Reference implementation for the Phoenix Architecture. Work in progress.
aicoding.leaflet.pub/
ai
coding
crazy
experiment: SAME_TYPE_REFINE_THRESHOLD 0.3→0.2 score=0.9640
D-rate now 9% average — well under the 20% target. Three specs at 0%.
Auth v1/v2 and Notifications have zero relates_to fallback edges.
Gateway still at 27% — largest spec with most same-type pairs.
+3
-1
+2
experiments/results.tsv
+2
experiments/results.tsv
···
16
16
2026-03-26T22:23:36.322Z 0.8791 100.0 91.7 95.5 60.7 100.0 6.2 rgwg4j
17
17
2026-03-26T22:23:50.652Z 0.8861 100.0 94.4 95.5 60.7 100.0 6.2 qwrndy
18
18
2026-03-26T22:46:38.390Z 0.9061 100.0 94.4 95.5 47.4 100.0 6.2 nq75ia
19
+
2026-03-26T23:12:45.269Z 0.9265 100.0 94.4 95.5 33.8 100.0 6.2 x0da3a
20
+
2026-03-26T23:13:07.799Z 0.9640 100.0 94.4 95.5 8.8 100.0 6.2 42knqt
+1
-1
src/experiment-config.ts
+1
-1
src/experiment-config.ts
···
12
12
JACCARD_DEDUP_THRESHOLD: 0.7,
13
13
FINGERPRINT_PREFIX_COUNT: 8,
14
14
DOC_FREQ_CUTOFF: 0.5,
15
-
SAME_TYPE_REFINE_THRESHOLD: 0.3,
15
+
SAME_TYPE_REFINE_THRESHOLD: 0.2,
16
16
17
17
// ─── canonicalizer.ts — scoring weights ───────────────────────────────────
18
18
CONSTRAINT_NEGATION_WEIGHT: 4,