| """Build the excerpts Lev was trained on. |
| |
| A decision is passed as a header line, its front matter, and its numbered paragraphs |
| ({number: text}, with the text starting "[N] "). Each function returns the `state` string to pass to |
| `model.system_one(state=..., questions=...)`. |
| |
| header "Smith v. Jones, 2025 ONCA 100 | Court of Appeal for Ontario (appeal court, Ontario) | decided 2025-02-03" |
| front the style of cause, coram, counsel and so on, as plain text |
| paragraphs {41: "[41] The respondent relies on ...", 42: "[42] ..."} |
| """ |
|
|
|
|
| def clip(text, limit): |
| return text if len(text) <= limit else text[:limit].rsplit(" ", 1)[0] + " [...]" |
|
|
|
|
| def citation_pair(header, front, paragraphs, cited_name, cited_citation, citing_paragraphs, budget=3400): |
| """Excerpt for the citator questions: treatment, history, reported, dissent. |
| |
| `citing_paragraphs` are the numbers of the paragraphs that name or cite the case. |
| """ |
| order = sorted(paragraphs) |
| index = {n: i for i, n in enumerate(order)} |
| cited = [n for n in dict.fromkeys(citing_paragraphs) if n in paragraphs] |
| if not cited: |
| raise ValueError("none of the citing paragraphs is in the decision") |
| chosen = list(cited) |
| for n in cited: |
| for j in (index[n] - 1, index[n] + 1): |
| if 0 <= j < len(order) and order[j] not in chosen: |
| chosen.append(order[j]) |
| kept, used = [], 0 |
| for n in chosen: |
| text = clip(paragraphs[n], 1300) |
| if used + len(text) > budget and kept: |
| continue |
| kept.append(n) |
| used += len(text) |
| kept.sort(key=index.get) |
| first = [n for n in order[:1] if n not in kept] |
| last = [n for n in order[-2:] if n not in kept] |
| lines = [f"Citing decision: {header}", f"Cited case: {cited_name}, {cited_citation}", "", |
| "Front matter of the citing decision:", clip(front, 500), ""] |
| if first: |
| lines += ["Opening paragraph:", clip(paragraphs[first[0]], 400), ""] |
| lines.append("Paragraphs that cite the case, with neighbours:") |
| lines += [clip(paragraphs[n], 1300) for n in kept] |
| if last: |
| lines += ["", "Closing paragraphs:"] + [clip(paragraphs[n], 350) for n in last] |
| return "\n".join(lines) |
|
|
|
|
| def decision(header, front, paragraphs): |
| """Excerpt for the decision-level questions: outcome, decision_type, proceeding, area and the three yes/no ones.""" |
| order = sorted(paragraphs) |
| first, last = order[:3], [n for n in order[-5:] if n not in order[:3]] |
| lines = [f"Decision: {header}", "", "Front matter:", clip(front, 900), "", "Opening paragraphs:"] |
| lines += [clip(paragraphs[n], 500) for n in first] |
| if last: |
| lines += ["", "Closing paragraphs:"] + [clip(paragraphs[n], 600) for n in last] |
| return "\n".join(lines) |
|
|
|
|
| def passage(header, lead, paragraphs, numbers, budget=3600): |
| """Excerpt for the passage-level questions. |
| |
| `lead` is ["Issue: ..."] for standard_of_review, or ["Legal test: ...", "Element: ..."] for element_finding. |
| `numbers` are the paragraphs that deal with the issue or the element. |
| """ |
| kept, used = [], 0 |
| for n in dict.fromkeys(numbers): |
| if n not in paragraphs: |
| continue |
| text = clip(paragraphs[n], 1300) |
| if used + len(text) > budget and kept: |
| break |
| kept.append(text) |
| used += len(text) |
| if not kept: |
| raise ValueError("none of the paragraphs is in the decision") |
| return "\n".join([f"Decision: {header}", *lead, "", "Paragraphs:", *kept]) |
|
|
