Star noise at 5k ticks: the stellar disk stops rotating by t=2500, the gas does not #66

New issue

Closed

opened 2026-08-07 06:46:58 +00:00 by coilyco-ops · 7 comments

coilyco-ops commented 2026-08-07 06:46:58 +00:00

Member

Copy link

Measured before tuning anything. ward exec debug-sim 5000 500 3 12345 2 — irregular=>spiral, size 500, seeds 12345/12346/12347, checkpoints to 5000.

The finding

svt (stellar mean tangential velocity) collapses to zero and stays there. Gas vt does not.

t	svt 12345	svt 12346	svt 12347	gas vt (12345)
250	+1.585	+1.549	+1.703	+1.645
900	+0.914	+0.912	+0.800	+1.290
1500	+1.251	+0.693	+0.593	+1.341
2000	+0.391	+0.172	+0.209	+1.322
2500	+0.251	+0.295	-0.009	+1.197
4000	-0.142	+0.255	+0.066	+1.171
5000	+0.114	-0.017	+0.063	+1.051

All three seeds, same shape: stellar rotation decays from ~+1.6 to ~0 between t=900 and t=2500, then stays at zero. A stellar population with mean tangential velocity of zero is not a disk. It is a pressure-supported random-motion cloud, and with 31k-52k resolved stars drawn on top of everything else, it is the dominant visual layer. That is the uniform mess.

The gas is fine. Gas vt holds +1.05 to +1.34 at t=5000, and gas spiral coherence at 5000 is 0.53 / 0.31 / 0.34 across the three seeds - as good as or better than at t=1000. The structure is still there. It is being buried under a non-rotating star field.

This reframes the problem: it is not "too many stars" and not "structure decays." It is one population losing its rotational support while the other keeps it.

Mechanism

Two asymmetries between how gas and stars are integrated, both in the same direction.

1. Gas has a dissipative relaxation term. Stars have nothing equivalent. Gas velocities decay toward the circular flow every tick via flow_drag (flow_decay at galaxy.rs:4001, per-scenario 0.015-0.035). That is an active energy sink continuously re-aiming gas onto circular orbits, which is why gas stays coherent indefinitely. Stars are collisionless by design and only get STAR_HALO_DRAG, and only while out in the halo band. Inside the disk nothing removes their velocity error, so any error integrates and the population randomizes.

2. Stars inherit gas velocities but read a much weaker field. The coarse field is built with v_flat_star = p.v_flat * 0.5 and then everything multiplied by STAR_FIELD_SCALE = 0.25 (galaxy.rs:2075-2101). So a star feels roughly a quarter of the centripetal support the gas feels - but it is born carrying its natal gas parcel's velocity, which is calibrated to the full gas rotation curve. Born too fast for its own potential, it climbs onto a large radial excursion, and the population phase-mixes within a few orbital times. The observed t=900 to t=2500 decay is about the right timescale for that.

The halved pace is deliberate ("so star orbits run at half the gas pace"), and that is fine as art direction. The problem is that the birth velocity was not halved to match it.

This is not a new discovery so much as an unfinished one. BIRTH_GAS_VEL_CAP already exists, and its comment says inheriting the raw gas velocity "launches newborns straight into the halo and empties the visible disk." Same failure mode, met once before and capped rather than resolved. The cap bounds the worst case; it does not make the inherited velocity consistent with the potential the star will actually feel.

Why this is probably also #65

#65 records a 2x stellar luminance bump pooling just outside disk_r at maturity. Stars born with too much energy for their potential are exactly the stars that climb out of the disk and pool in the halo band. One mechanism, two symptoms. Worth fixing once rather than twice.

Proposal

In order, measuring svt at t=2000/5000 after each step rather than stacking changes:

1. Scale birth velocity to the field stars actually read. Multiply the inherited orbital component by the same factor the star field is scaled by, so a newborn starts on a circular orbit in its potential rather than the gas's. This is the one-line hypothesis and should be tested first - if svt holds above ~1.0 at t=5000, everything below is unnecessary.
2. If that is not enough, add a weak orbital relaxation for disk stars. Far gentler than flow_drag, applied only inside disk_r, enough to bleed accumulated error without making the population collisional. This trades some physical purity for the thing the sim is actually for, which is looking like a galaxy.
3. Only then look at FIELD_RES = 64. At size 500 each field cell spans ~7.8 sim cells, so stars cannot resolve arm structure at all. This limits how much spiral detail the stellar layer can ever show, but it is not what is killing rotation, so it should not be touched until 1 and 2 are settled.

Do not start by cutting the birth rate. The star count is a symptom of nothing - 50k rotating stars would look like a galaxy.

Measured before tuning anything. `ward exec debug-sim 5000 500 3 12345 2` — irregular=>spiral, size 500, seeds 12345/12346/12347, checkpoints to 5000. ## The finding `svt` (stellar mean tangential velocity) collapses to zero and stays there. Gas `vt` does not. | t | svt 12345 | svt 12346 | svt 12347 | gas vt (12345) | |---|---|---|---|---| | 250 | +1.585 | +1.549 | +1.703 | +1.645 | | 900 | +0.914 | +0.912 | +0.800 | +1.290 | | 1500 | +1.251 | +0.693 | +0.593 | +1.341 | | 2000 | +0.391 | +0.172 | +0.209 | +1.322 | | 2500 | +0.251 | +0.295 | -0.009 | +1.197 | | 4000 | -0.142 | +0.255 | +0.066 | +1.171 | | 5000 | +0.114 | -0.017 | +0.063 | +1.051 | All three seeds, same shape: stellar rotation decays from ~+1.6 to ~0 between t=900 and t=2500, then stays at zero. A stellar population with mean tangential velocity of zero is not a disk. It is a pressure-supported random-motion cloud, and with 31k-52k resolved stars drawn on top of everything else, it is the dominant visual layer. **That is the uniform mess.** The gas is fine. Gas `vt` holds +1.05 to +1.34 at t=5000, and gas spiral coherence at 5000 is 0.53 / 0.31 / 0.34 across the three seeds - as good as or better than at t=1000. The structure is still there. It is being buried under a non-rotating star field. This reframes the problem: it is not "too many stars" and not "structure decays." It is one population losing its rotational support while the other keeps it. ## Mechanism Two asymmetries between how gas and stars are integrated, both in the same direction. **1. Gas has a dissipative relaxation term. Stars have nothing equivalent.** Gas velocities decay toward the circular flow every tick via `flow_drag` (`flow_decay` at galaxy.rs:4001, per-scenario 0.015-0.035). That is an active energy sink continuously re-aiming gas onto circular orbits, which is why gas stays coherent indefinitely. Stars are collisionless by design and only get `STAR_HALO_DRAG`, and only while out in the halo band. Inside the disk nothing removes their velocity error, so any error integrates and the population randomizes. **2. Stars inherit gas velocities but read a much weaker field.** The coarse field is built with `v_flat_star = p.v_flat * 0.5` and then everything multiplied by `STAR_FIELD_SCALE = 0.25` (galaxy.rs:2075-2101). So a star feels roughly a quarter of the centripetal support the gas feels - but it is born carrying its natal gas parcel's velocity, which is calibrated to the *full* gas rotation curve. Born too fast for its own potential, it climbs onto a large radial excursion, and the population phase-mixes within a few orbital times. The observed t=900 to t=2500 decay is about the right timescale for that. The halved pace is deliberate ("so star orbits run at half the gas pace"), and that is fine as art direction. The problem is that the birth velocity was not halved to match it. This is not a new discovery so much as an unfinished one. `BIRTH_GAS_VEL_CAP` already exists, and its comment says inheriting the raw gas velocity "launches newborns straight into the halo and empties the visible disk." Same failure mode, met once before and capped rather than resolved. The cap bounds the worst case; it does not make the inherited velocity consistent with the potential the star will actually feel. ## Why this is probably also #65 #65 records a 2x stellar luminance bump pooling just outside `disk_r` at maturity. Stars born with too much energy for their potential are exactly the stars that climb out of the disk and pool in the halo band. One mechanism, two symptoms. Worth fixing once rather than twice. ## Proposal In order, measuring `svt` at t=2000/5000 after each step rather than stacking changes: 1. **Scale birth velocity to the field stars actually read.** Multiply the inherited orbital component by the same factor the star field is scaled by, so a newborn starts on a circular orbit in *its* potential rather than the gas's. This is the one-line hypothesis and should be tested first - if `svt` holds above ~1.0 at t=5000, everything below is unnecessary. 2. **If that is not enough, add a weak orbital relaxation for disk stars.** Far gentler than `flow_drag`, applied only inside `disk_r`, enough to bleed accumulated error without making the population collisional. This trades some physical purity for the thing the sim is actually for, which is looking like a galaxy. 3. **Only then look at `FIELD_RES = 64`.** At size 500 each field cell spans ~7.8 sim cells, so stars cannot resolve arm structure at all. This limits how much spiral detail the stellar layer can ever show, but it is not what is killing rotation, so it should not be touched until 1 and 2 are settled. Do not start by cutting the birth rate. The star count is a symptom of nothing - 50k rotating stars would look like a galaxy.

coilyco-ops commented 2026-08-07 15:03:48 +00:00

Author

Member

Copy link

Measured it. The hypothesis was right in direction and badly understated in magnitude, and the mechanism is not the one I guessed.

Added two probes to debug-sim:

• scirc — mean disk-star tangential speed over the circular speed of the field those stars read. 1.0 is a balanced disk.
• bcirc — the same ratio a newborn is handed at 0.5 disk_r. Separates "born wrong" from "drifted wrong".

irregular=>spiral, size 500, seeds 12345 / 12346:

t	scirc 12345	bcirc 12345	scirc 12346	bcirc 12346
250	1.31	2.45	1.41	2.03
500	1.28	1.97	1.33	2.33
900	0.89	2.46	0.97	2.71
1200	0.78	2.34	0.84	2.51
1500	0.55	2.16	0.37	2.77
2000	0.29	2.11	0.13	2.63
2500	0.28	1.95	0.18	2.84

bcirc sits at 2.0-3.2 for the entire run. Newborn stars are handed two to three times the circular speed of the potential they have to orbit in. Past about 1.41x circular a star is above escape velocity for its own potential, so the disk has been launching the large majority of the stars it forms, continuously, from t=250 onward.

scirc then tells the rest of the story: the surviving disk population starts at 1.3x circular (already too fast) and decays to 0.28x by t=2500. The fast tail leaves. What stays is the low-angular-momentum remainder, which is sub-circular and falls inward. Neither group is on a disk orbit, so there is no rotating disk left — just a pressure-supported cloud of 40k points drawn over the still-perfectly-good gas structure.

Cause

Not the field scaling, and not the missing relaxation term. Both of those are fine. It is this line:

let target_tangential =
    (gas_tangential.max(0.0) + smooth_support).min(Galaxy::ASSOCIATION_ORBIT_SPEED_CAP);

ASSOCIATION_ORBIT_SPEED_CAP = 2.1 is an absolute speed, and it binds on essentially every birth. Circular speed in the star field at 0.5 disk_r is around 0.8, so a 2.1 ceiling is ~2.6x circular. A single absolute number cannot be right across radii and scenarios where circular speed varies by a factor of several — it is far too generous in the disk and would be too tight elsewhere.

The compensation stack above it is also additive: gas_tangential + 1.08 * v_circ + background_support. Since association_circular_speed already returns the true circular speed of the quarter-strength field stars actually read, adding a background halo/BH term on top double-counts the support. But the cap is what makes it extreme.

Fix

Clamp to a multiple of the local circular speed instead of to an absolute speed:

let target_tangential = (gas_tangential.max(0.0) + smooth_support)
    .min(Galaxy::ASSOCIATION_ORBIT_SPEED_CAP)
    .min(circular * Galaxy::BIRTH_ORBIT_RATIO_CAP);   // 1.06

1.06 keeps newborns just above circular — enough eccentricity that the disk looks alive rather than on rails, and far below the ~1.41 escape ratio. The absolute cap stays as a backstop for degenerate radii where the sampled field is unreliable.

Re-measuring now. This also predicts #65 goes away, see there.

Measured it. The hypothesis was right in direction and badly understated in magnitude, and the mechanism is not the one I guessed. Added two probes to `debug-sim`: - `scirc` — mean disk-star tangential speed over the circular speed of the field those stars read. 1.0 is a balanced disk. - `bcirc` — the same ratio a newborn is handed at 0.5 `disk_r`. Separates "born wrong" from "drifted wrong". irregular=>spiral, size 500, seeds 12345 / 12346: | t | scirc 12345 | bcirc 12345 | scirc 12346 | bcirc 12346 | |---|---|---|---|---| | 250 | 1.31 | 2.45 | 1.41 | 2.03 | | 500 | 1.28 | 1.97 | 1.33 | 2.33 | | 900 | 0.89 | 2.46 | 0.97 | 2.71 | | 1200 | 0.78 | 2.34 | 0.84 | 2.51 | | 1500 | 0.55 | 2.16 | 0.37 | 2.77 | | 2000 | 0.29 | 2.11 | 0.13 | 2.63 | | 2500 | 0.28 | 1.95 | 0.18 | 2.84 | **`bcirc` sits at 2.0-3.2 for the entire run.** Newborn stars are handed two to three times the circular speed of the potential they have to orbit in. Past about 1.41x circular a star is above escape velocity for its own potential, so the disk has been launching the large majority of the stars it forms, continuously, from t=250 onward. `scirc` then tells the rest of the story: the surviving disk population starts at 1.3x circular (already too fast) and decays to 0.28x by t=2500. The fast tail leaves. What stays is the low-angular-momentum remainder, which is sub-circular and falls inward. Neither group is on a disk orbit, so there is no rotating disk left — just a pressure-supported cloud of 40k points drawn over the still-perfectly-good gas structure. ## Cause Not the field scaling, and not the missing relaxation term. Both of those are fine. It is this line: ```rust let target_tangential = (gas_tangential.max(0.0) + smooth_support).min(Galaxy::ASSOCIATION_ORBIT_SPEED_CAP); ``` `ASSOCIATION_ORBIT_SPEED_CAP = 2.1` is an **absolute speed**, and it binds on essentially every birth. Circular speed in the star field at 0.5 `disk_r` is around 0.8, so a 2.1 ceiling is ~2.6x circular. A single absolute number cannot be right across radii and scenarios where circular speed varies by a factor of several — it is far too generous in the disk and would be too tight elsewhere. The compensation stack above it is also additive: `gas_tangential + 1.08 * v_circ + background_support`. Since `association_circular_speed` already returns the true circular speed **of the quarter-strength field stars actually read**, adding a background halo/BH term on top double-counts the support. But the cap is what makes it extreme. ## Fix Clamp to a multiple of the local circular speed instead of to an absolute speed: ```rust let target_tangential = (gas_tangential.max(0.0) + smooth_support) .min(Galaxy::ASSOCIATION_ORBIT_SPEED_CAP) .min(circular * Galaxy::BIRTH_ORBIT_RATIO_CAP); // 1.06 ``` 1.06 keeps newborns just above circular — enough eccentricity that the disk looks alive rather than on rails, and far below the ~1.41 escape ratio. The absolute cap stays as a backstop for degenerate radii where the sampled field is unreliable. Re-measuring now. This also predicts #65 goes away, see there.

coilyco-ops referenced this issue 2026-08-07 15:04:22 +00:00

Boundary ridge: gas confinement is load-bearing, and stars have their own edge #65

coilyco-ops commented 2026-08-07 15:09:45 +00:00

Author

Member

Copy link

Built the ratio cap, measured it, and reverted it. It works exactly as designed and it is still not landable. Two findings, one of them the important one.

The fix does what it claims

BIRTH_ORBIT_RATIO_CAP = 1.06, irregular=>spiral, seeds 12345 / 12346:

metric	before	after
bcirc (birth ratio)	2.0 – 3.2	1.06 exactly
mixed @ t=2500	3200 / 9009	114 / 149

Stars draining into the halo reservoir dropped by roughly 70x. That confirms the #65 mechanism outright: stars were being launched past escape, piling at the phase-mixing radius, and retiring. Stop launching them and the drain nearly stops.

But rotation still decays, so #66 has a second cause

scirc from a now-correct start:

t	250	900	1500	2000	2500
12345	1.16	0.36	0.33	0.01	0.08
12346	0.82	0.55	0.38	0.27	0.28

Born at exactly circular, the population still loses its rotational support within ~1000 ticks. So birth velocity was a real bug, but it is not what is killing the disk. Something is actively heating stars that are already sitting correctly.

Leading candidate is now numerical rather than physical: FIELD_RES = 64 against a 500-cell grid means each field cell spans ~7.8 sim cells, and stars integrate by sampling that coarse, clump-dominated field. Discrete-field scattering converts ordered rotation into random motion on about this timescale. This was item 3 on the original list and the data promotes it to item 1.

The important finding: the elliptical scenario is built on this bug

ward exec test-rust with the cap applied, 4 failures:

test_irregular_elliptical_relaxes_into_a_resolved_pressure_supported_spheroid
  stellar concentration ranged from (0.946, t=963) to (0.963, t=907)
  required: 0.45 <= c <= 0.85

test_spiral_remains_resolved_star_forming_and_coherent_for_100_ticks
  tick 1056 coherence was 0.2548   (required >= 0.3)

test_golden_mass_field_per_scenario
  only the 4th hash differs - irregular-elliptical. The other three
  scenarios are byte-identical.

test_ring_remains_hollow_resolved_and_star_forming_for_100_ticks

The elliptical goes to 0.95 stellar concentration — nearly every star inside 0.3 disk_r. Its "relaxation into a pressure-supported spheroid" is currently being produced by the over-fast births: stars launched at 2-3x circular are what populate the spheroid out to large radii. Give them correct orbits and they stay where they were born, which in that scenario is the middle.

So this is the same shape as the gas confinement ridge in #65: a bug that a scenario's visual identity depends on. Two for two. Fixing the birth velocity is not a constant change, it is a redesign of how the elliptical builds its spheroid — most likely giving it explicit velocity dispersion at birth, which is what a real pressure-supported system has, instead of borrowing the side effect of a bad cap.

Landed instead

Only the diagnostics, which are inert and pass all 83 tests:

• Galaxy::star_circular_ratio() and Galaxy::birth_circular_ratio(frac)
• scirc / bcirc columns in debug-sim
• a comment at the birth site recording the measured 2.0-3.2x and pointing here, so the next person does not "fix" it in isolation and get four red tests with no context

Suggested order now

1. Diagnose the residual heating (FIELD_RES first). If stars will not hold a disk orbit even when born on one, the birth fix cannot be evaluated properly anyway.
2. Give the elliptical explicit birth dispersion so its spheroid stops depending on the cap.
3. Only then apply the ratio cap, with the golden field regenerated in the same commit.

Built the ratio cap, measured it, and **reverted it**. It works exactly as designed and it is still not landable. Two findings, one of them the important one. ## The fix does what it claims `BIRTH_ORBIT_RATIO_CAP = 1.06`, irregular=>spiral, seeds 12345 / 12346: | metric | before | after | |---|---|---| | `bcirc` (birth ratio) | 2.0 – 3.2 | 1.06 exactly | | `mixed` @ t=2500 | 3200 / 9009 | 114 / 149 | Stars draining into the halo reservoir dropped by roughly **70x**. That confirms the #65 mechanism outright: stars were being launched past escape, piling at the phase-mixing radius, and retiring. Stop launching them and the drain nearly stops. ## But rotation still decays, so #66 has a second cause `scirc` from a now-correct start: | t | 250 | 900 | 1500 | 2000 | 2500 | |---|---|---|---|---|---| | 12345 | 1.16 | 0.36 | 0.33 | 0.01 | 0.08 | | 12346 | 0.82 | 0.55 | 0.38 | 0.27 | 0.28 | Born at exactly circular, the population still loses its rotational support within ~1000 ticks. So birth velocity was a real bug, but it is not what is killing the disk. Something is actively heating stars that are already sitting correctly. Leading candidate is now numerical rather than physical: `FIELD_RES = 64` against a 500-cell grid means each field cell spans ~7.8 sim cells, and stars integrate by sampling that coarse, clump-dominated field. Discrete-field scattering converts ordered rotation into random motion on about this timescale. This was item 3 on the original list and the data promotes it to item 1. ## The important finding: the elliptical scenario is built on this bug `ward exec test-rust` with the cap applied, 4 failures: ``` test_irregular_elliptical_relaxes_into_a_resolved_pressure_supported_spheroid stellar concentration ranged from (0.946, t=963) to (0.963, t=907) required: 0.45 <= c <= 0.85 test_spiral_remains_resolved_star_forming_and_coherent_for_100_ticks tick 1056 coherence was 0.2548 (required >= 0.3) test_golden_mass_field_per_scenario only the 4th hash differs - irregular-elliptical. The other three scenarios are byte-identical. test_ring_remains_hollow_resolved_and_star_forming_for_100_ticks ``` The elliptical goes to **0.95 stellar concentration** — nearly every star inside 0.3 `disk_r`. Its "relaxation into a pressure-supported spheroid" is currently being produced *by* the over-fast births: stars launched at 2-3x circular are what populate the spheroid out to large radii. Give them correct orbits and they stay where they were born, which in that scenario is the middle. So this is the same shape as the gas confinement ridge in #65: **a bug that a scenario's visual identity depends on.** Two for two. Fixing the birth velocity is not a constant change, it is a redesign of how the elliptical builds its spheroid — most likely giving it explicit velocity dispersion at birth, which is what a real pressure-supported system has, instead of borrowing the side effect of a bad cap. ## Landed instead Only the diagnostics, which are inert and pass all 83 tests: - `Galaxy::star_circular_ratio()` and `Galaxy::birth_circular_ratio(frac)` - `scirc` / `bcirc` columns in `debug-sim` - a comment at the birth site recording the measured 2.0-3.2x and pointing here, so the next person does not "fix" it in isolation and get four red tests with no context ## Suggested order now 1. Diagnose the residual heating (`FIELD_RES` first). If stars will not hold a disk orbit even when born on one, the birth fix cannot be evaluated properly anyway. 2. Give the elliptical explicit birth dispersion so its spheroid stops depending on the cap. 3. Only then apply the ratio cap, with the golden field regenerated in the same commit.

coilyco-ops referenced this issue from a commit 2026-08-07 15:10:23 +00:00

feat(debug): measure whether stars are on orbits their potential holds

coilyco-ops commented 2026-08-07 15:30:48 +00:00

Author

Member

Copy link

Correction to my previous comment. The intervention test does not support the causal story I gave, and one of my two probes is not measuring what I said it measures. Retracting both claims before anyone builds on them.

Full intervention matrix, scirc at t=2500, seeds 12345 / 12346

configuration	12345	12346
baseline (RES 64, no cap)	0.28	0.18
birth ratio cap only	0.08	0.28
FIELD_RES = 192 only	0.51	0.35
both together	0.09	0.29

The birth cap makes the metric worse or flat, in both field resolutions. It does not restore stellar rotation. My claim that over-fast births cause the rotation collapse is not supported — capping births to circular does not fix it, and combining the two fixes is no better than doing nothing.

FIELD_RES = 192 alone is the only configuration that improves anything, roughly doubling retained rotational support, at 9x the field-build cost.

Why the earlier conclusion was wrong

Two separate errors.

1. bcirc stopped tracking behaviour. When I reverted the cap I reverted it in the probe too, but then re-applied it at the birth site for the combined run. So bcirc in that last sweep reports the uncapped target ratio (1.76-2.84) while births were actually capped. The probe and the code disagreed and I read the probe. That is my bug, not the sim's.

2. scirc probably does not measure what I claimed. It is mean tangential speed over local circular speed across disk stars. That does not distinguish a star on a circular disk orbit from a star on an eccentric orbit passing through pericenter, where tangential speed is high by construction. So a population of fast eccentric interlopers scores well. That plausibly explains the whole inverted result: uncapped births produce eccentric orbits that inflate scirc while looking exactly like the mush being complained about.

If that reading is right, scirc is close to useless for this question and the t=250-to-2500 decay I built the whole diagnosis on may be measuring a change in orbit eccentricity rather than loss of rotation.

What still stands

Only the arithmetic, which is not inference: ASSOCIATION_ORBIT_SPEED_CAP is an absolute speed of 2.1 that binds on nearly every birth, against a local circular speed near 0.8 at 0.5 disk_r. Newborns really are handed 2-3x circular. That remains a defensible thing to call wrong on its own terms. What is now unsupported is that it causes the visual mush.

The elliptical finding also stands and is independent of all this: capping births to circular collapses that scenario to 0.95 stellar concentration, so its spheroid does depend on the current birth speeds.

What the next person should do

Do not tune anything against scirc until it is replaced. It needs a metric that separates ordered rotation from random motion properly — the standard one is v_rot / sigma, mean streaming velocity over velocity dispersion in radial bins. Dispersion is the thing that actually distinguishes a disk from a mush, and it is exactly what I failed to measure.

Concretely:

1. Add v_rot / sigma in radial bins. Confirm on a case with a known answer, e.g. the elliptical should score low and a fresh spiral high.
2. Re-run the intervention matrix above against the new metric. The FIELD_RES result is the one most likely to survive, since it improved a metric that was biased against it.
3. Treat everything in my previous comment as unverified until step 2 reproduces it.

Working tree is reverted; main carries only the inert probes at 916cf8e. Leaving scirc and bcirc in place for now, but the doc comments on both need the caveat above before anyone trusts them.

**Correction to my previous comment.** The intervention test does not support the causal story I gave, and one of my two probes is not measuring what I said it measures. Retracting both claims before anyone builds on them. ## Full intervention matrix, `scirc` at t=2500, seeds 12345 / 12346 | configuration | 12345 | 12346 | |---|---|---| | baseline (RES 64, no cap) | 0.28 | 0.18 | | birth ratio cap only | 0.08 | 0.28 | | `FIELD_RES = 192` only | **0.51** | **0.35** | | both together | 0.09 | 0.29 | The birth cap makes the metric *worse* or flat, in both field resolutions. It does not restore stellar rotation. My claim that over-fast births cause the rotation collapse is **not supported** — capping births to circular does not fix it, and combining the two fixes is no better than doing nothing. `FIELD_RES = 192` alone is the only configuration that improves anything, roughly doubling retained rotational support, at 9x the field-build cost. ## Why the earlier conclusion was wrong Two separate errors. **1. `bcirc` stopped tracking behaviour.** When I reverted the cap I reverted it in the probe too, but then re-applied it at the birth site for the combined run. So `bcirc` in that last sweep reports the *uncapped* target ratio (1.76-2.84) while births were actually capped. The probe and the code disagreed and I read the probe. That is my bug, not the sim's. **2. `scirc` probably does not measure what I claimed.** It is mean tangential speed over local circular speed across disk stars. That does not distinguish a star on a circular disk orbit from a star on an eccentric orbit passing through pericenter, where tangential speed is high by construction. So a population of fast eccentric interlopers scores *well*. That plausibly explains the whole inverted result: uncapped births produce eccentric orbits that inflate `scirc` while looking exactly like the mush being complained about. If that reading is right, `scirc` is close to useless for this question and the t=250-to-2500 decay I built the whole diagnosis on may be measuring a change in orbit *eccentricity* rather than loss of rotation. ## What still stands Only the arithmetic, which is not inference: `ASSOCIATION_ORBIT_SPEED_CAP` is an absolute speed of 2.1 that binds on nearly every birth, against a local circular speed near 0.8 at 0.5 `disk_r`. Newborns really are handed 2-3x circular. That remains a defensible thing to call wrong on its own terms. What is now unsupported is that it causes the visual mush. The elliptical finding also stands and is independent of all this: capping births to circular collapses that scenario to 0.95 stellar concentration, so its spheroid does depend on the current birth speeds. ## What the next person should do Do not tune anything against `scirc` until it is replaced. It needs a metric that separates ordered rotation from random motion properly — the standard one is `v_rot / sigma`, mean streaming velocity over velocity dispersion in radial bins. Dispersion is the thing that actually distinguishes a disk from a mush, and it is exactly what I failed to measure. Concretely: 1. Add `v_rot / sigma` in radial bins. Confirm on a case with a known answer, e.g. the elliptical should score low and a fresh spiral high. 2. Re-run the intervention matrix above against the new metric. The `FIELD_RES` result is the one most likely to survive, since it improved a metric that was biased *against* it. 3. Treat everything in my previous comment as unverified until step 2 reproduces it. Working tree is reverted; `main` carries only the inert probes at 916cf8e. Leaving `scirc` and `bcirc` in place for now, but the doc comments on both need the caveat above before anyone trusts them.

coilyco-ops referenced this issue from a commit 2026-08-07 16:35:05 +00:00

feat(debug): v_rot/sigma, the metric that can tell a disk from a mush

coilyco-ops commented 2026-08-07 16:39:14 +00:00

Author

Member

Copy link

Rebuilt the measurement and re-ran everything. The phenomenon is confirmed and neither candidate fix resolves it.

The metric now has a known-answer test

Galaxy::rotation_dispersion_ratio() — v_rot / sigma in radial bins, landed in 6c180eb. Calibration test: 400 stars on cold circular orbits must read above 1.5; the same positions and speeds with velocity directions scrambled must read below 0.7; the two must differ by 3x.

That test caught a bug in the metric on its first run. A perfectly cold disk returned 0 — zero dispersion tripped the small-sigma guard, so every bin was discarded as "no data" and maximal rotational support scored as none. Fixed by flooring sigma rather than dropping the bin. Without the test it would have produced another believable curve.

Confirmed: this is a real disk-to-spheroid transition

vsig, irregular=>spiral, baseline:

t	250	500	900	1500	2000	2500
12345	2.48	3.71	1.62	1.13	0.57	0.48
12346	2.76	3.14	1.59	0.70	0.38	0.31

Starts firmly rotation-dominated (>1.5), ends firmly dispersion-dominated (<0.7), crossing over around t=1500. Not a rendering artifact, not a star-count problem, not a metric artifact. The stellar disk is being heated into a spheroid.

FIELD_RES = 192 survives the better metric but does not fix it

t	900	1500	2000	2500
12345 baseline	1.62	1.13	0.57	0.48
12345 RES 192	1.19	1.04	0.60	0.76
12346 baseline	1.59	0.70	0.38	0.31
12346 RES 192	1.56	0.70	0.54	0.57

Late-time vsig improves ~1.6-1.8x, consistently across seeds. Early and mid times are neutral or slightly worse. So field coarseness is a real contributor to the late-time heating — my earlier prediction held up — but the disk still crosses into spheroid territory at the same time and still ends well below 0.7. It is not the cause. At 9x the field-build cost for a partial late-time effect, it is not worth landing on its own.

Confound worth noting: RES 192 also nearly doubles the star count (83802 vs 43656 at t=2500), so some of the vsig difference may be improved statistics rather than improved dynamics.

Where this leaves the two candidates

• Birth velocity (2-3x circular): real arithmetic bug, does not cause the transition, and capping it breaks the elliptical scenario. Documented at the birth site.
• Field resolution: real partial contributor to late-time heating, does not cause the transition, expensive.

Neither is the answer. The disk is being heated by something else, strongly enough to convert vsig 3.0 to 0.4 within ~2000 ticks.

Next diagnostic: ablation, not tuning

Do not tune another constant. Find the heat source by turning candidates off one at a time and watching whether the vsig crossover moves:

1. Transient clump scattering. Stars scatter off a lumpy, fast-evolving gas potential. This is a genuine physical effect (giant molecular cloud scattering) but the "clouds" here are enormous relative to the disk, so it is likely far too strong. Test by smoothing the field the stars read, azimuthally or in time, without changing its magnitude.
2. Pattern-speed resonance. Stars deliberately orbit at half the gas pace (STAR_FIELD_SCALE), so the spiral and ring structures they scatter off sweep past at roughly twice their orbital rate. That is a recipe for resonant heating and it is a design consequence, not a bug. Test by matching star orbital pace to the gas for one run.
3. Supernova and association kicks. ShockWave and the association internal potential both perturb nearby stars. Test by disabling each.

Whichever one moves the t=1500 crossover is the cause. Item 2 is my first guess, because the half-pace design puts the whole stellar population permanently off-resonance with the structures it lives in.

Working tree reverted to main; only the metric is landed.

Rebuilt the measurement and re-ran everything. The phenomenon is confirmed and neither candidate fix resolves it. ## The metric now has a known-answer test `Galaxy::rotation_dispersion_ratio()` — `v_rot / sigma` in radial bins, landed in 6c180eb. Calibration test: 400 stars on cold circular orbits must read above 1.5; the same positions and speeds with velocity *directions* scrambled must read below 0.7; the two must differ by 3x. That test caught a bug in the metric on its first run. A perfectly cold disk returned **0** — zero dispersion tripped the small-sigma guard, so every bin was discarded as "no data" and maximal rotational support scored as none. Fixed by flooring sigma rather than dropping the bin. Without the test it would have produced another believable curve. ## Confirmed: this is a real disk-to-spheroid transition `vsig`, irregular=>spiral, baseline: | t | 250 | 500 | 900 | 1500 | 2000 | 2500 | |---|---|---|---|---|---|---| | 12345 | 2.48 | 3.71 | 1.62 | 1.13 | 0.57 | 0.48 | | 12346 | 2.76 | 3.14 | 1.59 | 0.70 | 0.38 | 0.31 | Starts firmly rotation-dominated (>1.5), ends firmly dispersion-dominated (<0.7), crossing over around **t=1500**. Not a rendering artifact, not a star-count problem, not a metric artifact. The stellar disk is being heated into a spheroid. ## `FIELD_RES = 192` survives the better metric but does not fix it | t | 900 | 1500 | 2000 | 2500 | |---|---|---|---|---| | 12345 baseline | 1.62 | 1.13 | 0.57 | 0.48 | | 12345 RES 192 | 1.19 | 1.04 | 0.60 | **0.76** | | 12346 baseline | 1.59 | 0.70 | 0.38 | 0.31 | | 12346 RES 192 | 1.56 | 0.70 | 0.54 | **0.57** | Late-time `vsig` improves ~1.6-1.8x, consistently across seeds. Early and mid times are neutral or slightly worse. So field coarseness is a real contributor to the late-time heating — my earlier prediction held up — but the disk still crosses into spheroid territory at the same time and still ends well below 0.7. **It is not the cause.** At 9x the field-build cost for a partial late-time effect, it is not worth landing on its own. Confound worth noting: RES 192 also nearly doubles the star count (83802 vs 43656 at t=2500), so some of the `vsig` difference may be improved statistics rather than improved dynamics. ## Where this leaves the two candidates - **Birth velocity (2-3x circular)**: real arithmetic bug, does not cause the transition, and capping it breaks the elliptical scenario. Documented at the birth site. - **Field resolution**: real partial contributor to late-time heating, does not cause the transition, expensive. Neither is the answer. The disk is being heated by something else, strongly enough to convert `vsig` 3.0 to 0.4 within ~2000 ticks. ## Next diagnostic: ablation, not tuning Do not tune another constant. Find the heat source by turning candidates off one at a time and watching whether the `vsig` crossover moves: 1. **Transient clump scattering.** Stars scatter off a lumpy, fast-evolving gas potential. This is a genuine physical effect (giant molecular cloud scattering) but the "clouds" here are enormous relative to the disk, so it is likely far too strong. Test by smoothing the field the stars read, azimuthally or in time, without changing its magnitude. 2. **Pattern-speed resonance.** Stars deliberately orbit at half the gas pace (`STAR_FIELD_SCALE`), so the spiral and ring structures they scatter off sweep past at roughly twice their orbital rate. That is a recipe for resonant heating and it is a design consequence, not a bug. Test by matching star orbital pace to the gas for one run. 3. **Supernova and association kicks.** `ShockWave` and the association internal potential both perturb nearby stars. Test by disabling each. Whichever one moves the t=1500 crossover is the cause. Item 2 is my first guess, because the half-pace design puts the whole stellar population permanently off-resonance with the structures it lives in. Working tree reverted to `main`; only the metric is landed.

coilyco-ops commented 2026-08-09 03:36:10 +00:00

Author

Member

Copy link

Ran the ablation matrix. Found it. There are two heat sources, not one, and each is on its own sufficient to destroy the disk - which is exactly why one-at-a-time ablation came back empty.

First, the harness

Switches now live in the kernel (src/rust/ablation.rs), read from the environment, all off by default. debug-sim prints the resolved configuration as its first line. This is a direct response to the retraction above: a switch the sim itself reads cannot drift from what ran. Galaxy::birth_circular_ratio reads the same switch as the birth site for the same reason.

Baseline reproduces the published numbers exactly - vsig 2.48 / 3.71 / 1.62 / 1.13 / 0.57 / 0.48 for 12345 and 2.76 / 3.14 / 1.59 / 0.70 / 0.38 / 0.31 for 12346 - so the harness is measuring the same galaxy.

Worth recording that only two force paths reach a star at all: the coarse field, and the association binding potential. STAR_WAVE_COUPLING is 0.0 and supernovae kick gas, not stars. So the candidate list was closed, not open.

Single-factor sweep: everything is negative

vsig at t=2500, irregular=>spiral, size 500, seeds 12345 / 12346:

configuration	12345	12346
baseline	0.48	0.31
fresh field every tick (cadence 4 -> 1)	0.46	0.47
smoothed field (2 box passes)	0.36	0.19
axisymmetric field	0.30	0.25
no stellar self-gravity	0.27	0.26
no association binding	0.40	0.43
no birth dispersion	0.50	0.49

Not one of them prevents the crossover. Field staleness, clump scattering, non-axisymmetric structure, stars scattering off each other, the binding potential, and birth dispersion are all individually innocent.

The age split, which is what made the rest legible

Pooled vsig cannot tell post-birth heating from generations piling up on different birth orbits. Added rotation_dispersion_ratio_for_age, with a calibration test pinning the discrimination on both cases. It also pins the magnitude, which turned out to matter: two perfectly cold cohorts a factor of two apart in streaming speed still pool to 4.2, so generational offset alone cannot drive pooled vsig under 1.0. I had guessed otherwise and the test corrected me before it reached this thread.

Baseline cohorts, seed 12345:

cohort	t=900	t=1500	t=2000	t=2500
young, age < 150	2.91	2.22	2.44	2.12
mid, 150-500	1.11	0.98	0.45	0.74
old, age >= 500	-	0.56	0.38	0.28

The young cohort stays a disk for the entire run, at every checkpoint, in both seeds. Star formation is producing perfectly good disk stars at t=2500 just as it was at t=250. Stars are heated after birth, from ~2.5 to ~0.3 within about 500 sim-time units of age, and the pooled decline is the growing old fraction. Nothing is wrong with the disk's ability to make a disk.

Both factors together

The birth ratio cap was built and reverted before rotation_dispersion_ratio existed, and judged against star_circular_ratio, which the retraction above disqualified. So its effect on the disk had never actually been measured. Measuring it, together with the field ablations:

vsig at t=2500:

configuration	12345	12346
baseline	0.48	0.31
birth ratio cap 1.06 alone	0.19	0.44
axisymmetric field alone	0.30	0.25
both together	2.25	2.44

Old cohort, the population that was reading 0.28 / 0.24:

configuration	12345	12346
baseline	0.28	0.24
both together	2.07	2.21

There is no crossover. The disk is rotation-dominated at t=2500 and the old stars are rotation-dominated, which has not been true at any point in this investigation. With the probe now mirroring the site, bcirc reads 1.06 exactly, and vsig holds 4.5-5.3 through t=1500 at size 500.

Why neither works alone

Two independent mechanisms, each sufficient:

1. An over-fast birth becomes permanent eccentricity. An axisymmetric field is torque-free, so it conserves each star's angular momentum. A star handed 2-3x circular keeps that mismatch forever as a large radial excursion. Ablating the field alone leaves the birth error intact and converts it into pure radial dispersion - which is why axisymmetric-alone drives the old cohort to 0.02, the worst number in the entire matrix.
2. A lumpy field scatters a correct birth apart. Cap births to circular and leave the clumpy field in place, and the field randomizes them anyway.

Fix one and the other still finishes the job on the same schedule. That is the shape single-variable ablation is structurally blind to, and it is worth writing down for the next sweep that comes back empty.

Cost, measured

This is not landable as it stands, and it fails in the direction already predicted on this issue.

• Over-concentration. Star fraction inside 0.3 disk_r goes 0.03-0.14 to 0.16-0.17 for irregular=>spiral at size 500. At the size-150 test scale the four scenarios at t=1200 go: elliptical econ 0.76 -> 0.93 and sctr 0.66 -> 0.86; bang=>spiral coherence 0.36 -> 0.11; irregular=>spiral sctr 0.07 -> 0.45. The ring survives intact (ring 0.97 -> 0.98, hollow 0.99 -> 0.99). Treat the size-150 pass as indicative only.
• The elliptical starts rotating. Its vsig goes 0.51 -> 3.16. A pressure-supported spheroid that is rotation-dominated is the wrong object. Same finding as before, now with a metric that can see it.
• Resolved star count roughly halves, 43656 -> 24018. Worth understanding rather than accepting: the drain is nearly gone, so this is a drop in formation, not retention.
• The axisymmetric field is an ablation, not a design. It cannot ship as-is. Arm tracing does survive it better than I expected (arm 1.56 / 1.07 against baseline 1.02 / 1.12, since stars are born in the arms regardless), but the honest reading is that we now know what the stars need - a smooth potential for orbital support - not yet how to give them that and the arms too. STAR_WAVE_COUPLING exists for exactly this and is currently 0.0, which makes it the obvious first thing to try.

Suggested order now

1. Try smooth-field-plus-analytic-wave: axisymmetric field for orbital support, STAR_WAVE_COUPLING above zero for arm tracing, birth ratio cap for correct initial orbits. A coherent wave should not heat the way a clump-dominated 64-grid field does. Cheapest path from a diagnosis to a design.
2. Give the elliptical explicit birth dispersion so its spheroid stops depending on over-fast births, and find where the central over-concentration comes from - it appears in all four scenarios, so it is one mechanism, not four.
3. Land, with the golden fields regenerated in the same commit.

Landed on main: the harness, the age-resolved metric, calibration tests for both field filters and for the age split, ward exec ablation-sweep, and docs/ablation.md. Every switch is off by default, the wasm build reads no environment, and the golden mass field test is untouched and green - 91 tests pass. No physics has changed on main.

Stopping here rather than starting the scenario redesign: it is a deliberate multi-scenario retune with golden fields regenerated, and pushing to main rolls the public site.

Ran the ablation matrix. **Found it.** There are two heat sources, not one, and each is on its own sufficient to destroy the disk - which is exactly why one-at-a-time ablation came back empty. ## First, the harness Switches now live in the kernel (`src/rust/ablation.rs`), read from the environment, all off by default. `debug-sim` prints the resolved configuration as its first line. This is a direct response to the retraction above: a switch the sim itself reads cannot drift from what ran. `Galaxy::birth_circular_ratio` reads the same switch as the birth site for the same reason. Baseline reproduces the published numbers exactly - `vsig` 2.48 / 3.71 / 1.62 / 1.13 / 0.57 / 0.48 for 12345 and 2.76 / 3.14 / 1.59 / 0.70 / 0.38 / 0.31 for 12346 - so the harness is measuring the same galaxy. Worth recording that only two force paths reach a star at all: the coarse field, and the association binding potential. `STAR_WAVE_COUPLING` is 0.0 and supernovae kick gas, not stars. So the candidate list was closed, not open. ## Single-factor sweep: everything is negative `vsig` at t=2500, irregular=>spiral, size 500, seeds 12345 / 12346: | configuration | 12345 | 12346 | |---|---|---| | baseline | 0.48 | 0.31 | | fresh field every tick (cadence 4 -> 1) | 0.46 | 0.47 | | smoothed field (2 box passes) | 0.36 | 0.19 | | axisymmetric field | 0.30 | 0.25 | | no stellar self-gravity | 0.27 | 0.26 | | no association binding | 0.40 | 0.43 | | no birth dispersion | 0.50 | 0.49 | Not one of them prevents the crossover. Field staleness, clump scattering, non-axisymmetric structure, stars scattering off each other, the binding potential, and birth dispersion are all individually innocent. ## The age split, which is what made the rest legible Pooled `vsig` cannot tell post-birth heating from generations piling up on different birth orbits. Added `rotation_dispersion_ratio_for_age`, with a calibration test pinning the discrimination on both cases. It also pins the magnitude, which turned out to matter: two perfectly cold cohorts a factor of two apart in streaming speed still pool to 4.2, so generational offset alone cannot drive pooled `vsig` under 1.0. I had guessed otherwise and the test corrected me before it reached this thread. Baseline cohorts, seed 12345: | cohort | t=900 | t=1500 | t=2000 | t=2500 | |---|---|---|---|---| | young, age < 150 | 2.91 | 2.22 | 2.44 | 2.12 | | mid, 150-500 | 1.11 | 0.98 | 0.45 | 0.74 | | old, age >= 500 | - | 0.56 | 0.38 | 0.28 | The young cohort stays a disk for the entire run, at every checkpoint, in both seeds. Star formation is producing perfectly good disk stars at t=2500 just as it was at t=250. Stars are heated *after* birth, from ~2.5 to ~0.3 within about 500 sim-time units of age, and the pooled decline is the growing old fraction. Nothing is wrong with the disk's ability to make a disk. ## Both factors together The birth ratio cap was built and reverted before `rotation_dispersion_ratio` existed, and judged against `star_circular_ratio`, which the retraction above disqualified. So its effect on the disk had never actually been measured. Measuring it, together with the field ablations: `vsig` at t=2500: | configuration | 12345 | 12346 | |---|---|---| | baseline | 0.48 | 0.31 | | birth ratio cap 1.06 alone | 0.19 | 0.44 | | axisymmetric field alone | 0.30 | 0.25 | | **both together** | **2.25** | **2.44** | Old cohort, the population that was reading 0.28 / 0.24: | configuration | 12345 | 12346 | |---|---|---| | baseline | 0.28 | 0.24 | | **both together** | **2.07** | **2.21** | There is no crossover. The disk is rotation-dominated at t=2500 and the *old* stars are rotation-dominated, which has not been true at any point in this investigation. With the probe now mirroring the site, `bcirc` reads 1.06 exactly, and `vsig` holds 4.5-5.3 through t=1500 at size 500. ## Why neither works alone Two independent mechanisms, each sufficient: 1. **An over-fast birth becomes permanent eccentricity.** An axisymmetric field is torque-free, so it conserves each star's angular momentum. A star handed 2-3x circular keeps that mismatch forever as a large radial excursion. Ablating the field alone leaves the birth error intact and converts it into pure radial dispersion - which is why axisymmetric-alone drives the old cohort to 0.02, the worst number in the entire matrix. 2. **A lumpy field scatters a correct birth apart.** Cap births to circular and leave the clumpy field in place, and the field randomizes them anyway. Fix one and the other still finishes the job on the same schedule. That is the shape single-variable ablation is structurally blind to, and it is worth writing down for the next sweep that comes back empty. ## Cost, measured This is not landable as it stands, and it fails in the direction already predicted on this issue. - **Over-concentration.** Star fraction inside 0.3 `disk_r` goes 0.03-0.14 to 0.16-0.17 for irregular=>spiral at size 500. At the size-150 test scale the four scenarios at t=1200 go: elliptical `econ` 0.76 -> 0.93 and `sctr` 0.66 -> 0.86; bang=>spiral coherence 0.36 -> 0.11; irregular=>spiral `sctr` 0.07 -> 0.45. The ring survives intact (`ring` 0.97 -> 0.98, `hollow` 0.99 -> 0.99). Treat the size-150 pass as indicative only. - **The elliptical starts rotating.** Its `vsig` goes 0.51 -> 3.16. A pressure-supported spheroid that is rotation-dominated is the wrong object. Same finding as before, now with a metric that can see it. - **Resolved star count roughly halves**, 43656 -> 24018. Worth understanding rather than accepting: the drain is nearly gone, so this is a drop in formation, not retention. - **The axisymmetric field is an ablation, not a design.** It cannot ship as-is. Arm tracing does survive it better than I expected (`arm` 1.56 / 1.07 against baseline 1.02 / 1.12, since stars are born in the arms regardless), but the honest reading is that we now know *what the stars need* - a smooth potential for orbital support - not yet how to give them that and the arms too. `STAR_WAVE_COUPLING` exists for exactly this and is currently 0.0, which makes it the obvious first thing to try. ## Suggested order now 1. Try smooth-field-plus-analytic-wave: axisymmetric field for orbital support, `STAR_WAVE_COUPLING` above zero for arm tracing, birth ratio cap for correct initial orbits. A coherent wave should not heat the way a clump-dominated 64-grid field does. Cheapest path from a diagnosis to a design. 2. Give the elliptical explicit birth dispersion so its spheroid stops depending on over-fast births, and find where the central over-concentration comes from - it appears in all four scenarios, so it is one mechanism, not four. 3. Land, with the golden fields regenerated in the same commit. Landed on `main`: the harness, the age-resolved metric, calibration tests for both field filters and for the age split, `ward exec ablation-sweep`, and [docs/ablation.md](../src/branch/main/docs/ablation.md). Every switch is off by default, the wasm build reads no environment, and the golden mass field test is untouched and green - 91 tests pass. No physics has changed on `main`. Stopping here rather than starting the scenario redesign: it is a deliberate multi-scenario retune with golden fields regenerated, and pushing to `main` rolls the public site.

coilyco-ops referenced this issue 2026-08-09 03:36:36 +00:00

Boundary ridge: gas confinement is load-bearing, and stars have their own edge #65

coilyco-ops referenced this issue 2026-08-09 03:50:58 +00:00

Stars orbit a potential they were never given orbits for: one root cause behind #65 and #66 #70

coilyco-ops commented 2026-08-09 03:51:07 +00:00

Author

Member

Copy link

Compiled into #70, which owns the fix from here.

This issue and #65 turned out to be one root cause: stars are put on orbits the potential they read cannot hold, and two independently sufficient mechanisms convert that mismatch into random motion. #70 carries the evidence, the three-part change, the measured cost, and the ordered work items.

Nothing open here is left behind - every question on this thread is a work item there. Keeping this open for its history and for the measurement record until #70 lands.

Compiled into **#70**, which owns the fix from here. This issue and #65 turned out to be one root cause: stars are put on orbits the potential they read cannot hold, and two independently sufficient mechanisms convert that mismatch into random motion. #70 carries the evidence, the three-part change, the measured cost, and the ordered work items. Nothing open here is left behind - every question on this thread is a work item there. Keeping this open for its history and for the measurement record until #70 lands.

coilyco-ops referenced this issue 2026-08-09 03:51:12 +00:00

Boundary ridge: gas confinement is load-bearing, and stars have their own edge #65

coilyco-ops referenced this issue 2026-08-09 03:56:01 +00:00

Stars orbit a potential they were never given orbits for: one root cause behind #65 and #66 #70

coilyco-ops referenced this issue 2026-08-09 04:19:23 +00:00

Stars orbit a potential they were never given orbits for: one root cause behind #65 and #66 #70

coilyco-ops referenced this issue 2026-08-09 04:51:21 +00:00

Stars orbit a potential they were never given orbits for: one root cause behind #65 and #66 #70

coilyco-ops referenced this issue 2026-08-09 05:21:49 +00:00

The resolved star population has no bound once the halo drain is fixed, and dim stars are most of the cost for almost none of the light #72

coilyco-ops commented 2026-08-14 02:58:12 +00:00

Author

Member

Copy link

Closing as superseded by #70, on that issue's own accounting:

#66 keeps nothing but its history. Every open question there is an item above.

The three-step proposal here was overtaken by the ablation evidence #70 compiled. Step 1 - scale birth velocity to the field stars actually read - is #70's item 1, now expressed as a ratio cap against local circular speed rather than an absolute one. Step 2, weak orbital relaxation for disk stars, was tested and is not needed: every single-factor ablation is negative, and the disk holds only when the birth-orbit cap and a smooth field are applied together (vsig 0.48/0.31 baseline against 2.25/2.44 for both). Step 3, FIELD_RES, is subsumed by #70's item 2, designing the shipping smooth field.

The diagnosis here stands and is worth keeping visible: the heating is entirely post-birth, the star formation was never the problem, and the gas was always fine. #70's age-resolved measurement confirmed it - the young cohort reads 2.91 / 2.22 / 2.44 / 2.12 at t=900 through 2500, so the sim makes good disk stars at maturity exactly as it does at t=250.

No leftovers. #70 owns the change.

Closing as superseded by #70, on that issue's own accounting: > **#66** keeps nothing but its history. Every open question there is an item above. The three-step proposal here was overtaken by the ablation evidence #70 compiled. Step 1 - scale birth velocity to the field stars actually read - is #70's item 1, now expressed as a ratio cap against local circular speed rather than an absolute one. Step 2, weak orbital relaxation for disk stars, was tested and is not needed: **every single-factor ablation is negative**, and the disk holds only when the birth-orbit cap and a smooth field are applied together (`vsig` 0.48/0.31 baseline against 2.25/2.44 for both). Step 3, `FIELD_RES`, is subsumed by #70's item 2, designing the shipping smooth field. The diagnosis here stands and is worth keeping visible: the heating is entirely post-birth, the star formation was never the problem, and the gas was always fine. #70's age-resolved measurement confirmed it - the young cohort reads 2.91 / 2.22 / 2.44 / 2.12 at t=900 through 2500, so the sim makes good disk stars at maturity exactly as it does at t=250. No leftovers. #70 owns the change.

coilyco-ops closed this issue 2026-08-14 02:58:16 +00:00

Sign in to join this conversation.

No Branch/Tag specified

Branches Tags

main

aos/claude/mg96-fm

aos/claude/mg96

gg/claude/74-ci-in-dev-base

No results found.

Labels

Clear labels   

burndown-2026-06

Backlog burndown June 2026

  

autonomy

async-consult

A human needs to consult on the issue to upgrade it to headless

  

autonomy

epic

This issue has many units of sub work - its size makes it meaningfully exclusive with other autonomy types

  

autonomy

headless

The agent can perform the work on its own

  

autonomy

live-collab

The agent and the human need to work together in realtime

  

c#

Requires C# work, flagged b/c it requires a Eco server restart

  

priority

P0

priority tier

  

priority

P1

priority tier

  

priority

P2

priority tier

  

priority

P3

priority tier

  

priority

P4

priority tier

  

role/ai

requires work from the AI Engineer role

  

role/creator

requires work from Content Creator role

  

role/design

requires work from the design role

  

role/director

requires work from the director role

  

role/engineer

requires work from the engineer role

  

role/exec

requires work from the exec role

  

role/human

requires a person, and specifically not an agent seat

  

role/ops

requires work from the ops role

  

role/qa

requires work from the QA role

No labels

burndown-2026-06

autonomy

async-consult

autonomy

epic

autonomy

headless

autonomy

live-collab

c#

priority

P0

priority

P1

priority

P2

priority

P3

priority

P4

role/ai

role/creator

role/design

role/director

role/engineer

role/exec

role/human

role/ops

role/qa

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

coilyco-gaming/galaxy-gen#66

No description provided.