the archive-wide build falsifies three frozen rules

The rule (§2.7): DETCHANS=341, enforced by the M-V parser and the T5 builder.

OBSERVED: HEL1OS has two distinct PHA channel spaces:

Detector family	DETCHANS	CHANTYPE
CZT (CZT1, CZT2)	341	PHA
CdTe (CdTe1, CdTe2)	511	PHA

Every one of the 391 orbits contains CdTe spectra, so every orbit terminates and T3/T4/T5 are empty archive-wide.

Root cause — mine, and the fourth instance of one pattern. I wrote §2.7 after inspecting exactly one file (hel1os_czt_spectra_czt2.fits) and generalised its DETCHANS to all HEL1OS spectra. CONTRADICTION-001 (GTI convention), -003 (LC/spectrum relation), -004 Defect A (R-1 hypotheses) and now -005 A share a single root: asserting a property from a single reading without checking it across the population. The A-8/A-9/A-11/A-12 scoping discipline exists precisely because of this pattern; §2.7's DETCHANS was never given a scoping assumption, and that omission is what let it through.

Consequence for F-11. The contract treats channel-space separation as binary (SoLEXS PI 340 vs HEL1OS PHA 341). There are three incommensurable spaces: SoLEXS PI 340, HEL1OS CZT PHA 341, HEL1OS CdTe PHA 511. F-11's intent is unchanged and strengthened; its arithmetic is incomplete.

Proposed amendment (NOT applied)

1. §2.7 — replace the scalar DETCHANS=341 with a per-family allowlist: CZT → 341, CdTe → 511, both CHANTYPE='PHA'; an unlisted (family, DETCHANS) pair terminates via F-07, exactly as an unlisted band terminates via F-10.
2. §3 T5 — hel1os_spec_1min must carry a detchans column and must not stack CZT and CdTe spectra into one array column; the channel space is a property of the detector family.
3. §5 F-11 — restate over three spaces rather than two.
4. §8 — new assumption A-13: the per-family channel counts are VERIFIED on a sample of 8 orbits; Milestone VIII must verify across all 391; any deviation terminates validation.

---

The rule (§2.1 r2, owner-approved): NaN(COUNTS) set MUST equal the GTI-excluded second set exactly; mismatch → F-09. Scoped by A-9 to the reference archive, with Milestone VIII obliged to verify all 436 and terminate on any violation.

The T1 builder runs this check per day, so A-9's obligation was discharged at M-VII rather than M-VIII. The archive-wide result:

Relationship	Days
NaN == GTI-excluded — the r2 bijection holds	344
NaN ⊂ GTI-excluded — strict subset; bijection FAILS	70
NaN ⊃ GTI-excluded	0
Neither	0
NaN ⊆ GTI-excluded	414 / 414 (100%)

(414 days checked; 22 unparseable — 12 by Defect C, 10 by F-01.)

Excess GTI-excluded seconds carrying finite counts: min 2, median 758, max 43,199, total 266,919.

OBSERVED: the bijection is false archive-wide (70/414 = 17% violate it). The relationship that does hold on every single day is the one-directional NaN ⊆ GTI-excluded: every NaN second is GTI-excluded, but a GTI-excluded second may carry a finite count.

This is exactly the outcome A-9 was written to catch, and the reference day was unrepresentative: 2024-05-14 happens to sit in the 83% where the sets coincide.

Note the direction matters. NaN ⊆ GTI-excluded is still a strong, useful invariant — it forbids the dangerous case (a NaN inside good time, i.e. missing data silently treated as observed). What it does not support is equality, which the r2 rule asserts.

HYPOTHESIS (not for the contract): GTI may exclude seconds for reasons beyond data absence — the excess is large (median 758 s/day, max 43,199 s ≈ half a day) and looks structured rather than incidental. Establishing why is a scientific-validation question, not a parser one, and belongs with CONTRADICTION-003 in Milestone VIII. No mechanism is asserted here.

Proposed amendment (NOT applied)

1. §2.1 r2 / §7 — replace the bijection with the measured, one-directional invariant: NaN(COUNTS) ⊆ GTI-excluded (F-09 on any NaN inside good time). Do not replace it with a tolerance and do not encode the excess statistics.
2. §8 A-9 — mark DISCHARGED at M-VII and VIOLATED: record 344 equal / 70 strict-subset / 0 violations of the subset direction. Replace with A-14: the excess of GTI-excluded seconds over NaN seconds is unexplained; owner Milestone VIII (with CONTRADICTION-003).
3. §3 T1 — live_time_s continues to come from GTI and n_seconds_present from finite counts. They are already carried independently and neither is derived from the other, so no builder change follows from this defect.

---

OBSERVED: 12 of 436 archives terminate at F-19 (GTI row N has STOP <= START), e.g. AL1_SLX_L1_20240422_v1.0 row 2. A further 10 terminate at F-01 (unreadable/gzip).

No amendment proposed. F-19 is behaving exactly as designed: a GTI interval that ends before it starts is not a convention question, it is an archive defect. These 22 days are correctly excluded and individually logged. Milestone VIII should report them as an archive-quality finding.

---

SoLEXS T1/T2 built cleanly for 353 archives, and every invariant held on them: NaN⟺GTI (on those days), finite-only aggregation, live_time_s == EXPOSURE, 1440 rows/day, no duplicate minutes, complete provenance. T6: 2,130 GTI intervals. T7: 1,556 provenance rows. The Version Resolution Engine ran correctly (1,065,572 owned pairs, 48,604 conflicts resolved) — its output is simply unused because no HEL1OS product survived Defect A.

186/186 unit tests pass, including the D1 acceptance criterion: T1's rate_total peaks at 2024-05-14 16:49 UTC, within the frozen ±2 min of the GOES X8.7 at 16:51 — the project's first genuine Aditya-L1 observation.

Milestone VII is paused with the code committed and every rule terminating exactly as frozen. Nothing weakened: DETCHANS is still 341, the bijection is still equality, F-19 still fires. The build's own statistics are the evidence.

RESOLVED. All three defects closed by r5. Defect C required no amendment.

Milestone VIII is now the final validation milestone. It shall discharge A-8, A-11, A-12, A-13, A-14 and resolve CONTRADICTION-003 through archive-wide scientific validation. A-9 is already discharged (here).