#!/usr/bin/env python3 """Every number the Recommendations page quotes, printed by one script. .venv/bin/python model/recommendations.py Six blocks (plan: docs/plans/2026-09-07-site-findings-restructure-and-recommendations.md section 4, plus block F added by the coordinator 2026-09-07 on the author's childcare recommendation): A. The fifty-jurisdiction S1/S2 medians (Georgia held out; the ranked tier is 50), Massachusetts's rank and its distance from each median. B. Washington, California, New Jersey, Connecticut, New York at S1/S2, with each state's income basis and cap note as recorded in the ceilings dataset. C. Five candidate credits for equal parenting time at the worked example, priced against the fifty-state S1 median and against Washington and California. D. The Box 1 reduction that would put Massachusetts exactly at the median, at Washington's figure, and at California's figure. E. A corpus check for the "deferred, not decided" gross-vs-net claim. F. the author's childcare recommendation: remove childcare from the worksheet, split it 50-50 after base support, or split it on the post-transfer income mix -- each priced against the current rule at the worked example, with after-tax positions. Reuses model/worksheet.py's run(), model/box1_fix.py's run(), model/net_position.py's analyze(), and model/childcare_post_transfer.py's figures() for every worksheet and tax calculation; nothing here reimplements them. Every dollar figure printed below comes from a function call; only inputs (incomes, health premiums, overnight shares, child-care amounts) are literal. Numbers are rounded only at print. """ import json import os import re import statistics import sys sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) import box1_fix as bf # noqa: E402 import childcare_post_transfer as cpt # noqa: E402 import net_position as npos # noqa: E402 import worksheet as w # noqa: E402 REPO = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) TIER50_PATH = os.path.join(REPO, "data", "fifty-state", "tier-50-2026-09-05.json") CEILINGS_PATH = os.path.join(REPO, "data", "fifty-state", "ceilings-2026-09-06.json") EXTRACTED_DIR = os.path.join(REPO, "data", "extracted") GUIDELINES_FLOW = os.path.join(EXTRACTED_DIR, "Guidelines.flow.txt") # The worked example -- same fact pattern as submission_figures.py / box1_fix.py / # childcare_post_transfer.py: Parent B (payor) $201,000/yr gross, health premium $43; # Parent A (recipient) $570/wk gross ($29,640/yr), health premium $33; three children, # two under 13 for the MA Child and Family Tax Credit. PAYOR_GROSS_YR = 201_000.0 PAYOR_GROSS_WEEKLY = PAYOR_GROSS_YR / 52.0 RECIP_WEEKLY = 570.0 RECIP_GROSS_YR = RECIP_WEEKLY * 52.0 KIDS = 3 KIDS_UNDER_13 = 2 A_HEALTH, B_HEALTH = 33.0, 43.0 COMPARISON_STATES = ("Washington", "California", "New Jersey", "Connecticut", "New York") MONTHLY = 52.0 / 12.0 # weekly x MONTHLY = monthly, matching every fifty-state row def wk_to_mo(weekly): return weekly * MONTHLY # --------------------------------------------------------------------------- # Data loading # --------------------------------------------------------------------------- def load_tier50(): rows = json.load(open(TIER50_PATH, encoding="utf-8")) assert len(rows) == 50, ( f"expected the ranked fifty-jurisdiction tier to hold 50 rows, found {len(rows)} " f"-- Georgia is held out and must not appear here" ) assert not any(r["state"] == "Georgia" for r in rows), "Georgia is held out; it must not be in the ranked tier" return rows def load_ceilings(): rows = json.load(open(CEILINGS_PATH, encoding="utf-8")) return {r["state"]: r for r in rows} def state_row(rows, name): matches = [r for r in rows if r["state"] == name] return matches[0] if matches else None def median_rank_distance(rows, field): """Median of `field` among the ranked rows, Massachusetts's rank (1 = highest) and its dollar/percent distance above the median.""" values_desc = sorted((r[field] for r in rows), reverse=True) median = statistics.median(values_desc) ranked = sorted(rows, key=lambda r: r[field], reverse=True) ma_rank = next(i for i, r in enumerate(ranked, 1) if r["state"] == "Massachusetts") ma_value = next(r[field] for r in rows if r["state"] == "Massachusetts") dollars = ma_value - median pct = dollars / median if median else 0.0 return {"median": median, "ma_rank": ma_rank, "ma_value": ma_value, "distance_dollars": dollars, "distance_pct": pct, "n": len(values_desc)} # --------------------------------------------------------------------------- # Block C -- candidate credits for equal parenting time # --------------------------------------------------------------------------- def box2_order(): """Today's primary-custody order at the worked example: worksheet.run(box=2).""" return w.run(box=2, a_gross=RECIP_WEEKLY, b_gross=PAYOR_GROSS_WEEKLY, children_under18=KIDS, a_health=A_HEALTH, b_health=B_HEALTH) def box1_order(): """Today's shared-parenting order at the worked example: worksheet.run(box=1).""" return w.run(box=1, a_gross=RECIP_WEEKLY, b_gross=PAYOR_GROSS_WEEKLY, children_under18=KIDS, a_health=A_HEALTH, b_health=B_HEALTH) def variant_order(v, payor_overnight_share=0.5): """box1_fix.py's variant, at the given PAYOR overnight share (box1_fix.py's own convention: a_overnight_share is Parent A / the recipient's share).""" return bf.run(v, a_gross=RECIP_WEEKLY, b_gross=PAYOR_GROSS_WEEKLY, children_under18=KIDS, a_health=A_HEALTH, b_health=B_HEALTH, a_overnight_share=1.0 - payor_overnight_share) def linear_time_discount(payor_overnight_share, low=1.0 / 3.0, high=0.5, max_discount=0.5): """the author's sketch: 0% discount off today's Box 2 (primary) order at LOW payor overnights, scaling linearly to MAX_DISCOUNT at HIGH payor overnights. Clamped outside [low, high]. Returns the discount fraction, not a dollar figure.""" if payor_overnight_share <= low: return 0.0 if payor_overnight_share >= high: return max_discount return max_discount * (payor_overnight_share - low) / (high - low) def linear_discount_order(payor_overnight_share, base=None): base = box2_order()["7d"] if base is None else base return base * (1.0 - linear_time_discount(payor_overnight_share)) def candidate_credits(): """The five (six-row) candidates block C asks for, each as a dict with the weekly order and a label. All computed from worksheet.py / box1_fix.py; the linear rule is applied to worksheet.py's own Box 2 figure.""" b2 = box2_order()["7d"] rows = [] rows.append({"label": "(i) Variant A -- 6e limitation moved to the transfer", "weekly": variant_order("A")["7d"]}) rows.append({"label": "(ii) Variant B -- cross-credit at duplication 1.5", "weekly": variant_order("B")["7d"]}) rows.append({"label": "(iii) Variant C at equal time (payor 50% of overnights)", "weekly": variant_order("C", 0.5)["7d"]}) rows.append({"label": "(iv) Linear time-proportional discount at 50% of overnights", "weekly": linear_discount_order(0.5, base=b2)}) rows.append({"label": "(v-a) Linear time-proportional discount at 45% of overnights", "weekly": linear_discount_order(0.45, base=b2)}) rows.append({"label": "(v-b) Linear time-proportional discount at 40% of overnights", "weekly": linear_discount_order(0.40, base=b2)}) return rows, b2 # --------------------------------------------------------------------------- # Block E -- corpus check for the gross-vs-net deferral claim # --------------------------------------------------------------------------- SENTENCE_RE = re.compile(r"(?<=[.!?])\s+") COMMENTARY_MARKER_RE = re.compile(r"Comm\s*entary\s+(20\d\d)") def commentary_blocks(): """Split Guidelines.flow.txt at each 'Commentary ' marker (an OCR artifact sometimes inserts a space: 'Comm entary'). Returns a list of (year, block_text) in document order; the guidelines embed several blocks per year, one per section the commentary addresses, so a year can appear more than once.""" if not os.path.exists(GUIDELINES_FLOW): return {} text = open(GUIDELINES_FLOW, encoding="utf-8").read() matches = list(COMMENTARY_MARKER_RE.finditer(text)) blocks = [] for i, m in enumerate(matches): year = m.group(1) start = m.end() end = matches[i + 1].start() if i + 1 < len(matches) else len(text) blocks.append((year, text[start:end])) return blocks def gross_and_net_sentences_by_year(): """For each commentary year, every sentence (across all of that year's blocks) containing both 'gross' and 'net' (case-insensitive).""" by_year = {} for year, block in commentary_blocks(): by_year.setdefault(year, {"n_blocks": 0, "hits": []}) by_year[year]["n_blocks"] += 1 for s in SENTENCE_RE.split(block): s = s.strip() if re.search(r"gross", s, re.I) and re.search(r"\bnet\b", s, re.I): by_year[year]["hits"].append(s) return by_year def as_prior_task_forces_sentences(): """Every sentence containing 'as prior task forces' (case-insensitive) across every *.flow.txt in data/extracted, with the source file.""" hits = [] if not os.path.isdir(EXTRACTED_DIR): return [] for fname in sorted(os.listdir(EXTRACTED_DIR)): if not fname.endswith(".flow.txt"): continue path = os.path.join(EXTRACTED_DIR, fname) text = open(path, encoding="utf-8").read() for s in SENTENCE_RE.split(text): if "as prior task forces" in s.lower(): hits.append((fname, s.strip())) return hits # --------------------------------------------------------------------------- # Block F -- the author's childcare recommendation # --------------------------------------------------------------------------- def childcare_rules(): """Five childcare rules at the worked example ($300/wk paid by the recipient, $0 by the payor): the current worksheet, removing childcare from the worksheet entirely, a flat 50-50 split after base support, the Line 6b-1 post-transfer GROSS redline (childcare_post_transfer.py rule 2b), and the post-transfer NET version from the same module (rule 3/4, its fixed point). Each row reports the worksheet order (weekly/monthly), the payor's effective share of the $15,600/yr childcare cost in percent and dollars, and both parties' after-tax position from net_position.analyze(). Reuses childcare_post_transfer.figures() and net_position.analyze() throughout; the only new arithmetic here is applying an already-computed share to the $15,600 total, exactly as childcare_post_transfer.py itself does for its rule 2b. """ f = cpt.figures() base = f["base_7d"] # no-childcare Box 1 order cc_annual = f["cc_annual"] # $15,600/yr CC_WEEKLY = cpt.CC_WEEKLY # $300/wk def after_tax(support_weekly, payor_share_direct): return npos.analyze(PAYOR_GROSS_YR, RECIP_GROSS_YR, KIDS, support_weekly, CC_WEEKLY, payor_share_direct, kids_under_13=KIDS_UNDER_13) rows = [] # 1. Current worksheet: Line 6b on pre-order (Line 3c) shares. The order itself # carries the full childcare transfer; the payor pays nothing directly. order1 = f["current_7d"] share1 = f["rule1_share"] pos1 = after_tax(order1, 0.0) rows.append({"n": 1, "label": "Current worksheet (Line 6b on pre-order Line 3c shares)", "order_wk": order1, "share": share1, "pos": pos1}) # 2. Childcare removed from the worksheet entirely: the order reverts to the # no-childcare figure and each parent bears what they actually pay -- the payor # pays $0 of the recipient's $300/wk, so his effective share is zero. order2 = base share2 = 0.0 pos2 = after_tax(order2, 0.0) rows.append({"n": 2, "label": "Childcare removed from the worksheet (each parent bears what they pay)", "order_wk": order2, "share": share2, "pos": pos2}) # 3. Base support calculated first, then childcare split 50-50 regardless of # income share (the author's sketch). The payor reimburses half the $300/wk directly. order3 = base # the worksheet's support figure is unchanged share3 = 0.5 pos3 = after_tax(order3, share3) rows.append({"n": 3, "label": "Base support first, then childcare split 50-50", "order_wk": order3, "share": share3, "pos": pos3}) # 4. Post-transfer GROSS shares on Line 3a -- the Trial Court comments' Line # 6b-1 redline. childcare_post_transfer.py's rule 2b; order bakes in the # childcare transfer at the post-transfer share. order4 = f["rule2b_7d"] share4 = f["rule2b_share"] pos4 = after_tax(order4, 0.0) rows.append({"n": 4, "label": "Post-transfer GROSS shares on Line 3a (comments' Line 6b-1 redline)", "order_wk": order4, "share": share4, "pos": pos4}) # 5. Post-transfer NET shares -- the paper's version. The worksheet order stays # at the no-childcare figure; childcare is allocated directly at the # post-transfer NET income share, which childcare_post_transfer.py finds is a # fixed point equal to rule 3's (no-childcare) net share. order5 = base share5 = f["rule3_share"] pos5 = after_tax(order5, share5) rows.append({"n": 5, "label": "Post-transfer NET shares (paper's version, fixed point)", "order_wk": order5, "share": share5, "pos": pos5}) headline = { "before_share": f["rule1_share"], # Line 3c, "before the order" "after_gross_share": f["rule2_gross_share"], # post-transfer GROSS "after_net_share": f["rule3_share"], # post-transfer NET } return rows, cc_annual, headline # --------------------------------------------------------------------------- # Printing # --------------------------------------------------------------------------- def main(): print("=" * 78) print("RECOMMENDATIONS PAGE -- EVERY NUMBER, PRINTED FROM CODE") print("=" * 78) rows = load_tier50() ceilings = load_ceilings() # --- Block A ------------------------------------------------------------ print("\n" + "-" * 78) print("BLOCK A -- the fifty-jurisdiction medians and Massachusetts's distance") print("-" * 78) print("Source: data/fifty-state/tier-50-2026-09-05.json (Georgia held out; see") print("docs/2026-09-05-georgia-worked-example.md). S1 = equal parenting time,") print("S2 = the other parent has primary custody. Monthly figures throughout,") print("as the tier-50 dataset states them.\n") a_s1 = median_rank_distance(rows, "s1") a_s2 = median_rank_distance(rows, "s2") for label, d in (("S1 (equal time)", a_s1), ("S2 (primary custody)", a_s2)): print(f" {label}:") print(f" ranked jurisdictions {d['n']}") print(f" median ${d['median']:>10,.2f}/mo") print(f" Massachusetts ${d['ma_value']:>10,.2f}/mo rank {d['ma_rank']} of {d['n']}") print(f" MA above the median ${d['distance_dollars']:>9,.2f}/mo {d['distance_pct']:>7.1%}") print() # --- Block B ------------------------------------------------------------ print("-" * 78) print("BLOCK B -- Washington, California, New Jersey, Connecticut, New York") print("-" * 78) print("Basis and cap note: data/fifty-state/ceilings-2026-09-06.json.\n") for name in COMPARISON_STATES: r = state_row(rows, name) if r is None: print(f" {name:<12} NOT FOUND in the ranked tier -- excluded from any rank claim.") continue c = ceilings.get(name) basis = c["basis"] if c else "unrecorded" cap = f"${c['annual_ceiling']:,.0f}/yr ({c['period_as_stated']})" if c and c["annual_ceiling"] else \ (c["period_as_stated"] if c and c.get("period_as_stated") else "no stated ceiling") note = f" -- {c['note']}" if c and c.get("note") else "" print(f" {name:<12} S1 ${r['s1']:>9,.2f}/mo S2 ${r['s2']:>9,.2f}/mo basis: {basis:<5} cap: {cap}{note}") if r.get("note"): print(f" row note: {r['note']}") print() for name in ("Washington", "California"): present = state_row(rows, name) is not None print(f" {name} present at the ranked tier: {present}") print() # --- Block C ------------------------------------------------------------ print("-" * 78) print("BLOCK C -- candidate credits for equal parenting time, worked example") print("-" * 78) print("Payor $201,000/yr gross, health premium $43; recipient $570/wk gross,") print("health premium $33; three children; no child care. Sources: worksheet.py,") print("box1_fix.py (variants A, B, C), and the linear rule applied here to") print("worksheet.py's own Box 2 figure.\n") b2 = box2_order()["7d"] b1 = box1_order()["7d"] print(f" Today's Box 2 (primary) order ${b2:>9,.2f}/wk ${wk_to_mo(b2):>9,.2f}/mo") print(f" Today's Box 1 (shared) order ${b1:>9,.2f}/wk ${wk_to_mo(b1):>9,.2f}/mo") print() cand_rows, b2_check = candidate_credits() assert abs(b2_check - b2) < 0.005, "candidate_credits() must price off the same Box 2 figure printed above" print(f" {'candidate':<58} {'wk':>9} {'mo':>10} {'vs Box1':>9} {'vs median':>10} {'vs WA':>10} {'vs CA':>10}") wa_s1 = state_row(rows, "Washington")["s1"] ca_s1 = state_row(rows, "California")["s1"] for r in cand_rows: mo = wk_to_mo(r["weekly"]) pct_vs_b1 = (r["weekly"] - b1) / b1 print(f" {r['label']:<58} ${r['weekly']:>7,.2f} ${mo:>8,.2f} {pct_vs_b1:>8.1%} " f"{(mo - a_s1['median']) / a_s1['median']:>9.1%} {(mo - wa_s1) / wa_s1:>9.1%} {(mo - ca_s1) / ca_s1:>9.1%}") print("\n ('vs median/WA/CA' = the candidate's monthly figure, percent above (+) or") print(" below (-) the fifty-state S1 median / Washington's S1 / California's S1.)") va = variant_order("A") vb = variant_order("B") vc_equal = variant_order("C", 0.5) print(f"\n Check: Variant C at equal time reproduces Variant B exactly, as box1_fix.py's") print(f" own docstring says ('at equal time this is identical to B'): " f"{abs(vb['7d'] - vc_equal['7d']) < 0.005}") print() # --- Block D ------------------------------------------------------------ print("-" * 78) print("BLOCK D -- the Box 1 reduction that reaches the median, WA, and CA") print("-" * 78) targets = (("the fifty-state S1 median", a_s1["median"]), ("Washington's S1", wa_s1), ("California's S1", ca_s1)) print(f" Starting point: today's Box 1 order ${b1:,.2f}/wk (${wk_to_mo(b1):,.2f}/mo).\n") for label, target_mo in targets: target_wk = target_mo * 12.0 / 52.0 cut_wk = b1 - target_wk pct = cut_wk / b1 print(f" To reach {label} (${target_mo:,.2f}/mo): cut ${cut_wk:,.2f}/wk " f"(${cut_wk * 52:,.0f}/yr), {pct:.1%} off today's Box 1 order.") print() # --- Block E ------------------------------------------------------------ print("-" * 78) print("BLOCK E -- corpus check: was gross-vs-net deferred, and how many times") print("-" * 78) by_year = gross_and_net_sentences_by_year() if not by_year: print(" Corpus not present in this checkout (data/extracted/Guidelines.flow.txt);" " block E is reported in the committed run output only.") total_hits = 0 for year in sorted(by_year): info = by_year[year] total_hits += len(info["hits"]) print(f" Commentary {year}: {info['n_blocks']} block(s) in Guidelines.flow.txt, " f"{len(info['hits'])} sentence(s) containing both 'gross' and 'net':") for h in info["hits"]: print(f" \"{h}\"") print(f"\n Total sentences with both terms inside a Guidelines.flow.txt commentary " f"block: {total_hits}") prior_hits = as_prior_task_forces_sentences() print(f"\n 'as prior task forces' across data/extracted/*.flow.txt: {len(prior_hits)} hit(s)") for fname, sentence in prior_hits: print(f" {fname}: \"{sentence}\"") files_with_hit = sorted({fname for fname, _ in prior_hits}) print(f"\n CONCLUSION: {total_hits} sentence(s) mentioning both gross and net appear inside " f"the year-tagged commentary blocks embedded in the Guidelines PDF itself " f"({', '.join(sorted(by_year)) if by_year else 'none'} all checked, zero hits). The " f"explicit deferral -- \"as prior task forces have done ... decided not to recommend " f"a change from gross income to net income at this time\" -- lives in " f"{', '.join(files_with_hit) if files_with_hit else 'no file in this corpus'}, the " f"Brattle economic review, not in the Guidelines' own embedded commentary, and it does " f"not itself name which prior years deferred it. State it as: one document in this " f"corpus (the 2025 economic review) records a gross-vs-net deferral and says prior " f"task forces did the same; the Guidelines' own year-tagged commentary blocks contain " f"no sentence discussing gross vs. net at all.") print() # --- Block F -------------------------------------------------------- print("-" * 78) print("BLOCK F -- the author's childcare recommendation, worked example") print("-" * 78) print("Payor $201,000/yr gross ($43 health premium), recipient $570/wk gross") print("($29,640/yr, $33 health premium), three children (two under 13), Box 1,") print("recipient pays $300/wk child care, payor pays $0 directly. Source:") print("model/childcare_post_transfer.py (reused, not reimplemented) and") print("model/net_position.py's analyze().\n") f_rows, cc_annual, headline = childcare_rules() print(f" Total annual child care at issue: ${cc_annual:,.0f}/yr ($300/wk)\n") for r in f_rows: mo = wk_to_mo(r["order_wk"]) print(f" {r['n']}. {r['label']}") print(f" order ${r['order_wk']:>9,.2f}/wk ${mo:>9,.2f}/mo") print(f" payor's share of the ${cc_annual:,.0f}/yr child care: " f"{r['share']:>6.1%} (${r['share'] * cc_annual:>9,.2f}/yr)") print(f" after tax: payor keeps ${r['pos']['payor_after']:>9,.2f} " f"recipient household holds ${r['pos']['recip_after']:>9,.2f}") print() print(" The sentence the site makes (rule 1, current worksheet):") print(f" before the order, the payor's share of the parties' combined available") print(f" income (Line 3c) is {headline['before_share']:.1%};") print(f" after the order, his share of POST-TRANSFER GROSS income is " f"{headline['after_gross_share']:.1%};") print(f" after the order, his share of POST-TRANSFER NET income is " f"{headline['after_net_share']:.1%};") print(f" and child care is allocated on the {headline['before_share']:.1%} figure --") print(f" the income split BEFORE the order, not either figure after it.") print() print("=" * 78) print("END OF RUN") print("=" * 78) if __name__ == "__main__": main()