Conversation
Second correction to the same caveat, and the commit message says so rather than presenting v3 as if it were the first answer. #304 concluded theta_star is "protected" because the light curve pins rho * theta_E independently of the SED. ab194/av_clump landed and refutes it. Two ptde arms on event 194, Rhat 1.00, 78/78 chains: arm av prior av post theta*/truth R/truth D/truth av_true 9.01 +/- 0.3 8.99 0.983 0.403 0.410 av_clump 8.61 +/- 2.41 7.18 0.738 0.392 0.531 theta_star tracks av at 0.0688 dex/mag over those points -- STEEPER than the CSB relation's 0.042, because the colour term moves too. With `star_constrains_rho: true` the SED constrains rho, so av feeds straight into theta_star; "independent of the SED" was never available. TWO THINGS THE PAIR ESTABLISHES THAT NEITHER ARM COULD ALONE: * PRIOR WIDTH IS AS DECISIVE AS THE ANCHOR. Both priors come from the SAME event_info clump columns and differ only in inversion (monochromatic 8.61 vs integrated 9.01) and WIDTH (2.41 vs 0.3). The wide one let the SED pull av down to 7.18 -- the data's own av preference is LOW -- and theta_star fell with it. CONSEQUENCE FOR THE RUNNING SWEEP, which uses av = 9.01 +/- 1.83 precisely so the width carries the C29 systematic honestly: by this slope it should be expected nearer 0.74x than 0.983x on theta_star. That is an honest-width-vs-accuracy tradeoff and it is JDE's call, not something to re-tighten silently while the sweep runs. * R_source IS ~0.4x TRUTH AT BOTH av VALUES (0.403, 0.392). D_source and theta_star move with av; R_source does not. So the source-radius deficit is NOT an extinction problem, and that residual is ours. The docstring now carries the full v1/v2/v3 history, because a caveat that has been wrong twice should announce that to its next reader. v3 states only what these two arms measure and names what is open: av_band (prior 11.65) is still sampling and will confirm the slope or break it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
…hor to 0.1%
av_band landed, so the caveat now rests on three arms that differ ONLY in the
av prior (event 194, ptde, Rhat 1.00, 78/78 chains):
arm av prior av post theta*/truth R/truth D/truth
av_clump 8.61 +/- 2.41 7.18 0.738 0.392 0.531
av_true 9.01 +/- 0.30 8.99 0.983 0.403 0.410
av_band 11.65 +/- 1.40 9.42 1.086 0.425 0.391
THE RESULT WORTH THE WHOLE THREAD: a log-linear fit through those three
points crosses theta_star/truth = 1 at av = 9.00. The colour anchor -- the
clump's E(Z087-W149) inverted through our own integrated BC grid, computed
with no reference to any fit -- is 9.01. Two independent routes, one from
the light curve and source physics, one from a reddening measurement, agree
to 0.1%. That is the C29 colour-anchor argument confirmed rather than
asserted.
R_source IS ~0.4x TRUTH AT EVERY av (0.392, 0.403, 0.425 across 2.24 mag), so
the source-radius deficit is not an extinction problem on any of the three.
D_source and theta_star move with av; R_source does not. That residual is
ours and is now the thing to chase.
PREDICTION SCORING, since v3 made one before this arm ran: the numeric call
(theta_star ~ 1.31) FAILED, because I guessed av would land near 10.8 and it
came back 9.42 -- pulled 1.59 sigma below its prior, far harder than
av_clump's 0.59. Conditional on the actual av, the slope predicted 1.052
against a measured 1.086, good to 3.1%. So the slope held and the
prior-pull extrapolation did not.
AND v3's SWEEP FORECAST IS REVISED DOWNWARD IN SEVERITY: it said
av = 9.01 +/- 1.83 should give "nearer 0.74x". 0.74 was av_clump's value,
from a prior both wider (2.41) and 0.4 mag lower. Bracketing by the three
observed pulls gives 0.83-0.94x. The direction stands -- an honest width
costs theta_star accuracy -- but the size was overstated.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
jdeast
changed the title
…oint coincidence
Third correction to this caveat, from the run that was still sampling when v3
was written: the SWEEP's event-194 arm (av = 9.01 +/- 1.83, the honest width,
ptde, Rhat 1.00, 78 chains) finished overnight. It landed av = 7.29 (-0.94
sigma, pulled down like every other wide prior) and theta*/truth = 0.698 --
below v3's own 0.83-0.94x forecast. That is the third forecast this text has
gotten wrong, and the docstring now says so in those words.
WHAT THE NEW POINT COSTS v3'S HEADLINE CLAIM. v3 fit three arms, found the
theta*/truth = 1 crossing at av = 9.007 against an independent colour anchor
of 9.006, and called the agreement 0.1%. With every event-194 run that has
its own results.csv -- eleven of them, from av = 4.24 to 9.42 -- that
precision is not reproducible:
fit slope crossing
v3's three arms 0.0745 9.01 +/- 0.80
six runs with av > 5 0.0500 9.31 +/- 2.83
all eleven 0.0377 9.71 +/- 1.53
The relation is monotone but NOT log-linear: the five runs near av = 4.25 sit
at 0.63 where v3's line predicts 0.44, so a global fit is simply the wrong
estimator and its crossing drifts with the fit range. The LOCAL statement is
the defensible one -- the two runs that straddle 1 (8.99 -> 9.42) interpolate
to 9.06, and three runs at av = 6.8-7.3 scatter 0.037 dex at fixed av, i.e.
+/- 0.5 mag of crossing. So: 9.1 +/- 0.5 against an independent 9.01.
THE ANCHOR IS STILL CONFIRMED. A prior computed from a reddening measurement
with no reference to any fit lands within half a magnitude of where the fits
themselves cross truth, and that is the C29 argument. What is retracted is
the "two independent routes agree to 0.1%" phrasing, which was three points'
worth of luck.
fitu0te IS NOT the explanation for the sweep sitting low: at av = 4.26 the
u0te runs give 0.634/0.630 against 0.632/0.651 for u_0, under 0.5%, so the
sweep is comparable to the arms despite differing in that flag.
R_source IS UNCHANGED AND STRONGER: 0.35-0.43x truth at every av over 5.2 mag
and eleven runs. That residual is ours.
Two arms named in an earlier draft of this table (rpt_async, rpt_sync) are NOT
in it: they have no results.csv, so they were never measured and nothing may
be quoted from them.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
jdeast
changed the title
Owner
Author
|
Third correction, pushed as 5b368e4 -- and the PR title changed with it. The sweep arm that was still sampling when the previous commit was written finished overnight, and it retracts the "0.1%" headline.
What eleven runs do to the crossing claim
The relation is monotone but not log-linear -- the five runs near av = 4.25 sit at 0.63 where the three-arm line predicts 0.44 -- so a global fit is the wrong estimator and its crossing drifts with the fit range. The local statement survives: the two runs straddling theta*/truth = 1 (8.99 -> 9.42) interpolate to 9.06, and three runs at av = 6.8-7.3 scatter 0.037 dex at fixed av, i.e. +/- 0.5 mag of crossing. So 9.1 +/- 0.5 against an independently computed anchor of 9.01. The anchor is still confirmed -- a prior from a reddening measurement, computed with no reference to any fit, lands within half a magnitude of where the fits cross truth. What is retracted is "two independent routes agree to 0.1%", which was three points of luck. Also settled here: Two arms an earlier draft of this table named ( |
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.
Three
ptdearms on event 194 differing only in theavprior (Rhat 1.00, 78/78 chains in the good region):The result worth the whole thread
A log-linear fit through those three points crosses
theta_star/truth = 1at av = 9.00. The colour anchor — the clump'sE(Z087-W149)inverted through our own integrated BC grid, computed with no reference to any fit — is 9.01.Two independent routes, one from the light curve and source physics, one from a reddening measurement, agreeing to 0.1%. That is the C29 colour-anchor argument confirmed rather than asserted.
R_sourceis wrong independently ofav0.392, 0.403, 0.425 across 2.24 mag of
av.D_sourceandtheta_starmove withav;R_sourcedoes not. So the source-radius deficit is not an extinction problem on any of the three arms — that residual is ours, and it is now the thing to chase.This text has been wrong twice; the docstring says so
theta_starbiased 0.766x from the CSB relation.av_truemeasured 0.983x. Wrong.theta_staris "protected" by the light curve pinningrho * theta_E.av_clumpmeasured 0.738x at av = 7.18. Wrong —star_constrains_rho: truemeans the SED constrainsrho.Prediction scoring
v3 made a falsifiable prediction before
av_bandran. The numeric call (theta_star ~ 1.31) failed — I guessedavwould land near 10.8 and it came back 9.42, pulled 1.59 sigma below its prior againstav_clump's 0.59 sigma. Conditional on the actualav, the slope predicted 1.052 against a measured 1.086, good to 3.1%. The slope held; the prior-pull extrapolation did not.The sweep forecast, revised downward in severity
v3 said
av = 9.01 ± 1.83should give "nearer 0.74x". That 0.74 wasav_clump's value, from a prior both wider (2.41) and 0.4 mag lower. Bracketing by the three observed pulls gives 0.83-0.94x. The direction stands — an honest width coststheta_staraccuracy — but the size was overstated.Script only — no library code, no fixtures.