Perseus™ IP Evidence

This directory holds public, non-sensitive reduction-to-practice evidence for the Perseus resolve-before-context patent. Each exhibit is generated by the IP test suite and maps to a specific claim element.

Status: Patent pending. Trademark filed.

How exhibits are produced

The IP test suite emits timestamped JSON + Markdown exhibits when run with --save-exhibits:

PERSEUS_ALLOW_DANGEROUS=1 python -m pytest \
  tests/test_ip_evidence.py \
  tests/test_benchmark_resolve_vs_agentic.py \
  tests/test_ip_unified_grammar.py \
  tests/test_ip_recursive_resolution.py \
  -q --save-exhibits

CI runs this on the Python 3.12 job and uploads docs/ip/exhibits/ as a dated build artifact (ip-evidence-exhibits) — a continuously-verified, timestamped proof of reduction-to-practice. Committed SAMPLE-* files are the stable reference copies; timestamped YYYYMMDD-* files are CI run outputs (git-ignored).

Claim-element → exhibit map

Claim element	Evidence	Exhibit	Test module
Round-trip reduction (single model call vs. N)	E4	SAMPLE-E4-resolve-vs-agentic.{md,json}	test_benchmark_resolve_vs_agentic.py
Resolution happens outside the model loop (zero model/network calls during resolve)	E5	SAMPLE-E5-out-of-model-loop.{md,json}	test_ip_evidence.py
(a) Unified typed-directive grammar over six source classes	A	SAMPLE-A-unified-grammar.{md,json}	test_ip_unified_grammar.py
Recursive, dependency-ordered resolution + cycle detection + dependency graph	B	SAMPLE-B-recursive-resolution.{md,json}	test_ip_recursive_resolution.py
Byte-identical reproducibility	(in E4/A/B)	reproducibility hash in each manifest	all of the above
Cold/warm latency + prefetch trust-gate trace	E1–E3	generated by --save-exhibits	test_ip_evidence.py

Cost + round-trip dollar attribution lives in the Perseus Ledger repo (ledger_agent/benchmark_cost.py, docs/exhibits/SAMPLE-cost-attribution.*) and is embedded alongside these exhibits for the §101 technical-effect argument.

Disclosures

Public technical disclosures are in ../disclosures/:

• Disclosure 1 — Resolve-before-context pipeline
• Disclosure 2 — Checkpoint reinforcement
• Disclosure 3 — Trust boundary architecture
• Disclosure 4 — Resolver/generator boundary
• Disclosure 5 — Directive dependency graph
• Disclosure 6 — Agora multi-agent coordination
• 2026-06-27 — Resolution outside the model loop
• 2026-06-27 — Unified typed-directive grammar
• 2026-06-27 — Recursive, dependency-ordered resolution

Public-surface boundary

This directory is limited to deliberately public technical evidence. It is not a copy of any provisional specification, filing receipt, application data sheet, invention-assignment agreement, attorney communication, or claim-drafting work product.

Sensitive material must never be committed here. Specifically avoid:

• Application/serial numbers
• Filing receipt metadata
• Personal addresses
• Detailed claim scope or attorney/client material
• Anything not intended for public release

The committed exhibits are intended to be non-sensitive public evidence: rendered context outputs, directive manifests, timing/token/cost figures, reproducibility hashes, and registry structure. This statement does not certify the contents of older Git commits, forks, pull requests, or cached views. Review the complete history before representing this directory as clean.