The visual record includes evidence maps, research pipelines, runtime architecture, and method history. Each diagram states what it represents and whether it describes current work or an earlier method.
01 · Evidence Sankey
This generated D3 diagram maps tagged signal mentions from Beta 1.0 manual evals through outcomes and signal categories into OCR weighting.
Link width represents evidence weight.
Use Evidence to read the counts | Open the generated Sankey source
02 · OCR research pipeline
The OCR progress funnel shows source episodes becoming strict cases, replayed OCR checks, stability results, and follow-up groups.
Use Research to read the OCR lane | Open the Evidence and OCR diagram set
03 · Beta evidence pipeline
This pipeline links manual and screenshot-backed evidence, the shift into binary gates, current eval reports, the Sankey payload, and curated visual outputs.
Use Evidence to read the source notes | Open the tracked evidence-map definition
04 · Runtime pipeline
Polinko is also a runtime system. Its architecture connects the server and configuration layers to the API, core chat and retrieval behaviour, local databases, and binary eval reports.
Use Method to read the gate logic | Open the runtime architecture
05 · Frozen last-beta pipeline
This tracked Beta 2.3 pipeline describes case or row judgement and the post-failure decision to keep a valid failure or remove malformed, duplicate, or noisy source material.
This remains part of Polinko's evidence history. The current method follows a source-first research model.
Use Method to compare current and historical gates | Open the frozen eval contract diagrams
06 · Historical fail-pressure pulse proposal
This proposal used small, bounded chatbot experiments to test whether timed checkpoints could preserve a behavioural signal before it was summarised as a pattern.
One mini chatbot and one exact behavioural boundary.
Behaviour emerges with source context and corrections preserved.
The checkpoint records only pass or fail.
Repeated, changed, recovered, or newly failed signals are compared.
One meaningful binary checkpoint attached to its source interaction.
A behavioural shape begins to recur or change across multiple pulses, sessions, or toys.
The research model determines whether the evidence earns a formal case, question, or bounded claim.
Polinko keeps this visual as method history. Current work uses the source-first method described on the method page.
Use Method to read the current method | Read the historical pulse hypothesis
