ChatterjeeLab/KoopDoob-MDM2-P53

KoopDoob -- MDM2 / p53-MDM2 data package

0fa668bb6563f031732bd26347f2a153ac9abe18 | 2987 files | 4.49 GB | generated 2026-08-21T09:58:55+00:00

MDM2(25-109) unbiased MD, the p53(17-29)+MDM2 complex campaigns, the BioEmu MDM2 draws, and the committed artifacts that describe them.

Layout

path	role	files	size	what
md/mdm2_systems	production	1010	3.425 GB	MDM2(25-109) unbiased MD: 33 chemical systems x 2 frozen states, APO__ plus 50 MDM2PEP__ peptide complexes
md/mdm2_features	derived	132	0.188 GB	per-trajectory feature npz for the systems above
md/mdm2_features_scout	derived	156	0.229 GB	scout-run features (50 ns), the pre-production sizing set
md/mdm2_joint	production	128	0.026 GB	joint-state MD
md/mdm2_joint_logs	logs	64	0.000 GB	joint-run logs
md/mdm2_smoke	smoke	25	0.037 GB	MDM2 MD smoke -- sizing only, never an estimand
md/mdm2_test	test	9	0.026 GB	harness test runs
md/mdm2_buildtest	test	5	0.004 GB	system-build tests
md/p53mdm2	production	911	0.518 GB	p53(17-29)+MDM2(25-109) complex: pull replicas, sources/sources_v2, the shooting campaigns (v1-v4), the Doob campaigns, benches and throughput
bioemu/mdm2_raw	production	351	0.019 GB	BioEmu draws for MDM2_CORE_25_109 and MDM2_LID_1_118
artifacts/metadata	artifacts	153	15.9 MB	committed metadata/ files matching mdm2*, p53mdm2*, MDM2_*
artifacts/docs	artifacts	38	1.0 MB	committed docs/ files matching mdm2*, MDM2_*
artifacts/configs	artifacts	5	0.1 MB	committed configs/ files matching mdm2*

Read this before quoting a number

Every MDM2 round in this project ended without a token, and four were overturned by their own adversarial audits. The data here is the evidence for a ledger of negative results; it is not a set of validated measurements. Where a reconciliation or amendment exists it SUPERSEDES the report, and must be read first.

round	outcome	read first
mdm2_stage0	no token; the ITS ladder was a static between-trajectory offset at 89-90 % of the leading coordinate's variance	docs/mdm2_stage0_audit_reconciliation.md
mdm2_stage0_shared	no token; the M2->M0 Doob ceiling is 3.93x/step, BELOW the 5x target -- the third consecutive round with a ceiling error	docs/mdm2_stage0_audit_A.md
mdm2_complex_koopdoob	MDM2_COMPLEX_KOOPMAN_NOT_VALIDATED + C2 FAIL; discrete Doob VALIDATED	docs/mdm2_koopdoob_closure_complete_account.md
mdm2_koopdoob_engineering_closure	ENGINEERING: FULLY_REPAIRED on DEV; C2 passes 1 of 5 confirmatory lags, so MDM2_COMPLEX_C2_VALIDATED is withheld	docs/mdm2_koopdoob_closure_complete_account.md
mdm2_koopdoob_final_closure	MDM2_COMMITTOR_ROUTE_LIMITED_BY_COMPOSITION_DEPTH; neither committor token emitted, but the C1 positive control passes 4/4	docs/mdm2_closure_audit2_blocker.md
mdm2_koopdoob_one_shot_repair	MDM2_DOOB_LIMITED_BY_THE_DIFFERENTIABLE_CV_INTERFACE; the deployed control field was at or below a random direction	docs/MDM2_KOOPDOOB_ONE_SHOT_REPAIR.md (read its AMENDMENT banner first)

The repo's CLAUDE.md carries a DO-NOT-QUOTE table of claims that were published inside the project and then refuted. Several of them were derived from exactly these files.

Integrity

MANIFEST.csv lists every file with its SHA256. The package is hardlinked from the live trees, so a manifest hash is a hash of the real data. No PDF is present, checked over the assembled tree.