https://kytosapp.netlify.app/ Public experiment tracker for Kytos: cell-eval metrics, biological audit flags, literature evidence, provenance, and VEED Fabric briefings — failure modes visible during the 2026 Virtual Cell Challenge. https://kytosapp.netlify.app/runs/k013-context-scale-validation/ https://kytosapp.netlify.app/runs/k013-context-scale-validation/ Fri, 18 Sep 2026 11:40:00 +0000 per-context delta_scale {A:1.7, B:2.65, C:0.75} scored +0.0312 (rank 560) — clean negative: redistributing the magnitude budget by lineage ratios lost vs uniform x1.7 (+0.0596) on every component https://kytosapp.netlify.app/runs/k011-delta-scale-x2p0-validation/ https://kytosapp.netlify.app/runs/k011-delta-scale-x2p0-validation/ Fri, 18 Sep 2026 09:44:00 +0000 delta_scale=2.0 — overall +0.0566 (rank 515); the scale curve has bent: pds best-yet 0.356 and fid ~0, but nmae -0.125 outweighs the gains. x1.7 (+0.0596) stands as champion; the scalar magnitude axis is now bracketed https://kytosapp.netlify.app/runs/k011-delta-scale-x1p7-validation/ https://kytosapp.netlify.app/runs/k011-delta-scale-x1p7-validation/ Thu, 17 Sep 2026 13:55:00 +0000 delta_scale=1.7 — overall +0.0596 (rank 486), new best; fid nearly closed (-0.006) and pds back to 0.339, but nmae cratered to -0.076: overscaled DE magnitudes are now the cost https://kytosapp.netlify.app/runs/k011-delta-scale-x1p3-validation/ https://kytosapp.netlify.app/runs/k011-delta-scale-x1p3-validation/ Thu, 17 Sep 2026 13:35:00 +0000 explicit delta_scale=1.3 on the champion config — overall +0.0559 (rank 502), new best score; fid improved to -0.031 but nmae flipped negative (-0.023) and pds slipped to 0.304 https://kytosapp.netlify.app/runs/k012-lineage-score/ https://kytosapp.netlify.app/runs/k012-lineage-score/ Thu, 17 Sep 2026 00:00:00 +0000 context identity resolved on discriminative genes: A is clearly Jurkat-like (r=0.65 vs next best 0.40), B weakly RPE1-leaning (0.37), C unresolved/hESC-leaning — global Pearson is inflated by housekeeping+assay similarity and misleads https://kytosapp.netlify.app/runs/k012-transfer-loo/ https://kytosapp.netlify.app/runs/k012-transfer-loo/ Thu, 17 Sep 2026 00:00:00 +0000 LOO eval on 47 K562/hESC paired deltas: raw transfer cosine is only ~0.13 — context transfer is fundamentally weak; low-rank map lifts sign accuracy to 0.60 but destroys magnitude (deP 0.10); fitted scalar s=0.44 says hESC deltas are WEAKER than K562, opposite the 2026-context direction https://kytosapp.netlify.app/runs/k010-mean-corrected-kd-s4p0-validation/ https://kytosapp.netlify.app/runs/k010-mean-corrected-kd-s4p0-validation/ Wed, 16 Sep 2026 07:22:34 +0000 k010 s4.0 scored overall +0.0122 (rank 564): nmae +0.019, pds 0.290, fid -0.292, reach 0.077 — marginal gain over s2.0 (+0.0090) but far below the uncorrected champion (+0.0511). The k009/k010 pair is now conclusive: at mean-1, no eta shape or spread tested recovers what the clipped trunc-normal's ~40% mean inflation delivered. The effective driver is applied delta magnitude — borrowed K562/Atlas signatures appear systematically weak in the 2026 contexts, and the 'bias' was compensating. Next honest lever: explicit delta-scale calibration (or context-conditioned Layer A scaling), not more eta-distribution search. https://kytosapp.netlify.app/runs/k010-mean-corrected-kd-s2p0-validation/ https://kytosapp.netlify.app/runs/k010-mean-corrected-kd-s2p0-validation/ Wed, 16 Sep 2026 06:44:01 +0000 k010 s2.0 scored overall +0.0090 (rank 565): nmae +0.004 -> +0.017 (mean correction worked), but pds 0.336 -> 0.288 and fid -0.111 -> -0.305. Removing the trunc-normal's clipped-mean inflation (eta mean 1.40 -> 1.00) cost more in fid/pds than it gained in nmae — the champion's apparent bias was load-bearing. Two readings: (a) the metric rewards dispersion magnitude per se, or (b) our borrowed deltas are systematically ~40% too weak and the inflation was compensating. s4.0 (mean 1.0, std 1.24) disambiguates: if it still loses, spread-at-mean-1 can't buy back what inflated mean provided, pointing at delta-scale calibration as the real lever. https://kytosapp.netlify.app/runs/k009-gamma-kd-s2p0-validation/ https://kytosapp.netlify.app/runs/k009-gamma-kd-s2p0-validation/ Tue, 15 Sep 2026 10:30:58 +0000 gamma s2.0 scored overall +0.0075 (rank 544): nmae +0.020 -> +0.022 (best yet), fid -0.272 -> -0.233, pds 0.204 -> 0.208, reach 0.065 -> 0.072. More spread helps directionally but the gamma A/B is conclusive: at matched kd_std=2.0 the mean-preserving gamma scores +0.0075 vs the biased trunc-normal's +0.0511. The leaderboard rewards symmetric dispersion centered on eta~1 — gamma's shape-0.25 density piles mass near eta~0 (unperturbed-looking cells) which directly costs discrimination. The nmae lesson is real though: remove the trunc-normal's mean inflation while KEEPING its symmetric shape — mean-corrected trunc-normal is the next variant. https://kytosapp.netlify.app/runs/k009-gamma-kd-s1p4-validation/ https://kytosapp.netlify.app/runs/k009-gamma-kd-s1p4-validation/ Tue, 15 Sep 2026 10:14:51 +0000 gamma s1.4 scored overall -0.0011 (rank 565): nmae +0.004 -> +0.020 (best yet — the mean-preserving fix worked as designed), but pds 0.336 -> 0.204 and fid -0.111 -> -0.272 regressed hard, reach 0.091 -> 0.065, jac -0.014 -> -0.024. Interpretation: at kd_std=1.4 the capped gamma has shape 0.51, concentrating ~40%+ of eta mass near zero — those cells are effectively unperturbed, which directly erodes discrimination (pds) and direction fidelity (fid). The trunc-normal keeps mass centered on eta~1, which is what the metric rewards. Net: mean-preservation helps nmae but the skewed shape costs more than the bias did. https://kytosapp.netlify.app/runs/k008-kd-s2p0-validation/ https://kytosapp.netlify.app/runs/k008-kd-s2p0-validation/ Sun, 13 Sep 2026 16:40:17 +0000 kd_std=2.0 scored overall +0.0511 (rank 462): fid -0.148 -> -0.111 (+0.037, down from +0.061/step), pds 0.326 -> 0.336, reach 0.087 -> 0.091, jac -0.016 -> -0.014 — all still improving but the curve is bending. nmae eroded +0.008 -> +0.004 for the second consecutive step: overspread is blurring the DE signal. The scalar-eta optimum sits around 2.0-2.3; further scalar pushes trade nmae for fid at a worsening rate. The right next move is per-target spread, not a bigger global value. https://kytosapp.netlify.app/runs/k008-kd-s1p7-validation/ https://kytosapp.netlify.app/runs/k008-kd-s1p7-validation/ Sun, 13 Sep 2026 16:39:48 +0000 kd_std=1.7 scored overall +0.0429 (rank 477): fid -0.209 -> -0.148 (+0.061, same step size as before), pds 0.311 -> 0.326, reach 0.079 -> 0.087, jac improving. nmae dipped +0.011 -> +0.008 — the first sign of a trade-off: overspread is starting to blur the DE log-FC signal. The optimum is near; 2.0 is the ceiling probe. https://kytosapp.netlify.app/runs/k008-kd-s1p3-validation/ https://kytosapp.netlify.app/runs/k008-kd-s1p3-validation/ Sun, 13 Sep 2026 15:56:44 +0000 kd_std=1.3 scored overall +0.0291 (rank 490): fid -0.270 -> -0.209, pds 0.297 -> 0.311, reach 0.073 -> 0.079, jac -0.019 -> -0.018; nmae flat (+0.011). fid gains per step: 0.4->0.7 gave +0.054, 0.7->1.0 gave +0.064, 1.0->1.3 gave +0.061 — still linear, not bending. The heavy right tail in measured eta (global std 2.32) explains why the optimum sits above the median; next probes 1.7 and ~2.0. https://kytosapp.netlify.app/runs/k008-kd-s1p0-validation/ https://kytosapp.netlify.app/runs/k008-kd-s1p0-validation/ Sat, 12 Sep 2026 16:45:00 +0000 kd_std=1.0 scored overall +0.0152 (rank 509): fid -0.334 -> -0.270 (largest gain yet), pds 0.282 -> 0.297, reach 0.067 -> 0.073, nmae +0.011. Every component improved again — the mechanism is not exhausted at the Atlas median. Note the measured eta distribution is heavy-tailed (global std 2.32, p75 2.51) so values >1 may still help; s1.3 submitted as the bracket point. https://kytosapp.netlify.app/runs/k008-kd-s0p7-validation/ https://kytosapp.netlify.app/runs/k008-kd-s0p7-validation/ Sat, 12 Sep 2026 16:00:26 +0000 Identical pipeline to k008 s0.4 except kd_std=0.7. Overall crossed zero: -0.0113 -> +0.0007, rank 519. fid -0.388 -> -0.334 (the largest single-run fid gain so far), nmae +0.007 -> +0.010, pds 0.272 -> 0.282, reach 0.062 -> 0.067. The scalar-eta mechanism has real headroom — the Atlas-measured spread (~1.1 median) suggests trying 1.0 next before concluding the mechanism is exhausted. https://kytosapp.netlify.app/runs/k008-kd-heterogeneity-validation/ https://kytosapp.netlify.app/runs/k008-kd-heterogeneity-validation/ Fri, 11 Sep 2026 20:54:06 +0000 k007 priors unchanged (dispatch real 816 / neighbor 54 / fallback 30); only the sampler changed: perturbed_i = basal_i + eta_i * delta + eps, eta ~ N(1, 0.4) clipped at 0, eps ~ N(0, 0.05). Overall -0.0159 -> -0.0113. fid improved modestly (-0.403 -> -0.388), nmae +0.002 -> +0.007, pds 0.269 -> 0.272. Hypothesis partially confirmed: scalar KD heterogeneity is real signal but small — it models spread along the delta axis only. The residual fid deficit likely lives in off-direction covariance, which needs the full Layer B sampler. https://kytosapp.netlify.app/runs/k007-neighbor-prior-validation/ https://kytosapp.netlify.app/runs/k007-neighbor-prior-validation/ Fri, 11 Sep 2026 19:53:48 +0000 The 28 fallback genes are genuinely unscreened (absent from K562 GWPS, K562 essential, and RPE1 screens; no alias recovery). Neighbor imputation from STRING partners recovered real signature shape for 18 of them (PSMB9->proteasome, TAF4->TFIID, MAPK7->MAP2K5/MEF2, MLKL->necroptosis). Score -0.0159 vs k006 -0.0210; nmae swung -0.074 -> +0.002; fid worsened slightly (-0.36 -> -0.40), sharpening the case that distribution shape — not signature means — is the deficit. https://kytosapp.netlify.app/runs/k006-replogle-prior-validation/ https://kytosapp.netlify.app/runs/k006-replogle-prior-validation/ Fri, 11 Sep 2026 06:16:22 +0000 Replogle K562 genome-wide Perturb-seq — real CRISPRi signatures for 272 of 300 targets. Score -0.021, rank 534, pds 0.265. The 28 targets outside Replogle still use the hand-tuned prior. https://kytosapp.netlify.app/runs/k005-atlas-prior-validation/ https://kytosapp.netlify.app/runs/k005-atlas-prior-validation/ Fri, 11 Sep 2026 01:51:42 +0000 Atlas-prior attempt — the 2025 validation set only overlaps 4 of 300 2026 targets. Built and verified but never submitted: coverage, not the prior, was the bottleneck. This run is why we went broad. https://kytosapp.netlify.app/runs/k004-layer-a-b-validation/ https://kytosapp.netlify.app/runs/k004-layer-a-b-validation/ Fri, 11 Sep 2026 00:15:44 +0000 First target-specific model — a context-conditioned knockdown prior with log1p transport over real control cells. pds went positive for the first time (0.002); score reached -0.149. https://kytosapp.netlify.app/runs/k004-real-resampling-validation/ https://kytosapp.netlify.app/runs/k004-real-resampling-validation/ Thu, 10 Sep 2026 22:32:27 +0000 Swapped the sparse slice for real single-cell resampling — 360k control cells with natural dispersion. Score jumped to -0.304: dispersion matters, but the targets were still indistinguishable. https://kytosapp.netlify.app/runs/k003-mean-shift-validation/ https://kytosapp.netlify.app/runs/k003-mean-shift-validation/ Sat, 05 Sep 2026 12:48:42 +0000 First real submission — a sparse top-300 mean-shift slice proved the pipeline end-to-end. Scored at the degenerate floor (-0.948) exactly as predicted: every target shares one control slice, so perturbation discrimination is zero. https://kytosapp.netlify.app/runs/k001-mean-shift-baseline/ https://kytosapp.netlify.app/runs/k001-mean-shift-baseline/ Sat, 22 Aug 2026 00:00:00 +0000 Basal mean-shift baseline — ceiling headroom probe https://kytosapp.netlify.app/runs/k002-vcc2025-validation-mean-shift/ https://kytosapp.netlify.app/runs/k002-vcc2025-validation-mean-shift/ Sat, 22 Aug 2026 00:00:00 +0000 First real cell-eval run — mean-shift floor vs VCC 2025 validation (H1 hESC, 50 targets)