Day 3 of 78 · Run #1 · 2026-08-22
Basal mean-shift baseline — ceiling headroom probe
probe · mock dataDE gene recall 27% ceiling · Pearson Δ 26% ceiling · 2 audit warns
Audit & metrics DE gene recall 27% ceiling · Pearson Δ 26% ceiling · 2 audit warns
- !housekeeping_shiftHousekeeping genes shifted up to +2.10 log2FC (threshold ±1.0); peak ACTB.#ACTB #GAPDH💬 Challenge
- !pathway_coherencePathway 'interferon_response' shows mixed directionality (2 up, 2 down among measured genes).#ISG15 #IFIT1 #MX1 #OAS1💬 Challenge
Metrics vs ceiling (all scores)
Differential expression volcano plot (log2FC vs -log10 p)
Pre-registered hypotheses (2)
- Basal co-expression alone explains <20% of cross-context transfer on DESigGenesRecall
- Mean-shift baseline will sit below 50% of ceiling on pearson_delta
Evidence 6 genes · 77 entities · Tavily + Pioneer
Literature
6 genesTavily · auxiliary, not scored
ACTB2/3
- Decoding heterogeneous single-cell perturbation responses(h), and correctly estimated cells in low MOI (i) and high MOI (j). k, A representative gene (ACTB) where PS correctly e…
- Perturbation effect is not an on-off switchCRISPRi effects are inherently variable, not uniform: Knockdown effects span a spectrum, with many targets showing non-d…
- + 1 more source(s)
GAPDH2/2
- Isoform-specific single-cell perturb-seq reveals distinct functions of alternative promoters in drug responseTo implement our Isoform-Specific Perturb-Seq approach, we generated MCF-7 breast cancer cells stably expressing dCas9-C…
- CRISPR interferenceexpression. Cells were harvested 72 hours post-transfection, total RNA was isolated, and relative gene expression was me…
IFIT12/3
- IFIT1 is rapidly evolving and exhibits disparate antiviral activities across 11 mammalian orders | eLifea lid over the 3′ exit of the tunnel, and helps link subdomain II to subdomain III (Figure 2A–C; Abbas et al., 2017). L1…
- Perturb-Seq Combines CRISPR and Single-Cell RNA-SeqIn the UCSF-led study (“A Multiplexed Single-Cell CRISPR Screening Platform Enables Systematic Dissection of the Unfolde…
- + 1 more source(s)
ISG152/3
- ISG15 Gene - Ma'ayan Lab – Computational Systems BiologyISG15 in regulating immune responses, protein quality control, autophagy, and tumor biology, while linking cellular stre…
- CRISPR interferenceMultiplexed gene knockdown CRISPRi synthetic sgRNA offers a unique method for the simultaneous knockdown of multiple gen…
- + 1 more source(s)
MX12/2
- CRISPR interferenceCRISPR interference (CRISPRi) is a genetic perturbation technique that allows for sequence-specific repression of gene e…
- Perturb-seq – Enables Large-Scale Analysis of Complex Genetic Interactions Using CRISPR-Based Gene Perturbation and Single-Cell RNA Sequencing | RNA-Seq BlogIn the UCSF-led experiments, researchers used CRISPR-based transcriptional interference (CRISPRi) to simultaneously repr…
OAS12/3
- Gene - OAS1LINCS L1000 CMAP Chemical Perturbation Consensus Signatures small molecule perturbations changing expression of OAS1 gen…
- OAS1 Gene - 2'-5'-Oligoadenylate Synthetase 1 | Function, Mutations & Disease – EDITGENEOAS1 is a 363-amino acid protein (canonical isoform) that belongs to the 2-5A synthetase family. It contains a nucleotid…
- + 1 more source(s)
Field context
VCC / perturbation research- Arc Institute Launches Virtual Cell Challenge to Accelerate AI Model DevelopmentSTATE is a transformer-based model for predicting perturbation effects across sets of cells. [Arc Institute] #### Data-…
- NeurIPS Competition Single-cell perturbation prediction: generalizing experimental interventions to unseen contextsSingle-cell sequencing technologies have revolutionized our understanding of the heterogeneity and dynamics of cells and…
- The 2026 Virtual Cell Challenge: predicting perturbation responses in cell contexts a model has never seen | Arc InstituteStack — A single-cell foundation model that uses in-context learning to predict cellular responses to perturbations neve…
- + 3 more
Biomedical NER
77 entitiesPioneer regex fallback · 77 entities across 6 genes · 34 gene, 29 perturbation type, 8 pathway, 3 cell type
ACTB
GAPDH
IFIT1
ISG15
MX1
OAS1
Narrative digest run digest · traces to facts.json
> Fallback digest rendered deterministically from `facts.json` (no LLM call).
Full digest
Headline
Basal mean-shift baseline — ceiling headroom probe
Metrics
| metric | value | ceiling | |---|---|---| | DESigGenesRecall | 0.12 | 0.45 | | pearson_delta | 0.08 | 0.31 |
Provenance
- commit: `2fe7bd7f4f0902b454d0f2face73f917e6e065d5` - seed: 0 - code hash: `k001-mean-shift-v0` - hypotheses pre-registered: ['Basal co-expression alone explains <20% of cross-context transfer on DESigGenesRecall', 'Mean-shift baseline will sit below 50% of ceiling on pearson_delta']
Trust & provenance Self-tests + reproduce command
6 pipeline steps · sourced from committed artifacts only
Evaluated 4 cell-eval metrics vs 2 ceiling bounds · data_status=probe.
Ran 5 deterministic rules · 2 flags raised (housekeeping_shift, pathway_coherence).
TAVILY_API_KEY not set — enrichment not attempted.
Regex fallback extractor — 6 files enriched (no API key or all model calls failed).
Deterministic fallback digest (no API key or LLM call failed).
Passed: planted-signal 13/13 caught · holo-agent 4/4 passed · narrative-check 4/4 checks passed.
PASSPlanted-signal self-test13/13 caught · 13/13
We plant known-answer failures through our own audit rules. If it can't catch what we planted, it can't be trusted to catch what we didn't. 13/13 caught (13/13 cases).
python tools/planted_signal.py
PASSIndependent agent audit4/4 passed · 4/4
An autonomous agent (h/web-surfer-flash) browsed the live page and verified 4/4 passed (4/4 fields) against facts.json — the surface is honest, not just claimed.
python tools/holo_audit.py --run experiments/k001-mean-shift-baseline
PASSNarrative grounding check4/4 checks passed · 4/4
The LLM digest is grounded by construction: every number in it must trace back to facts.json, verified deterministically. 4/4 checks passed (4/4 checks).
python tools/check_narrative.py --run experiments/k001-mean-shift-baseline