feat: Milliquas bin v5 + ΛCDM distances + observable-universe horizon shell#186
Open
rulkens wants to merge 23 commits into
Open
feat: Milliquas bin v5 + ΛCDM distances + observable-universe horizon shell#186rulkens wants to merge 23 commits into
rulkens wants to merge 23 commits into
Conversation
Factor the J<RA><Dec> coord-suffix builder out of iauName so any consumer that needs to glue a non-survey prefix onto an IAU coord string can share one byte-identical emitter. Pins the historical output via a regression test against iauName. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The case comment named parentSurveyByte / OTHER sentinel / galaxyInfoBuilder machinery that doesn't yet exist at this commit. Trim to a forward-neutral note about the upstream catalogue convention so this commit stands on its own under bisect. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Two new per-record uint8 typed-array fields, ready to be threaded through the binary format (Task 3) and the build pipeline (Task 4). Standalone type-only commit so the format bump that follows is a focused diff. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…y bytes Two new per-record uint8 slots at offset 52/53, carved from the v4 tail padding without changing the 64-byte stride. v4 files are now rejected with the documented "regenerate via npm run build-tiers" error. classByte carries the per-source class enum (Milliquas class letter today); parentSurveyByte carries the Milliquas parent-survey enum so the InfoCard can reconstruct the historical "<PARENT> J<RA><Dec>" display name without the JSON sidecar. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
makeCloud / makeOrientCloud and the four inline GalaxyCatalog literals now
carry zero-filled classByte + parentSurveyByte so the v5 type is satisfied.
The cross-version rejection list extends from {v1, v2, v3} to {v1, v2, v3,
v4}, and the regenerate-message assertion follows the docstring's
"build-tiers" wording.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Replaces "v3-specific" / "v4-specific" / "v4 (orientation round-trip)" describe titles with timeless, behaviour-focused names (orientation round-trip; diameterKpc + stride + rejection). The tests themselves already work against the current v5 encoder, so the rename is purely documentation hygiene. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…sToCloud Defaults are 0 across the board; the next task wires up each parser to populate the new fields (zero for everyone except Milliquas, which carries the AGN class + parent-survey prefix). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…n-Milliquas sources Adds the two new bytes (zero-defaulted) at every non-Milliquas GalaxyCatalog construction site: the three ParsedRecord-building parsers (sdssCsv, twoMrs, glade), the Famous compiler (buildFamous constructs GalaxyCatalog directly, not via ParsedRecord), the runtime synthetic-cloud generator, and the Worker-transfer clone. Future per-survey class signals (e.g. GLADE morphology) can override these defaults locally without touching the pipeline. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…kups sourceClassLabel(source, byte) maps the bin's classByte slot to a human string (Milliquas Q/A/B/K/N/S today; null for every other source). milliquasParentSurveyPrefix(byte) maps the parentSurveyByte to a display string (SDSS/2MASX/GAIA/…) for the InfoCard's display-name reconstruction. Constants are reused by the Milliquas parser. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Drop the parallel names/classes sidecar return shape in favour of per-record uint8 bytes that round-trip through the v5 .bin format. The InfoCard reconstructs "<PARENT> J<RA><Dec>" at hover time from the parentSurveyByte + iauRaDecSuffix(ra, dec), with the IAU "MQ J<RA><Dec>" fallback for literature designations. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Per-record classByte + parentSurveyByte now ride on the .bin (Task 6), so the build no longer emits the parallel JSON sidecar. subsampleByAbsMag's value-returning variant preserves the bytes through the brightest-N subset; the indices-returning variant becomes dead code and its import goes too. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Reconciles buildAllBins.milliquas smoke test with the v5 byte fields landing in Task 6/7: the parser no longer returns `names`, so the "sidecar invariant" assertion is replaced with a per-record class + parent-survey byte round-trip through recordsToCloud and encode/decode. Other pipeline coverage (NaN-able photometry, morphology fallbacks, diameterKpc default) is preserved. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The v5 .bin format carries the Milliquas display-name ingredients (class byte + parent-survey byte) per record, so the JSON sidecar + fetcher + slot + EngineSourceState fields + companion-asset registry row are all dead weight. Drop them, leaving the buildGalaxyInfo signature one parameter shorter — that parameter is replaced by the byte-driven reconstruction in the next task. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Drop the milliquasNames runtime parameter; instead, read the v5 .bin's per-record classByte (for the InfoCard's AGN class row) and parentSurveyByte (for "<PARENT> J<RA><Dec>" display-name reconstruction). GalaxyInfo grows an optional `agnClass` field for the AGN class row. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…veat GalaxyInfo's agnClass JSDoc now describes the absent-row consumer contract as an expectation rather than asserting current InfoCard code; the consumer gate (`info.agnClass && (…)`) is the intended pattern, not a present-day claim. galaxyInfoBuilder's milliquasDisplayName comment now notes that the J-suffix is recomputed from the row's stored coords and may differ in least-significant digits from the upstream catalogue's Name column, heading off "why doesn't this match Milliquas v8 byte-for- byte?" investigations. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Replace the milliquasNames-arg assertions with classByte + parentSurveyByte assertions that exercise the full v5 byte vocabulary (six class letters, seven parent-survey prefixes, the OTHER fallback, and the non-Milliquas no-op). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…xtures Sweep across every test that constructs a GalaxyCatalog or ParsedRecord literal directly, defaulting the two new v5 bytes to zero (or to the distinguishing fill values used by the cloneGalaxyCatalogForTransfer test). Also extends the transfer-list count assertion from 10 to 12, flips the galaxyCatalogFetcher mock header from v4 to v5, and adds a classByte/parentSurveyByte default to the subsampleByAbsMag test rec factory. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The v5 .bin format carries the Milliquas class + parent-survey metadata per record, so the JSON sidecar is dead. Drop it from the ALLOW filter and add an idempotent delete step that removes any leftover small/medium/large names.json keys from R2 on every sync. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
famous.bin is a 75-row curated catalogue checked into git (the bulk survey bins are gitignored). The v4→v5 format bump invalidates the old bytes — re-encoded via `npm run build-famous` against the freshly rebuilt 2mrs.bin and glade-large.bin. Same 4816-byte stride; only the version byte at offset 4 and the new classByte/parentSurveyByte slots (zero-filled for Famous) changed. famous_xrefs.json picks up a few cross-match shifts from the new GLADE row order. Also commits the implementation plan doc that drove this branch. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Deploying with
|
| Status | Name | Latest Commit | Preview URL | Updated (UTC) |
|---|---|---|---|---|
| ✅ Deployment successful! View logs |
skymap | f09ea31 | Commit Preview URL Branch Preview URL |
May 21 2026, 02:56 PM |
…rvable volume Replace the linear-Hubble redshift→distance approximation with a flat-ΛCDM Simpson integral (and a LUT-backed bisection inverse), gated behind USE_LCDM_DISTANCES so the linear mode stays one edit away. Positions past z≈0.3 were off by tens of percent; the Milliquas quasar tail now sits at its correct ~9 Gpc instead of ~30 Gpc. Split cartesianToRaDec out of cartesianToRaDecZ so the per-galaxy bake hot paths skip the (now non-trivial) redshift inverse they always discarded. Bump FAR_CLIP_MPC and the zoom envelope so the camera can reach and render the full volume. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add a translucent shell marking the comoving particle-horizon radius (~14.3 Gpc), drawn as an analytic ray-marched sphere: a fullscreen quad plus per-pixel ray-sphere intersection in the fragment, with Fresnel-rim shading. Rays are built from the camera basis + FOV (no view-projection inverse, which is unstable at the shell's huge near:far ratio) and the intersection runs in Gpc units so the quadratic stays inside fp32's precise range — eliminating the per-triangle dropouts and size jitter a rasterised mesh suffered at these scales. Wire it as an HDR pass after volume-upsample, with its own GPU-timing slot (20/21). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The observable-universe shell now ramps in as the camera retreats instead of drawing every frame. The fade band is a fraction of the shell radius (5%..40%), the mirror image of the Milky-Way impostor's near-volume fade, and both share a new smoothstep primitive. Gating lives in the pass's enabled() so a faded-out shell skips its full-screen ray-march pass and its GPU-timings slot entirely. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…n-class-byte # Conflicts: # src/@types/engine/state/EngineAssetSlots.d.ts # src/@types/engine/subsystems/CreateSelectionSubsystemInput.d.ts # src/data/colourIndex.ts # src/data/galaxyCatalogFormat.ts # src/services/engine/camera/resolveFocusTarget.ts # src/services/engine/engine.ts # src/services/engine/helpers/galaxyInfoBuilder.ts # src/services/engine/subsystems/selectionSubsystem.ts # src/utils/math/iauName.ts # tests/@types/engineState.test.ts # tests/services/engine/subsystems/selectionSubsystem.test.ts
rulkens
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Bundles three pieces of work on top of the Milliquas bin-v5 change. Merged with
main(#182–#185 SOURCE_REGISTRY + unified selection refactors).1. Milliquas bin v5 (AGN class + parent survey)
classByte+parentSurveyByte) carved from the existing tail padding; the 64-byte stride is unchanged."<PARENT> J<RA><Dec>"Milliquas display name in the InfoCard purely from the bin via a sharediauRaDecSuffix(ra, dec)helper; addagnClass?toGalaxyInfo._names.jsonsidecar pipeline end-to-end (fetcher, slot, payload type, engine wiring, R2 ALLOW entry).2. ΛCDM comoving distances
USE_LCDM_DISTANCESflag for quick A/B toggling against the old linear variant.FAR_CLIP_MPC6000→50000,MAX_DISTANCE_MPC5000→30000) so the full observable volume — high-z quasars + horizon shell — stays in frustum.cartesianToRaDecto avoid the inverse-LUT cost per galaxy.3. Observable-universe horizon shell
smoothstepprimitive. Gated in the pass'senabled()so a faded-out shell skips its full-screen pass and GPU-timings slot.File-format impact
encodeGalaxyCatalogwrites version=5. Existing v4.binfiles fail loudly withunsupported version: 4 — please regenerate via "npm run build-tiers". Do NOT merge before runningnpm run sync-r2against a freshly-rebuiltpublic/data/(the R2 sync also deletes the orphanedmilliquas-*_names.jsonkeys).Regenerated data sizes (v5 + ΛCDM positions)
sdss-medium.binsdss-large.binglade-small.binglade-medium.binglade-large.binmilliquas-medium.binmilliquas-large.bin2mrs.binfamous.bin(Dropping the
milliquas-*_names.jsonsidecars saves ~34 MB of per-session network transfer.)Test plan
npm run typecheckpasses (both src and tools tsconfigs).npm testpasses — 1697/1697 across 270 files.npm run build-tiersregenerates everypublic/data/*.binand emits no_names.jsonfiles.main(refactor(sources): SOURCE_REGISTRY as the single registry of every data source #182–refactor(volumes): close VolumeFieldId set + unify volumes under SOURCE_REGISTRY #185) resolved; SOURCE_REGISTRY + unified galaxy/POI selection adopted, this PR's deltas (v5 bytes, ΛCDM, horizon shell) ported on top.npm run sync-r2uploads new v5 bins and deletes themilliquas-*_names.jsonkeys. Deferred — run before merging."SDSS J…"(not the"MQ J…"IAU fallback); known class letter Q → AGN class row shows"Quasar".🤖 Generated with Claude Code