Dontopedia

ConstellationDecoder

From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)

ConstellationDecoder has 52 facts recorded in Dontopedia across 12 references, with 7 live disagreements.

52 facts·39 predicates·12 sources·7 in dispute

Mostly:has parameter count(4), rdf:type(4), dominates parameter count(2)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (11)

Other subjects in dontopedia point AT this entity as a value. These are inverse relationships — e.g. "X motherOf this subject" — and answer questions the forward facts can't. Grouped by predicate.

involvesEntityInvolves Entity(2)

conditionalOnDecoderConditional on Decoder(1)

hasClassHas Class(1)

lacksLacks(1)

mattersOnlyWithMatters Only With(1)

mentionsComponentMentions Component(1)

needsBuildingNeeds Building(1)

proposedAsDecoderProposed As Decoder(1)

relevanceConditionRelevance Condition(1)

usesUses(1)

Other facts (49)

The long tail: predicates that appear too rarely to warrant their own section. Filter or scroll to find a specific one. Each row links to its source.

49 facts
PredicateValueRef
Has Parameter Count32000[4]
Has Parameter Count32000[8]
Has Parameter Count32000[9]
Has Parameter Count23808[12]
Rdf:typeDesign Principle[7]
Rdf:typeModel Component[8]
Rdf:typeMode or Component[10]
Rdf:typeModel Component[12]
Dominates Parameter Counttrue[2]
Dominates Parameter CountModel[4]
Present in VariantsModel Variant a Global[6]
Present in VariantsModel Variant C Coherence Gated[6]
Constitutes Percentage of94[8]
Constitutes Percentage of84[12]
Part of ModelModel Variant a[11]
Part of ModelModel Variant C[11]
Is Most Interesting Part{}[1]
Has One PerByte[1]
AvoidsLinear D to 256[1]
Avoids Flat Projection{}[1]
Is Manifold Aware{}[1]
ReplacesLinear Readout[1]
ComputesHermitian Similarity to Byte Templates[1]
Computes Hermitian Similarity BetweenOutput State and 256 Learned Constellation Points[1]
Has Dimensions256 × 125[2]
Param Dominant Componenttrue[2]
Constitutes Percentage of Model94%[2]
Has Memory ComplexityO(T × 256)[3]
Implies Vocab Size256null[3]
Has Compute ComplexityO(256 × readout_dim)[3]
Constitutes94 Percent ofModel Parameters[4]
Exhibits Design SmellDesign Smell[4]
Dominates Total Params0.84[5]
Has Parameters23808[5]
Uses Readout Dim93[5]
Produces Higher Entropy Outputs onSynthetic Patterns[6]
At Expense ofSynthetic Patterns[6]
Part ofDelta Memory Variants[6]
Allocates Capacity toNatural Text Distribution[6]
Specialized forLm Text[6]
Uses MethodHermitian Similarity[7]
Excludes ComponentLinear D 256[7]
Characterized Asmost interesting part[7]
Computation Methodcompute Hermitian similarity between the output state and 256 learned constellation points[7]
Propertymanifold-aware rather than a flat projection[7]
Constitutes Percentage of Model Parameters94[9]
Described Asa design smell[9]
ContrastsSymmetric Mode[10]
Has Readout Dimension256 × 93[12]

Timeline

Timeline axis is valid_time — when each source says the fact was true in the world, not when Dontopedia learned about it. Retracted rows are kept for provenance; coloured stripes indicate the context kind.

isMostInterestingPartblah/watt-activation/part-384
{}
hasOnePerblah/watt-activation/part-384
ex:byte
avoidsblah/watt-activation/part-384
ex:linear-d-to-256
avoidsFlatProjectionblah/watt-activation/part-384
{}
isManifoldAwareblah/watt-activation/part-384
{}
replacesblah/watt-activation/part-384
ex:linear-readout
computesblah/watt-activation/part-384
ex:hermitian-similarity-to-byte-templates
computesHermitianSimilarityBetweenblah/watt-activation/part-384
ex:output-state-and-256-learned-constellation-points
hasDimensionsblah/watt-activation/part-385
256 × 125
paramDominantComponentblah/watt-activation/part-385
true
constitutesPercentageOfModelblah/watt-activation/part-385
94%
dominatesParameterCountblah/watt-activation/part-385
true
hasMemoryComplexityblah/watt-activation/part-401
O(T × 256)
impliesVocabSize256blah/watt-activation/part-401
null
hasComputeComplexityblah/watt-activation/part-401
O(256 × readout_dim)
hasParameterCountblah/watt-activation/part-402
32000
dominatesParameterCountblah/watt-activation/part-402
ex:model
constitutes94PercentOfblah/watt-activation/part-402
ex:model-parameters
exhibitsDesignSmellblah/watt-activation/part-402
ex:design-smell
dominatesTotalParamsblah/watt-activation/part-607
0.84
hasParametersblah/watt-activation/part-607
23808
usesReadoutDimblah/watt-activation/part-607
93
producesHigherEntropyOutputsOnblah/watt-activation/part-606
ex:synthetic-patterns
atExpenseOfblah/watt-activation/part-606
ex:synthetic-patterns
partOfblah/watt-activation/part-606
ex:delta-memory-variants
presentInVariantsblah/watt-activation/part-606
ex:model-variant-a-global
presentInVariantsblah/watt-activation/part-606
ex:model-variant-c-coherence-gated
allocatesCapacityToblah/watt-activation/part-606
ex:natural-text-distribution
specializedForblah/watt-activation/part-606
ex:lm-text
typeblah/watt-activation/382
ex:DesignPrinciple
usesMethodblah/watt-activation/382
ex:hermitian-similarity
excludesComponentblah/watt-activation/382
ex:linear-d-256
characterizedAsblah/watt-activation/382
most interesting part
computationMethodblah/watt-activation/382
compute Hermitian similarity between the output state and 256 learned constellation points
propertyblah/watt-activation/382
manifold-aware rather than a flat projection
typeblah/watt-activation/383
ex:ModelComponent
labelblah/watt-activation/383
constellation decoder
constitutesPercentageOfblah/watt-activation/383
94
hasParameterCountblah/watt-activation/383
32000
labelblah/watt-activation/400
ConstellationDecoder
constitutesPercentageOfModelParametersblah/watt-activation/400
94
hasParameterCountblah/watt-activation/400
32000
describedAsblah/watt-activation/400
a design smell
typeblah/watt-activation/406
ex:ModeOrComponent
contrastsblah/watt-activation/406
ex:symmetric-mode
partOfModelblah/watt-activation/603
ex:model-variant-a
partOfModelblah/watt-activation/603
ex:model-variant-c
typeblah/watt-activation/604
ex:ModelComponent
labelblah/watt-activation/604
Constellation (decoder)
hasParameterCountblah/watt-activation/604
23808
hasReadoutDimensionblah/watt-activation/604
256 × 93
constitutesPercentageOfblah/watt-activation/604
84

References (12)

12 references
  1. [1]Part 3848 facts
    ctx:discord/blah/watt-activation/part-384
  2. [2]Part 3854 facts
    ctx:discord/blah/watt-activation/part-385
  3. [3]Part 4013 facts
    ctx:discord/blah/watt-activation/part-401
  4. [4]Part 4024 facts
    ctx:discord/blah/watt-activation/part-402
  5. [5]Part 6073 facts
    ctx:discord/blah/watt-activation/part-607
  6. [6]Part 6067 facts
    ctx:discord/blah/watt-activation/part-606
  7. [7]3826 facts
    ctx:discord/blah/watt-activation/382
    • full textwatt-activation-382
      text/plain3 KBdoc:agent/watt-activation-382/7400f8cd-0f1a-470c-9260-c96918758eae
      Show excerpt
      [2026-03-19 01:37] xenonfun: ⏺ This is gold. Let me check the sweep results while I synthesize the architecture idea: ``` Here's the stripped-down principled design, pulling from the harmonic-gpt findings: "Resonant Wire" — Helmholtz+Cli
  8. [8]3834 facts
    ctx:discord/blah/watt-activation/383
  9. [9]4004 facts
    ctx:discord/blah/watt-activation/400
    • full textwatt-activation-400
      text/plain2 KBdoc:agent/watt-activation-400/bfd3ae1a-a87b-4ef9-bd0b-548fd78cc0cb
      Show excerpt
      [2026-03-19 05:11] xenonfun: ⏺ The ConstellationDecoder is 94% of the model's parameters (32K of 27K dynamics). That's a design smell. ``` The most elegant option: use the encoding table itself as the decoder. The BPSK table maps each byt
  10. [10]4062 facts
    ctx:discord/blah/watt-activation/406
    • full textwatt-activation-406
      text/plain1 KBdoc:agent/watt-activation-406/d69009ec-32be-45d6-b3d2-0b3143e59d42
      Show excerpt
      [2026-03-19 06:36] xenonfun: ⏺ Here's what each knob does: G (groups) = 8 — How many independent oscillator groups. Each group reads a different bit of the byte. More groups = more parallel information channels, but more Givens coupling
  11. [11]6032 facts
    ctx:discord/blah/watt-activation/603
    • full textwatt-activation-603
      text/plain3 KBdoc:agent/watt-activation-603/7aa9fa83-a058-43b0-b700-073a69d3e610
      Show excerpt
      [2026-04-10 04:25] xenonfun: ``` ⏺ Now we have real eval results on the trained models. Key observations: Results Summary (10K steps, 50% curriculum, trained model eval) ┌─────────────────┬────────────┬───────────────┬────────────────
  12. [12]6045 facts
    ctx:discord/blah/watt-activation/604
    • full textwatt-activation-604
      text/plain2 KBdoc:agent/watt-activation-604/5619f2b9-2e60-495c-a6e6-9fe7ca80a696
      Show excerpt
      [2026-04-10 04:27] xenonfun: ``` From the training output logs, the models we've been testing have 28,525 total parameters: - Constellation (decoder): 23,808 params (256 × 93 readout dim) - Readout: 153 params (mode amplitudes, pro

See also

Keep researching

Missing something or suspicious of what's here? Kick off a research session — a Claude agent will investigate, cite its sources, and file new facts into a dedicated context you can review before accepting into the shared view.