Know Your Starting Point

Why 'what file is this?' is a scientific question

Processing assumptions depend on provenance and state, not on a file's extension or how it looks on screen. The same FITS container can hold a raw light subframe, a native telescope stack, a NOVA pipeline stack, a partially processed intermediate, or a finished result -- and each one requires different handling. Reading the file is not the same as knowing what it is.

The product ladder

Data moves through a real sequence: light subframes are calibrated, registered, and integrated into a native SeeStar stack or a NOVA pipeline stack; linear processing and restoration produce linear intermediates, including a star/starless split; a stretch produces nonlinear intermediates; finishing and recombination produce the processed final. A mosaic is a badge that can span any of these rows -- it is not a separate fourth branch of the ladder.

The state card

A complete description of an image's state names its source, product class, filter, calibration provenance, channel/spectral regime, linearity, WCS and registration status, mosaic/geometry attributes, stack provenance, and processing history so far. Product class itself has a boundary worth naming explicitly: a processed scientific/ intermediate artifact kept for further processing or comparison is not the same class as a presentation export made for viewing, even when both were rendered from the same result. Any one of these left unknown is itself a fact worth recording, not something to guess past.

Broadband, dual-band, and mono

Broadband/UV-IR-cut and Ha/OIII dual-band SeeStar data are different spectral inputs, and mono/discrete-narrowband data is different again. Downstream color and metric assumptions that hold for one spectral regime do not automatically transfer to another, even when the target and telescope are the same.

Linear is not 'dark'

A screen transfer function or autostretch can display linear data brightly for viewing without changing the stored pixels at all -- appearance is a display choice, not state evidence. Judging whether data is linear by how it looks on screen is exactly the mistake this page exists to prevent.

WCS present versus WCS trusted

WCS metadata is a claim about how pixels map to sky coordinates, not proof that the claim is correct for this specific image and field. A WCS-dependent operation needs independent confidence that the recorded solution actually belongs to the data it's attached to before relying on it.

Native stack versus pipeline stack

A native SeeStar stack and a NOVA pipeline stack can represent the same target while being genuinely different provenance classes -- the integration engine, frame selection, registration, rejection, and normalization history are not guaranteed to match. They are not interchangeable controls in an experiment; comparing them directly is a strategy comparison, not a same-input comparison.

Experiment Mode fairness checklist

Before trusting any Experiment Mode winner language, check that the candidate inputs actually share the relevant scientific state -- or that the differing state is explicitly what the experiment is testing. A comparison that silently mixes native and pipeline stacks, or broadband and dual-band data, is not measuring what it claims to measure.

When NOVA is uncertain

Provenance and state facts fall into honest categories: known, source-declared, heuristically inferred, and unknown. A heuristic like a filename pattern or a pixel percentile can support an inference, but it is not the same claim as a recorded, source-confirmed fact -- and 'unknown' is a legitimate, useful answer rather than something to paper over.

When to check, and when to stop rather than infer

A state check matters before choosing a processing workflow, reprocessing an existing file, starting Experiment Mode, comparing results quantitatively, or treating an external or manually supplied artifact as a raw pipeline input. Stop rather than infer when product provenance is unknown, when linearity is ambiguous ahead of a linear-only step, when a calibration source is unknown for a calibration-sensitive decision, or when the only evidence for raw status is that a filename happens to contain the word 'stack.'

Auditable claims

1. A FITS file describes a container, not a processing state. Knowing what produced an artifact and what transformations it has already undergone matters more than its file extension or on-screen appearance.

   source-confirmed · none

2. A native SeeStar stack (produced by the telescope/app during capture) and a NOVA pipeline stack (produced by NOVA's own stacker) are different provenance classes even when they represent the same target -- the integration engine, frame selection, registration, rejection, and normalization can all differ between them.

   source-confirmed · none

3. Mosaic is a geometry and provenance attribute, not a second astronomical target and not a fourth product class alongside subframe, stack, and processed result. Treating a mosaic suffix as a separate canonical target fragments provenance that belongs together.

   source-confirmed · none

4. Broadband/UV-IR-cut and Ha/OIII dual-band SeeStar data are different spectral inputs. Their channel ratios and the downstream color assumptions built on them are not directly interchangeable, even when both are nominally the same filter product.

   source-confirmed · none

5. Linear data can be displayed brightly with a screen transfer function or autostretch without the stored pixels becoming nonlinear. On-screen appearance alone is not evidence of processing state.

   source-confirmed · none

6. WCS metadata states how pixels are claimed to map to sky coordinates -- it does not by itself prove that claim is correct for this image and field. A WCS-dependent operation still needs independent confidence that the recorded solution actually belongs to the data it's attached to.

   vendor-documented · none

7. A raw-input safety predicate and a user-facing provenance label answer different questions. The label describes what a product is; the safety rule decides whether it is allowed to enter a raw processing pipeline -- NOVA's own `is_raw_stack()` check exists specifically because a plausible-looking label is not sufficient grounds to skip that decision.

   execution-confirmed · none

8. A processed scientific/intermediate artifact retained for further processing or comparison is a different product class from a presentation export (JPEG, PNG, or a color-managed TIFF) generated for viewing -- the two can look similar or even come from the same result, but clipping, normalization, resizing, gamma/display transforms, or color-profile conversion applied for presentation make that export unsafe to substitute as the parent for a later scientific comparison.

   source-confirmed · none

9. An Experiment Mode comparison is interpretable only when its candidate inputs share the relevant scientific state, or when the differing state is explicitly the strategy being tested -- the same same-input-branching fairness principle Experiment Mode itself is built around, applied specifically to the input data's own state.

   source-confirmed · none