Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
{
"decision_id": "p2-date-grounding-dev",
"plan": "P2",
"stage": "dev",
"eligible_for_default": false,
"eligibility_note": "Dev verdicts come from development data and never set a default. A held-out verdict comes from the custodian after a committed preregistration.",
"verdict_type": "quality",
"overall": "inconclusive",
"decided_at": "2026-10-04T22:20:28.151Z",
"sources": [
{
"source": "locomo-dev-temporal-facts",
"verdict": "inconclusive",
"arms": {
"baseline": {
"package": "gbrain",
"declared_pin": "github:garrytan/gbrain#739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"declared_sha": "739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"version": "0.60.48.0",
"package_sha256": "5f9974dcaa7b13d1077ff6fdad3c24fd2b975cb4536efb4693663d0e4bc3c84c",
"loaded_git_head": "5bd9e84978126689b11cf65107c0a9aa0e91e11b"
},
"candidate": {
"package": "gbrain",
"declared_pin": "github:garrytan/gbrain#739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"declared_sha": "739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"version": "0.60.48.0",
"package_sha256": "7adf6c1f0d9df80f9bd2f5f6c75a32a83d330b9b8377ca86f11590a322dcc2c1",
"loaded_git_head": "5d5375d438da490ad4d865fa93e3218a57e6bbdd"
}
},
"family": {
"family_id": "p2-date-grounding-dev:locomo-dev-temporal-facts",
"verdict": "inconclusive",
"reasons": [
"temporal-qa-from-facts: inconclusive"
],
"alpha": 0.05,
"holm_family": [
"facts-unresolved-relative",
"temporal-qa-from-facts",
"recall-all-5"
],
"comparisons": [
{
"id": "facts-unresolved-relative",
"metric": "facts_unresolved_share",
"gate": "superiority",
"status": "pass",
"reasons": [],
"n_pairs": 100,
"n_clusters": 3,
"n_excluded": 0,
"stats": {
"n_pairs": 100,
"n_clusters": 3,
"mean_a": 0.06163688425139521,
"mean_b": 0.012235578453962004,
"delta": -0.0494013057974333,
"se_cluster": 0.011790246180836328,
"se_iid": 0.0017730644807014318,
"design_effect": 44.21775478915512,
"effective_n": 2.261534998256539,
"ci95": [
-0.0733333333333333,
-0.027287195563057618
],
"p_two_sided": 0.25,
"p_greater": 1,
"p_less": 0.125,
"p_method": "exact-sign-flip"
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 0.033031379426796154,
"min_one_sided_discordant": null,
"note": "with 100 pairs in 3 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 0.0330; the observed delta is at least that size; clustering inflates variance by 44.22x (effective n about 2.3 of 100)"
},
"direction": "lower",
"p_holm": 0.00019998000199980003,
"min_effect": 0,
"p_superiority": 0.00009999000099990002
},
{
"id": "temporal-qa-from-facts",
"metric": "qa_score",
"gate": "superiority",
"status": "inconclusive",
"reasons": [
"McNemar treats items as independent; the clustered sign-flip p-value is the primary test here",
"an improvement larger than 0 is not shown (Holm p = 0.2603 > 0.05); this is not evidence of a loss"
],
"n_pairs": 100,
"n_clusters": 3,
"n_excluded": 0,
"stats": {
"n_pairs": 100,
"n_clusters": 3,
"mean_a": 0.64,
"mean_b": 0.66,
"delta": 0.02,
"se_cluster": 0.040320714279387465,
"se_iid": 0.03171845844395038,
"design_effect": 1.615966265060239,
"effective_n": 61.88247995156771,
"ci95": [
-0.058823529411764705,
0.08333333333333333
],
"p_two_sided": 1,
"p_greater": 0.5,
"p_less": 0.875,
"p_method": "exact-sign-flip"
},
"mcnemar": {
"losses": 4,
"wins": 6,
"both_pass": 60,
"both_fail": 30,
"discordant": 10,
"p_two_sided": 0.75390625
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 0.11296191713847759,
"min_one_sided_discordant": 6,
"note": "with 100 pairs in 3 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 0.1130; the observed delta is below that size, so a null result here is not evidence of no effect; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05; clustering inflates variance by 1.62x (effective n about 61.9 of 100)"
},
"direction": "higher",
"p_holm": 0.2602739726027397,
"min_effect": 0,
"p_superiority": 0.2602739726027397
},
{
"id": "temporal-qa-from-facts-by-question",
"metric": "qa_score",
"gate": "exploratory",
"status": "report_only",
"reasons": [],
"n_pairs": 100,
"n_clusters": 100,
"n_excluded": 0,
"stats": {
"n_pairs": 100,
"n_clusters": 100,
"mean_a": 0.64,
"mean_b": 0.66,
"delta": 0.02,
"se_cluster": 0.03171845844395039,
"se_iid": 0.03171845844395038,
"design_effect": 1.0000000000000004,
"effective_n": 99.99999999999996,
"ci95": [
-0.04,
0.08
],
"p_two_sided": 0.7508249175082492,
"p_greater": 0.3808619138086191,
"p_less": 0.8235176482351765,
"p_method": "monte-carlo-sign-flip"
},
"mcnemar": {
"losses": 4,
"wins": 6,
"both_pass": 60,
"both_fail": 30,
"discordant": 10,
"p_two_sided": 0.75390625
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 0.08886196434117832,
"min_one_sided_discordant": 6,
"note": "with 100 pairs in 100 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 0.0889; the observed delta is below that size, so a null result here is not evidence of no effect; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05"
},
"direction": "higher"
},
{
"id": "recall-all-5",
"metric": "recall_all_at_5",
"gate": "noninferiority",
"status": "pass",
"reasons": [
"McNemar treats items as independent; the clustered sign-flip p-value is the primary test here"
],
"n_pairs": 100,
"n_clusters": 3,
"n_excluded": 0,
"stats": {
"n_pairs": 100,
"n_clusters": 3,
"mean_a": 0.79,
"mean_b": 0.79,
"delta": 0,
"se_cluster": 0,
"se_iid": 0,
"design_effect": null,
"effective_n": null,
"ci95": [
0,
0
],
"p_two_sided": 1,
"p_greater": 1,
"p_less": 1,
"p_method": "degenerate"
},
"mcnemar": {
"losses": 0,
"wins": 0,
"both_pass": 79,
"both_fail": 21,
"discordant": 0,
"p_two_sided": 1
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": null,
"min_one_sided_discordant": 6,
"note": "no paired variation was observed, so the detectable effect cannot be estimated from this data; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05"
},
"direction": "higher",
"p_holm": 0,
"tolerance": 0.01,
"p_noninferiority": 0
},
{
"id": "facts-count",
"metric": "facts_count",
"gate": "exploratory",
"status": "report_only",
"reasons": [],
"n_pairs": 100,
"n_clusters": 3,
"n_excluded": 0,
"stats": {
"n_pairs": 100,
"n_clusters": 3,
"mean_a": 315.64,
"mean_b": 313.6,
"delta": -2.04,
"se_cluster": 2.044483074031184,
"se_iid": 0.28565713714171415,
"design_effect": 51.22439999999994,
"effective_n": 1.9521946572336644,
"ci95": [
-6,
0
],
"p_two_sided": 1,
"p_greater": 1,
"p_less": 0.5,
"p_method": "exact-sign-flip"
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 5.727793560388261,
"min_one_sided_discordant": null,
"note": "with 100 pairs in 3 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 5.7278; the observed delta is below that size, so a null result here is not evidence of no effect; clustering inflates variance by 51.22x (effective n about 2.0 of 100)"
},
"direction": "higher"
}
]
}
},
{
"source": "locomo-dev-nontemporal-facts",
"verdict": "inconclusive",
"arms": {
"baseline": {
"package": "gbrain",
"declared_pin": "github:garrytan/gbrain#739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"declared_sha": "739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"version": "0.60.48.0",
"package_sha256": "5f9974dcaa7b13d1077ff6fdad3c24fd2b975cb4536efb4693663d0e4bc3c84c",
"loaded_git_head": "5bd9e84978126689b11cf65107c0a9aa0e91e11b"
},
"candidate": {
"package": "gbrain",
"declared_pin": "github:garrytan/gbrain#739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"declared_sha": "739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"version": "0.60.48.0",
"package_sha256": "7adf6c1f0d9df80f9bd2f5f6c75a32a83d330b9b8377ca86f11590a322dcc2c1",
"loaded_git_head": "5d5375d438da490ad4d865fa93e3218a57e6bbdd"
}
},
"family": {
"family_id": "p2-date-grounding-dev:locomo-dev-nontemporal-facts",
"verdict": "inconclusive",
"reasons": [
"nontemporal-qa-from-facts: inconclusive"
],
"alpha": 0.05,
"holm_family": [
"nontemporal-qa-from-facts",
"recall-all-5"
],
"comparisons": [
{
"id": "nontemporal-qa-from-facts",
"metric": "qa_score",
"gate": "noninferiority",
"status": "inconclusive",
"reasons": [
"McNemar treats items as independent; the clustered sign-flip p-value is the primary test here",
"non-inferiority within 0.01 not shown (Holm p = 0.3732 > 0.05); this is not evidence of a regression or of equivalence"
],
"n_pairs": 150,
"n_clusters": 3,
"n_excluded": 0,
"stats": {
"n_pairs": 150,
"n_clusters": 3,
"mean_a": 0.52,
"mean_b": 0.5133333333333333,
"delta": -0.006666666666666667,
"se_cluster": 0.02866273874898546,
"se_iid": 0.02757404782305726,
"design_effect": 1.0805237609520058,
"effective_n": 138.8215654488163,
"ci95": [
-0.05555555555555555,
0.046511627906976744
],
"p_two_sided": 1,
"p_greater": 0.75,
"p_less": 0.5,
"p_method": "exact-sign-flip"
},
"mcnemar": {
"losses": 9,
"wins": 8,
"both_pass": 69,
"both_fail": 64,
"discordant": 17,
"p_two_sided": 1
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 0.08030110521082548,
"min_one_sided_discordant": 6,
"note": "with 150 pairs in 3 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 0.0803; the observed delta is below that size, so a null result here is not evidence of no effect; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05; clustering inflates variance by 1.08x (effective n about 138.8 of 150)"
},
"direction": "higher",
"p_holm": 0.3731626837316268,
"tolerance": 0.01,
"p_noninferiority": 0.3731626837316268
},
{
"id": "nontemporal-qa-by-question",
"metric": "qa_score",
"gate": "exploratory",
"status": "report_only",
"reasons": [],
"n_pairs": 150,
"n_clusters": 150,
"n_excluded": 0,
"stats": {
"n_pairs": 150,
"n_clusters": 150,
"mean_a": 0.52,
"mean_b": 0.5133333333333333,
"delta": -0.006666666666666667,
"se_cluster": 0.027574047823057266,
"se_iid": 0.02757404782305726,
"design_effect": 1.0000000000000004,
"effective_n": 149.99999999999994,
"ci95": [
-0.06,
0.04666666666666667
],
"p_two_sided": 1,
"p_greater": 0.6942305769423057,
"p_less": 0.4968503149685031,
"p_method": "monte-carlo-sign-flip"
},
"mcnemar": {
"losses": 9,
"wins": 8,
"both_pass": 69,
"both_fail": 64,
"discordant": 17,
"p_two_sided": 1
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 0.0772510448048524,
"min_one_sided_discordant": 6,
"note": "with 150 pairs in 150 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 0.0773; the observed delta is below that size, so a null result here is not evidence of no effect; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05"
},
"direction": "higher"
},
{
"id": "facts-count",
"metric": "facts_count",
"gate": "exploratory",
"status": "report_only",
"reasons": [],
"n_pairs": 150,
"n_clusters": 3,
"n_excluded": 0,
"stats": {
"n_pairs": 150,
"n_clusters": 3,
"mean_a": 312.14666666666665,
"mean_b": 314.28,
"delta": 2.1333333333333333,
"se_cluster": 1.7499177546811018,
"se_iid": 0.20325465159004416,
"design_effect": 74.12322185318905,
"effective_n": 2.023657313454278,
"ci95": [
-1,
5
],
"p_two_sided": 0.5,
"p_greater": 0.25,
"p_less": 0.875,
"p_method": "exact-sign-flip"
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 4.90254371571218,
"min_one_sided_discordant": null,
"note": "with 150 pairs in 3 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 4.9025; the observed delta is below that size, so a null result here is not evidence of no effect; clustering inflates variance by 74.12x (effective n about 2.0 of 150)"
},
"direction": "higher"
},
{
"id": "recall-all-5",
"metric": "recall_all_at_5",
"gate": "noninferiority",
"status": "pass",
"reasons": [
"McNemar treats items as independent; the clustered sign-flip p-value is the primary test here"
],
"n_pairs": 150,
"n_clusters": 3,
"n_excluded": 0,
"stats": {
"n_pairs": 150,
"n_clusters": 3,
"mean_a": 0.6066666666666667,
"mean_b": 0.6066666666666667,
"delta": 0,
"se_cluster": 0,
"se_iid": 0,
"design_effect": null,
"effective_n": null,
"ci95": [
0,
0
],
"p_two_sided": 1,
"p_greater": 1,
"p_less": 1,
"p_method": "degenerate"
},
"mcnemar": {
"losses": 0,
"wins": 0,
"both_pass": 91,
"both_fail": 59,
"discordant": 0,
"p_two_sided": 1
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": null,
"min_one_sided_discordant": 6,
"note": "no paired variation was observed, so the detectable effect cannot be estimated from this data; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05"
},
"direction": "higher",
"p_holm": 0,
"tolerance": 0.01,
"p_noninferiority": 0
}
]
}
},
{
"source": "lme-s-dev-temporal-facts",
"verdict": "inconclusive",
"arms": {
"baseline": {
"package": "gbrain",
"declared_pin": "github:garrytan/gbrain#739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"declared_sha": "739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"version": "0.60.48.0",
"package_sha256": "5f9974dcaa7b13d1077ff6fdad3c24fd2b975cb4536efb4693663d0e4bc3c84c",
"loaded_git_head": "5bd9e84978126689b11cf65107c0a9aa0e91e11b"
},
"candidate": {
"package": "gbrain",
"declared_pin": "github:garrytan/gbrain#739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"declared_sha": "739e5cc89ca43b9b9351f0f203c7b12a7c0c571c",
"version": "0.60.48.0",
"package_sha256": "7adf6c1f0d9df80f9bd2f5f6c75a32a83d330b9b8377ca86f11590a322dcc2c1",
"loaded_git_head": "5d5375d438da490ad4d865fa93e3218a57e6bbdd"
}
},
"family": {
"family_id": "p2-date-grounding-dev:lme-s-dev-temporal-facts",
"verdict": "inconclusive",
"reasons": [
"temporal-qa-from-facts: inconclusive"
],
"alpha": 0.05,
"holm_family": [
"facts-unresolved-relative",
"temporal-qa-from-facts",
"recall-all-5"
],
"comparisons": [
{
"id": "facts-unresolved-relative",
"metric": "facts_unresolved_share",
"gate": "superiority",
"status": "pass",
"reasons": [],
"n_pairs": 11,
"n_clusters": 11,
"n_excluded": 1,
"stats": {
"n_pairs": 11,
"n_clusters": 11,
"mean_a": 0.05460439068683047,
"mean_b": 0.016964034601117436,
"delta": -0.03764035608571304,
"se_cluster": 0.004495448790958246,
"se_iid": 0.004495448790958245,
"design_effect": 1.0000000000000004,
"effective_n": 10.999999999999995,
"ci95": [
-0.04584398442519193,
-0.028906612246604926
],
"p_two_sided": 0.0009765625,
"p_greater": 1,
"p_less": 0.00048828125,
"p_method": "exact-sign-flip"
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 0.012594382884831402,
"min_one_sided_discordant": null,
"note": "with 11 pairs in 11 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 0.0126; the observed delta is at least that size"
},
"direction": "lower",
"p_holm": 0.00019998000199980003,
"min_effect": 0,
"p_superiority": 0.00009999000099990002
},
{
"id": "temporal-qa-from-facts",
"metric": "qa_score",
"gate": "superiority",
"status": "inconclusive",
"reasons": [
"an improvement larger than 0 is not shown (Holm p = 1.0000 > 0.05); this is not evidence of a loss"
],
"n_pairs": 11,
"n_clusters": 11,
"n_excluded": 1,
"stats": {
"n_pairs": 11,
"n_clusters": 11,
"mean_a": 0.5454545454545454,
"mean_b": 0.5454545454545454,
"delta": 0,
"se_cluster": 0,
"se_iid": 0,
"design_effect": null,
"effective_n": null,
"ci95": [
0,
0
],
"p_two_sided": 1,
"p_greater": 1,
"p_less": 1,
"p_method": "degenerate"
},
"mcnemar": {
"losses": 0,
"wins": 0,
"both_pass": 6,
"both_fail": 5,
"discordant": 0,
"p_two_sided": 1
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": null,
"min_one_sided_discordant": 6,
"note": "no paired variation was observed, so the detectable effect cannot be estimated from this data; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05"
},
"direction": "higher",
"p_holm": 1,
"min_effect": 0,
"p_superiority": 1
},
{
"id": "temporal-qa-from-facts-by-question",
"metric": "qa_score",
"gate": "exploratory",
"status": "report_only",
"reasons": [],
"n_pairs": 11,
"n_clusters": 11,
"n_excluded": 1,
"stats": {
"n_pairs": 11,
"n_clusters": 11,
"mean_a": 0.5454545454545454,
"mean_b": 0.5454545454545454,
"delta": 0,
"se_cluster": 0,
"se_iid": 0,
"design_effect": null,
"effective_n": null,
"ci95": [
0,
0
],
"p_two_sided": 1,
"p_greater": 1,
"p_less": 1,
"p_method": "degenerate"
},
"mcnemar": {
"losses": 0,
"wins": 0,
"both_pass": 6,
"both_fail": 5,
"discordant": 0,
"p_two_sided": 1
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": null,
"min_one_sided_discordant": 6,
"note": "no paired variation was observed, so the detectable effect cannot be estimated from this data; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05"
},
"direction": "higher"
},
{
"id": "recall-all-5",
"metric": "recall_all_at_5",
"gate": "noninferiority",
"status": "pass",
"reasons": [],
"n_pairs": 11,
"n_clusters": 11,
"n_excluded": 1,
"stats": {
"n_pairs": 11,
"n_clusters": 11,
"mean_a": 0.8181818181818182,
"mean_b": 0.8181818181818182,
"delta": 0,
"se_cluster": 0,
"se_iid": 0,
"design_effect": null,
"effective_n": null,
"ci95": [
0,
0
],
"p_two_sided": 1,
"p_greater": 1,
"p_less": 1,
"p_method": "degenerate"
},
"mcnemar": {
"losses": 0,
"wins": 0,
"both_pass": 9,
"both_fail": 2,
"discordant": 0,
"p_two_sided": 1
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": null,
"min_one_sided_discordant": 6,
"note": "no paired variation was observed, so the detectable effect cannot be estimated from this data; an exact McNemar test needs at least 6 wins with zero losses to reach 0.05"
},
"direction": "higher",
"p_holm": 0,
"tolerance": 0.01,
"p_noninferiority": 0
},
{
"id": "facts-count",
"metric": "facts_count",
"gate": "exploratory",
"status": "report_only",
"reasons": [],
"n_pairs": 11,
"n_clusters": 11,
"n_excluded": 1,
"stats": {
"n_pairs": 11,
"n_clusters": 11,
"mean_a": 225.8181818181818,
"mean_b": 223.45454545454547,
"delta": -2.3636363636363638,
"se_cluster": 1.7229611258883109,
"se_iid": 1.7229611258883106,
"design_effect": 1.0000000000000004,
"effective_n": 10.999999999999995,
"ci95": [
-5.7272727272727275,
0.5454545454545454
],
"p_two_sided": 0.2294921875,
"p_greater": 0.9072265625,
"p_less": 0.11474609375,
"p_method": "exact-sign-flip"
},
"power": {
"alpha": 0.05,
"power": 0.8,
"mde": 4.827022422936356,
"min_one_sided_discordant": null,
"note": "with 11 pairs in 11 clusters, a two-sided 0.05 test has 80% power for an absolute change of about 4.8270; the observed delta is below that size, so a null result here is not evidence of no effect"
},
"direction": "higher"
}
]
}
}
]
}