Frozen multi-survey association and temporal-support audit

Frozen multi-survey association and temporal-support audit

12 October 2026. This is the next bounded preparation unit of the accepted multi-survey program, not a model experiment or reopening of E22/E28. The eligibility checkpoint established identities but not one-to-one associations.

Question and decision

Can the cached nearest-counterpart products be treated as distinct clumps with one unambiguous molecular record and a usable anchored temporal view? The falsifiable geometric hypothesis is that every nearby-parent clump with MALT90 support has exactly one record within 30 arcsec and that record’s position has no other reliable Hi-GAL clump within 30 arcsec. One counterexample rejects this hypothesis. This is a deterministic statement about the cached catalogues, not a random-sky significance test or a test of physical protostar membership. Report the full denominator, multiple velocity records, shared positions and neighbouring catalogue objects rather than silently selecting nearest rows.

Temporal quality/visit counts and view intersections are secondary readiness diagnostics. They have no model pass threshold or claimed population recall. They will define missing-view and grouping requirements for a later frozen comparison; they do not authorize a fit by themselves. No outcomes are opened.

Population, inputs and roles

Retain every one of the 19,328 exact designation strings in the verified nearby Hi-GAL catalogue-prestellar ledger. Reconstruct evol_flag=1 and finite 0<dist<=4 from the cached parent. Other reliable catalogue rows (evol_flag 0,1,2), regardless of distance, form the neighbour/confusion context. Exact input/code/doc hashes are in the separate registration. Goal/program policy is bound to Git revision 9bbd4924, preserving that historical scope if later policy text changes.

Read MALT90 recno, ID, Name, GLON and GLAT only. Preserve every native string and source row ordinal, including repeated coordinates and multiple components. Do not read S, Vc, line intensities, detection flags, old activity, fragmentation, CLIP scores or SPICY classes. Read only IDs from historical E22 membership via the prior eligibility ledger. The source paper describes separate A/B velocity components at the same sky location; do not collapse these to independent beams or pick a velocity component by ordering. No old chemistry outcome is evaluated.

Retain derived AllWISE coordinates, their input row ordinal and separation only; native IDs, position errors and discarded alternatives are absent. The original cache builder already deduplicated its chunk products, so this audit cannot recover alternatives or certify unique WISE identity. No WISE colours are used.

NEOWISE input is the exact historical target order plus all cached single-exposure rows. Query target coordinates must agree to 1e-6 degree. Detection identity is an input hash and 1-based data-row ordinal, not an invented native source ID. Retain raw bytes as the complete alternatives record; export compact per-target counts and visit summaries, not a duplicate 113 MB observation table.

Fixed association and quality rules

  1. Use great-circle distances, not a planar RA subtraction. MALT90 and Hi-GAL matching uses native Galactic coordinates. Export every reliable catalogue neighbour within 30 arcsec for every MALT90 record, including outside-parent neighbours. Also report all neighbours within 19 arcsec, half the nominal 38-arcsec beam FWHM. These are geometric radii, not beam edges or ownership probabilities. Same-coordinate molecular rows share a sky-position group; distinct velocity components stay separate. Link parent clumps sharing any molecular sky-position group within 30 arcsec into a deterministic connected component for subsequent leakage prevention. Position errors are unknown.
  2. AllWISE separation is recomputed from the retained celestial coordinates. A retained match within 3 arcsec is an anchor only. Missing alternatives and native IDs remain explicit. Neither a nearest anchor nor agreement between epochs establishes clump membership.
  3. A NEOWISE row passes the available-field screen iff valid finite RA/Dec, MJD, W2 magnitude and positive finite magnitude error; W2 SNR>=5; qual_frame>0; nb=1; and cc_flags has four characters with its W2 character equal to ‘0’. These choices are conservative analysis rules, not a complete reliability certificate. Missing reduced chi-square, moon/solar-system/saturation flags and positional uncertainties remain unverified. Count each failure predicate separately (they overlap), including upper-limit-like SNR<2 rows.
  4. For an existing anchor within 3 arcsec of its clump, screen-passing detections within 2 arcsec of the anchor form the anchored sample. Report passing rows outside that radius as competing/unanchored evidence, without declaring them different stars. A 2-arcsec rule is geometric, not a fitted astrometric sigma.
  5. Deduplicate anchored measurements by the exact (MJD,RA,Dec,W2 magnitude, error) string tuple; do not call this proof of unique exposure identity. Sort by numeric MJD and start a new visit when the successive gap exceeds 30 days. A supported visit contains >=5 retained tuples. Temporal support requires

    =3 supported visits and an overall anchored time span >=365 days. The span uses all retained anchored tuples. Do not compute variability, events or a classifier in this unit. Export visit counts, time span and observed MJD endpoints so the rule is independently reconstructible.

  6. Four-band FIR/error support reuses the prior presence states OOOOEE; 350/500 errors remain absent. Intersections include all anchored temporal support, any molecular support, unique-record/unshared molecular support and four-band/error presence, with all missing cases retained. No missing view is a negative physical label. Do not impute or use 70-micron stage as input.

Estimator, checks and limits

Primary is the exact number of molecular-supported parent clumps violating the one-record/unshared-position hypothesis, divided by all molecular-supported parent clumps. No confidence interval is appropriate for this exhaustive cached geometry test. Report each condition separately and at 19 versus 30 arcsec; retain all alternatives and same-position component multiplicity. Recompute edges independently using Astropy spherical separation against each catalogue position, independently stream temporal screens and visits, and verify all output hashes. The checker imports no generator or historical research module. Use synthetic boundary/RA-wrap/component/duplicate tests before real readout.

One offline worker and one independent checker, each <=600 seconds, one CPU thread, <=256 MiB sampled RSS, no GPU. Combined worker/checker/report stages <=4 MiB, below the program’s 64 MiB ceiling; live resource guard enforces each stage and the final combined check. Keep 5 GiB free. No network observations, downloads, input deletion, cloud jobs, fitting, model sweep, N20/c6828 restart or archive work. A missing input, identity mismatch, resource violation or failed verification stops that attempt; preserve its logs before any separately frozen repair.

Sources and limitations

Rathborne et al., MALT90 line catalogue, sections 2 and 4 describes the 38-arcsec resolution and distinct A/B velocity entries. The official NEOWISE schema defines magnitudes/errors, MJD, nb, frame quality and W2 flag position. IRSA reliability notes identify additional checks absent from this cache. No physical validation cohort is introduced. Old consumed labels cannot become fresh by a spatial split. The 333/800 occurrence gate stays failed and unopened. No detector, occurrence fraction, physical lifetime or stellar birth is established by this audit.