v0.1.2

What's Changed

• A README you can read in a minute, and releases that cut themselves by @sbryngelson in #213
• Refuse a radius runaway instead of returning it by @sbryngelson in #214
• Choose serial or pooled by measuring, not by counting by @sbryngelson in #215
• Give the stiffening axis headroom, and say what the grid is for by @sbryngelson in #217
• Retire PLAN.md by @sbryngelson in #219
• Ask whether the residuals are white, since chi-squared will not by @sbryngelson in #220
• Track the model-selection writeup by @sbryngelson in #221
• Pareto design search, and three corrections to T-optimality by @sbryngelson in #223
• Design-measure mathematics and a proper fit-quality assessment by @sbryngelson in #224
• Figure currency: label grid values, draw the certificate by @sbryngelson in #225
• Draw the trial-variation result (#222) by @sbryngelson in #226
• Model comparison: new models, the machinery to rank them, and what this record cannot determine by @sbryngelson in #227
• Write up the operator and design results, with schematics for the machinery by @sbryngelson in #228
• Operators, model screening, and how to choose the next N experiments by @sbryngelson in #229
• One place for the argument checks, and fewer tests saying the same thing by @sbryngelson in #230
• The thermal treatment is favored on every record, and still not established by @sbryngelson in #231
• One criterion in nats: what would this experiment change my mind about by @sbryngelson in #233
• The trial spread is neither noise nor the model's own parameters by @sbryngelson in #234
• Model-form error: the visible part, a bound on the invisible part, and the coordinates where it exists by @sbryngelson in #236
• The model-form budget, woven into how selection, estimation and design are read by @sbryngelson in #237
• A grid node is not an estimate, so stop storing it under a name that says it is by @sbryngelson in #238
• Ogden's first exponent was an Arruda-Boyce chain count, so Ogden never ran by @sbryngelson in #239
• The next term of the strain-energy expansion, and an oracle for it by @sbryngelson in #240
• One Zener construction for every elastic law, and a screen that respects signs by @sbryngelson in #242
• Constrain the measure, not the rounding, to get a testable batch by @sbryngelson in #243
• yeohSLS: a good fit from the wrong point, at any budget by @sbryngelson in #244
• Write up what could remove the discrepancy, and persist the sign by @sbryngelson in #246
• Refit results.json, and follow the trial-scatter mode into the rebound by @sbryngelson in #249
• Stop a collapse the model cannot survive, instead of grinding to the step limit by @sbryngelson in #248
• Upload test results to Codecov by @sbryngelson in #251
• Review of T. Chen's surrogate-BOED note by @sbryngelson in #252
• What the noise model and the allocation do to the design chapter by @sbryngelson in #250
• The #245 band is the sonic line, and max_wall_mach names it by @sbryngelson in #254
• Six changes to how the experiments are optimised by @sbryngelson in #255
• Point at the new results from the places that should be citing them by @sbryngelson in #256
• Five referee items: two conclusions narrowed, one defence tested, one new suspect by @sbryngelson in #257
• Fit how the records begin, with the stretch treated as unknown by @sbryngelson in #258
• The acquisition data arrives, and #222 turns out to be largely the clock by @sbryngelson in #259
• The clock the pipeline uses, and why one scale factor cannot fix the rebound by @sbryngelson in #260
• #222 explained: the clock has two hands, and aligning both removes the mode by @sbryngelson in #261
• Where R_max and R_eq come from, and what that eliminates by @sbryngelson in #262
• Not resolution and not replicates: the two datasets are not the same phenomenon by @sbryngelson in #263
• Design over the setpoint, because the geometry is not a knob by @sbryngelson in #264
• Price the control: account for the scatter, don't pay to remove it by @sbryngelson in #265
• Is the allocation gain the clock? Asked, and the answer is no by @sbryngelson in #266
• Make the design robust to the bias bound, which is the prior §measure lacked by @sbryngelson in #267
• The certificate is smaller than it looked, and the compound front is asymmetric by @sbryngelson in #268
• The material explanation, tested; and the setpoint widths, measured by @sbryngelson in #269
• Retitle and reorganise the design half; refine the certificate; audit the numbers and the prose by @sbryngelson in #271
• Answer the referee pass: state the noise constants, sweep them, and name the double penalty by @sbryngelson in #272
• Fix a factor of two in setpoint_design.py and update the numbers it fed by @sbryngelson in #278
• The density mismatch is real, is not the u0 cause, and unifying it fits worse (#273) by @sbryngelson in #280
• The clock is right; the medium density is a sink for a 5-9% collapse-time deficit (#273) by @sbryngelson in #281
• The depth bias and the collapse-time deficit are separate defects (#273, #274) by @sbryngelson in #282
• An argmin at a bracket edge is not an optimum (#273, #274) by @sbryngelson in #283
• The jittered fit target is real and does not explain the correlated residual (#274) by @sbryngelson in #284
• Restructure around the design work, and make the paper read straight by @sbryngelson in #285
• Alpha tracks whichever box wall binds, and it is never the data (#279) by @sbryngelson in #286
• Half the temperature trend is the vapour pressure this package leaves off (#277) by @sbryngelson in #287
• sec:thermal scores incomplete physics, and completing it costs eleven hours a cell (#277) by @sbryngelson in #288
• Say what gelatin is, and stop calling its viscosity trend unphysical by @sbryngelson in #289
• The vapour and mass-transfer model is not resolved at Nt=11, which is why its fits failed by @sbryngelson in #290
• Gelatin refuses time-temperature superposition, so the Arrhenius axis assumes what is false by @sbryngelson in #291
• Two rheological candidates refuted, and what that says about delta-hat by @sbryngelson in #292
• The writeup and its analysis move to a private papers repository; PyIMR keeps the machinery by @sbryngelson in #293
• Retire docs/open-work.md; what was live is now issues #294-#301 by @sbryngelson in #302
• A relative tolerance at a zero-valued optimum fails on the runs that search best by @sbryngelson in #304
• The docs describe the package as it was in July; bring them to what it is by @sbryngelson in #305
• Fix nightly CI: pyright error with numpy 2.5 stubs by @sbryngelson in #306
• Compare the guards by steps taken, not wall time by @sbryngelson in #307

Full Changelog: v0.1.1...v0.1.2