Tight counterpart positions do not supply red Class-I colours
12 October 2026. The registered, retrospective IRAC diagnostic is independently verified. For the same three consumed positional counterparts, the central [3.6]-[4.5] colours are -0.089, +0.134 and -0.011 mag. None exceeds the >0.7 mag necessary colour cut in the declared Gutermuth et al. (2009) Phase-I scheme. Under the frozen three-times-sum-of-quoted-errors sensitivity, the first and third remain below this cut; the second is unresolved. Zero of three has the four-band photometric availability needed to apply that full scheme.
This is no physical Class-I exclusion, field-star classification, independent truth or probability. The selection and magnitudes were consumed before this registration; the diagnostic is not a preregistered discovery. A lack of red colour support for a particular point source does not exclude another protostar inside its Hi-GAL clump. The earlier tight 21/21 positional alignment stays valid.
Fixed records and measured colours
| Clump | Fixed best geometric Catalog record | [3.6]-[4.5] mag | Upper 3x quoted-error sensitivity | Missing bands | Assessment |
|---|---|---|---|---|---|
| HIGALBM330.8814-0.3656 | SSTGLMC G330.8829-00.3661 | -0.089 | 0.292 | 8.0 micron | Below the necessary cut throughout the declared sensitivity |
| HIGALBM351.2258+0.6937 | SSTGLMC G351.2258+00.6940 | +0.134 | 1.124 | 5.8, 8.0 micron | Unresolved under quoted-error sensitivity |
| HIGALBM353.4272-0.3734 | SSTGLMC G353.4273-00.3723 | -0.011 | 0.208 | 8.0 micron | Below the necessary cut throughout the declared sensitivity |
The first record has native 3.6/4.5 magnitudes 8.027/8.116 and quoted magnitude errors 0.055/0.072. Its known catalogue position is 0.168–0.336 arcsec from all 21 supported query-visit medians and 5.719 arcsec from the clump centre. Positionally it remains a strong inspection case. Its short-wave colour does not supply the red infrared support required by this chosen Class-I colour component. Its evolutionary stage and clump membership still remain unknown.
The first and third also have measured [4.5]-[5.8] colours +0.194 and +0.268 mag; the second lacks that colour. All twelve native product records (six Catalog records and their six Archive aliases) and all alternative colours are retained. The full record table includes 34 calculable band pairs and every unavailable pair as null. Archive/Catalog agreement is not independent observational truth.
What the published comparator does and what we implemented
Gutermuth et al. (2009), Appendix A.1, DOI 10.1088/0067-0049/184/1/18, uses four IRAC detections with magnitude errors <0.2, contaminant screening, and two Class-I colour inequalities: [3.6]-[4.5]>0.7 and [4.5]-[5.8]>0.7. This diagnostic implements only availability and those necessary colour checks, not the complete classifier. Other phases require additional near-IR or 24-micron evidence not established here. The rule is not a universal physical protostar definition, nor a consensus class for these particular sources.
The missing 8-micron values cannot be converted into upper limits or physical nondetections. Extinction, per-band reliability, calibration/covariance, decomposition, epoch differences and emitter ownership remain unresolved. A corrected-photometry flag need not invalidate the source, but it does not establish calibrated colour errors either. No full standard Class-I/II decision is reported and no physical starless label is assigned.
Unknown covariance and the exact conditional scope
For c=m_a-m_b, quoted marginal errors s_a,s_b and unknown correlation rho,
Var(c) = s_a^2 + s_b^2 - 2*rho*s_a*s_b
|s_a-s_b| <= sigma(c) <= s_a+s_b, for -1 <= rho <= 1.
The frozen primary diagnostic is c +/- 3*(s_a+s_b); the one-times envelope and independent-error reference sqrt(s_a^2+s_b^2) are secondary. The upper c12 values are 0.292, 1.124 and 0.208 mag. They are quoted-error sensitivity values, not confidence bounds, physical exclusion limits or calibrated probabilities. Unmodelled systematics, non-Gaussian errors and extraction problems may exceed these envelopes. The two colours share 4.5-micron measurements; separate marginal envelopes do not supply joint probabilities or a jointly compatible fit.
No source is newly admitted to the earlier strict zero-flag association test. That original 0/3 gate, the 21/21 positional geometry, the failed 3/27 cadence majority gate and the 14/27 versus 15/27 ranking result all remain unchanged. The old 333/800 uncertainty gate also remains failed; occurrence is unmeasured.
Our system and the source-level physical question
All three clumps were already selected by all four old ranking methods at the 982-review all-requested budget. The embedding scores remain 0.9564869181, 0.6959411573 and 1.3350584606. They rank clump chemistry proxies, not the youth of a particular Spitzer point source, and they are not probabilities. No model was fitted, changed or rerun and no new colour input was added to it.
The new comparison cards retain those historical ranks beside the published colour-component assessment, the original geometry, every native alternative, and physical class/membership unknown. It would be incorrect to interpret a high clump rank as a high probability that the matched blue point source is a protostar. Conversely, that point source’s colour does not rule out an embedded object elsewhere within the molecular beam or dust clump.
All 19,328 parent identities and 19,066 unknown qualifying proxy states remain unchanged. These three selected, chemistry-positive cases provide neither representative physical precision/recall nor independent held-out validation. No new star, birth, robust detector or improved ranking is established.
Verification, resources and stopped preflight
The registration and synthetic controls were committed at a2eb4d5b before the
new diagnostic worker. The first guard stopped with elevated memory pressure
before spawning a worker; its terminal receipt is retained at 4880314a.
A separate Proof Atlas Lean build finished on its own. No other project was
cancelled. Resumption amendment c889cc5e then bound fresh stages after normal
memory pressure and sufficient free disk space returned. All inputs, rules and
limits stayed fixed. This operational stop/resumption is not a scientific result.
The resumed analysis and independent verifier both completed under the original one-CPU, ten-minute, 256-MiB sampled-RSS, 1-MiB stage guards. Each took about 1.05 seconds including monitoring. Minimum observed free space was about 40.59 GiB. Their very short lifetimes mean small sampled RSS does not establish actual transient peaks. No guard violation, new observations, network worker, model fit or paid computation occurred. The combined exact evidence/report/code packet is below 4 MiB; original observations remain intact.
The checker imports no generator functions. It reads all six original native responses directly, uses exact Fraction arithmetic instead of the worker’s decimal differences/envelopes, and verifies every source/counter/ordinal, native field, available and missing colour, sensitivity, case/card linkage and count. It checked all twelve records and 34 calculable pairs. This is independent software recomputation, not fresh physical validation or human adjudication.
Conclusion and stop
The important change in interpretation is that tight positional agreement and a high clump-chemistry rank do not provide red protostar-colour evidence for the best-aligned point source. Its geometric lead remains useful, while physical youth remains unestablished. The other cases also provide no robust necessary red-colour support under this diagnostic; incomplete bands prevent full classification. The physical detector objective remains active and unmet.
This bounded diagnostic stops here. Native-image ownership and independent membership/stage evidence are still needed; no automatic source substitution, weaker colour/flag cuts, band acquisition, new model fit or successor test follows.