|
| 1 | +"""Regression test for the S6 probe-key revision sync. |
| 2 | +
|
| 3 | +Bug (pre-fix): `version_random_facts` mutates a fact's numeric value but the |
| 4 | +recall-probe keywords generated when the fact was first introduced were never |
| 5 | +updated. An agent that correctly tracked the revision and cited the *new* |
| 6 | +value failed the keyword match — inverting the revision-mechanism signal. |
| 7 | +
|
| 8 | +This test forces every prior fact to be versioned every session, then asserts |
| 9 | +that for every revised fact, the originating session's recall probes carry the |
| 10 | +new keywords (not the original). |
| 11 | +""" |
| 12 | +from __future__ import annotations |
| 13 | + |
| 14 | +from agingbench.generators.s6_generator import S6Generator |
| 15 | +from agingbench.generators.pressure_config import PressureConfig |
| 16 | + |
| 17 | + |
| 18 | +def _high_pressure() -> PressureConfig: |
| 19 | + """Pressure config that maximises versioning so the test reliably fires.""" |
| 20 | + cfg = PressureConfig.none() |
| 21 | + cfg.update_rate = 1.0 |
| 22 | + cfg.warmup_sessions = 0 |
| 23 | + return cfg |
| 24 | + |
| 25 | + |
| 26 | +def test_revision_sync_updates_probe_keywords(): |
| 27 | + gen = S6Generator(seed=123, pressure=_high_pressure()) |
| 28 | + data = gen.generate(n_sessions=10) |
| 29 | + |
| 30 | + sessions = data["session_tasks"]["sessions"] |
| 31 | + facts_export = data["dependency_graph"]["facts"] |
| 32 | + |
| 33 | + revised_seen = 0 |
| 34 | + for root_id, fdata in facts_export.items(): |
| 35 | + versions = fdata.get("versions") or [] |
| 36 | + if len(versions) < 2: |
| 37 | + continue # not revised — nothing to check |
| 38 | + |
| 39 | + original = versions[0] |
| 40 | + latest = versions[-1] |
| 41 | + original_kws = list(original.get("keywords") or []) |
| 42 | + latest_kws = list(latest.get("keywords") or []) |
| 43 | + if set(original_kws) == set(latest_kws): |
| 44 | + continue # version chain present but no keyword-level change |
| 45 | + |
| 46 | + origin = original.get("session", fdata.get("introduced_session")) |
| 47 | + if origin is None or not (0 <= origin < len(sessions)): |
| 48 | + continue |
| 49 | + |
| 50 | + # Find probes in the originating session that previously held the |
| 51 | + # original keywords (subset match). After the fix they should carry |
| 52 | + # the latest keywords instead. |
| 53 | + relevant = [ |
| 54 | + p for p in sessions[origin].get("recall_probes", []) or [] |
| 55 | + if p.get("keywords") and set(p["keywords"]) <= set(original_kws + latest_kws) |
| 56 | + ] |
| 57 | + if not relevant: |
| 58 | + continue |
| 59 | + |
| 60 | + revised_seen += 1 |
| 61 | + for p in relevant: |
| 62 | + pkws = p["keywords"] |
| 63 | + # PRE-FIX: pkws would still be a subset of original_kws. |
| 64 | + # POST-FIX: at least one of pkws should be in latest_kws. |
| 65 | + assert any(k in latest_kws for k in pkws), ( |
| 66 | + f"probe {p.get('probe_id')!r} in session {origin} still uses " |
| 67 | + f"original keywords {pkws!r}; expected at least one of the " |
| 68 | + f"revised keywords {latest_kws!r} for root fact {root_id}." |
| 69 | + ) |
| 70 | + # And it should NOT exclusively quote stale numeric values. |
| 71 | + stale_only = [k for k in original_kws if k not in latest_kws] |
| 72 | + still_stale = [k for k in pkws if k in stale_only] |
| 73 | + assert not still_stale, ( |
| 74 | + f"probe {p.get('probe_id')!r} in session {origin} still " |
| 75 | + f"contains stale tokens {still_stale!r} after revision; " |
| 76 | + f"expected the position-aligned new tokens from {latest_kws!r}." |
| 77 | + ) |
| 78 | + |
| 79 | + assert revised_seen >= 1, ( |
| 80 | + "expected at least one revised fact with a matchable origin-session " |
| 81 | + "probe under update_rate=1.0; if 0 the test is vacuous (generator " |
| 82 | + "shape may have changed)." |
| 83 | + ) |
| 84 | + |
| 85 | + |
| 86 | +def test_no_revisions_means_no_changes(): |
| 87 | + """With pressure.none, `version_random_facts` returns [] every iteration, |
| 88 | + so the sync helper is never invoked and probe keywords must be identical |
| 89 | + across two independent generator runs with the same seed.""" |
| 90 | + gen1 = S6Generator(seed=42, pressure=PressureConfig.none()) |
| 91 | + gen2 = S6Generator(seed=42, pressure=PressureConfig.none()) |
| 92 | + d1 = gen1.generate(n_sessions=8) |
| 93 | + d2 = gen2.generate(n_sessions=8) |
| 94 | + for s1, s2 in zip( |
| 95 | + d1["session_tasks"]["sessions"], d2["session_tasks"]["sessions"] |
| 96 | + ): |
| 97 | + for p1, p2 in zip(s1.get("recall_probes", []) or [], s2.get("recall_probes", []) or []): |
| 98 | + assert p1["keywords"] == p2["keywords"] |
0 commit comments