Study 03 Solar Flare SIDs

Study 3 — Half the Sky Steps Up at Once: Solar-Flare Sudden Ionospheric Disturbances

The eclipse study (Eclipse 2026 — Overview) watched a small shadow travel across the ground and carve the ionosphere down along a track you can compute to the minute. This study is its exact mirror. When a solar flare erupts, X-rays and extreme ultraviolet arrive at light speed and light the entire sunlit hemisphere simultaneously — no track, no travel, no waiting. The ionosphere over half the planet steps up within minutes, in lockstep with an X-ray clock a satellite reads once per minute, and the response is ordered by a single geometric quantity: how high the Sun stands in each patch of sky.

Same grid. Same integer quantizer. Opposite sign. Complementary geometry. The sky writes geometry, and geometry cannot lie.

---

The forcing and its clock

The clock is the GOES XRS 0.1–0.8 nm soft X-ray channel — the instrument whose scale defines flare classes (an "X9" is a reading of the 0.1–0.8 nm flux). It is public in two forms:

• Live, 1-minute cadence: NOAA SWPC JSON at services.swpc.noaa.gov/json/goes/primary/xrays-*.json (fields time_tag/satellite/flux/energy; VERIFIED by fetch 2026-07-16, currently GOES-18 primary).
• Archival, science-quality: NCEI GOES-R XRS Level-2 netCDF (xrsf-l2-avg1m_science, 2017–present) and the GOES 1–15 XRS science archive (goes08–goes15 present, 1995–2020). Flare begin/max/end times to the minute come from SWPC edited event lists (XRA rows) and the xrsf-l2-flsum_science flare-summary netCDF (VERIFIED: the GOES-18 summary file runs through 2026-07-15); for events before mid-2015 — the floor of the SWPC FTP events archive — they come from the NGDC GOES XRS flare reports (ngdc.noaa.gov/stp/space-weather/solar-data/solar-features/solar-flares/x-rays/goes/xrs/; VERIFIED by fetch 2026-07-16: the yearly report file contains the 2003-10-28 row).

Clock precision: 1 minute, with a 1-second science product (xrsf-l2-flx1s_science, VERIFIED in the NCEI directory) if finer timing is ever needed. The published 1/e rise time of the ionospheric response is ~2.5 minutes (Tsurutani et al. 2005), so a 1-minute clock resolves the onset.

The track is not a moving curve — it is a cap. At flare-maximum minute t, compute the subsolar point (solar declination from the date, equation of time for the subsolar longitude), then the solar zenith angle χ for every cell of the global TEC grid:

Zone	Definition	Role
SUNLIT	χ < 80°	where the step must appear, ordered ~cos χ (Chapman geometry)
TERMINATOR	80° ≤ χ ≤ 100°	excluded band — grazing geometry, no verdict taken here
NIGHT	χ > 100°	control — must stay silent

The eclipse program's obscuration track asked where the shadow is now. This study asks where the Sun is up right now — a computation of the same kind, run in reverse: flares step total electron content up across the whole instantaneous sunlit cap at once; eclipses carve it down inside a small traveling shadow.

---

The response archive

Raw, public, named to the URL. Nothing scored here comes from a private feed.

Archive	URL	Format	Cadence	Auth	Sourcing
NCEI GOES-R XRS L2 1-min irradiance (GOES-16/18)	https://data.ngdc.noaa.gov/platforms/solar-space-observing-satellites/goes/goes16/l2/data/xrsf-l2-avg1m_science/ (GOES-18 under .../goes18/...)	netCDF-4, yearly files + full-mission file; XRS-A (0.05–0.4 nm) and XRS-B (0.1–0.8 nm) in W/m²	1 min; GOES-16 2017-02-07 → 2025-04-06, GOES-18 continues	none (open HTTPS directory)	VERIFIED by fetch 2026-07-16; sibling products flsum (flare class/begin/peak), flloc (disk location), flx1s (1-s flux) also VERIFIED
NOAA SWPC real-time GOES X-ray JSON	https://services.swpc.noaa.gov/json/goes/primary/xrays-6-hour.json	JSON array {time_tag, satellite, flux, observed_flux, electron_correction, energy}	1 min, rolling 6-h/1-d/3-d/7-d windows, updated ~every minute	none	VERIFIED by fetch 2026-07-16 (live GOES-18 records)
NOAA SWPC edited event reports	https://services.swpc.noaa.gov/text/solar-geophysical-event-reports.txt (daily files at ftp://ftp.swpc.noaa.gov/pub/indices/events/)	fixed-width text; XRA rows carry Begin/Max/End to the minute, class, active region	updated every 5 min; FTP archive reaches back to 2015-06-29	none	VERIFIED by fetch 2026-07-16 (live and FTP)
NCEI GOES 1–15 XRS science archive (pre-GOES-R)	https://www.ncei.noaa.gov/data/goes-space-environment-monitor/access/science/xrs/	netCDF per satellite (goes08…goes15) + 2025-reprocessing readme PDF	1-min averages + high-cadence, 1995–2020	none	VERIFIED by fetch 2026-07-16; needed for the 2003-10-28 fossil
Madrigal gridded GNSS TEC (MIT Haystack, CEDAR)	http://cedar.openmadrigal.org	HDF5/netCDF vertical TEC, 1°×1° grid, TECU, via madrigalWeb API	5 min, global, 1998–present, few-day latency	identification only (name/email/affiliation, no password)	already ingested by this program's pipeline (the eclipse study's Madrigal client); same integer-TECU quantizer reused
Stanford SuperSID VLF network (secondary, D-region)	https://solar-center.stanford.edu/SID/sidmonitor/	CSV per station, VLF signal strength vs UT, ~5-s sampling	continuous daily files, hundreds of stations	none for program page	program page VERIFIED (HTTP 200); central browser REPORTED-degraded (TLS error + 403 on 2026-07-16) — secondary corroboration only, e.g. via the DIAS Dunsink archive (dunsink.dias.ie/supersid)

Note

The live SWPC JSON is the watch clock; the NCEI science files are the crucible archive. They are independent copies of the same instrument, which is itself independent of every response-side archive above.

---

The adversary

Every study in this program must name the thing that mimics magnitude but not shape — the adversary that fools any model scoring only on "how big." Here there are two.

The primary adversary: CME-driven geomagnetic storms. Storms produce large positive dayside TEC excursions — storm-enhanced density, prompt-penetration electric fields — and they arrive within a day of the very flares this study scores, because the same active region that flares also launches the CME. The 2017-09-06 X9.3 flare was followed by the Sep 7–8 storm; the 2003-10-28 X17 was followed by the Halloween superstorm (Dst −383 nT). Any pipeline that attributes TEC excursions by geomagnetic index cannot tell them apart. This is the same adversary class that defeated magnitude-models in the eclipse study — the Gannon shear, measured raw in Eclipse 2026 — Model shear — now attacking from the opposite side of the geometry.

The second-order adversary: solar radio bursts that fake the measurement itself. On 2006-12-06, a burst reaching ~1,000,000 SFU at the GPS L1 frequency put 25 dB fades on sunlit receivers, dropping many below 4 tracked satellites (Cerruti et al. 2008, doi:10.1029/2007SW000375). That disturbance is sunlit-confined and simultaneous with the flare — the one adversary that shares the cap.

Why shape defeats both. Three geometric discriminants, none of them a magnitude:

1. Clock. The flare TEC step follows the GOES X-ray peak with near-zero lag (~2.5 min 1/e rise — Tsurutani 2005). Storm TEC lags CME arrival by hours and has no X-ray clock at all.
2. Confinement. The flare response is physically confined to the sunlit hemisphere (χ < 90°) and ordered by cos χ; it is scored only at χ < 80°, per the zone table. Storm response is global — night and day — and organizes by magnetic latitude and local time, not solar zenith angle.
3. Shape. A flare is an impulsive step-and-decay lasting roughly the flare duration up to ~3 h. A storm is a slow multi-hour ramp. And the radio-burst artifact does not produce a coherent SZA-ordered TEC step — it produces loss-of-lock and data gaps, a negative-quality signature, while the VLF SuperSID witness (which never touches L-band) still records the real D-region disturbance.

The adversary matches the number. It cannot match the cone, the clock, and the cap at once.

---

The blind spots this study targets

Five failure modes, each documented in the published record, each one this charter is built to close.

The indices are flare-blind. Kp is a 3-hour number; Dst is a 1-hour number. Apart from the small magnetic crochet, neither registers a flare at all — so index-driven space-weather pipelines file every TEC excursion under "storm" (Tsurutani et al. 2009, Radio Science, doi:10.1029/2008RS004029). Flare-vs-storm attribution requires the X-ray clock cross-reference this study formalizes.

The biggest flare on record produced the smaller response. The 2003-11-04 X28 — the largest soft-X-ray flare ever measured — struck at the limb and moved the ionosphere less than the 2003-10-28 X17 at disk center (25 TECU), because the EUV that actually ionizes the E/F region is attenuated toward the limb while X-rays are not (Tsurutani et al. 2005, doi:10.1029/2004GL021475). Class alone is a magnitude, not a geometry. Every threshold in this study is gated on flare disk position (the flloc product, VERIFIED to exist).

The measurement dies exactly when the signal peaks. The 2006 radio burst put receivers below 4 satellites; the 2017 X9.3 tripled precise-positioning error from ~0.15–0.2 m to ~0.57 m (Yasyukevich et al. 2018, doi:10.1029/2018SW001932). GNSS-TEC archives carry a built-in selection bias against the largest events. Here, sunlit-cell dropout-rate is itself a scored channel, and the VLF witness stands outside L-band entirely.

The standard maps smear the signal into invisibility. Global ionosphere maps at 2-hour (even 15-minute rapid) cadence cannot hold a ~2.5-minute rise. Operational products built on them systematically under-report sudden ionospheric disturbances. This study scores raw 5-minute Madrigal grids only.

The clock changed its own calibration. GOES-R-era science fluxes run ~1.4× higher than the historical SWPC-scaled GOES 1–15 values (the scaling factor was dropped in 2020, per the NCEI GOES 1–15 readme — archive VERIFIED). A 2003 "X17" and a 2024 "X9.0" are not directly commensurable. Thresholds are sealed per calibration era.

---

The shear metric

Confinement analog — the reciprocal of the eclipse obscuration gate. The flare step must be confined to the sunlit cap and ordered by cos χ, as exact integer inequalities in TECU × 100:

band(χ<40°)  >  band(40–80°)  >  0    AND    |band(χ>100°)| ≤ sealed negative floor

All four quantities are exact integers (TECU × 100); the negative floor is the sealed integer from the derivation below. Eclipse: down, inside a small moving shadow. Flare: up, across the entire instantaneous sunlit cap. Same grid, same quantizer, opposite sign.

Traveling-lag analog — an anti-traveling signature. In the eclipse study the response traveled with the umbra, lag ordered by ground-track arrival time. Here the response must be simultaneous at the archive's cadence: the step is present in the first 5-minute Madrigal epoch at or after the GOES peak minute, in every sunlit band, regardless of longitude, with no propagation ordering. The discriminant is onset-epoch uniformity — eclipse: onset ordered by ground-track arrival; flare: the same first epoch everywhere sunlit; storm: incoherent (hours of lag from the X-ray clock and traveling structure — SED plumes, TIDs). Storms fail the clock twice.

Quantization — exact integers only, in the eclipse study's own quantizer: TEC as TECU × 100 exact-integer strings; X-ray flux as a log-class integer (flare class × 10 — X9.3 → 93, from the flsum netCDF class field); solar zenith angle in integer degrees; onset as an integer Madrigal epoch index (5-min epochs). No floating point touches anything sealed.

Threshold derivation plan — frozen from the historical corpus, then sealed before any 2026 event is scored:

1. Ingest Madrigal 5-min TEC for the 5 fossil flare days and the 3 adversary windows below.
2. Per event, compute the step per SZA band, with every choice fixed here and sealed with the threshold: baseline = integer mean of the three 5-min epochs ending at T−5 (T = the GOES peak minute); step = the maximum of the three epochs T…T+10, minus the baseline; band membership frozen at minute T; band statistic = the integer median of cell values, in TECU × 100.
3. Anchor the class-response curve on the published points (X2.2 → 2–4 TECU; X9.3 → 8–16 TECU; X17 → 25 TECU) and set the detection threshold as the largest step observed in any sunlit band on the storm-only negative windows — the false-positive floor. Per-band floors are published raw before any interpretation. Pre-registered branch: if the negative-window floor exceeds any published anchor, that outcome is itself the recorded finding, published raw, and the confinement and clock gates alone carry the discrimination. A flare detection must exceed the floor and pass the SZA-ordering and first-epoch clock gates.
4. Seal the threshold — together with the baseline/step/band definitions above and the coverage-gate integers N and M from the pre-registration — before the first pre-registered flare is scored. The standing X-flare watch is the falsification arm.

A recorded failure at any of these gates is a finding. It is written raw and never renamed.

---

Historical corpus

Five flares (the crucible) and three adversary windows (the negative floor). All times UT.

Flare fossils

Event	Date	Clock (GOES XRS)	Published response	Citation	Sourcing
X17.2, AR 10486 — largest disk-center TEC event on record	2003-10-28	start 09:51 (impulsive stage ~11:01), max 11:10, end 11:24	~25 TECU (~30% above background) subsolar step, ~2.5 min 1/e rise, ~3 h persistence; larger than the limb X28's response	Tsurutani et al. 2005, GRL 32, L03S09, doi:10.1029/2004GL021475; peak corroborated by Simnett 2005, doi:10.1029/2004JA010789	VERIFIED via abstract text; begin/max/end VERIFIED against the NGDC goes-xrs-report yearly file (fetched 2026-07-16)
X9.3, AR 12673 — largest of Solar Cycle 24	2017-09-06	peak 12:02	~8–10 TECU midlatitude / 15–16 TECU low-latitude step; GPS PPP error 0.15–0.2 m → 0.57 m (~3×); dayside HF fadeout	Yasyukevich et al. 2018, Space Weather 16, doi:10.1029/2018SW001932	VERIFIED by full-text fetch 2026-07-16
X2.2, AR 12673 — same-day class-scaling control	2017-09-06	peak 09:10	2–4 TECU dayside; no significant positioning degradation — a built-in calibration pair three hours before the X9.3	Yasyukevich et al. 2018, doi:10.1029/2018SW001932	VERIFIED by full-text fetch
X8.7, AR 13664 — limb flare on a storm-recovery background (control)	2024-05-14	start 16:46, peak 16:51, end 17:02	strongest flare since 2017 at the time; American-sector dayside SID/HF blackout; 3.5 days after the Gannon superstorm from the same region, which by then had rotated to the west limb (~W89 — REPORTED; to be confirmed against flloc) — outside the 60° disk-position gate, scored as a limb/EUV-attenuation control, not a required detection; no published TEC anchor — exploratory row, not S1-scored	NOAA SWPC news; The Watchers 2024-05-14; NASA SVS 14592	REPORTED (times consistent across three sources)
X9.0, AR 13842 — largest flare of Solar Cycle 25	2024-10-03	start 12:08, peak 12:18, end 12:27	subsolar point over Africa/Atlantic; dense European/African Madrigal coverage — the primary modern crucible fossil; no published TEC anchor — exploratory row, not S1-scored	SIDC "Strongest solar flare of SC25"; NASA SVS 14701; The Watchers 2024-10-03	REPORTED (times consistent across sources)

Adversary windows

Event	Date	Why it attacks this study	Sourcing
Gannon (Mother's Day) superstorm, G5	2024-05-10 → 05-12	Peak Dst −412 nT at 02:00 UT May 11, strongest storm since 1989/2003. Global TEC restructuring, day and night, hours-long ramps, no X-ray step at onset (CME arrival ~17:05 UT May 10). The detector must return zero flare-detections through the main phase; the X8.7 of May 14 from the same active region — by then at the west limb, outside the disk-position gate — serves as a limb control, not a required detection.	Dst −412 nT at 02:00 UT VERIFIED (published Gannon-storm analysis, gc.copernicus.org 2024); CME arrival time REPORTED
Halloween storms after the X17	2003-10-29 → 10-31	CME/shock arrival ~19 h after the flare (~06:00–06:30 UT Oct 29); the twin Dst minima of −353 nT and −383 nT followed on Oct 30. The storm raised dayside TEC >200% in places (Mannucci et al. 2005, doi:10.1029/2004GL021467). Flare step (minutes, 25 TECU) and storm enhancement (hours, larger) sit adjacent in time — the canonical case where index-based attribution smears them into one. The SZA-lock plus the X-ray-clock lag must cut them apart.	Storm timing and Dst minima REPORTED (published Dst/SYM-H records); TEC enhancement REPORTED via Mannucci et al. 2005 citation
Extreme radio burst with X6.5 flare	2006-12-06	~1,000,000 SFU at GPS L1, 25 dB fades, sunlit receivers <4 satellites (Cerruti et al. 2008, doi:10.1029/2007SW000375). The one adversary sharing the sunlit-cap confinement — defeated by data-quality shape (lock-loss and gaps, not a coherent SZA-ordered step) and by the VLF witness.	REPORTED via Cerruti et al. 2008 citation

---

Pre-registration

Target. The next GOES XRS flare of class ≥ X1.0 (science-flux basis) with disk position within 60° of disk center, occurring while Madrigal coverage passes a sealed integer gate: at minute T, ≥ N cells with valid TEC in the χ<40° band and ≥ M cells in the 40–80° band, with N and M sealed alongside the detection threshold before the first event. Solar Cycle 25 remains capable in its declining phase — the 2024–2025 maximum produced repeated X-class flares, including two of this study's own fossils (SWPC/NCEI event lists) — so the expectation horizon is weeks-to-months, not years.

Cadence. Live watch polls the SWPC 1-minute JSON on a 5-minute heartbeat, mirroring the eclipse-watch daemon. A flare is confirmed against the SWPC edited-events XRA row — ground truth independent of our own peak detection. TEC scoring runs automatically when the Madrigal window lands (few-day latency).

The standing law, sealed before scoring:

Note

IF GOES XRS 0.1–0.8 nm flux peaks ≥ 1×10⁻⁴ W/m² (X1) at minute T, with disk-center distance ≤ 60°, THEN gridded TEC at cells with χ < 80° shows a positive step ≥ the sealed detection threshold (a single exact integer, TECU × 100) in the first 5-min Madrigal epoch at or after minute T, SZA-ordered (band medians monotonic in cos χ), AND cells with χ > 100° show |step| ≤ the sealed negative floor. CONVERSELY, on any storm interval (Dst < −100 nT) containing no X-ray peak, the detector stays silent. Either failed leg falsifies the law — the same two-sided structure as the eclipse 2026-08-12 pre-registration.

Why the law is two-sided

A one-sided detector that only ever finds flares can never lose to the adversary — it just files storm days under "noise." The converse leg makes the storm windows scoring events in their own right: a single flare-detection fired during the Gannon or Halloween main phase falsifies the law exactly as loudly as a missed qualifying X-flare. Both legs are decidable from public archives by anyone.

---

Success criteria

Each criterion is decidable from the named public archives and the sealed integer thresholds — yes or no, no adjudication.

• S1 — Crucible replication. The three anchored fossil flares (2017 X2.2, 2017 X9.3, 2003 X17), ingested from raw Madrigal 5-min TEC, produce exact-integer sunlit steps inside explicit acceptance intervals sealed before ingestion, per SZA band, in TECU × 100: X2.2 → [200, 400] dayside; X9.3 → [800, 1000] midlatitude and [1500, 1600] low-latitude; X17 → [2000, 3000] subsolar (the published "~25 TECU" widened to an explicit interval, sealed before ingestion) — each SZA-ordered and each present in the first 5-min epoch at or after the GOES peak minute in the flare-report/event-list clock. The two 2024 fossils have no published TEC anchors and are ingested as exploratory rows, not S1-scored.
• S2 — Adversary silence. Zero flare-detections during the Gannon main phase (2024-05-10→12) and the Halloween storm days (2003-10-29→31); the 2006-12-06 radio burst scores as a data-quality event (lock-loss/gap channel), not a TEC step; the 2024-05-14 X8.7 — at the west limb, outside the 60° disk-position gate per its flloc record — is scored as a limb/EUV-attenuation control on its storm-recovery background: its sunlit-band steps are recorded raw, a reduced or absent step is the expected outcome, and no detection is required.
• S3 — Threshold sealed before scoring. The detection threshold equals the largest sunlit-band step on the negative windows, is recorded as an exact integer with its derivation, and is frozen before any 2026 flare is scored. Any post-hoc adjustment voids the study.
• S4 — Pre-registered live event. The next qualifying ≥X1 disk-center flare passes both legs of the standing law: sunlit step ≥ the sealed threshold in the first 5-min epoch at or after the peak minute, SZA-ordered, night control silent. If either leg fails, the failure is published raw as the study's finding.

Honest limits

• This is not a flare forecaster. Nothing here predicts when a flare erupts — only what the ionosphere must do in the minutes after one, and what it must not do when there is none.
• This is not a storm model. Storms appear here only as the adversary. Their inner structure (SED plumes, penetration fields) is out of scope beyond "the detector stays silent."
• Cross-era classes are not commensurable. GOES-R science fluxes run ~1.4× the pre-2020 SWPC-scaled values; thresholds are sealed per calibration era, and 2003-vs-2024 class comparisons are never made on a single scale.
• The measurement can die at the peak. An extreme flare or radio burst may leave gaps instead of a step. Dropout-rate is a co-detection channel and the VLF witness stands by, but a false negative from receiver blinding remains possible and would be recorded as such.
• A limb X-flare may legitimately under-shoot. EUV center-to-limb attenuation means the 60° disk-position gate excludes some real flares from scoring — a narrower target set, accepted openly.
• The wait may be long. SC25 is past its maximum interval; the standing law persists until the target arrives. It does not expire and it is not relaxed.

Status: OPEN — charter published, corpus not yet ingested. Nothing here is sealed until the corpus runs.