Fiesta 1 — second pass on papers with open items
What this is. The paper→dex repoint campaign flips each shipped paper onto dex-sourced data now and keeps the published snapshot intact. Every flip that surfaces something the flip itself shouldn't resolve — a conclusion-shifting adoption, a prose rewrite, a fact-check, a degraded source, a deferred flip — leaves an open item on the paper. Fiesta 1 is the second editorial pass that works through those open items, paper by paper. It is not just a rewrite pass: a paper's second pass may re-run its flipped extract to adopt better data, rewrite the sections the Δ forces, re-ingest a degraded source, complete a flip that was gated, or fact-check a live-written explainer — whatever its open items are.
Parked here (rather than done mid-flip) per [[feedback_flip_accept_better_data]]: flip now, second pass later, so the flip campaign stays a clean plumbing sweep and the editorial judgment happens in one deliberate pass. First of a series — later flips will accrue their own open items; those roll into Fiesta 1 while it's open, or a Fiesta 2 once it closes.
Sequencing (Mike): Fiesta 1 does not begin until the first pass is finished — i.e. every
remaining shipped paper is flipped onto its dex (the plumbing sweep in docs/dex-repoint-plan.md). We
finish the first pass first; this list just accumulates the open items so nothing is lost when the
second pass starts.
Order (Mike): both passes run in paper publish-date order (oldest first), per
web/src/data/features.ts. The open items below are listed in that order, and the second pass works
them the same way.
Rule of the second pass: each paper restores from its committed snapshot, resolves its open items,
rewrites only what the data forces, and re-clears the hed/dek + review gate
([[feedback_review_hed_dek_before_publish]]). Nulls stay honest; conclusions that shift are stated as
shifts, not smoothed over. See docs/dex-repoint-plan.md for the flip mechanics and
docs/dex-backfill-backlog.md for the gated backfills.
Shared method — seismic date audit (cascadia / kermadec / sanriku). The reno flip (54de6f8)
exposed that a hand-authored breaking-news explainer, written live, carried a wrong mainshock date
(stated Apr 12 23:22 UTC; measured Apr 14 01:29 UTC, ~26 h off) and a fabricated foreshock timeline
(narrated M2.9/M3.4 foreshocks; measured zero in the 6 h before). The other live-written seismic
explainers have not been checked for the same class of error. The audit cross-checks every cited
headline event (mainshock date/time/mag/depth/location, any narrated foreshock/aftershock sequence, the
snapshot counts) against the eq dex (usgs_earthquake, UTC-pinned conn, deduped by USGS event_id) —
the reno procedure. Headline/mainshock events are M4.5+ and already in the eq spine, so the reno-class
date check runs now without the micro-tier backfill; the full micro-catalog flip for cascadia
(EQ-2) and kermadec (EQ-3) stays backfill-gated. Outcome rule: a reno-class error → correct it (snapshot
framing) and deploy; clean → record the pass. Stated honestly either way.
Open items by paper (publish-date order)
sky-anomalies — 2026-04-04 — degraded-source re-ingest (NF-1) (parked at src#1, bd42408)
The NUFORC source coarsened since publish, so src#1 was not adopted; the starlink sub-workspace's
inline PG ufo_sighting reads are parked until a re-ingest. → Re-ingest a clean NUFORC, flip src#1,
reconcile any count Δ. Tracked as NF-1 in docs/dex-backfill-backlog.md. (The paper's other four
sources are flipped; adoptions on src#3/#5 were applied inline, conclusion untouched.)
solar-watch-april-2026 — 2026-04-05 — adoption + date-labeling (flip via verify-extract, fcc0aa2)
30 cited claims: 19 match, 11 Δ, conclusion HOLDS. Two causes:
- Date convention — the article's daily figures use US-Eastern (EDT) day boundaries; the dex is UTC. Under EDT the Apr-3/Apr-4 wind timeline reproduces exactly and the "Apr 3" M7.5 flare is the dex's 2026-04-04 01:07 UTC event (same flare). Labeling fixes, not measurement disputes.
- Adopt-better — Kp 6.67 realtime → 6.33 GFZ definitive (still G2); Forbush troughs deeper on the complete curve through the Apr 7-11 recovery (−2.3..−3.4%) than the mid-event partials the article froze on Apr 5 (−0.9..−1.8%) — the signal firms. → Adopt deeper-trough / definitive-Kp numbers, add a UTC-vs-EDT note, rewrite the affected daily rows. Residuals (Apr-2 partial wind, CME 8-vs-9, two far-sided comet-CME speeds) are DONKI-cataloging fixes; none reverses a claim.
wspr-solar-cycle-bug — 2026-04-07 — refresh on drifted raw (flip 7dcfa77)
Drift, not error. /mnt/ursa WSPR raw was backfilled in place in the 7 months since publish
(n_spots up to +1.16 M; mean_snr up to +0.20 dB), and SILSO revised some SSN values — so 7 published
months no longer bit-reproduce from the working surface. Adopting the revised catalog moves the per-band
correlations by |Δr| ≤ 0.003 (every band's sign + magnitude holds).
→ Re-run on the current catalog, note the revision, confirm the conclusion. Snapshot preserved unchanged.
cascadia-swarm-apr2026 — 2026-04-16 — seismic date audit + micro-flip (EQ-2 gated)
Breaking-news swarm explainer — run the shared seismic date audit against the eq spine now. The full
micro-catalog statistical flip (the swarm population is M<4.5) waits on EQ-2 (9 of 23 swarm events
never ingested by the live feed). → Audit the headline dates/sequence now; complete the flip when EQ-2
clears.
wspr-aircraft-detection — 2026-04-16 — complete the deferred WSPR-side flip (aircraft side done, 8993501)
The ADS-B side is flipped onto the aircraft MovingPlatformDex (18/18 numbers reproduce exactly). The
WSPR side is deferred (option B): the raw monthly archive (wspr_raw_202604.parquet) ends Apr 5 and
doesn't yet cover the Apr 12-13 study window, so the study's WSPR spots live only in the frozen workspace
pull. → Once the monthly backfill covers the window (WS-4), repoint the WSPR read onto the archive
pointer and re-run the verify — the WSPR side flips for free. (The "5.3M → 6,626,465" spot-count fix was
a plain error correction, already done + deployed; not an open item.)
kermadec-doublet-apr2026 — 2026-04-19 — seismic date audit + flip (EQ-3 gated)
Breaking-news doublet explainer — run the shared seismic date audit against the eq spine now. The
cross-network flip waits on EQ-3 (EMSC/GFZ/ISC not associated). → Audit the headline dates/sequence
now; complete the flip when EQ-3 clears.
sanriku-m66-may2026 — 2026-05-18 — seismic date audit (doubles as flip)
Single M6.6 mainshock, fully in the eq spine — ungated. It's a foreshock-signature paper, the
exact class where reno's fabricated foreshocks bit, so the audit is high-value. Here the flip and the
audit are the same act: repoint onto read_source_rows("eq", source="usgs_earthquake", …), reproduce
the published numbers, and verify the foreshock sequence against the dex.
bz-dst-lead-time — 2026-06-11 — adoption + rewrite (flip 7512973)
Re-running the flipped extract adopts a much larger window (dst_index backfilled to 1963; DSCOVR now
2025-10 → 2026-06): N 1,359 → 4,343 (~2 mo → ~8.5 mo). Headline HOLDS and strengthens (Bz→Dst
peak r=0.42 @ 5 h; Granger survives all lags; strict-null fails; Burton R²=0.098). Two things need a
real rewrite, not a number-swap:
- The secondary stratification claim ("harder driving → faster response") no longer cleanly holds (strong strata ~4 h vs moderate ~2 h).
- The old "45-hour wrong turn" dissolves (now 143 h, and powered). → Adopt the larger window, rewrite the stratification section + abstract. Snapshot (data + article + paper.tex) restored intact.