{
 "release_id": "gi-promoter-atlas-designs-2026-08-05",
 "why_this_file_exists": "The pairs that were attempted and dropped are published with the measurements that stopped them. These runs happened, they are real records, and they are not part of the shipped set.",
 "not_shipped": [
  {
   "id": "A4",
   "pair_id": "A4_cns_neuron__vs__hepatocyte",
   "on_cell": "CNS neuron",
   "off_cell": "Hepatocyte (liver parenchymal cell)",
   "status": "attempted, not shipped",
   "designs_generated": 128,
   "designs_shipped": 0,
   "one_line": "Both candidate design environments fail, and they fail differently. FAM47C's admission for this pair is carried by one control, SYT1, by 0.0014 log units; the AAVS1_plus candidates collapse to a median 0.802 when re-scored in a real gene TSS window. 0 shipped.",
   "environments": [
    {
     "scaffold": "FAM47C",
     "run_id": "A4_FAM47C_margin",
     "admitted": true,
     "why_not_shipped": [
      "The scaffold's admission for this pair is carried by exactly one neuronal control promoter, SYT1, and by 0.0014 log units: its margin is 0.8014 against a floor of 0.8. Every other member of the eight-member pre-named neuronal panel fails its own floor in this scaffold.",
      "That single carrier is also A4 seed 7. An environment admitted by the optimiser's own starting material is not an independent test of the environment.",
      "The unchanged selection rule removed all 64 designs at the off-arm criterion: every one of them is louder in hepatocyte than this run's own random-DNA negatives.",
      "A4 is the first pair whose designs are positively rank-correlated between its two arms -- the ON side and the OFF side of the same on/off comparison (Spearman ON/OFF +0.503): pushing the neuronal (ON) arm up pulls the hepatocyte (OFF) arm up with it. Concretely, the 64 designs sit at +0.0069 .. +0.0693 in hepatocyte against this run's negatives' OFF max of -0.0091. For A1 and A2 the off arm is flat and below the negatives."
     ],
     "off_arm_note": "The A4 seeds themselves span -0.034 .. +0.019 in hepatocyte, so as with A1 and A2 there was no off-target gradient to descend -- but unlike A1 and A2, the designs ended up louder in the off-target cell than random DNA is.",
     "best_natural_in_this_scaffold": {
      "symbol": "NCAM2",
      "margin": 2.016
     },
     "best_design_margin": 4.2512,
     "admission_carried_by": [
      "syt1__region_600bp"
     ],
     "on_standard_panel": [
      {
       "variant_label": "hsyn1__published_cassette",
       "on_score": 0.6758,
       "off_score": -0.0168,
       "margin": 0.6926,
       "margin_floor": 0.8,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "hsyn1__region_600bp",
       "on_score": 0.707,
       "off_score": -0.0081,
       "margin": 0.7151,
       "margin_floor": 0.8,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "stmn2__region_600bp",
       "on_score": 0.0669,
       "off_score": -0.0354,
       "margin": 0.1023,
       "margin_floor": 0.8,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "syt1__region_600bp",
       "on_score": 0.7852,
       "off_score": -0.0162,
       "margin": 0.8014,
       "margin_floor": 0.8,
       "margin_verdict": "reportable",
       "carries_admission": true
      },
      {
       "variant_label": "eno2__region_600bp",
       "on_score": 0.2422,
       "off_score": -0.0198,
       "margin": 0.262,
       "margin_floor": 0.8,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "tubb3__region_600bp",
       "on_score": 0.4004,
       "off_score": -0.0015,
       "margin": 0.4019,
       "margin_floor": 0.8,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "snap25__region_600bp",
       "on_score": 0.0518,
       "off_score": -0.0052,
       "margin": 0.057,
       "margin_floor": 0.8,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "rbfox3__region_600bp",
       "on_score": 0.2256,
       "off_score": -0.0245,
       "margin": 0.2501,
       "margin_floor": 0.8,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      }
     ],
     "selection_steps": [
      {
       "step": "S1",
       "rule": "drop unmodified natural seeds",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S2",
       "rule": "banded per-context margin gate (grade.margin_verdict == reportable), unmodified",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S3",
       "rule": "ON score >= pre-derived in-scaffold on-target floor 0.0518",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S4",
       "rule": "OFF score <= max OFF of this run's own composition-matched negatives (-0.0091)",
       "kept": 0,
       "dropped": 64
      },
      {
       "step": "S5",
       "rule": "greedy by descending margin, minimum pairwise edit distance 60 bp between selected designs (uncapped: this is the run's diversity ceiling)",
       "kept": 0,
       "dropped": 0
      },
      {
       "step": "S6",
       "rule": "take the first 50",
       "kept": 0,
       "shortfall": 50,
       "diversity_ceiling": 0
      }
     ],
     "designs_selected": 0
    },
    {
     "scaffold": "AAVS1_plus",
     "run_id": "A4_AAVS1_margin",
     "admitted": true,
     "why_not_shipped": [
      "The admission is properly carried here -- ENO2 clears its own floor -- but the environment itself is loud. The EMPTY scaffold, with no design in it at all, scores 0.5039 on the neuronal arm, against 0.0024 for the empty FAM47C scaffold. Signal that the design did not put there is signal we cannot attribute to the design.",
      "Re-scored in FAM47C -- a real gene TSS window -- the AAVS1_plus designs collapse from 6.249 / 6.331 / 6.970 (min / median / max where they were evolved) to 0.153 / 0.802 / 3.146. The median 0.802 is below FAM47C's best NATURAL neuronal promoter (NCAM2, +2.016) and the worst is indistinguishable from the noise floor. The design-over-nature ratio falls 3.088x -> 1.561x on that re-read. A design whose advantage does not survive a change of environment is not a design we can put in front of a wet lab.",
      "A safe-harbour integration site is also further out of distribution than a gene TSS window: the insert sits at an integration coordinate, not at a transcription start site, and the cost of that has not been measured."
     ],
     "empty_scaffold_on_score": 0.5039,
     "empty_scaffold_off_score": 0.0151,
     "best_design_margin": 6.9696,
     "ratio_in_this_scaffold": 3.088,
     "ratio_when_re_read_in_a_gene_tss_window": 1.561,
     "on_standard_panel": [
      {
       "variant_label": "hsyn1__published_cassette",
       "on_score": 0.7656,
       "off_score": -0.0154,
       "margin": 0.781,
       "margin_floor": 0.8,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "hsyn1__region_600bp",
       "on_score": 1.0703,
       "off_score": -0.0195,
       "margin": 1.0898,
       "margin_floor": 1.25,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "stmn2__region_600bp",
       "on_score": 0.6172,
       "off_score": 0.0035,
       "margin": 0.6137,
       "margin_floor": 0.8,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "syt1__region_600bp",
       "on_score": 1.0391,
       "off_score": -0.0012,
       "margin": 1.0403,
       "margin_floor": 1.25,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "eno2__region_600bp",
       "on_score": 1.4453,
       "off_score": -0.0175,
       "margin": 1.4628,
       "margin_floor": 1.25,
       "margin_verdict": "reportable",
       "carries_admission": true
      },
      {
       "variant_label": "tubb3__region_600bp",
       "on_score": 2.2969,
       "off_score": 0.0249,
       "margin": 2.272,
       "margin_floor": 1.25,
       "margin_verdict": "reportable",
       "carries_admission": true
      },
      {
       "variant_label": "snap25__region_600bp",
       "on_score": 1.0,
       "off_score": 0.0708,
       "margin": 0.9292,
       "margin_floor": 1.25,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "rbfox3__region_600bp",
       "on_score": 0.5508,
       "off_score": -0.0292,
       "margin": 0.58,
       "margin_floor": 0.8,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      }
     ],
     "selection_steps": [
      {
       "step": "S1",
       "rule": "drop unmodified natural seeds",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S2",
       "rule": "banded per-context margin gate (grade.margin_verdict == reportable), unmodified",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S3",
       "rule": "ON score >= pre-derived in-scaffold on-target floor 0.5508",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S4",
       "rule": "OFF score <= max OFF of this run's own composition-matched negatives (0.0309)",
       "kept": 45,
       "dropped": 19
      },
      {
       "step": "S5",
       "rule": "greedy by descending margin, minimum pairwise edit distance 60 bp between selected designs (uncapped: this is the run's diversity ceiling)",
       "kept": 34,
       "dropped": 11
      },
      {
       "step": "S6",
       "rule": "take the first 50",
       "kept": 34,
       "shortfall": 16,
       "diversity_ceiling": 34
      }
     ],
     "designs_selected": 34
    }
   ]
  },
  {
   "id": "A5",
   "pair_id": "A5_cns_neuron__vs__astrocyte",
   "on_cell": "CNS neuron",
   "off_cell": "Astrocyte",
   "status": "attempted, not shipped",
   "designs_generated": 64,
   "designs_shipped": 0,
   "one_line": "The only environment of twelve that admitted this pair is a safe-harbour integration site, and the margin the run produced there does not exist anywhere else. Same shape as A4/AAVS1_plus, worse.",
   "attempted_2026_08_14": true,
   "never_call_this_validated": "CLAUDE.md fact 6 records A5 as `not demonstrated` in the held-out-only re-ranking, and nothing in this run changes that. Both arms are also ontology subtype narrowings of their registry cell types -- thalamic neuron for `cns_neuron`, striatal astrocyte for `astrocyte`.",
   "environments": [
    {
     "scaffold": "Keppel19_minus",
     "run_id": "A5_Keppel19_margin",
     "admitted": true,
     "why_not_shipped": [
      "THE MARGIN DOES NOT LEAVE THIS SCAFFOLD. Re-scored in FAM47C, the scaffold of record, the same 64 designs fall from a margin span of 3.812 / 3.961 / 4.508 (min / median / max where they were evolved) to -0.005 / 0.090 / 0.546. Every one of them is below the 0.80 reportable threshold in a real gene TSS window.",
      "The ORDERING does not survive either: the Spearman correlation of the margin ranking across the two environments is 0.0724. A4/AAVS1_plus at least kept its ordering while its magnitudes collapsed; this run keeps neither. The margin is a property of the Keppel19 integration coordinate, not of the designs.",
      "HOLLOW ADMISSION. The admission is carried by exactly one member of the pre-named eight-member neuronal panel, ENO2. ENO2 is at least not one of A5's own seeds and is a held-out gene, which is why this run happened at all -- but the environment is admitted at the strength of one promoter and no more.",
      "NO GENE-TSS SCAFFOLD ADMITS A5. In FAM47C, FOXB2, OR4C13 and GSX1 the ON panel fails criterion C2 -- including hSYN1, the field's standard neuron-restricted promoter, whose neuron-over-astrocyte margin in FAM47C is 0.4619 as a published cassette and 0.4902 as a genomic window, both `not_a_finding`. On this checkpoint the field's own neuronal standard does not separate neuron from astrocyte at reportable resolution.",
      "A safe-harbour integration site is further out of distribution than a gene TSS window: the insert sits at an integration coordinate, not at a transcription start site."
     ],
     "mechanism_note": "CLAUDE.md fact 7's claim rule, re-checked for this pair. A5's off arm was NOT quiet at generation zero (astrocyte median 0.734, max 2.031), so this is neither the A1/A2 case nor the off-arm-seeded descent. Over 40 generations the ON median rose 1.367 -> 5.359 while the OFF median rose 0.734 -> 1.379 and then flattened. The margin comes from differential gain, not from silencing. Nothing here taught the model to repress the astrocyte arm.",
     "empty_scaffold_on_score": 0.5117,
     "empty_scaffold_off_score": 0.3984,
     "best_design_margin": 4.5079,
     "best_natural_in_this_scaffold": {
      "symbol": "THY1",
      "margin": 2.3204
     },
     "ratio_in_this_scaffold": 1.943,
     "margin_rank_spearman_vs_fam47c": 0.0724,
     "same_best_design_in_both_scaffolds": false,
     "admission_carried_by": [
      "eno2__region_600bp"
     ],
     "on_standard_panel": [
      {
       "variant_label": "hsyn1__published_cassette",
       "on_score": 1.3594,
       "off_score": 0.7891,
       "margin": 0.5703,
       "margin_floor": 1.25,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "hsyn1__region_600bp",
       "on_score": 1.3984,
       "off_score": 0.7852,
       "margin": 0.6132,
       "margin_floor": 1.25,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "stmn2__region_600bp",
       "on_score": 0.9297,
       "off_score": 0.7656,
       "margin": 0.1641,
       "margin_floor": 0.8,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "syt1__region_600bp",
       "on_score": 1.8672,
       "off_score": 0.8203,
       "margin": 1.0469,
       "margin_floor": 1.25,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "eno2__region_600bp",
       "on_score": 2.4531,
       "off_score": 0.9961,
       "margin": 1.457,
       "margin_floor": 1.25,
       "margin_verdict": "reportable",
       "carries_admission": true
      },
      {
       "variant_label": "tubb3__region_600bp",
       "on_score": 2.1406,
       "off_score": 0.8984,
       "margin": 1.2422,
       "margin_floor": 1.25,
       "margin_verdict": "provisional_within_noise",
       "carries_admission": false
      },
      {
       "variant_label": "snap25__region_600bp",
       "on_score": 1.2266,
       "off_score": 0.6445,
       "margin": 0.5821,
       "margin_floor": 1.25,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      },
      {
       "variant_label": "rbfox3__region_600bp",
       "on_score": 1.5391,
       "off_score": 0.8438,
       "margin": 0.6953,
       "margin_floor": 1.25,
       "margin_verdict": "not_a_finding",
       "carries_admission": false
      }
     ],
     "selection_steps": [
      {
       "step": "S1",
       "rule": "drop unmodified natural seeds",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S2",
       "rule": "banded per-context margin gate (grade.margin_verdict == reportable), unmodified",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S3",
       "rule": "ON score >= pre-derived in-scaffold on-target floor 0.5938",
       "kept": 64,
       "dropped": 0
      },
      {
       "step": "S4",
       "rule": "OFF score <= max OFF of this run's own composition-matched negatives (1.625)",
       "kept": 57,
       "dropped": 7
      },
      {
       "step": "S5",
       "rule": "greedy by descending margin, minimum pairwise edit distance 60 bp between selected designs (uncapped: this is the run's diversity ceiling)",
       "kept": 45,
       "dropped": 12
      },
      {
       "step": "S6",
       "rule": "take the first 50",
       "kept": 45,
       "shortfall": 5,
       "diversity_ceiling": 45
      }
     ],
     "designs_selected": 45
    }
   ]
  },
  {
   "id": "CCR5-A1 / CCR5-A2",
   "pair_id": "A1 and A2, re-run in the CCR5site_plus safe-harbour scaffold",
   "status": "superseded",
   "designs_shipped": 0,
   "one_line": "Higher raw margins, WORSE design-over-nature ratio in both pairs (A1 1.351x -> 1.168x, A2 2.308x -> 1.256x): the scaffold lifted the natural baseline more than it lifted the designs, so the ratio and not the raw margin is the quantity.",
   "measurements": [
    "A1 design-over-nature ratio 1.351x in FAM47C -> 1.168x in CCR5site_plus, even though the best raw design margin rose from 6.3228 to 8.2705: the scaffold lifted the natural baseline (4.6801 -> 7.0806) more than it lifted the designs.",
    "A2 ratio 2.308x -> 1.256x on the same comparison.",
    "The CCR5site_plus designs collapse when re-read in a real gene TSS window, and the scaffold's off arm degrades.",
    "FAM47C remains the scaffold of record. The CCR5 selections are kept as valid records of real runs and are not shipped or quoted."
   ]
  }
 ]
}