Listening for events…
← All Docs

Paper Revisits

Created 2026-05-20. Registry of pre-committed paper V2/V3 revisits with explicit trigger conditions. Companion to paper-coverage-gaps.md, dream-papers.md (legendary-tier long-horizon pitches), and research-arcs-and-idle-projects.md.

Why this exists

A V2 paper is a strong honesty signal only if the trigger condition was committed before the V1 result was known. This file tracks those commitments. When a trigger fires (N threshold met, calendar date reached, infra landed, event observed), the V2 paper is queued with the original methodology — no methodology shopping, no post-hoc justification.

Each entry must contain enough to re-run V2 mechanically: original paper, trigger condition, pre-registered method, and what would change the conclusion.

Trigger flavors

  • N-gated — wait until sample size crosses a stated power threshold. Trigger is a SQL query against observations.
  • Coverage-gated — wait until a new data source, sensor, or constituent comes online. Trigger is the presence of a new datasource_id or metric.
  • Event-gated — wait for the next instance of an external event (Forbush, X-class flare, named storm, large earthquake near sensor). Trigger is event detection in the relevant feed.
  • Infra-gated — wait for fetcher / pipeline work to land. Trigger is a closed platform issue.
  • Calendar-gated — wait for a specific date, typically an anniversary or seasonal window. Trigger is just the clock.

A single revisit can be multi-gated (e.g., AQI V2 is N-gated AND calendar-gated; CNEOS V2 is N-gated with a calendar estimate).

Per-entry fields

Every entry below uses the same skeleton:

  • Original — workspace slug + commit(s) + explainer pointer
  • Flavor(s) — one or more of the five above
  • Trigger condition — specific, falsifiable, machine-checkable when possible
  • Earliest expected fire — date or conditional estimate
  • Pre-registered method — pointer to V1 §M.x or supplemental; methodology is locked
  • What would change the conclusion — one sentence per possible outcome
  • Status — pending / fired / superseded / cancelled

Active revisits

V2 of #203 — AQI anomaly characterization

  • Original: workspace aqi-anomaly-characterization, paper commits f0efe77→e4bcb3a, explainer docs/paper-explainers/203-aqi-anomalies.md
  • Flavor(s): N-gated + Calendar-gated
  • Trigger condition: ≥60 ≥3σ anomalies in full-constituent (post-2026-03-16) ingest. Check via SELECT COUNT(*) of anomaly-flagged rows where timestamp_utc >= '2026-03-16'.
  • Earliest expected fire: ~2026-09-01 (~5.5 months of full-constituent data at current cadence projects to 60–80 anomalies).
  • Pre-registered method: 6-D KMeans clustering across the 6 AQI constituents (originally pre-registered in #203 V1). V1 reported structural finding instead because N=28 was insufficient for stable cluster recovery.
  • What would change the conclusion: Stable cluster recovery at k=2 or k=3 with silhouette ≥0.4 would upgrade V1's structural finding to a real anomaly typology. Failure to recover stable clusters confirms the modest bimodality (V1 ΔBIC=4.64) is noise, not a population structure.
  • Status: pending

V2 of #205 — CNEOS fireballs × meteor showers

  • Original: workspace pma-cneos-fireball-day-of-year-vs-meteor, paper commits 1518211→9406b37, explainer docs/paper-explainers/205-cneos-fireball-showers.md
  • Flavor(s): N-gated (with calendar estimate)
  • Trigger condition: N ≥ 600 unique CNEOS fireballs, deduped on minute-resolution timestamp + 0.1° rounded coordinates. At N=600 the Bonferroni-12 detection floor drops to rate ratio ~1.5, matching the optical-bolide Geminid literature range.
  • Earliest expected fire: ~2034 at current ingest rate (~22 deduped events/year post-2014). Re-check yearly; could fire earlier if CNEOS sensor program expands.
  • Pre-registered method: Schuster test + per-shower Poisson rate-ratio at h ∈ {3,5,7,10}d with Bonferroni across 12 named showers, identical to V1 §3. Pred. column (A=asteroidal, C=cometary) carries forward.
  • What would change the conclusion: Geminids (or any asteroidal-parent shower) surviving Bonferroni at any h with rate ratio in 1.3–1.5 resolves V1's ambiguity toward Story B (real but previously sub-floor signature). Continued Bonferroni-null at the lower floor resolves toward Story A (genuinely sporadic kt-energy population, distinct from optical meteors).
  • Status: pending

V2 of #202 — Safecast × Forbush radiation modulation

  • Original: workspace safecast-forbush-radiation-modulation, paper commits f6619db→6a2ec46, explainer docs/paper-explainers/202-safecast-forbush.md
  • Flavor(s): Infra-gated (partially satisfied; see below)
  • Trigger condition: NMDB fetcher merged to main AND producing rows for ≥3 stations spanning a useful latitude range (must include ≥1 equatorial |φ|<30° AND ≥1 southern-hemisphere monitor). V1 was bounded by Safecast CPM detection physics — gamma-only, ground-level, no polar/equatorial coverage. NMDB neutron-monitor data is the correct instrument, but inherits the same global-coverage requirement.
  • Status of trigger (verified 2026-05-20): Partially met. NMDB fetcher is live (nmdb_cosmic_rays.py, 4,469 rows, 2026-04-02 → 2026-05-20). Station selection is 4 NH-only sites — Jungfraujoch (46.55°N), Oulu (65.05°N), Apatity (67.57°N), Thule (76.5°N) — polar/mid only, no equatorial, no southern. The spirit of the original condition (replicate Safecast's latitude-coverage gap with the correct instrument) is NOT met. Two paths forward: (a) expand NMDB station selection to include equatorial + southern monitors (Tsumeb, Princess Sirindhorn, Kerguelen, etc. are public NMDB stations); (b) honestly scope V2 as a NH-only paired-SEA with explicit acknowledgement that polar/equatorial decomposition isn't possible at current station selection.
  • Earliest expected fire: Depends on chosen path. (a) requires a fetcher-config change + ~few weeks of new station data accumulating; (b) is runnable today with the caveat documented up front.
  • Pre-registered method: Same paired SEA design as V1 §4, swapping Safecast CPM for NMDB neutron count. Bonferroni across the same 5 candidate drivers. Latitude stratification using the NMDB station network as configured at fire time.
  • What would change the conclusion: A Bonferroni-significant Forbush signal in NMDB data closes out the "CPM channel was the wrong tool" hypothesis with a positive result and supersedes V1. Continued null in NMDB points at driver selection or coupling mechanism, not the sensor, and bounds the search differently.
  • Status: pending — awaiting station-network decision (see "Status of trigger" above)

Workflow when a trigger fires

  1. Verify the trigger condition still holds (re-run the N query; confirm fetcher is live and stable; etc.).
  2. Open a GitHub issue for V2 (numbered separately, e.g., "V2 of #203 — ...").
  3. Run extraction with the V1 script. Only date-range or metric updates allowed.
  4. Run analysis with the pre-registered V1 method. Do not add new tests. A new test = V3, not a methodology change.
  5. Write V2 paper as "what V1 said → what changed in the data → what V2 says now." Cite V1 prominently and link the explainer.
  6. If V2 contradicts V1, that's more interesting, not less. Lead with the contradiction.

Maintenance

  • Add a new entry whenever a published paper has a pre-registered V2 condition. Do this at ship time, not later.
  • Update Status when a trigger fires; never delete fired/superseded entries — keep the chain so V2 can cite the registry.
  • Calendar-gated and near-N-threshold entries: re-check at the start of each session if within 30 days of expected fire.
  • Yearly review: re-estimate N-gated fire dates from current ingest rates.
  • If a V1 paper should have had a V2 condition pre-registered but didn't, note it here under a ## Retroactive candidates section — but treat retroactive entries as weaker honesty signal than pre-registered ones.
Live Feed