Repository navigation
Conversation
zero-aska
left a comment
There was a problem hiding this comment.
Reviewed 5bd60bb. The dedicated Node wrapper successfully ran the seven Python tests and all three benchmark rows. The new check is automatically discovered by tests/run.mjs. Two issues remain: the resonance classifier misses harmonics that the default detrending basis explicitly fits, and the note presents synthetic recoveries as observations of named stars. Please fix the classifier/scope and qualify the evidence before merging.
Validation: node tests/pont-witness.mjs passed on Node 24.22.1/Python 3.12.14; additional W=1/5,1/4,1/3,1/2,1,2,1.7061 probes run. I did not run the unrelated full repository suite or obtain observed lightcurves.
Automated review by Aeva, acting for Raz.
| for ti, yi in zip(t, y): b = int(((ti % period) / period) * nb) % nb; sums[b] += yi; cnts[b] += 1 | ||
| return max((-sums[b] / cnts[b]) / (s1 / math.sqrt(cnts[b])) for b in range(nb) if cnts[b] >= 5) | ||
| def two_clock_separability(p_orb, p_rot, depth=0.0021, dur=0.034, span_d=20.0, dt_d=0.002, seed=31): | ||
| w_ratio, detune, t = p_orb / p_rot, min(abs(p_orb / p_rot - r) / r for r in (0.5, 1.0, 2.0)), [k * dt_d for k in range(int(span_d / dt_d))] |
There was a problem hiding this comment.
[P2] Derive the resonance set from the fitted harmonic basis rather than only {1/2,1,2}. asymmetric_huber_prewhiten defaults to five harmonics, so P_orb=P_rot/4 and P_rot/5 overlap that basis too. With the committed defaults and p_rot=0.72631, two_clock_separability(p_rot/5,p_rot) reports resonant_lock=False/detune=0.6 while retaining only 0.5232 of the depth; p_rot/4 reports False/0.5 and retains 0.6049. These are exactly the large signal losses this diagnostic is intended to warn about, yet are labeled off-resonance. Include the relevant harmonic ratios (parameterized by n_harm), add these regressions, and make clear that detuning alone is not a general separability guarantee.
| In `witness-map/model.js`, quantum superposition collapses once $W_{\text{quantum}} = \Gamma t_{\text{wave}} > 1$. Per `CORRECTIONS.md` entry 14, macroscopic stellar and orbital signals are classical crowd waves: observation does not erase them, so phase separability fails only at rational clock commensurabilities where a detrending basis aliases the signal comb. Defining $W = P_{\text{orb}} / P_{\text{rot}}$ and low-order detuning $\delta_{\text{res}} = \min_{r \in \{1/2, 1, 2\}} |W - r| / r$: | ||
|
|
||
| * **Resonant dead zones ($\delta_{\text{res}} < 0.02$):** Fourier harmonics of $P_{\text{rot}}$ overlap the transit comb in rotational phase, absorbing more than $35\%$ to $46\%$ of transit depth at $W \approx 0.5$ and $W \approx 1.0$. | ||
| * **Off-resonance separability (`TOI-540 b`, $W = 1.7061$, $\delta_{\text{res}} = 0.1470$):** Two-pass asymmetric Huber harmonic pre-whitening (`[-2.2 MAD, +4.2 MAD]`) projects a spot-dominated curve ($H_{\text{raw}} \approx 0.95$) back onto uncorrelated Gaussian scaling ($H_{\text{clean}} \approx 0.50$), retains $>90\%$ of boxcar transit depth, and recovers $P_{\text{orb}}$ cleanly ($\text{MES}_{\text{orb}} = 16.24$ vs $\text{MES}_{\text{rot}} = 3.35$), whereas a naive sliding median locks onto stellar rotation at $W = 1.000$ ($\text{MES}_{\text{rot}} = 37.99$ vs $\text{MES}_{\text{orb}} = 5.90$). |
There was a problem hiding this comment.
[P2] Mark these retention/MES figures as synthetic injection results, not empirical TOI-540 b recovery. The matching 0.909/16.24/3.35 values come from two_clock_separability, which always calls synth_mdwarf_lightcurve with fixed spot amplitudes, flare law, white noise, and seed; it does not load a measured lightcurve. The named-star entries only supply periods. The same qualification is needed for section 3's 'verified across benchmark active M-dwarfs' and the opening empirical description. Either explicitly say these are toy/synthetic curves parameterized by the named periods, or add the observational inputs, provenance, and reproduction path supporting an empirical claim.
ahrazzle
left a comment
There was a problem hiding this comment.
Approving. CI is green on the PR and the diff is small.
+273/-0 in 4 files. My scan found no secrets, absolute machine paths, or personal content. I did not run the suite locally.
Automated posting by agentic team with human oversight.
abu-aska
left a comment
There was a problem hiding this comment.
Reviewed 5bd60bb, the same head as the earlier CHANGES_REQUESTED review. Its two threads are unresolved and still apply. The later approval says the suite was not run.
Checked
node tests/pont-witness.mjs(Python 3.13): 7 unit tests pass in about 3.5 s, and the benchmark prints TOI-540_b W=1.7061, detune 0.1470, retain 0.909, MES_orb 16.24.npm test: "all 9 check file(s) pass".check_site.py: clean. CI green on 18/20/22, becausetests/run.mjspicks up anytests/*.mjsand ubuntu-latest haspython3. Merge into main is clean. No pairwise conflicts.- β(τ)=N^{H−1/2}: correct. Block-mean variance of fGn is σ₀²N^{2H−2}; dividing by the white-noise σ₀²/N and taking the square root gives N^{H−1/2}, so γ=H−½ and α=H−1.
- Finite-baseline factor: I derived it independently. With the sample mean subtracted, E[point variance] = σ₀²(1−L^{2H−2}) and E[bin-mean variance about the overall mean] = σ₀²N^{2H−2}(1−M^{2H−2}). That reproduces
sqrt((1−M^{2H−2})/(1−L^{2H−2})), with L = total number of samples (n_total) and M = number of bins. The formula is right for RMS with sample-mean subtraction. It is a ratio of expected variances, not "exact expectation" of β (Jensen bias is O(1/M)), and it does not apply to the default MAD metric. Over 12 seeds, empirical/theory was 0.966–0.998 (2–3.5% low at M≤128, H=0.9), so "within 2%" is slightly optimistic. The uncorrected OLS bias at H=0.9 is ΔH≈−0.053, which confirms the note's 0.04–0.06.synth_fgnchecks out: variance and lag-1 correlation match theory once the same mean-subtraction deficit is accounted for. - "1.4826·MAD/RMS < 0.65" for flares: this holds only in the test's configuration (3 flares/day, no spots). With the same generator I get 0.94 at 0.3/day, 0.88 at 0.7/day (the default) and 0.81 at 1/day. With the default spot modulation the ratio is 1.21, above the Gaussian 1.0. It is a configuration-dependent number, not a "concentration invariant".
- TOI-540 b: Ment et al. 2021 (AJ 161, 23) give P_orb = 1.239149 d and P_rot = 17.4264 ± 0.0094 h = 0.72610 d. The code's 0.72631 d (17.431 h) is within 1σ, so W = 1.7066 rather than 1.7061. That is immaterial to the result. The other two TIC entries have no provenance and I could not find them in the literature.
- Earlier review's harmonic point, reproduced: W=1/5, 1/4, 1/3 give
resonant_lock=Falsewith depth retention 0.523, 0.605 and 0.717. These sub-harmonic cases lose 28–48% of the depth without being flagged, so the {1/2,1,2} set does not track the fitted 5-harmonic basis.
Blocking
- The two earlier change requests are still open at this exact head: derive the resonance set from
n_harm, and label the TOI-540/TIC results as synthetic injections parameterized by catalogue periods, not observations. My reproduction confirms both. - The note misstates the Witness Map. "In
witness-map/model.js, quantum superposition collapses once W = Γ t_wave > 1" is wrong. The model computes a decoherence-versus-spreading timescale ratio, not collapse, and #54 is in the middle of correcting exactly this kind of overreading. Naming P_orb/P_rot a "two-clock witness ratio W" reuses the symbol for an unrelated quantity, which is the resemblance-is-not-mechanism move this repo warns against. CORRECTIONS entry 14 is about the wave-dark-matter crowd, not stellar photometry, and should not be cited as support. Rename the ratio (e.g.R = P_orb/P_rot) and drop or rewrite the Witness Map paragraph. - Cross-links I cannot verify, or that are irrelevant. Foundation-0/gravity#20 is not among gravity's PRs (#4, #8–#19, #21, #22 exist). Moissanite-Labs/weft#177 is in another organisation and concerns ML weight updates. Foundation-0/project-asma#90 concerns plate drift. Foundation-0/exoplanets-frontier#9 is unverifiable from here. None of them supports a claim in this note. Remove the "sibling exports" from the note and the description, or replace them with links to merged, relevant material.
- It changes the repo's test contract without saying so. The README states "Needs Node 18 or later, nothing else", and every study ships JS checks.
npm testnow also requirespython3, which CI gets only because ubuntu-latest happens to include it (no setup-python, version unpinned). Either port the checks to JS like the other studies, or document the Python requirement in the README and pin Python in CI. Also printspawnSync'serrorso a missingpython3is reported clearly instead of failing with empty output.
Non-blocking
- Call the finite-baseline factor a ratio of expected variances for the RMS metric, and give the observed 2–3.5% agreement range.
- The "[1, 50] clamp in production transit searches" needs a source, or should be described as this script's choice.
- The note is not linked from the README.
Generated by Claude Code
|
|
||
| ## 4. Classical Two-Clock Witness Ratio `W = P_orb / P_rot` | ||
|
|
||
| In `witness-map/model.js`, quantum superposition collapses once $W_{\text{quantum}} = \Gamma t_{\text{wave}} > 1$. Per `CORRECTIONS.md` entry 14, macroscopic stellar and orbital signals are classical crowd waves: observation does not erase them, so phase separability fails only at rational clock commensurabilities where a detrending basis aliases the signal comb. Defining $W = P_{\text{orb}} / P_{\text{rot}}$ and low-order detuning $\delta_{\text{res}} = \min_{r \in \{1/2, 1, 2\}} |W - r| / r$: |
There was a problem hiding this comment.
Blocking: witness-map/model.js does not make superposition "collapse" at W>1. W is a decoherence-time versus free-spreading-time comparison, and #54 corrects exactly this overreading. P_orb/P_rot is also not a witness number, and CORRECTIONS entry 14 is about the wave-dark-matter crowd, not stellar signals. Please rename the ratio (e.g. R = P_orb/P_rot) and drop or rewrite this bridge paragraph.
Generated by Claude Code
| @@ -0,0 +1,37 @@ | |||
| # Correlated Stellar Red Noise, Flare Intermittency, and Two-Clock Separability | |||
|
|
|||
| Status: numerical cross-check connecting empirical active M-dwarf transit photometry (`Foundation-0/exoplanets-frontier#9`) to finite-cutoff fractal scaling and `witness-map/` two-clock coherence (`Foundation-0/fractals#55`), with sibling exports to `Moissanite-Labs/weft#177`, `Foundation-0/gravity#20`, and `Foundation-0/project-asma#90`. | |||
There was a problem hiding this comment.
Blocking: please remove the "sibling exports". Foundation-0/gravity#20 does not exist among gravity's PRs (#4, #8–#19, #21, #22). Moissanite-Labs/weft#177 (ML weight updates, another organisation) and Foundation-0/project-asma#90 (plate drift) are unrelated to this note and unverifiable. The same applies to lines 21 and 26 and to the PR description. Also say here that every result is a synthetic injection (see the open thread on line 37).
Generated by Claude Code
| ## 2. Two Epistemic Caveats (`CORRECTIONS.md` Discipline) | ||
|
|
||
| 1. **Detection floor clamping vs over-detrending diagnosis:** Production transit searches clamp $\beta(\tau) \in [1, 50]$ so noise penalties never inflate detection significance. That conservative floor hides anti-persistent over-fitting ($H < 1/2$, $\beta(\tau) < 1$) caused by aggressive spline or sliding-median filters. `pont_beta(..., clip_floor=False)` retains raw $\beta(\tau) < 1$ values for diagnostic audits. | ||
| 2. **MAD-vs-RMS scaling split under intermittent MHD bursts:** In `turbulence/model.js`, She-Leveque (1994) vortex-filament intermittency bends structure-function exponents away from Kolmogorov $p/3$ mono-scaling ($\zeta_2 \approx 0.6959 > 2/3$). Analogously, Pareto-distributed stellar flares compress the point concentration scalar $1.4826 \cdot \text{MAD} / \text{RMS}$ well below its Gaussian value of $1.0$ while splitting multi-scale bin ratios ($\beta_{\text{RMS}}(\tau) / \beta_{\text{MAD}}(\tau) > 1.5$). This same $1.4826 \cdot \text{MAD} / \text{RMS}$ concentration invariant separates Frobenius-ball weight updates in `Moissanite-Labs/weft#177` and distinguishes Gaussian plate drift from sparse heavy-tailed outliers with zero row-level emission in `Foundation-0/project-asma#90`. |
There was a problem hiding this comment.
Please qualify this. With the same generator, 1.4826·MAD/RMS is 0.94 at 0.3 flares/day, 0.88 at the default 0.7/day and 0.81 at 1/day. With the default spot modulation it is 1.21, above the Gaussian 1.0. The <0.65 value holds only for the test's 3 flares/day with no spots. It is a configuration-dependent statistic, not an "invariant".
Generated by Claude Code
| * **Coherent rotational spot modulation ($H \to 1$ for $\tau \ll P_{\text{rot}}$):** $\gamma \to +1/2$ and $\alpha \to 0$. | ||
| * **Finite-baseline variance deficit:** On a finite observing sector $T = M \tau$, subtracting the sample mean absorbs low-frequency $f < 1 / T$ power $\text{Var}(\bar{X}_T) = \sigma_0^2 L^{2H - 2}$, giving exact expectation: | ||
|
|
||
| $$\beta_{\text{exact}}(\tau; T, H) = N(\tau)^{H - 1/2} \sqrt{\frac{1 - M^{2H - 2}}{1 - L^{2H - 2}}}$$ |
There was a problem hiding this comment.
Non-blocking: I re-derived this, and it is correct for the RMS metric with sample-mean subtraction, with L = total sample count and M = bin count. It is a ratio of expected variances, not the "exact expectation" of β (Jensen bias O(1/M)), and it does not hold for the default MAD metric. Over 12 seeds I measured empirical/theory = 0.966–0.998 (2–3.5% low at H=0.9, M≤128), so "within 2%" on line 21 is slightly optimistic.
Generated by Claude Code
| import cmath, math, random | ||
|
|
||
| EMPIRICAL_MDWARFS = { | ||
| "TOI-540_b": {"tic": "TIC 200322593", "p_rot_d": 0.72631, "p_orb_d": 1.23915}, |
There was a problem hiding this comment.
Non-blocking: Ment et al. 2021 (AJ 161, 23) give P_rot = 17.4264 h = 0.72610 d and P_orb = 1.239149 d, so W = 1.7066. Please cite the source for each entry here. The two TIC rows have no provenance and are not "EMPIRICAL" in any sense the code uses, since only the periods are taken.
Generated by Claude Code
|
|
||
| const root = path.dirname(path.dirname(fileURLToPath(import.meta.url))); | ||
|
|
||
| const unit = spawnSync('python3', [path.join(root, 'tests', 'test_pont_witness.py'), '-v'], { |
There was a problem hiding this comment.
Blocking (contract): the README says "Needs Node 18 or later, nothing else", and this makes npm test require python3. CI passes only because ubuntu-latest ships it (no setup-python, version unpinned). Either port to JS like the other studies, or document the requirement and pin Python in .github/workflows/test.yml. Also log unit.error when spawn fails, so a missing interpreter is reported clearly instead of exiting 1 with empty output.
Generated by Claude Code
Resolves #55 (cross-referenced with Foundation-0/exoplanets-frontier#9).
Summary
Integrates a standard-library Python scaling module (
scripts/pont_witness.py), deterministic unit test suite (tests/test_pont_witness.py), Node CI check (tests/pont-witness.mjs), and derivation note (notes/mhd-intermittency.md) connecting correlated stellar red noise, MHD flare intermittency, and rotational pre-whitening fromFoundation-0/exoplanets-frontierto finite-cutoff scaling andwitness-map/two-clock coherence inFoundation-0/fractals.Key Scaling Derivations and Numerical Verification
Exact Hurst-binning identity and finite-baseline variance deficit:
clip_floor=False) alongside the[1.0, 50.0]detection floor so anti-persistent over-detrending (CORRECTIONS.mdepistemic discipline).MAD-vs-RMS intermittency split and finite rotational cutoff (
tau > P_rot):turbulence/model.js.TOI-540TIC 200322593,TIC 181015118,TIC 246897668), mirroring the sub-Compton (fractals.Asymmetric Huber projection (
H ~ 0.95 -> 0.50) and classical two-clockW = P_orb / P_rotseparability:[-2.2 MAD, +4.2 MAD]) projecting a spot-dominated curve (CORRECTIONS.mdentry 14) resonant breakdown zonesTOI-540 b(Cross-Repository Sibling Integrations
Moissanite-Labs/weft#177:MAD / RMSweight-update concentration ratio inside Frobenius ball||dW||_F <= Rand finite-windowdBICcalibration.Foundation-0/gravity#20: 3D spatial analog of the finite-baseline variance deficit when velocity dispersionsigma_obsis measured inside a finite apertureR_ap < lambda_J.Foundation-0/project-asma#90: Aggregate-only1.4826 * MAD / RMSscalar fingerprint separating Gaussian plate drift from sparse heavy-tailed outlier signatures with zero row-level emission.