.

    Why: Inbox intake: a public X post by Beatriz Villarroel, supplied by the founder, naming three works. The post itself is its supplier's words and is not a source. Of the three works: (1) the machine-learning paper 'accepted and in press' is, by the hotspot preprint's own reference 10 (title and DOI 10.1038/s41598-026-66148-2), the work the ledger already holds as SRC-BRUEHL-2026-ML — dispositioned duplicate; (2) arXiv:2609.05105 was retrieved and its title, authors and content confirm it is the Earth-shadow altitude/modelling paper the post describes — proposed as a new preprint source; (3) arXiv:2609.09461 was retrieved and confirmed as the Earth-projected hotspot paper — proposed as a new preprint source. Both preprints are VASCO-group work built on the same 107,875-row working catalog and the same group's ML probabilities, so their independence group is the proponents'; both are model-dependent (the modelling paper states it proceeds from the assumption that the transients are real Earth-orbit sources; the hotspot paper projects onto a fixed 35,786 km shell). I catalog their distinct testable propositions as unfeatured claims (altitude estimates, declination excess, size and rotation inferences, transverse double flashes, no meteor-shower association, dusty-plasma exclusion, NHI toy-model argument, hotspot excess and clustering, control-sample clustering, nuclear-window concentration on one night, UAP-lore coincidence, location specificity), each anchored to shown text. The most consequential observations for existing claims are recorded as qualifications: the shadow-deficit magnitude reported now varies with assumed altitude and definition (bears on TRN-C005); 203 of the 231 highest-probability nuclear-window transients derive from a single night before a single test, with location specificity untested statistically by the authors' own statement (bears on TRN-C006); random control samples also exceed the paper's own 0.50 'non-random' silhouette threshold (bears on the new clustering claim); and the hotspot preprint misquotes the altitude range of its companion paper. Negative evidence (meteor showers) is recorded. One secondary work named by the modelling paper (Ahlberg 2026 GitHub replication) is blocked with its route. One correction: a url for SRC-YANG-VILLARROEL-2026, which the ledger holds without one, derived from its existing arxiv key as cited by the hotspot preprint. No edition change: the new material sits on the attribution rung and is model-conditional; the current telling ('the catalog is the battleground') is not shown wrong, and the one-night concentration finding is a qualification within an already 'mixed' verdict rather than a load-bearing reversal.

  4. aletheia verify (claude-sonnet-5 second reader; drafter claude-fable-5-1; proposal 2026-09-10-draft-transients-173620, verification 2026-09-10-verify-transients-174405) · AI-assisted

    Correction applied from proposals/2026-09-10-draft-transients-173620: SRC-YANG-VILLARROEL-2026.url — "null" → "https://arxiv.org/abs/2608.26371".

    Why: SRC-YANG-VILLARROEL-2026: The ledger holds this source without a url; its keys already include arxiv:2608.26371, and the hotspot preprint (SRC-BRUEHL-2026, reference 11, [p. 25]) cites the same work as 'Preprint at: https://arxiv.org/abs/2608.26371 (2026)'. Adding the url gives the verifier something to read; the arXiv text itself was not shown in this pass, so the record's content is unchanged.

  5. scripts/migrations/2026-09-09-hash-locators.ts (mechanical re-stamp) · AI-assisted

    Ledger hash redefined to exclude source locators (url, archivedUrl); 1 edition/assessment basis stamp(s) recomputed mechanically, content unchanged: edition-2026-09-08-migration.yaml.

    Why: Founder decision 2026-09-09: a locator says where a record is read, not what it says; a by-hand URL addition had set a fresh panel aside and made an edition due that could only re-adopt.

  6. Cursor Grok 4.6 (Cursor cloud agent), directed by the founder; runId title-punctuation-2026-09-08; promptVersion none (founder-directed session, not a pipeline prompt) · AI-assisted

    Case title punctuation restored after the pre-overhaul publication restore: "Before Sputnik?" is again "Before Sputnik". The trailing question mark is dropped so the published name is a statement. Subtitle and framing questions are unchanged.

    Why: Founder direction (first given 2026-09-06, reaffirmed 2026-09-08 after the restore): case names should not carry question marks. The interrogative belongs in the subtitle and the research questions, not on the card and in the header.

  7. scripts/migrations/2026-09-08-editions.ts (mechanical), founder-directed session · AI-assisted

    Migrated to editions (docs/AUTOMATION.md, build step 2). The article, the 14-claim featured set, and the displayed judgment now live in edition edition-2026-09-08-migration, which adopts the migration assessment 2026-09-08-migration-transients. Claim records carry propositions and anchors only; evaluative fields moved into the adopted assessment's per-claim treatment; the dossier header, research priority, and component verdicts moved from case.yaml into the assessment; claims-catalog.yaml merged into claims.yaml (tiers ended). No judgment changed. Where a claim record's hand-written credibility differed from the draft overlay's verdict, the record's value (what claim pages displayed) was carried: TRN-C023: record presently_untestable, draft unresolved; TRN-C001: record well_supported, draft established.

    Why: One reader-facing unit that binds verdict, selection, and article, so they cannot drift apart; one home for judgment; featuring becomes a recorded decision.

  8. Fable (AI agent), maintainer under AGENTS.md 3.15, directed by the founder · AI-assisted

    Two connected updates. (1) New featured claim TRN-C024: the nine April 1950 sources are physically present on the original telescope negative, not only on copy material. The proposition was already reported by TRN-E022 (the Yang-Villarroel microscopy examination, carried at reported-not-established weight), but no atomic claim stated it, so the copy-stage defect pathway (TRN-C007) had nothing in the ledger it directly falsifies. TRN-E022 now also anchors the new claim. Graded provisionally_supported / high diagnosticity: a single unreplicated, non-blind proponent examination, but the one record that cleanly separates the copy-defect family from everything else. (2) The manuscript became public: the preprint was posted to arXiv on 2026-08-26 (arXiv:2608.26371), surfaced by the 2026-08-31 literature watch, and matched to the existing source by the duplicate guard. As SRC-YANG-VILLARROEL-2026's verification note pre-registered, the identifier is updated and every quoted passage in TRN-E022/E023/E024 was re-checked verbatim against the arXiv text — all present. The supplied-provenance and permission history stays on the record; citation no longer depends on it. Unpublished-manuscript caveats across the records become preprint/not-peer-reviewed caveats; the blind-examination gap (TRN-R001) is unchanged.

    Why: Adopted from the 2026-08-31 agenda run (transients proposal 1) by founder direction on 2026-09-01; the source upgrade follows the procedure the record itself specified for the day a public version appeared.

  9. Fable (AI agent), maintainer under AGENTS.md 3.15; permission obtained by the founder · AI-assisted

    Permission recorded on the removed correspondence (runId trn-villarroel-2026-08-24): Beatriz Villarroel granted the founder permission for her email to be in the repository, on 2026-08-26, after the removal; relayed by the founder to the maintainer the same day, with the correspondence recording the grant held by the founder. Per the 3.15 standard, the grant is recorded here with who, when, channel, and custody. The files stay out of the tree regardless - the email carries a third party's personal address and remarks about colleagues, whose permission has not been sought, and nothing in the ledger cites the correspondence. What changes is the standing of the old commits: her correspondence remaining in repository history is no longer a 3.15 violation on her account.

    Why: Permission converts supplied material only when recorded with the same rigor as any provenance (AGENTS.md 3.15, as amended 2026-08-25). A verbal all-clear that lives nowhere is exactly the unverifiable attestation the panel rejected on the microscopy intake.

  10. Fable (AI agent), maintainer under AGENTS.md 3.15 · AI-assisted

    Correction: the raw correspondence behind the 2026-08-24 intake (runId trn-villarroel-2026-08-24) — a private email from the VASCO group's principal investigator, with personal email addresses and a reference to an unpublished manuscript — had been committed to this public repository under inbox/processed/, despite the intake's own history entry stating the correspondence "is not published." No permission to publish it was ever granted or recorded. The files are now removed from the tree and replaced with a tombstone note; the editor retains the correspondence privately. Removal from repository history requires a history rewrite, which is recorded as pending founder authorization in docs/DECISIONS.md. No case content changes: the intake never quoted or cited the correspondence, and all sources it pointed to were independently verified against public records.

    Why: AGENTS.md 3.15 forbids publishing material supplied in confidence, and the 2026-08-24 history entry promised the correspondence was not published. The repository state contradicted both. Discovered during a provenance audit of repository history on 2026-08-26; corrected the same day, before any other maintenance work, per the correction-first rule.

  11. Fable (AI agent), maintainer under AGENTS.md 3.15; permission obtained by the founder · AI-assisted

    Citation permission recorded on SRC-YANG-VILLARROEL-2026: the corresponding author's group confirmed by email that the editor may reference the unpublished microscopy manuscript and its findings ahead of publication. The source's verification note now says so; nothing else about the record changes.

    Why: The constitutional panel's trial vote on the microscopy intake parked it 3-2, both dissenting seats converging on the same gap - supply provenance was recorded but permission to cite was not. This closes that gap the way the panel implied it should be closed: not by argument, but by obtaining and recording the permission.

  12. Fable (AI agent), maintainer under AGENTS.md §3.15 · AI-assisted

    TRN-C023 (extraterrestrial artifacts) reclassified from unresolved to presently untestable. The statement itself is unchanged; what changed is the honest label for its status. The claim's own summary had already conceded that no test on the table can reach it; the label now matches the summary.

    Why: Two independent signals converged: the 2026-08-24 scheduled reassessment proposed the reclassification on its own reasoning, and all five cross-model judges in the 2026-08-24 panel graded the claim presently_untestable against the house label — the panel's only unanimous disagreement with the displayed assessment. Under the stage-3 governance this is exactly the review path: the change ships as a pull request and stands or falls on whether a fresh panel ratifies the resulting assessment.

  13. Fable (AI agent), directed by Eugene · AI-assisted

    The microscopy manuscript arrived. Yang & Villarroel's unpublished "Microscopy assessment of two multiple-transient candidates" — supplied by the authors' group ahead of publication, read in full, deliberately not committed to this repository — is recorded as SRC-YANG-VILLARROEL-2026 with three evidence records. TRN-E022: all nine 1950 sources are present on the original glass at Caltech, and the 1952 triple is not confined to copy material. TRN-E023: the copy-stage defect mechanism therefore cannot explain the two canonical events, though it keeps its force over the uninspected catalog. TRN-E024: microscopy cannot distinguish a sub-second flash from an intrinsic emulsion defect, and the manuscript's own microscopist — directed to the field by Hambly — judged emulsion defects a reasonable attribution. The overview's centerpiece section was rewritten from "the observation nobody has made" to "the observation, finally made," recording that the decisive test returned neither of its two anticipated outcomes: the crux narrows from "original or copy" to "flash or flaw," which single-source microscopy cannot answer. TRN-R001's information gain is revised downward accordingly, and TRN-R009 (multi-station simultaneous imaging, the manuscript's own prospective proposal) is added to the agenda.

    Why: This is the event the case was structured around, and the ledger's job is to record what the test actually returned rather than the verdict either camp hoped for. Everything is labeled reported-not- established: the manuscript is unpublished, the examination was not blind, and no outsider has repeated the registration.

  14. Fable (AI agent), directed by Eugene · AI-assisted

    Two sources and two evidence records added from the literature watch: Cann's third preprint (SRC-CANN-2026C, TRN-E020), which preregisters a post-storm transient-recovery profile and proposes a natural mechanism — magnetospherically trapped icy dust at geosynchronous altitude — and Zhitnitsky's dark-matter interpretation (SRC-ZHITNITSKY-2026, TRN-E021). Both are recorded weakly and neither is entered as support: TRN-E020 qualifies, because a natural near-Earth source would leave the near-Earth rung (TRN-C021) standing while removing the reason to believe the artificial one (TRN-C022); TRN-E021 is context only, because it takes the transients' reality as its premise and so inherits the crux (TRN-C020) rather than bearing on it. Three further items in the same watch run were checked and already held: Doherty 2605.01190, Cann 2604.16470 and Bruehl et al. 2604.18799.

    Why: The case had been framed as plate chemistry versus artificial reflectors. Cann's proposal is a third family — real near-Earth objects that are natural — and a map of a controversy that omits a live third option is not neutral, whatever its verdicts say. Recording it at low weight, with its dependence on the same author's earlier work stated, is the honest form of that correction.

  15. Fable (AI agent), directed by Eugene · AI-assisted

    Plate III added — a three-panel figure showing the POSS-I copy chain (original negative → glass positive → distributed second-generation copy), with the same two scratches inherited through every generation. It is placed in the article beside the copy-ancestry argument (TRN-C007) and appears on the pages for TRN-C007 and TRN-C020. Adding it required amending the image rules: plate provenance previously had to include a public sourceUrl, which no figure supplied directly by a researcher can have. The schema now accepts a second, equally strict form — suppliedBy plus permission, both required together — and the credit line names the supplier on the plate itself.

    Why: The figure makes visible the mechanism the whole case turns on, which a paragraph of prose conveys poorly. The old rule would have excluded it not because its provenance was unknown but because its provenance was direct, which selects for whatever has already been published rather than for what is checkable. The replacement rule forbids unrecorded provenance just as absolutely, and requires the reader be told the figure came from a party to the dispute.

  16. Fable (AI agent), directed by Eugene · AI-assisted

    Overview article updated on its central paragraph. It previously stated that the original telescope negatives "have never been examined for these candidates" — no longer accurate given the proponent group's report that it has now examined them, with a microscopy manuscript in preparation. The paragraph now records that every *published* analysis still works from copies, that the examination is reported but unseen and unpublished, and that the decisive test's remaining value lies in blinding, generation-recording and outside repeatability. A second edit adds the independence footnote on the replication statistician to the population-statistics paragraph.

    Why: The narrative may not assert as fact something the ledger now records as superseded; a case's centerpiece claim is exactly where staleness does the most damage.

  17. Fable (AI agent), directed by Eugene · AI-assisted

    All six assessments re-run against the updated case file: the house draft (reassess-changed) and all five cross-model checks (Opus 5, GPT-5.1, Gemini 3.1 Pro, Grok 4.6, Kimi K3). The case verdict held at unresolved in every one, and the house draft moved no claim verdicts. Four vendor-level claim moves occurred, all at the attribution rung: GPT-5.1 moved TRN-C022 from contradicted and TRN-C021 from weakly supported, both to unresolved; Gemini moved TRN-C023 from presently untestable to unresolved; Kimi moved TRN-C001 from established to well supported. Inter-judge agreement rose — 9 of 13 claims exact (from 7), with split claims falling from four to two (TRN-C011, TRN-C023). Honest caveat: with one run per model there is no repeat-sampling baseline, so these moves cannot be cleanly attributed to the new evidence rather than to run-to-run variation.

    Why: Re-assessing after a material content change is the point of the overlay design; recording what moved — and that some of it may be noise — is what keeps the comparison honest.

  18. Fable (AI agent), directed by Eugene · AI-assisted

    TRN-R001 (blind examination of the original negatives) updated with a research-status note: the proponent group reports that the original Palomar plates are themselves negatives, that they have already examined those originals for the nine April 1950 transients, and that a microscopy manuscript is in preparation. That manuscript has not been seen and no result from it is recorded. No claim credibility changed. The project's framing shifts from "has anyone examined the originals" to whether the examination was blind, recorded the generation at which each mark first appears, and can be repeated by a party outside the group — which argues for preregistering that protocol before the manuscript appears.

    Why: Observation before interpretation: a reported examination is not a result, and the decisive test's value depends on blinding and independent repeatability rather than on the examination having happened.

  19. Fable (AI agent), directed by Eugene · AI-assisted

    Independence note sharpened on SRC-DOHERTY-2026, volunteered by the proponent side against interest: the author of the replication was initially unconvinced by the transient claims, produced the replication from that position, and subsequently joined the group while continuing to work semi-independently. Recorded because the sequence cuts both ways — it strengthens the replication itself, which was performed from outside and while skeptical, and it means later output under that name is group-affiliated rather than independent. The publication status reported for Busko 2026b (accepted at Scientific Reports, in press) is recorded as proponent-reported and explicitly NOT confirmed against the journal.

    Why: Independence rules: how many genuinely independent evidential events exist is the load-bearing question in this case, and a correction to an independence characterization matters more than another paper.

  20. Fable (AI agent), directed by Eugene · AI-assisted

    Intake from correspondence with the VASCO group's principal investigator (runId trn-villarroel-2026-08-24). The correspondence is not published, not quoted, and is not cited as a source; it was used as a pointer, and every citation was verified independently against the arXiv or repository record before entry. Most of the literature suggested was already in the ledger — Watters et al. 2026 and the group's response, Doherty 2026, Hayes 2026, both Busko papers, Cann 2026 and the Bruehl machine-learning paper were already carried. Three sources were new and are now added: Doherty 2026b (alignments, arXiv 2605.01190), Cann 2026b (geomagnetic-invariance control, arXiv 2604.16470), and the Sinkkoinen open Bayesian reanalysis (GitHub euxoa/plates, CC BY 4.0), with evidence records TRN-E017–E019.

    Why: Fairness and completeness: a proponent's own reading list is a legitimate pointer to literature, and the items it surfaces enter on the strength of the verified sources, not on the correspondence.

  21. Fable (AI agent), directed by Eugene · AI-assisted

    Fifth independent judge added to the cross-model check: Kimi K3 (Moonshot's open-weights model, served via Venice) re-assessed the case blind under the same protocol as the four frontier judges. Its case verdict and per-claim verdicts appear on the concurrence panel and claim pages alongside the others; concurrence statistics recomputed over five judges.

    Why: An open-weights judge from a different training lineage strengthens the independence of the check — agreement across five vendors, including one outside the Western frontier labs, is harder to attribute to shared disposition.

  22. Fable (AI agent), directed by Eugene · AI-assisted

    Cross-model check run (scripts/cross-model-check.mjs, promptVersion aletheia-check-v1): four independent judges — Opus 5 (Anthropic), GPT-5.1 (OpenAI), Gemini 3.1 Pro (Google), Grok 4.6 (xAI) — each re-assessed the case blind to all prior assessments. Result: unanimous case verdict (unresolved) matching the displayed draft; claim agreement 7/13 exact, 2 within one step, 4 split (TRN-C011, TRN-C021, TRN-C022, TRN-C023 — the coma-aberration discriminant and the entire attribution rung). Split claims are linked on the concurrence panel as review entry points. Check runs never display as the case narrative.

    Why: A verdict corroborated (or contested) blind by four independent vendors is information about the evidence file itself; both outcomes belong on the public page.

  23. Eugene (direction); applied by Fable (AI agent) · AI-assisted

    Site v1.1: the case now carries two outputs instead of one — an evidence state and a research priority (level plus stated reason) — and, where a single verdict would mislead by compression, component verdicts for the separable parts of the question. Display governance tightened site-wide: only a human-endorsed assessment run may present as the editorial assessment; unreviewed AI runs are labeled drafts and never silently replace it.

    Why: "Weak evidence, strong reason to investigate" is this site's signature category; a single verdict cannot express it. Priorities and component rows were set editorially in this change and are open to revision on review.

  24. Eugene (review decisions); applied by Fable (AI agent) · AI-assisted

    First human review (PR #4): four open questions decided. TRN-C023 (extraterrestrial-artifact rung) relabeled from presently_untestable to unresolved for consistency with the other open attribution rungs, with the observation that no currently proposed test bears on it retained in the claim's credibility summary and the overview; TRN-C006 (nuclear-window timing) confirmed as mixed; all 13 featured claims and 16 sources retained despite exceeding the medium-depth guideline; the Pleiades survey plate kept, with its caption made explicitly context-only (it depicts no transient candidate and is evidence for nothing in the case).

    Why: Resolve the review questions flagged at import so labels reflect a human decision rather than an AI default.

  25. Fable (AI agent), directed by Eugene · AI-assisted

    First AI assessment overlay generated (runId 2026-08-22-fable-transients-1) and labeled as an unreviewed AI draft: case verdict unresolved, with the incident-light question (TRN-C020) identified as the crux and the original-negative examination (TRN-R001) as the highest-information outstanding test.

    Why: Provide an argued structural roll-up over the ladder without implying human review.

  26. Fable (AI agent), directed by Eugene · AI-assisted

    Case imported into Aletheia: 13 featured claims (6 observation, 3 mechanism/forensics rungs, 4 attribution) plus 4 catalog-tier claims, 16 evidence records, 16 sources, and 8 research opportunities. All claims labeled ai_extracted pending human review; all sources labeled ai_verified. Supporting and undermining evidence imported with equal structure: the Hambly & Blair copy-genealogy mechanism and the Watters dataset critique are carried as first-class records, and the nuclear-window correlation's direction-ambiguity (fallout contamination of plates predicts it as well as glints do) is stated on the claim.

    Why: Launch the case on real, provenance-labeled content with the conventional explanation steelmanned.

  27. Fable (AI agent), directed by Eugene · AI-assisted

    Citation verification batch: all 16 sources checked independently against Crossref, arXiv, and publisher pages before import; none of the research report's citations failed verification. Full texts of the six key preprints (Hambly & Blair, Hayes, Doherty, both Busko papers, Cann) were read directly; journal papers verified by DOI with abstracts read. Two citation-hygiene findings recorded on source records: Cann 2026 mis-cites the Bruehl & Villarroel article number (correct DOI, wrong number), and the Watters critique is confirmed as published in PASA (DOI 10.1017/pasa.2026.10230), not preprint-only.

    Why: Real citations only: research reports can hallucinate references, so nothing enters the case without an independent check.

  28. Fable (AI agent), directed by Eugene · AI-assisted

    Source-mining layer produced: an AI research report ("Adapting the Villarroel transient research to Aletheia," research/ folder) mapped the VASCO/POSS-I literature onto the Aletheia structure — a proposed claim ladder separating five propositions (real plate mark, real incident light, near-Earth reflection, artificial object, extraterrestrial artifact), a replication-independence taxonomy, and a draft research agenda.

    Why: Structure the topic before importing anything, so evidence for low rungs cannot silently support high ones.