{ "acceptance_version": "0.1", "object_path": "examples/mpb-xlsx", "manifest": "agent/tools.json", "published_results_path": "results.json", "notes": "Every published unit reproduces at the defaults, including all eight bootstrap CIs, which requires replaying the published run's single random stream (see stream_offset in tools.json). Off-reference calls draw 2,000 resamples per bootstrap block against the published 10,000 (the 99% interval and the one-sided p reuse the published block), so the suite runs in under two minutes; rho, p and n do not depend on resampling, and reseeded-bootstrap-off-reference measures what it does to CI endpoints (about 0.002). The off-reference calls qualify the object more than they confirm it. The 2010 association's negative sign depends on financial costs being deducted from profit: with profit before financial costs it reverses to rho = +0.053 (p = 0.0016), and claim 3's year-by-year replication splits by firm panel (2008 stays negative; 2009 and 2010 turn positive). The workbook's column labelled EBIT cannot show this, because it is struck after financial costs (profit-columns-identity). Total debt combines components whose 2010 associations with ROA have opposite signs (long-term -0.116, short-term +0.038); long-term debt stays negative before financial costs (-0.038) and under 5/95% trimming, while the total-debt estimate loses significance under 5/95% trimming. The workbook is two firm panels whose associations differ about threefold, so the strong-2008, weak-2010 pattern tracks the panel at least as much as the year. With net profit, the negative sign survives a revenue denominator (the shared-Assets concern), a Holm family of all seven reported tests, and excluding the 2009b batch from the firm means; the firm-level aggregation stays negative before financial costs too, at under a third of its published size. Literal values were produced by calling these tools, never typed.", "calls": [ { "call_id": "source-md5-published", "tool": "dataset_summary", "args": {}, "select": "/observed_md5", "expect": { "kind": "published_pointer", "result_path": "/meta/observed_md5" }, "recipe_id": "mpb-financials-source-md5", "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "sample-profile-published", "tool": "dataset_summary", "args": {}, "select": "/per_year", "expect": { "kind": "published_pointer", "result_path": "/sample_profile/per_year" }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "total-rows-published", "tool": "dataset_summary", "args": {}, "select": "/total_rows", "expect": { "kind": "published_pointer", "result_path": "/meta/total_rows" }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "nan-year-rows-published", "tool": "dataset_summary", "args": {}, "select": "/nan_year_rows", "expect": { "kind": "published_pointer", "result_path": "/meta/nan_year_rows" }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "krs-overlap-panels", "tool": "dataset_summary", "args": {}, "select": "/krs_overlap", "expect": { "kind": "literal", "value": { "2008": { "2008": 3264, "2009": 554, "2009b": 3224, "2010": 561 }, "2009": { "2008": 554, "2009": 2947, "2009b": 548, "2010": 2926 }, "2009b": { "2008": 3224, "2009": 548, "2009b": 3244, "2010": 555 }, "2010": { "2008": 561, "2009": 2926, "2009b": 555, "2010": 2969 } } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The workbook is two firm panels, not four cross-sections. By registry number, 3,224 of the 3,244 firms labelled 2009b also appear in 2008, and 2,926 of the 2,969 labelled 2010 also appear in 2009; between the panels the overlap is 548 to 561 firms. The narrative's basis for keeping 2009b apart (few KRS matches with 2009) is correct, but 2009b is the 2008 firms observed again. The published estimates line up by panel, not by year: 2008 and 2009b give rho = -0.212 and -0.200, 2009 and 2010 give -0.068 and -0.058. This does not change claim 2 (2010 is weak), but the contrast between a strong 2008 and a weak 2010, which the Usage notes read as a feature of the most recent year, is also a contrast between panels: the 2010 panel's own 2009 is just as weak." }, { "call_id": "profit-columns-identity", "tool": "dataset_summary", "args": {}, "select": "/profit_columns", "expect": { "kind": "literal", "value": { "rows": 14421, "share_labelled_ebit_eq_labelled_pre_tax_less_financial_costs": 0.9753137785174398, "share_labelled_pre_tax_eq_labelled_ebit_less_net_financial_costs": 0.00672630192080993, "net_profit_ratio": { "labelled_ebit": { "rows_net_and_column_positive": 11469, "median_net_to_column": 0.8057544514737894, "rows_net_financial_costs_at_least_half": 1971, "median_net_to_column_there": 0.7883329604965349 }, "labelled_pre_tax": { "rows_net_and_column_positive": 11565, "median_net_to_column": 0.7253852432183683, "rows_net_financial_costs_at_least_half": 1084, "median_net_to_column_there": 0.21514481212537562 } } } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The labels of the workbook's two gross-profit columns appear to be swapped. In 97.5% of the 14,421 rows the column labelled EBIT equals the column labelled 'Profit before income tax' less Financial Costs, to the grosz; the identity the labels imply (pre-tax profit = EBIT - financial costs + financial revenues) holds in 0.7%. Net profit is a median 0.806 of the column labelled EBIT, about what Poland's 19% corporate income tax leaves, and still 0.788 in the 1,971 rows where net financial costs are at least half that column, which a figure struck before financial costs would not allow; for the column labelled 'Profit before income tax' the same ratio falls to 0.215 (1,084 rows). So the column labelled EBIT is profit before tax but after financial costs, and the column labelled 'Profit before income tax' is profit before financial costs and tax. The profit codes before_tax and before_financial_costs name them by content." }, { "call_id": "primary-published", "tool": "spearman_cross_section", "args": {}, "expect": { "kind": "published_pointer", "result_path": "/primary" }, "recipe_id": "mpb-financials-primary", "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "stream-offset-2009-published", "tool": "spearman_cross_section", "args": { "year": 2009, "stream_offset": 2 }, "select": "/ci_low", "expect": { "kind": "published_pointer", "result_path": "/robustness/temporal/per_year/2009/ci_low" }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "This tool, at year 2009 and stream_offset 2, reproduces the published 2009 CI exactly. That is the claim stream_offset makes: the published run drew every CI from one generator, and the 2009 block was its third draw." }, { "call_id": "one-sided-h1-off-reference", "tool": "spearman_cross_section", "args": { "alternative": "less" }, "select": "/p", "expect": { "kind": "literal", "value": 0.0002516373368919785 }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "H1 is directional (rho < 0), but the pre-registered and reported test is two-sided (p = 5.03e-04). The one-sided p in H1's direction is 2.52e-04, so the reported p is conservative by a factor of two. The conclusion is unchanged." }, { "call_id": "long-term-debt-off-reference", "tool": "spearman_cross_section", "args": { "debt": "long_term", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": -0.11625856742214648, "p": 2.9871516313325224e-12, "n": 3581, "ci_low": -0.1489126141739876, "ci_high": -0.08334846870647819, "ci_level": 95, "effect_size": -0.11625856742214648, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Leverage from long-term debt alone: rho = -0.116, 95% CI [-0.149, -0.083], twice the published -0.058. Still small: the rankings share about 1.4% of variance (rho squared), against claim 2's under 1% for total debt." }, { "call_id": "short-term-debt-off-reference", "tool": "spearman_cross_section", "args": { "debt": "short_term", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": 0.038355483010801515, "p": 0.02171643962750753, "n": 3581, "ci_low": 0.003709178590300075, "ci_high": 0.07085318871950222, "ci_level": 95, "effect_size": 0.038355483010801515, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Short-term debt alone runs the other way in 2010: rho = +0.038, 95% CI [+0.004, +0.071], p = 0.022 (the CI's lower end is within resampling error of zero; the rank test's p is exact). The published leverage measure sums two components whose associations with ROA have opposite signs, and a rank correlation of the sum is not an average of the two, so the published -0.058 is best read as partly offsetting components rather than one weak relation. That qualifies claim 2's reading of the weak rho as leverage being only a minor rank-predictor of profitability: long-term debt tracks ROA twice as strongly, and short-term debt runs the other way." }, { "call_id": "labelled-ebit-off-reference", "tool": "spearman_cross_section", "args": { "profit": "before_tax", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": -0.04617501670654151, "p": 0.005715186182129145, "n": 3581, "ci_low": -0.07908833188817303, "ci_high": -0.01260194882563445, "ci_level": 95, "effect_size": -0.04617501670654151, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The workbook's column labelled EBIT, which profit-columns-identity shows is struck after financial costs: rho = -0.046, 95% CI [-0.079, -0.013], p = 0.0057. This is profit before tax, not EBIT, so it tests only whether tax produces the association (it does not; the rho is about a fifth weaker and the same sign). It cannot test the interest channel; before-financial-costs-off-reference does." }, { "call_id": "before-financial-costs-off-reference", "tool": "spearman_cross_section", "args": { "profit": "before_financial_costs", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": 0.05268720303083367, "p": 0.0016106624683187981, "n": 3581, "ci_low": 0.019223104309829393, "ci_high": 0.08732127630233795, "ci_level": 95, "effect_size": 0.05268720303083367, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Net profit is struck after financial costs, which rise with debt, so debt lowers it by construction. With profit before financial costs and tax (the column labelled 'Profit before income tax'; see profit-columns-identity) the 2010 association reverses: rho = +0.053, 95% CI [+0.019, +0.087], p = 0.0016. Claim 1's negative sign therefore depends on deducting financial costs, consistent with the mechanical channel the object's Usage notes do not mention. Financial Costs covers more than interest (for example exchange-rate losses), so this removes the whole financing charge, not interest alone, and it keeps financial revenues in profit. Claim 1 is correct as a statement about net-profit ROA; it does not show that more-indebted firms were less profitable before paying for their debt." }, { "call_id": "before-financial-costs-long-term-debt-off-reference", "tool": "spearman_cross_section", "args": { "profit": "before_financial_costs", "debt": "long_term", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": -0.038089823932037824, "p": 0.022644372042720823, "n": 3581, "ci_low": -0.0685211355587474, "ci_high": -0.004271681444776721, "ci_level": 95, "effect_size": -0.038089823932037824, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Long-term debt stays negatively associated with profit before financial costs: rho = -0.038, 95% CI [-0.069, -0.004], p = 0.023, a third of its net-profit value (-0.116). So part of the long-term-debt association is not the financing charge. Whether the total-debt sign reversal comes from short-term debt is not tested here: no call computes short-term debt against profit before financial costs, and a rank correlation of a sum is not an average of its components' correlations." }, { "call_id": "revenue-scale-off-reference", "tool": "spearman_cross_section", "args": { "scale": "revenues", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": -0.2581936571678443, "p": 2.7568616167322234e-55, "n": 3558, "ci_low": -0.29044642913635305, "ci_high": -0.22353836130929353, "ci_level": 95, "effect_size": -0.2581936571678443, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The object's Usage notes say a mechanical component from the shared Assets denominator cannot be excluded. Scaling net profit by Revenues instead, so profitability shares no denominator with leverage, gives rho = -0.258, 95% CI [-0.290, -0.224] (n = 3,558 with positive revenues), more than four times the published value. A negative sign does not need the shared denominator. Margin is not ROA (they differ by asset turnover), so this bounds the concern rather than isolating it, and net margin is still struck after financial costs, so it says nothing about the interest channel." }, { "call_id": "reseeded-bootstrap-off-reference", "tool": "spearman_cross_section", "args": { "seed": 1, "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": -0.05810823833185246, "p": 0.000503274673783957, "n": 3581, "ci_low": -0.09140741574040462, "ci_high": -0.022003919620344512, "ci_level": 95, "effect_size": -0.05810823833185246, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "A different seed, a fresh stream and 2,000 resamples, the count every off-reference call uses, give [-0.0914, -0.0220] against the published 10,000-resample [-0.0924, -0.0241]: the endpoints move by 0.0010 and 0.0021. The published interval is not a lucky draw, and the off-reference CIs carry resampling error of about 0.002 at their ends." }, { "call_id": "ci-99-off-reference", "tool": "spearman_cross_section", "args": { "ci": 99 }, "expect": { "kind": "literal", "value": { "test": "Spearman rank correlation (two-sided)", "year": 2010, "rho": -0.05810823833185246, "p": 0.000503274673783957, "n": 3581, "ci_low": -0.1028515788935507, "ci_high": -0.013154203658132448, "ci_level": 99, "effect_size": -0.05810823833185246, "effect_size_name": "Spearman rho" } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The 99% interval from the published stream's 10,000 resamples, [-0.103, -0.013], still excludes zero." }, { "call_id": "temporal-published", "tool": "temporal_replication", "args": {}, "expect": { "kind": "published_pointer", "result_path": "/robustness/temporal" }, "recipe_id": "mpb-financials-robust-temporal", "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "temporal-before-financial-costs-off-reference", "tool": "temporal_replication", "args": { "profit": "before_financial_costs", "batch_year": null, "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "correction": "Holm-Bonferroni across [2008, 2009, 2010]", "per_year": { "2008": { "rho": -0.07810656004869178, "p": 2.1720823141416387e-06, "n": 3669, "ci_low": -0.11407617152463252, "ci_high": -0.04200381247511682, "ci_level": 95, "p_holm": 6.516246942424916e-06 }, "2009": { "rho": 0.038531104088940595, "p": 0.02226909308772694, "n": 3519, "ci_low": 0.004621657476171772, "ci_high": 0.07382317201507663, "ci_level": 95, "p_holm": 0.02226909308772694 }, "2010": { "rho": 0.05268720303083367, "p": 0.0016106624683187981, "n": 3581, "ci_low": 0.017786872225081192, "ci_high": 0.08675644295106769, "ci_level": 95, "p_holm": 0.0032213249366375963 } } } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Claim 3 does not survive the interest-channel check. With profit before financial costs, 2008 stays negative (rho = -0.078, p_Holm = 6.5e-06) but 2009 and 2010 turn positive (rho = +0.039, p_Holm = 0.022; rho = +0.053, p_Holm = 0.0032). The split follows the firm panels (see krs-overlap-panels): the 2008 panel's association survives adding back financial costs, the 2009/2010 panel's reverses." }, { "call_id": "temporal-one-panel-off-reference", "tool": "temporal_replication", "args": { "family_years": [ 2009, 2010 ], "batch_year": null, "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "correction": "Holm-Bonferroni across [2009, 2010]", "per_year": { "2009": { "rho": -0.06827400141301414, "p": 5.04820023180644e-05, "n": 3519, "ci_low": -0.1020298286601395, "ci_high": -0.03308096541170983, "ci_level": 95, "p_holm": 0.0001009640046361288 }, "2010": { "rho": -0.05810823833185246, "p": 0.000503274673783957, "n": 3581, "ci_low": -0.08998670671351074, "ci_high": -0.025659758183043564, "ci_level": 95, "p_holm": 0.000503274673783957 } } } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The published family mixes the two firm panels (2008 is one, 2009 and 2010 the other; see krs-overlap-panels). The 2009/2010 panel is the only one observed in two family years, and within it the association replicates in both after Holm (p_Holm = 1.0e-04 and 5.0e-04). These equal the published corrected values, because 2008's p is so small that Holm multiplies 2009's p by two and 2010's by one either way. So claim 3, as published, does not depend on mixing panels." }, { "call_id": "outlier-published", "tool": "outlier_sensitivity", "args": {}, "expect": { "kind": "published_pointer", "result_path": "/robustness/winsorization" }, "recipe_id": "mpb-financials-robust-winsorization", "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "tails-2.5-off-reference", "tool": "outlier_sensitivity", "args": { "lower": 0.025, "upper": 0.975, "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "Raw": { "rho": -0.05810823833185246, "p": 0.000503274673783957, "n": 3581, "ci_low": -0.09313578627248537, "ci_high": -0.024302105573211902, "ci_level": 95 }, "Winsorized 2.5/97.5%": { "rho": -0.05845583344475602, "p": 0.0004655132529030295, "n": 3581, "ci_low": -0.09053615572497709, "ci_high": -0.025223940252967203, "ci_level": 95 }, "Trimmed 2.5/97.5%": { "rho": -0.058555702015139946, "p": 0.0008035294464419578, "n": 3273, "ci_low": -0.09349615151565772, "ci_high": -0.023040902944165217, "ci_level": 95 } } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Trimming 2.5% from each tail of both ratios (n = 3,273) leaves rho = -0.059, p = 8.0e-04." }, { "call_id": "tails-5-off-reference", "tool": "outlier_sensitivity", "args": { "lower": 0.05, "upper": 0.95, "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "Raw": { "rho": -0.05810823833185246, "p": 0.000503274673783957, "n": 3581, "ci_low": -0.09313578627248537, "ci_high": -0.024302105573211902, "ci_level": 95 }, "Winsorized 5/95%": { "rho": -0.059133777010997916, "p": 0.00039934795455066674, "n": 3581, "ci_low": -0.0910968935234773, "ci_high": -0.02613310270068157, "ci_level": 95 }, "Trimmed 5/95%": { "rho": -0.02724449042442035, "p": 0.1387650506953203, "n": 2954, "ci_low": -0.06323033693937334, "ci_high": 0.00934613370821935, "ci_level": 95 } } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "At 5/95% the trimmed estimate halves to rho = -0.027 and its 95% CI [-0.063, +0.009] includes zero (p = 0.14, n = 2,954, so 627 of 3,581 rows are dropped), while winsorizing at the same level barely moves it (-0.059), as expected for a rank statistic, whose ranks winsorizing changes only by tying the clipped tails. Claim 4, stated for 1/99%, holds, and still holds at 2.5/97.5%, but the 2010 total-debt association is not significant after 5/95% trimming. Trimming both variables also restricts their range, which attenuates any correlation, so part of the drop is expected from that alone." }, { "call_id": "tails-5-long-term-debt-off-reference", "tool": "outlier_sensitivity", "args": { "lower": 0.05, "upper": 0.95, "debt": "long_term", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "Raw": { "rho": -0.11625856742214648, "p": 2.9871516313325224e-12, "n": 3581, "ci_low": -0.1489126141739876, "ci_high": -0.08334846870647819, "ci_level": 95 }, "Winsorized 5/95%": { "rho": -0.11672231112623986, "p": 2.446076201696024e-12, "n": 3581, "ci_low": -0.15052481719653954, "ci_high": -0.0840939141736625, "ci_level": 95 }, "Trimmed 5/95%": { "rho": -0.08774555495557089, "p": 1.089544848877474e-06, "n": 3077, "ci_low": -0.12282836606171117, "ci_high": -0.054456610414305884, "ci_level": 95 } } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The long-term-debt association survives the same 5/95% trim: rho = -0.088, 95% CI [-0.123, -0.054], p = 1.1e-06. What loses significance at 5% is the total-debt composite, consistent with its short-term component running the other way." }, { "call_id": "firm-level-published", "tool": "firm_level_association", "args": {}, "expect": { "kind": "published_pointer", "result_path": "/robustness/firm_level" }, "recipe_id": "mpb-financials-robust-firm-level", "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "firm-level-family-years-off-reference", "tool": "firm_level_association", "args": { "years": [ 2008, 2009, 2010 ], "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "note": "One observation per firm (mean across firm-years); 'brak' KRS excluded. Year labels: [2008, 2009, 2010].", "rho": -0.1728763021987235, "p": 2.1932269599259054e-39, "n": 5685, "ci_low": -0.20079601296478805, "ci_high": -0.14521705999359938, "ci_level": 95, "n_firms": 5685 } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The methods say 2009b is never pooled with the primary family, and the published firm-level aggregation pools it anyway. Restricted to the family years, rho = -0.173 over 5,685 firms against the published -0.186. Claim 5 does not depend on that inconsistency." }, { "call_id": "firm-level-2008-panel-off-reference", "tool": "firm_level_association", "args": { "years": [ 2008, "2009b" ], "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "note": "One observation per firm (mean across firm-years); 'brak' KRS excluded. Year labels: [2008, '2009b'].", "rho": -0.24784687593569632, "p": 3.5734721192680676e-47, "n": 3285, "ci_low": -0.2826730284072595, "ci_high": -0.2153329613282774, "ci_level": 95, "n_firms": 3285 } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Within the 2008/2009b panel: rho = -0.248 over 3,285 firms." }, { "call_id": "firm-level-2010-panel-off-reference", "tool": "firm_level_association", "args": { "years": [ 2009, 2010 ], "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "note": "One observation per firm (mean across firm-years); 'brak' KRS excluded. Year labels: [2009, 2010].", "rho": -0.0780144339291418, "p": 1.9492601321240555e-05, "n": 2990, "ci_low": -0.1155513274849805, "ci_high": -0.038755742734736906, "ci_level": 95, "n_firms": 2990 } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "Within the 2009/2010 panel: rho = -0.078 over 2,990 firms. The published -0.186 blends two panels whose firm-level associations differ about threefold. The pseudoreplication guard holds in both, but claim 5's magnitude is carried mainly by the 2008 panel." }, { "call_id": "firm-level-2010-panel-long-term-debt-off-reference", "tool": "firm_level_association", "args": { "years": [ 2009, 2010 ], "debt": "long_term", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "note": "One observation per firm (mean across firm-years); 'brak' KRS excluded. Year labels: [2009, 2010].", "rho": -0.17245706870106361, "p": 2.1518688846392717e-21, "n": 2990, "ci_low": -0.21108785053727822, "ci_high": -0.13507764125721017, "ci_level": 95, "n_firms": 2990 } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "In the panel where the published association is weakest, the long-term-debt association is more than twice as strong: rho = -0.172 over the same 2,990 firms, against -0.078 for total debt." }, { "call_id": "firm-level-before-financial-costs-off-reference", "tool": "firm_level_association", "args": { "profit": "before_financial_costs", "n_boot": 2000 }, "expect": { "kind": "literal", "value": { "note": "One observation per firm (mean across firm-years); 'brak' KRS excluded.", "rho": -0.05215357189467876, "p": 8.054212659136172e-05, "n": 5710, "ci_low": -0.07999370206896346, "ci_high": -0.025037516154863156, "ci_level": 95, "n_firms": 5710 } }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The published firm-level aggregation with profit before financial costs: rho = -0.052, 95% CI [-0.080, -0.025], p = 8.1e-05 over 5,710 firms, against the published -0.186. Claim 5's negative sign survives, at under a third of its published size; most of the published magnitude comes with deducting financial costs." }, { "call_id": "holm-2008-published", "tool": "holm", "args": {}, "select": "/0", "expect": { "kind": "published_pointer", "result_path": "/robustness/temporal/per_year/2008/p_holm" }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "holm-2010-published", "tool": "holm", "args": {}, "select": "/2", "expect": { "kind": "published_pointer", "result_path": "/robustness/temporal/per_year/2010/p_holm" }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true }, { "call_id": "holm-every-reported-test-off-reference", "tool": "holm", "args": { "p_values": [ 1.8148407556115628e-38, 5.04820023180644e-05, 0.000503274673783957, 4.878900109859027e-34, 0.0004895087779839368, 0.000416956622627233, 1.943566614107522e-45 ] }, "expect": { "kind": "literal", "value": [ 1.0889044533669378e-37, 0.0002019280092722576, 0.001250869867881699, 2.4394500549295135e-33, 0.001250869867881699, 0.001250869867881699, 1.3604966298752654e-44 ] }, "rel_tolerance": 1e-06, "abs_tolerance": 1e-09, "enabled": true, "notes": "The published correction covers three tests; the object reports seven (2008, 2009, 2010, the 2009b batch, winsorized, trimmed, firm-level; p-values from results.json in that order). Correcting across all seven, 2010 rises to p_Holm = 1.25e-03, and the step-down's monotone clamp, which the published family never triggers, ties it with the winsorized and trimmed tests. Every reported test still clears 0.05." } ] }