Repository navigation

PR for 0.5.0 release - #18

Merged

harmsm merged 230 commits into

harmslab:mainfrom

harmsm:main

Oct 25, 2026

Merged

PR for 0.5.0 release#18
harmsm merged 230 commits into
harmslab:mainfrom
harmsm:main

Commits on Sep 23, 2026
- Sample per-genotype latents at library size under autoguides
  
  committed
  9f7d562
- Add SVI guide selection (--guide_type) with numpyro autoguides
  
  harmsm
  and
  claude
  committed
  8db2c76
- Keep test-suite output files out of the working directory
  harmsm
  committed
  d118c5c
- Merge branch 'claude/gallant-lederberg-468b50'
  harmsm
  committed
  3561e08
- updated docs and reports
  harmsm
  committed
  7893bdf
- added "future" directory
  harmsm
  committed
  15ddddb
- Add tfs-summarize-calibration; fix summarize-fit globs
  
  harmsm
  and
  claude
  committed
  641f691
- Split a co-transformed cell's abundance among its plasmids
  
  harmsm
  and
  claude
  committed
  cf3726e
- Build co-transformed cells' growth from cell-level physics
  
  harmsm
  and
  claude
  committed
  abec95f
- Carry congression purity as GrowthData.bulk_fraction; record 3.0 study
  
  harmsm
  and
  claude
  committed
  fc0f2b9
- Rename future/ to planning/; add planning/studies
  
  harmsm
  and
  claude
  committed
  18f5ed8
- cleanup commit. files should have gone in with:
  harmsm
  committed
  7b59d75
- Fix batch-order bug in mut activity and _simple theta; add order check
  
  harmsm
  and
  claude
  committed
  d46af88
- Stop committing test reports and badges
  claude
  committed
  4315be2
- Merge pull request #1 from harmsm/claude/charming-ritchie-gebxj3
  harmsm
  authored
  cfa549b
- Let dk_geno and activity components return library-ordered values
  
  harmsm
  and
  claude
  committed
  6338ee4
- Merge branch 'main' of github.com:harmsm/tfscreen
  harmsm
  committed
  943d903
- Keep categorical_geno theta library-ordered (step 3.3a)
  
  harmsm
  and
  claude
  committed
  69105a2
- Draw fixed co-resident sets for the congression mixture (step 3.3b)
  
  harmsm
  and
  claude
  committed
  5b442fa
Commits on Sep 24, 2026
- Keep library-sized latents whole in chunked posterior forward passes
  
  harmsm
  and
  claude
  committed
  fbba986
- Mix clean and congressed cells at the observable level (step 3.3c)
  
  harmsm
  and
  claude
  committed
  2b50492
- Match congression weights to the simulator; retire Stage 1.5 (step 3.4)
  
  harmsm
  and
  claude
  committed
  c3354b8
- Fix growth prediction for library-file models; set up 3.5 grid
  
  harmsm
  and
  claude
  committed
  6e4735e
- Merge branch 'main' of github.com:harmslab/tfscreen
  harmsm
  committed
  87d9125
- Record the binary-transformant bench result (step 4 input)
  
  harmsm
  and
  claude
  committed
  1870553
- Record the 3.5 grid results and detection-floor diagnosis
  
  harmsm
  and
  claude
  committed
  ebcf5b1
- File per-sample offset and count-likelihood ideas
  
  harmsm
  and
  claude
  committed
  d2523b2
Commits on Sep 25, 2026
- Resolve nested file paths in tfs-setup-sim-grid
  
  harmsm
  and
  claude
  committed
  9665b99
- Merge pull request #2 from harmsm/claude/stoic-moser-869046
  harmsm
  authored
  f2de521
- Record masked results; set up binding-weight grid (step 3.5)
  
  harmsm
  and
  claude
  committed
  14f1bc4
- Make tfs-setup-sim-grid output directories movable
  
  harmsm
  and
  claude
  committed
  3098fa2
- Make tfs-setup-grid output directories movable
  
  harmsm
  and
  claude
  committed
  5b7789d
- Diagnose the unanchored k/dk_geno slide; set up anchored grid (3.5)
  
  harmsm
  and
  claude
  committed
  c7b228d
- Set up lambda-0 controls for the mixture anomaly (step 3.5)
  
  harmsm
  and
  claude
  committed
  2d86049
- cleaned up some spuriously added files
  harmsm
  committed
  c6b75d4
- Review step 3.5 fits: SVI unconverged; file convergence study
  
  harmsm
  and
  claude
  committed
  c5061de
- Merge branch 'claude/congression-calibration-debug-ad50bb'
  harmsm
  committed
  847edb0
- Integrate the convergence review; set up resume runs (step 3.5)
  
  harmsm
  and
  claude
  committed
  05870c9
- Count only designed codons as library sites
  
  harmsm
  and
  claude
  committed
  bff2c51
- Merge branch 'claude/great-leavitt-c8287b'
  harmsm
  committed
  e998f7a
- Stop SVI/MAP on noise-referenced convergence, cut step size on stalls
  
  harmsm
  and
  claude
  committed
  5bf5b78
- Merge branch 'claude/gallant-goldwasser-2329c8'
  harmsm
  committed
  c7cde80
- Record converged 3.5 fits: transient confirmed, lambda pulled up
  
  harmsm
  and
  claude
  committed
  56045e6
- Set up baseline and lambda-profile grids (step 3.5)
  
  harmsm
  and
  claude
  committed
  a2d3441
Commits on Sep 26, 2026
- Stop clipping Adam gradients; gate convergence on loss skew
  
  harmsm
  and
  claude
  committed
  9666951
- Record baseline and profile grids on the new convergence code (3.5)
  
  harmsm
  and
  claude
  committed
  985783d
- Merge branch 'claude/priceless-gauss-369bc1'
  harmsm
  committed
  501bc00
- Add a simulated Hill-coefficient limit; realistic study binding noise
  
  harmsm
  and
  claude
  committed
  ec47f11
- Record converged 3.5 re-run: mixture beats single under congression
  
  harmsm
  and
  claude
  committed
  f76254d
- Set up the 3.5 confirming round; record open items (step 3.5)
  
  harmsm
  and
  claude
  committed
  811018c
- Build a congressed cell's theta by partition function (step 4)
  
  harmsm
  and
  claude
  committed
  cdfd5db
- Record the 3.5 confirming round and close step 3.5
  
  harmsm
  and
  claude
  committed
  6696510
- Set up the step 4 cross-rule validation grid
  
  harmsm
  and
  claude
  committed
  510639b
- Add planning idea for wt-relative growth-only fitting
  
  harmsm
  and
  claude
  committed
  3f628bf
- Merge branch 'claude/growth-only-data-fitting-2882bf'
  harmsm
  committed
  9caf070
- Record the cross-rule grid: theta rule not identifiable (step 4)
  
  harmsm
  and
  claude
  committed
  3d60dfd
- Add soft-min congression dk rules and a sensitivity grid (step 5)
  
  harmsm
  and
  claude
  committed
  8254642
- Add an analysis roadmap; document the sampling protocol
  
  harmsm
  and
  claude
  committed
  b949095
- Record the dk-rule grid: rule irrelevant at realistic dk spread (5)
  
  harmsm
  and
  claude
  committed
  dc61879
- Add a post-count-likelihood re-validation step to the congression plan
  
  harmsm
  and
  claude
  committed
  55eb562
- Add step 0 study scripts; provisional first results
  
  harmsm
  and
  claude
  committed
  25638d3
- Merge branch 'main' into claude/physics-improvements-roadmap-03a273
  harmsm
  committed
  3f7fe49
- Allow growth-only models with no binding data (roadmap step 1)
  
  harmsm
  and
  claude
  committed
  98fb6db
Commits on Sep 27, 2026
- Add sampling noise and OD600 to the simulator (roadmap step 4)
  
  harmsm
  and
  claude
  committed
  590687d
- Fit read counts directly (roadmap step 7)
  
  harmsm
  and
  claude
  committed
  775f8b7
- Add a growth-only fit on a wt-relative scale (roadmap step 5)
  
  harmsm
  and
  claude
  committed
  5820c19
- Fix the pre-fit Hessian point and count-likelihood underflow
  
  harmsm
  and
  claude
  committed
  c65390a
- Pool stalled windows before cutting; overflow-safe horseshoe
  
  harmsm
  and
  claude
  committed
  740ebc1
- Record count-likelihood validation results (roadmap step 7)
  
  harmsm
  and
  claude
  committed
  1ef283d
- Default tfs-configure-model to the count likelihood
  
  harmsm
  and
  claude
  committed
  2620a5b
- Add tfs-calibrate-od600 (roadmap step 3)
  
  harmsm
  and
  claude
  committed
  7fe75ed
- Start SVI at the pre-MAP point with narrow guide scales
  
  harmsm
  and
  claude
  committed
  c57d9ee
Commits on Sep 28, 2026
- Add low-rank and MAP+Laplace arms to the relative-fit grid
  
  harmsm
  and
  claude
  committed
  e29c739
- Fix NUTS start; record step 5 and the SVI-overconfidence pilot
  
  harmsm
  and
  claude
  committed
  8e244cf
- Non-centered ln_cfu0 tube offset, runaway-loss stop, SVI study results
  
  harmsm
  and
  claude
  committed
  32a7ec1
- Refuse factored ln_cfu0 across libraries; rerun grids with hierarchical
  
  harmsm
  and
  claude
  committed
  722393c
- Record reruns with ln_cfu0 hierarchical (count-likelihood, relative-fit, SVI)
  
  harmsm
  and
  claude
  committed
  5103d40
Commits on Sep 29, 2026
- Add a two-stage fit that carries k/m uncertainty to genotypes
  
  harmsm
  and
  claude
  committed
  beaab7b
- adding gitignore to make sure pilot data does not end up in repo
  harmsm
  committed
  f21e9cc
- Unfreeze SVI hyper scales; floor Laplace at prior curvature
  
  harmsm
  and
  claude
  committed
  9db2828
- Stop clipping simulated binding observations to [0, 1]
  
  harmsm
  and
  claude
  committed
  6e2903d
Commits on Sep 30, 2026
- Fix ln_cfu0 extraction; sampling flags for full-library fits
  
  harmsm
  and
  claude
  committed
  079680b
- Add sigma_fixed to level tube offsets
  
  harmsm
  and
  claude
  committed
  d451faa
- Add tfs-fit-model --init_from to start at an earlier MAP point
  
  harmsm
  and
  claude
  committed
  faa5b11
Commits on Oct 1, 2026
- Allow per-condition m prior scales in linear growth
  
  harmsm
  and
  claude
  committed
  e4320c7
- Add a fixed log(hill_n) population SD to the Hill theta components
  
  harmsm
  and
  claude
  committed
  beacad0
Commits on Oct 2, 2026
- Flag the relative-fit guide verdict; set up a post-fix rerun
  
  harmsm
  and
  claude
  committed
  0e220e7
- Rerun the relative-fit guide comparison after the scale and Laplace fixes
  
  harmsm
  and
  claude
  committed
  f762b13
- Add a per-genotype Laplace for full-library uncertainty
  
  harmsm
  and
  claude
  committed
  cd208d1
- Add shared-parameter uncertainty to the block Laplace; hold Hill SDs
  
  harmsm
  and
  claude
  committed
  a874ac3
Commits on Oct 3, 2026
- Close pre-merge gaps: study commits, Laplace script, stale docs
  
  harmsm
  and
  claude
  committed
  41eaddd
Commits on Oct 4, 2026
- Add contributing guide, PR template and CI coverage check
  harmsm
  committed
  6502972
- Track the real-data fit scripts; plan the experiment pipeline
  
  harmsm
  and
  claude
  committed
  c5de9f2
- Merge branch 'claude/physics-improvements-roadmap-03a273'
  harmsm
  committed
  8c7e780
- Keep test_nan_explosion working under numpyro 0.22
  
  harmsm
  and
  claude
  committed
  be2f90f
- Clean up the CLI interface and docs for 0.5.0
  
  harmsm
  and
  claude
  committed
  caeefdc
- small fixes:
  harmsm
  committed
  bb4f17b
- Stage the level-offset MAP inside tfs-fit-model
  
  harmsm
  and
  claude
  committed
  8a7abbb
- Judge MAP convergence windows by the exact full-batch loss
  
  harmsm
  and
  claude
  committed
  6671763
Commits on Oct 5, 2026
- Record the offset-mode finding; re-plan the pipeline's last steps
  
  harmsm
  and
  claude
  committed
  27109e2
- Write the simulated experiment in the raw formats real data come in
  
  harmsm
  and
  claude
  committed
  9ebaa4a
- Set up the full-size simulation of the dev-data design
  
  harmsm
  and
  claude
  committed
  bd837ba
- Flag offset-mode fits in tfs-summarize-fit; write the real-data recipe
  
  harmsm
  and
  claude
  committed
  47b02e6
Commits on Oct 7, 2026
- Keep the simulated wt on its reference curve under hill_geno
  
  harmsm
  and
  claude
  committed
  8399fb6
Commits on Oct 8, 2026
- Record the recipe and full-size simulation runs; close step 7
  
  harmsm
  and
  claude
  committed
  bf4a182
- Calibrate the simulator's read depth to the real screen
  
  harmsm
  and
  claude
  committed
  554ef44
- Fix --set_priors under pandas 3; run CI tests in parallel
  
  harmsm
  and
  claude
  committed
  2287207
- plateau counter
  harmsm
  committed
  b79ffdc
Commits on Oct 9, 2026
- Cover every line the PR changes; drop a duplicate __main__ block
  
  harmsm
  and
  claude
  committed
  c3c5482
- Calibrate the simulator's read depth to the real screen
  
  harmsm
  and
  claude
  committed
  222d147
- Plan honest intervals for the deep genotypes; set up the k/m anchor arms
  
  harmsm
  and
  claude
  committed
  85c354e
- Record the k/m anchor results; set up the m-bias factorial
  
  harmsm
  and
  claude
  committed
  34b8143
Commits on Oct 10, 2026
- Start the PnnC structure MLPs from a random W1 so they can learn
  
  harmsm
  and
  claude
  committed
  2bdce17
- Carry learned model-side params into posteriors, predictions and hand-offs
  
  harmsm
  and
  claude
  committed
  845b0c9
- Write the thermodynamic models' full simulation truth and score it
  
  harmsm
  and
  claude
  committed
  46f13b6
- Simulate binding data for the thermodynamic theta models
  
  harmsm
  and
  claude
  committed
  11de71f
- Project ddG_H onto K_U in the U1 PddG and PnnC models
  
  harmsm
  and
  claude
  committed
  b4e5994
- Merge branch 'claude/pnnc-nn-init' into claude/thermo-sim-truth
  harmsm
  committed
  434ea81
- Record the m-bias factorial's first round; set up fit-side subsets
  
  harmsm
  and
  claude
  committed
  1a0c300
- Read titrant concentrations in mM for every thermodynamic theta model
  
  harmsm
  and
  claude
  committed
  9f25202
- Check units, columns and numbering of the thermo structural inputs
  
  harmsm
  and
  claude
  committed
  845821e
- Keep the fitted mutation index when predicting on genotype subsets
  
  harmsm
  and
  claude
  committed
  d94bc8a
- Extract and score the per-structure ddG of the PddG and PnnC models
  
  harmsm
  and
  claude
  committed
  cf99296
- Record the m-bias subsets: the doubles steepen m; set up a real arm
  
  harmsm
  and
  claude
  committed
  7fcef80
- Record the real no-doubles fit; set up the held-k/m library fit
  
  harmsm
  and
  claude
  committed
  fed6391
- Use the standard MWC dimer in both thermodynamic theta families
  
  harmsm
  and
  claude
  committed
  09e4f7e
- Widen the M3 effector selectivity and reject unknown priors rows
  
  harmsm
  and
  claude
  committed
  4c14099
- Record m-bias round 3 (held k/m fixes X); set up kanR-only holds
  
  harmsm
  and
  claude
  committed
  75611b0
- Add ddG_truth_sd to set the SD of the simulated per-structure ddG
  
  harmsm
  and
  claude
  committed
  1bbae40
Commits on Oct 11, 2026
- Pin linear growth's k and m per condition; record m-bias round 4
  
  harmsm
  and
  claude
  committed
  58b83b2
- Record m-bias round 5: pheS's free k reopens the k/dk_geno slide
  
  harmsm
  and
  claude
  committed
  629ed2a
- Let tfs-simulate subsample the library it simulates
  
  harmsm
  and
  claude
  committed
  7984d7e
- Drop undetectable-OD600 tubes from the direct growth table too
  
  harmsm
  and
  claude
  committed
  c714103
- Merge branch 'claude/sim-subsample' into claude/thermo-sim-truth
  
  harmsm
  and
  claude
  committed
  410ac6a
- Make the per-structure ddG scale a centered hierarchical prior
  
  harmsm
  and
  claude
  committed
  c4f03d4
- Merge branch 'claude/sigma-s-centered' into claude/thermo-sim-truth
  
  harmsm
  and
  claude
  committed
  fca6106
- Compute Laplace prior SDs at the MAP and fail loudly without them
  
  harmsm
  and
  claude
  committed
  acc6973
- Put the structural ddG prior on the constants
  
  harmsm
  and
  claude
  committed
  f8fa461
- Merge branch 'claude/ddg-constant-prior' into claude/thermo-sim-truth
  
  harmsm
  and
  claude
  committed
  f53d24d
- Compute MWC theta in log space so extreme constants stay finite
  
  harmsm
  and
  claude
  committed
  89845dd
- Merge branch 'claude/laplace-numerics' into claude/thermo-sim-truth
  
  harmsm
  and
  claude
  committed
  f39a471
- Regenerate the CLI reference
  
  harmsm
  and
  claude
  committed
  f4669f8
- Record m-bias rounds 5-6: pin all four; real X stretches 1.47
  
  harmsm
  and
  claude
  committed
  b8be0d2
- Set up step 2 (k/m draws) and a second simulation seed
  
  harmsm
  and
  claude
  committed
  128044e
- Add a held per-structure ddG scale to the PddG and PnnC models
  
  harmsm
  and
  claude
  committed
  e6035ba
- Merge branch 'claude/ddg-sigma-fixed' into claude/thermo-sim-truth
  
  harmsm
  and
  claude
  committed
  b3864eb
- Record the stage-1 m bias across seeds; set up a no-congression test
  
  harmsm
  and
  claude
  committed
  187e4f9
- Find congression behind stage 1's pheS bias; set up a spikes-only stage 1
  
  harmsm
  and
  claude
  committed
  e712f77
- Record the spikes-only stage 1; set up a tight-control real rerun
  
  harmsm
  and
  claude
  committed
  2e382c0
- Record the spikes-only stage 1: it works on the simulation, not real data
  
  harmsm
  and
  claude
  committed
  f2e8918
Commits on Oct 12, 2026
- Record option 2: deep singles bring congression's pheS bias back
  
  harmsm
  and
  claude
  committed
  aee1e6d
- Separate the doubles' pull from congression; set up a mixture stage 1
  
  harmsm
  and
  claude
  committed
  5a77214
- Model congression in stage 1: pheS bias gone; set up stage 2 tests
  
  harmsm
  and
  claude
  committed
  332e99a
- Record the first stage 2 after a mixture stage 1; finish mixture fits
  
  harmsm
  and
  claude
  committed
  72bb96d
- Settle the two-stage recipe; set up the real mixture stage 1
  
  harmsm
  and
  claude
  committed
  95d4a57
- Record the real mixture stage 1: pheS gap is not congression
  
  harmsm
  and
  claude
  committed
  d38d549
- Record the real data's stage 2 under the full two-stage recipe
  
  harmsm
  and
  claude
  committed
  a11ffde
Commits on Oct 13, 2026
- Plan the intervals and pipeline; set up the kanR-only check and jackknife
  
  harmsm
  and
  claude
  committed
  d574e13
- kanR-only check, seed-1 jacknife
  
  harmsm
  and
  claude
  committed
  639012a
- Add a control-growth check for a library-specific dk_geno
  
  harmsm
  and
  claude
  committed
  881cb78
- Carry stage 1's k uncertainty into X: deep coverage near nominal
  
  harmsm
  and
  claude
  committed
  1e79e4b
- Plan the two-stage fit; add its functions and the shared X shift
  
  harmsm
  and
  claude
  committed
  a8abff6
- Add tfs-fit-two-stage: the two-stage relative fit as one command
  
  harmsm
  and
  claude
  committed
  4923375
- Add tfs-fit-two-stage: the two-stage relative fit as one command
  
  harmsm
  and
  claude
  committed
  3ce04cc
Commits on Oct 14, 2026
- Add a k-shift control refit to the two-stage validation
  
  harmsm
  and
  claude
  committed
  4f52e68
Commits on Oct 15, 2026
- Add a one-library stage 2 run to the two-stage validation
  
  harmsm
  and
  claude
  committed
  9477a57
- Record the two-stage validation results: the k shift fails
  
  harmsm
  and
  claude
  committed
  7d01721
- Let hill_relative hold wt's curve; queue a pheS-only fit with it held
  
  harmsm
  and
  claude
  committed
  31e3362
- Record the wt-held pheS fit: the pheS compression is real
  
  harmsm
  and
  claude
  committed
  b665567
- Add a titrant-dependent k to linear growth; test it in stage 1
  
  harmsm
  and
  claude
  committed
  f28eb10
- Make the titrant-dependent k jit-safe; add the stage 1 comparison
  
  harmsm
  and
  claude
  committed
  dc9862d
- Record the titrant-dependent k test: log-linear only, partial
  
  harmsm
  and
  claude
  committed
  aea7f6f
- Drop the free titrant k; check stage 1 against the within-tube data
  
  harmsm
  and
  claude
  committed
  d806051
- Propose taking m from relative information only
  
  harmsm
  and
  claude
  committed
  aff855d
- Set up relative stage 1; read k and m from a MAP when pinned
  
  harmsm
  and
  claude
  committed
  ad25c6b
- Record relative stage 1: totals are not the cause; pheS is dropped
  
  harmsm
  and
  claude
  committed
  e5501ab
Commits on Oct 16, 2026
- Add the per-marker midpoint test: the markers differ in steepness
  
  harmsm
  and
  claude
  committed
  6aa9ae4
- Trace the 0-IPTG pheS anomaly to one row of tubes; add tube QC
  
  harmsm
  and
  claude
  committed
  3d5fbc4
- Add a per-condition warp of theta to linear growth; set up the test
  
  harmsm
  and
  claude
  committed
  d540edf
- Add tfs-check-tubes: flag tubes whose composition disagrees with label
  
  harmsm
  and
  claude
  committed
  76ce30d
- Merge branch 'main' into claude/thermo-sim-truth
  harmsm
  committed
  8be1bd1
- Merge branch 'claude/amazing-mcnulty-a57ada'
  harmsm
  committed
  e9eb8ee
- Add tfs-correct-tubes: documented corrections to a tube table
  
  harmsm
  and
  claude
  committed
  f15f5b4
- Merge branch 'claude/agitated-mcnulty-2fbe65'
  harmsm
  committed
  feb092b
- Record the warp runs: kanR's warp helps, pheS's gap remains
  
  harmsm
  and
  claude
  committed
  a167070
- Add the likelihood attribution for stage 1 fits
  
  harmsm
  and
  claude
  committed
  ba1752b
- Record the likelihood attribution: wt's own rows resist the within-tube m
  
  harmsm
  and
  claude
  committed
  9a0a2f8
- Record wt's trajectory against the library: the scale of X is the issue
  
  harmsm
  and
  claude
  committed
  b86c0ec
- Remove a stray conflict marker from a configuration_io test
  
  harmsm
  and
  claude
  committed
  643145d
- Merge claude/thermo-sim-truth: thermodynamic simulation truth and fixes
  
  harmsm
  and
  claude
  committed
  7a76d8b
Commits on Oct 17, 2026
- Plan the return to a joint fit of growth and binding
  
  harmsm
  and
  claude
  committed
  8eac5fe
- Add tfs-process-binding: lab binding table to a binding_df
  
  harmsm
  and
  claude
  committed
  d614b11
- Default binding_weight to 1: each binding observation counts once
  
  harmsm
  and
  claude
  committed
  45ffb7f
- Add a binding-to-growth titrant scale s to hill_geno
  
  harmsm
  and
  claude
  committed
  be172c2
- Add genotype_offset binding noise: one offset per binding genotype
  
  harmsm
  and
  claude
  committed
  a408a63
- Test the joint fit's model pieces together; extract the warp
  
  harmsm
  and
  claude
  committed
  47f3f9c
- Simulate the joint fit's pieces; set up the joint stage 1 study
  
  harmsm
  and
  claude
  committed
  5224a2a
- Record joint-fit simulation run 1: pieces work; null finds a wrong mode
  
  harmsm
  and
  claude
  committed
  ba949f3
- Joint-fit simulation run 2: the null fit was trapped; start from plain
  
  harmsm
  and
  claude
  committed
  9cf28ae
- Joint-fit simulation run 3: start from the plain fit recovers both arms
  
  harmsm
  and
  claude
  committed
  1981197
- Joint-fit simulation: set up the pinned-truth run; jobs on gpu, 6 h
  
  harmsm
  and
  claude
  committed
  3505230
- Joint-fit simulation run 4: the steep m sits in pheS's slow growers
  
  harmsm
  and
  claude
  committed
  776838c
Commits on Oct 18, 2026
- Joint-fit simulation run 5: half the m bias is MAP shrinkage of theta
  
  harmsm
  and
  claude
  committed
  cb71c73
- Hold hill_geno's population SDs; set up joint-fit simulation run 6
  
  harmsm
  and
  claude
  committed
  c0d38db
- Joint-fit simulation run 6: the m bias is an affine squeeze of theta
  
  harmsm
  and
  claude
  committed
  c137646
- Joint-fit simulation: set up run 7 (weaker pooling, more binding)
  
  harmsm
  and
  claude
  committed
  333aaba
- Joint-fit simulation run 7: binding singles are the lever on m
  
  harmsm
  and
  claude
  committed
  1a3d3d4
- Joint-fit simulation: set up run 8 (binding singles with congression)
  
  harmsm
  and
  claude
  committed
  5f45e87
- Joint-fit simulation run 8: congression is now the largest m bias
  
  harmsm
  and
  claude
  committed
  f40e234
- Add tfs-configure-model --congression_sets; set up run 9
  
  harmsm
  and
  claude
  committed
  e2dda3b
- Joint-fit simulation run 9: more co-resident draws help, then plateau
  
  harmsm
  and
  claude
  committed
  34fd22d
- Joint-fit simulation: record theta so far; set up run 10
  
  harmsm
  and
  claude
  committed
  43c2126
- Joint-fit simulation run 10: theta error falls as 1/sqrt(binding singles)
  
  harmsm
  and
  claude
  committed
  dff58cd
- Set up the joint stage 1 on the real screen (joint-fit plan step 6)
  
  harmsm
  and
  claude
  committed
  a76a657
- Add a findings ledger; record joint-fit-real run 1; pause the joint plan
  
  harmsm
  and
  claude
  committed
  bf319f8
Commits on Oct 19, 2026
- x-vs-binding run 2: M42I's inverted order survives the two-stage X
  
  harmsm
  and
  claude
  committed
  f36df1d
- Add the 0.5.0 close-out checklist; drop the reversed pheS row
  
  harmsm
  and
  claude
  committed
  76881c5
- Keep the two-stage k-shift refits at stage 2's MAP
  
  harmsm
  and
  claude
  committed
  d8161bc
- Release checklist: keep the release at 0.5.0
  
  harmsm
  and
  claude
  committed
  37dd8d1
- Refuse experimental pieces in tfs-fit-two-stage; apply tube corrections
  
  harmsm
  and
  claude
  committed
  9639145
Commits on Oct 20, 2026
- Add tfs-fit-two-stage's summary step: one results table, tfs_x.csv
  
  harmsm
  and
  claude
  committed
  6fd184a
- Fix --shared_shift under pandas 3; test with copy-on-write locally
  
  harmsm
  and
  claude
  committed
  ccd5544
- Add the "Analyzing a screen" docs page and its SLURM example
  
  harmsm
  and
  claude
  committed
  796a2f0
- Point CLAUDE.md and the pipeline page at the release route
  
  harmsm
  and
  claude
  committed
  f30c2ef
- Ship the two-stage fit without the k shift (validation run 2)
  
  harmsm
  and
  claude
  committed
  4396cbb
- Floor the dependencies, add a cuda12 extra and the 0.5.0 constraints
  
  harmsm
  and
  claude
  committed
  3fe6941
- Require Python 3.12, test 3.12-3.14; drop the constraints file
  
  harmsm
  and
  claude
  committed
  7109137
Commits on Oct 25, 2026
- Screen-analysis docs: make the growth table step by step
  
  harmsm
  and
  claude
  committed
  68b958f
- Fold the CHANGELOG into 0.5.0
  
  harmsm
  and
  claude
  committed
  8eb4af9
- Cover every changed line for diff-cover; fix a ΔΔG loader crash
  
  harmsm
  and
  claude
  committed
  d5e636a

Commits

Commits on Sep 23, 2026

Commits on Sep 24, 2026

Commits on Sep 25, 2026

Commits on Sep 26, 2026

Commits on Sep 27, 2026

Commits on Sep 28, 2026

Commits on Sep 29, 2026

Commits on Sep 30, 2026

Commits on Oct 1, 2026

Commits on Oct 2, 2026

Commits on Oct 3, 2026

Commits on Oct 4, 2026

Commits on Oct 5, 2026

Commits on Oct 7, 2026

Commits on Oct 8, 2026

Commits on Oct 9, 2026

Commits on Oct 10, 2026