Fountain-code decoding: count, rank and peeling Hesela, 2026-11-21 This exact, synthetic example uses three source symbols a, b and c, with known-correct XOR equations and known coefficients from one source block. An equation's integer mask selects inputs: 1=a, 2=b, 4=c. We inspect coefficient matrices, not received network packets. Received equations Count Rank Inputs peeled a^b, a^c, b^c 3 2 0 a^b, a^c, a^b^c 3 3 0 a^b, a^c, a^b^c, a 4 3 3 a^b, a^c, b^c, a^b 4 2 0 a, b, c 3 3 3 The first set is dependent: its three equations XOR to zero. In the second, a is the XOR of all three outputs, b is the XOR of outputs two and three, and c is the XOR of outputs one and three. The solution exists uniquely, but pure degree-one peeling cannot start. An extra equation a permits peeling, even though rank stays three. A duplicate of a^b adds neither independent information nor a starting singleton to the first set. These are five deliberately chosen sets, not a random sample. No rate estimate or confidence interval follows from them. Zero or duplicate equations are accepted by the demonstrator to test algebraic behavior. Reproduce using Node.js 22 or later, with no packages: node model.mjs --check node --test model.test.mjs To print the observations: node model.mjs Only --write updates observations.json. The tests independently enumerate the null space for all 512 ordered 3-by-3 binary matrices, including zero rows, and check all eight binary payloads for the full-rank stalled example. The 168 full-rank matrices include 18 with no initial singleton. This finite enumeration is not an LT or RaptorQ reception experiment. Sources and reading map: - Michael Luby, LT Codes, FOCS 2002, sections 1.1-1.2, 2 and 3. https://pages.cs.wisc.edu/~suman/courses/740/papers/luby02lt.pdf - Amin Shokrollahi, Raptor Codes, IEEE TIT 52(6), 2551-2567, 2006, sections III-IV (rank, decoder and precode). https://www.qualcomm.com/content/dam/qcomm-martech/dm-assets/documents/Raptor_Codes_IEEE_technical_analysis.pdf - RFC 6330, sections 5.3.1, 5.8 and 6 (RaptorQ padding, conditional recovery requirements, integrity and authentication). https://www.rfc-editor.org/rfc/rfc6330.html The original calculation and tests are Hesela's work, with no copied paper code or figures. No degree distribution, transmission process, production codec or physical storage failure is simulated.