"""Personal Dimension Benchmark: the subject, the facts, and the ground-truth engine. One synthetic person. Every fact is authored once, here, with a validity interval, a source and a natural-language note. Every representation used in the study is RENDERED from this single table, so no representation can be missing a fact that another one has. That is the design correction the review of the previous study forced on us: completeness is guaranteed by construction and is verified in code, not by inspection. TODAY is fixed so the study is reproducible. """ from datetime import date TODAY = date(2026, 9, 6) SUBJECT = "Mira Vantaa" # ------------------------------------------------------------------ sources # tier 1 is the most trusted. The conflict policy reads these tiers. SOURCES = { "S-DOC-01": ("document_scan", 1, date(2019, 2, 24)), "S-DOC-02": ("document_scan", 1, date(2022, 7, 20)), "S-DOC-03": ("document_scan", 1, date(2025, 5, 12)), "S-DOC-04": ("document_scan", 1, date(2021, 11, 18)), "S-DOC-05": ("document_scan", 1, date(2025, 2, 21)), "S-DOC-06": ("document_scan", 1, date(2016, 4, 12)), "S-DOC-07": ("document_scan", 1, date(2017, 2, 3)), "S-DOC-08": ("document_scan", 1, date(2025, 9, 2)), "S-DOC-09": ("document_scan", 1, date(2026, 1, 31)), "S-DOC-10": ("document_scan", 1, date(2024, 11, 9)), "S-DEV-01": ("device_sync", 2, date(2024, 7, 1)), "S-DEV-02": ("device_sync", 2, date(2026, 8, 30)), "S-EML-01": ("email", 3, date(2018, 8, 20)), "S-EML-02": ("email", 3, date(2024, 2, 20)), "S-EML-03": ("email", 3, date(2024, 7, 1)), "S-EML-04": ("email", 3, date(2021, 4, 28)), "S-EML-05": ("email", 3, date(2022, 12, 27)), "S-EML-06": ("email", 3, date(2022, 8, 30)), "S-EML-07": ("email", 3, date(2024, 3, 15)), "S-EML-08": ("email", 3, date(2025, 2, 26)), "S-SLF-01": ("self_reported", 4, date(2019, 1, 10)), "S-SLF-02": ("self_reported", 4, date(2020, 1, 4)), "S-SLF-03": ("self_reported", 4, date(2023, 6, 2)), "S-SLF-04": ("self_reported", 4, date(2025, 6, 20)), "S-SLF-05": ("self_reported", 4, date(2022, 9, 5)), "S-SLF-06": ("self_reported", 4, date(2024, 5, 19)), "S-SLF-08": ("self_reported", 4, date(2025, 3, 30)), "S-SLF-07": ("self_reported", 4, date(2026, 2, 10)), } TIER_NAMES = {1: "document_scan", 2: "device_sync", 3: "email", 4: "self_reported"} POLICY_ID = "CP-01" POLICY_TEXT = ( "CP-01: when two facts about the same attribute cover the same date and disagree, " "the fact from the lower source tier wins (document_scan 1, device_sync 2, email 3, " "self_reported 4). Within one tier, the fact captured later wins." ) # ------------------------------------------------------------------ facts # (fid, attr, value, valid_from, valid_to, source, note) # valid_to None means open-ended. F = [] def fact(fid, attr, value, vfrom, vto, source, note): F.append({"fid": fid, "attr": attr, "value": value, "valid_from": vfrom, "valid_to": vto, "source": source, "note": note}) fact("F-001", "home_address", "12 Harbour Lane, Limassol", date(2019, 3, 1), date(2022, 7, 31), "S-DOC-01", "Signed the lease for 12 Harbour Lane, Limassol and moved in on 2019-03-01; " "stayed until 2022-07-31. The lease scan is filed as S-DOC-01.") fact("F-002", "home_address", "48 Aristotelous, Nicosia", date(2022, 8, 1), date(2025, 5, 14), "S-DOC-02", "Moved to 48 Aristotelous, Nicosia on 2022-08-01 and lived there until " "2025-05-14. Lease scan filed as S-DOC-02.") fact("F-003", "home_address", "7 Kalymnou, Larnaca", date(2025, 5, 15), None, "S-DOC-03", "Moved to 7 Kalymnou, Larnaca on 2025-05-15 and still live there. Lease scan " "filed as S-DOC-03.") fact("F-004", "home_address", "48 Aristotelous, Nicosia", date(2025, 5, 15), None, "S-SLF-04", "Wrote down on 2025-06-20 that my address is 48 Aristotelous, Nicosia, and that " "this holds from 2025-05-15 onwards. This is just my own note, S-SLF-04.") fact("F-010", "employer", "Halvard Systems", date(2018, 9, 1), date(2021, 11, 30), "S-EML-01", "Started at Halvard Systems on 2018-09-01 and left on 2021-11-30. The offer email " "is S-EML-01.") fact("F-011", "employer", "Tessera Analytics", date(2021, 12, 1), date(2025, 2, 28), "S-DOC-04", "Worked at Tessera Analytics from 2021-12-01 to 2025-02-28. Signed contract scan " "S-DOC-04.") fact("F-012", "employer", "Kestrel Data Works", date(2025, 3, 1), None, "S-DOC-05", "Joined Kestrel Data Works on 2025-03-01 and still work there. Contract scan " "S-DOC-05.") fact("F-020", "mobile_phone", "+357 99 100 224", date(2019, 1, 10), date(2024, 6, 30), "S-SLF-01", "My mobile number was +357 99 100 224 from 2019-01-10 until 2024-06-30. I noted " "it myself, S-SLF-01.") fact("F-021", "mobile_phone", "+357 96 442 018", date(2024, 7, 1), None, "S-DEV-01", "New mobile number +357 96 442 018 from 2024-07-01, synced from the handset, " "S-DEV-01.") fact("F-022", "mobile_phone", "+357 99 100 224", date(2024, 7, 1), None, "S-SLF-06", "Wrote on 2024-05-19 that my number stays +357 99 100 224 from 2024-07-01. Own " "note, S-SLF-06.") fact("F-030", "monthly_salary_eur", "4200", date(2025, 3, 1), date(2025, 12, 31), "S-EML-08", "The offer email of 2025-02-26 put my monthly salary at EUR 4200 from 2025-03-01 " "until 2025-12-31. Email S-EML-08.") fact("F-031", "monthly_salary_eur", "4410", date(2026, 1, 1), None, "S-DOC-09", "The January 2026 payslip scan shows a monthly salary of EUR 4410 from " "2026-01-01. Scan S-DOC-09.") fact("F-032", "monthly_salary_eur", "4200", date(2026, 1, 1), None, "S-SLF-07", "On 2026-02-10 I wrote down that my salary is EUR 4200 from 2026-01-01. Own note " "S-SLF-07.") fact("F-040", "blood_type", "A+", date(2015, 1, 1), None, "S-SLF-03", "I have always said my blood type is A+, and that it has held since 2015-01-01. " "Own note from 2023-06-02, S-SLF-03.") fact("F-041", "blood_type", "0+", date(2015, 1, 1), None, "S-DOC-10", "The laboratory report of 2024-11-09 gives my blood type as 0+, holding since " "2015-01-01. Scan S-DOC-10.") fact("F-050", "diet", "vegetarian", date(2023, 6, 2), date(2025, 3, 29), "S-SLF-03", "From 2023-06-02 until 2025-03-29 I ate vegetarian. Own note S-SLF-03.") fact("F-051", "diet", "pescatarian", date(2025, 3, 30), None, "S-SLF-08", "From 2025-03-30 I eat pescatarian. Own note S-SLF-08.") fact("F-060", "passport_number", "P4417832", date(2016, 4, 12), date(2026, 4, 11), "S-DOC-06", "Passport P4417832 was issued on 2016-04-12 and expired on 2026-04-11. Scan " "S-DOC-06.") fact("F-061", "driving_licence", "DL-772104", date(2017, 2, 3), date(2027, 1, 31), "S-DOC-07", "Driving licence DL-772104 runs from 2017-02-03 to 2027-01-31. Scan S-DOC-07.") fact("F-062", "health_insurance_policy", "HI-88231", date(2025, 9, 1), date(2026, 8, 31), "S-DOC-08", "Health insurance policy HI-88231 covered 2025-09-01 to 2026-08-31. Scan " "S-DOC-08.") fact("F-063", "bank_iban", "CY17 0020 0128 0000 0012 0052 7600", date(2021, 12, 1), None, "S-DOC-04", "Salary goes to IBAN CY17 0020 0128 0000 0012 0052 7600, on file since " "2021-12-01 with the contract scan S-DOC-04.") fact("F-064", "drug_allergy", "penicillin", date(2015, 1, 1), None, "S-DOC-10", "The laboratory file of 2024-11-09 records a penicillin allergy, known since " "2015-01-01. Scan S-DOC-10.") fact("F-070", "device_laptop", "Aurora 14", date(2024, 2, 20), None, "S-EML-02", "Bought an Aurora 14 laptop on 2024-02-20; the receipt email is S-EML-02.") fact("F-071", "laptop_warranty_until", "2027-02-20", date(2024, 2, 20), date(2027, 2, 20), "S-EML-02", "The Aurora 14 laptop warranty runs to 2027-02-20 according to the receipt " "S-EML-02.") fact("F-072", "device_phone", "Corvus 8", date(2024, 7, 1), None, "S-EML-03", "Bought a Corvus 8 handset on 2024-07-01; receipt email S-EML-03.") fact("F-073", "phone_warranty_until", "2026-07-01", date(2024, 7, 1), date(2026, 7, 1), "S-EML-03", "The Corvus 8 handset warranty ran to 2026-07-01, per receipt S-EML-03.") fact("F-080", "subscription_cloudline", "9", date(2021, 5, 1), None, "S-EML-04", "Cloudline storage costs EUR 9 a month, running since 2021-05-01. Billing email " "S-EML-04.") fact("F-081", "subscription_cloudline", "12", date(2021, 5, 1), None, "S-SLF-05", "I noted on 2022-09-05 that Cloudline storage costs EUR 12 a month from " "2021-05-01. Own note S-SLF-05.") fact("F-082", "subscription_trainline", "19", date(2023, 1, 1), date(2024, 12, 31), "S-EML-05", "The Trainline pass cost EUR 19 a month from 2023-01-01 and was cancelled on " "2024-12-31. Billing email S-EML-05.") fact("F-083", "subscription_gym", "35", date(2022, 9, 1), date(2025, 5, 31), "S-EML-06", "The Nicosia gym cost EUR 35 a month from 2022-09-01 until I cancelled it on " "2025-05-31 after the move. Billing email S-EML-06.") fact("F-084", "subscription_journal", "22", date(2024, 3, 15), None, "S-EML-07", "Journal access costs EUR 22 a month since 2024-03-15. Billing email S-EML-07.") fact("F-085", "subscription_music", "11", date(2020, 1, 1), None, "S-SLF-02", "The music service costs EUR 11 a month since 2020-01-01. Own note S-SLF-02.") fact("F-090", "review_due_home_address", "2027-05-15", date(2025, 5, 15), None, "S-DOC-03", "The lease at 7 Kalymnou is up for review by 2027-05-15, per the scan S-DOC-03.") fact("F-091", "review_due_health_insurance", "2026-08-31", date(2025, 9, 1), None, "S-DOC-08", "The health insurance needs renewing by 2026-08-31, per the scan S-DOC-08.") # ------------------------------------------------------------------ people # (name, role, city, phone, note) PEOPLE = [ ("Ilona Vantaa", "sister", "Nicosia", "+357 99 774 310", "My sister Ilona Vantaa lives in Nicosia; her number is +357 99 774 310."), ("Alex Reiner", "colleague at Kestrel Data Works", "Larnaca", "+357 96 220 145", "Alex Reiner is a colleague at Kestrel Data Works, based in Larnaca, " "+357 96 220 145."), ("Alex Doran", "general practitioner", "Larnaca", "+357 24 812 660", "Alex Doran is my GP in Larnaca, +357 24 812 660."), ("Alex Kovacs", "landlord of 7 Kalymnou", "Larnaca", "+357 99 018 442", "Alex Kovacs is the landlord of 7 Kalymnou in Larnaca, +357 99 018 442."), ("Nadja Lind", "former colleague at Tessera Analytics", "Nicosia", "+357 97 553 208", "Nadja Lind is a former colleague from Tessera Analytics, now in Nicosia, " "+357 97 553 208."), ("Petros Sarris", "dentist", "Larnaca", "+357 24 330 917", "Petros Sarris is my dentist in Larnaca, +357 24 330 917."), ("Georgia Mensah", "accountant", "Limassol", "+357 25 447 002", "Georgia Mensah is my accountant in Limassol, +357 25 447 002."), ] # (date, destination, companion or None, note) TRIPS = [ (date(2023, 4, 10), "Athens", "Ilona Vantaa", "Travelled to Athens on 2023-04-10 with Ilona Vantaa."), (date(2024, 2, 3), "Nicosia", "Nadja Lind", "Went to Nicosia on 2024-02-03 with Nadja Lind."), (date(2024, 9, 17), "Vienna", None, "Flew to Vienna alone on 2024-09-17."), (date(2025, 6, 21), "Athens", "Alex Reiner", "Travelled to Athens on 2025-06-21 with Alex Reiner."), (date(2025, 11, 8), "Nicosia", "Ilona Vantaa", "Went to Nicosia on 2025-11-08 with Ilona Vantaa."), (date(2026, 1, 22), "Nicosia", "Nadja Lind", "Went to Nicosia on 2026-01-22 with Nadja Lind."), (date(2026, 3, 14), "Berlin", "Alex Reiner", "Travelled to Berlin on 2026-03-14 with Alex Reiner."), (date(2026, 7, 2), "Athens", None, "Flew to Athens alone on 2026-07-02."), ] # ------------------------------------------------------------------ engine def tier(fid_or_source): src = fid_or_source if fid_or_source in SOURCES else None return SOURCES[src][1] if src else 99 def captured(source): return SOURCES[source][2] def covers(f, when): if f["valid_from"] > when: return False return f["valid_to"] is None or f["valid_to"] >= when def candidates(attr, when): return [f for f in F if f["attr"] == attr and covers(f, when)] def resolve(attr, when=TODAY): """Apply CP-01: lowest tier wins, then latest captured. Returns the winning fact.""" cand = candidates(attr, when) if not cand: return None cand.sort(key=lambda f: (tier(f["source"]), -captured(f["source"]).toordinal())) return cand[0] def value_at(attr, when=TODAY): f = resolve(attr, when) return f["value"] if f else None def conflicted(attr, when=TODAY): """True when more than one distinct value covers `when`.""" return len({f["value"] for f in candidates(attr, when)}) > 1 def expired(attr, when=TODAY): """The attribute has a fact whose validity ended before `when` and no later one.""" same = [f for f in F if f["attr"] == attr] if not same: return False if candidates(attr, when): return False return any(f["valid_to"] is not None and f["valid_to"] < when for f in same) CONFLICT_ATTRS = ["home_address", "mobile_phone", "monthly_salary_eur", "blood_type", "subscription_cloudline"] def naive_value(attr, when=TODAY): """What a summary without provenance produces: the most recently captured fact that covers the date. This is the rule a plain assistant memory actually uses.""" cand = candidates(attr, when) if not cand: return None cand.sort(key=lambda f: -captured(f["source"]).toordinal()) return cand[0]["value"] def person(name): for p in PEOPLE: if p[0] == name: return p return None def people_named(first): return [p for p in PEOPLE if p[0].split()[0] == first] def trips_to(city): return sorted([t for t in TRIPS if t[1] == city], key=lambda t: t[0]) def all_values(): """Every value that appears anywhere, for the fabrication check.""" out = {str(f["value"]) for f in F} for p in PEOPLE: out.update({p[0], p[2], p[3]}) for t in TRIPS: out.add(t[1]) if t[2]: out.add(t[2]) return out def self_check(): """Fail loudly rather than publish a study built on an inconsistent world.""" problems = [] for f in F: if f["source"] not in SOURCES: problems.append(f"{f['fid']}: unknown source {f['source']}") if f["valid_to"] is not None and f["valid_to"] < f["valid_from"]: problems.append(f"{f['fid']}: valid_to before valid_from") # the six planted conflicts must actually be conflicts today for attr in ("home_address", "mobile_phone", "monthly_salary_eur", "blood_type", "subscription_cloudline"): if not conflicted(attr): problems.append(f"{attr}: expected a live conflict, found none") if conflicted("diet"): problems.append("diet: expected the later self-report to stand alone") # expiries the study depends on for attr in ("passport_number", "health_insurance_policy", "phone_warranty_until"): if not expired(attr): problems.append(f"{attr}: expected to be expired as of TODAY") if expired("driving_licence"): problems.append("driving_licence: expected to still be valid") if len(people_named("Alex")) != 3: problems.append("expected exactly three people called Alex") disagree = {a for a in CONFLICT_ATTRS if naive_value(a) != value_at(a)} if disagree != {"home_address", "monthly_salary_eur", "subscription_cloudline"}: problems.append(f"conflict balance drifted: naive differs on {sorted(disagree)}") return problems if __name__ == "__main__": bad = self_check() print("facts:", len(F), "people:", len(PEOPLE), "trips:", len(TRIPS), "sources:", len(SOURCES)) print("current address:", value_at("home_address")) print("current salary:", value_at("monthly_salary_eur")) print("blood type:", value_at("blood_type")) print("address on 2023-01-01:", value_at("home_address", date(2023, 1, 1))) print("passport expired:", expired("passport_number")) print("self check:", "OK" if not bad else bad)