Date: 2026-10-27. Status: proposed implementation for human review.
Inspected README, AGENTS, handoff/roadmap/buildable steps, historical project profile, both research charters, continuity register/lineage/evidence manifest, RFC catalog and RFC-00D0/00D1, existing evidence validators, experiments, Python packaging and CI conventions. Existing structured capabilities include the continuity hypothesis register and source pins, experiment-specific results and hash manifests, RFC-00D1/00D3 JSON schemas, and AI execution evidence validators. Searches of scripts, tools, research docs, workflows and pyproject found no MCP server or common research interaction interface. Existing agent/TUI/runtime APIs are execution surfaces, not a scientific evidence registry.
GitHub PR metadata and merged repository artifacts were inspected:
827d2accbf8ccb56afe3fe8d96dfb7251f249d7d into
research/investigation09-tisc-language-design, not main. Its E1 packet
under experiments/investigation09_e1/ reports NOT SUPPORTED for the scoped
executable-footprint discriminator. Its protocol is at
f1054438b805b541c4acfd294e6dba5a71711c05; implementation baseline is
c930b8c2ee7478782bf3c6d231bdd59984595ef2. No measurements rerun here.18bd6c78b1ec2c78a500a8e290cb004a062f513e on main:
CAPABILITY_BASELINE.md, RFC_EVIDENCE_MATRIX.csv, TISC_T81LANG_COVERAGE.md,
RESEARCH_GAPS.md and AUDIT_REPRODUCIBILITY.md. These are reported audit
claims, not independently verified test runs. The packet describes the earlier
c930b8c baseline, omits merged E1 evidence, mixes ISA versions, cites missing
paths, and promotes some comparative claims beyond identified controls.274421fd8e85c5b9bb4f165fce190715530f4e85 on main.
HUMAN_AI_RESEARCH_INITIATIVE.md supplies T81-ORI collaboration intent, but its
section 4.2 reassigns canonical H01–H08, calls H07 comparatively validated,
and conflates H08 with H-S1. Its April date predates its October merge;
percentages and AI methodological benefits are unvalidated assessments.The continuity charter and original register remain authoritative for H01–H08. ROOT_DOCUMENTATION_ALIGNMENT_AUDIT.md qualifies comparative assurance claims: selected output parity is not equivalent computation or assurance. Existing records remain intact; conflicts become discoverable discrepancy entries.
Add a versioned JSON index of existing evidence. Each source has a full Git commit, repository path, SHA-256 and immutable URL. Read Git objects rather than mutable checkout files, including #605’s separate merged history. Missing objects fail closed and require explicit human setup; the interface never fetches. Canonical claims, results, confidence and falsifiers are returned verbatim from the pinned register, with supplementary evidence and unresolved conflicts linked separately. Hashes prove byte identity, not correctness or authenticity.
Evaluate MCP as the preferred future integration transport. For this initial foundation choose a justified equivalent: six allowlisted offline JSON CLI operations with strict versioned requests and deterministic response envelopes. No MCP server exists to extend; adding a server SDK, lifecycle and transport would increase scope before the evidence contract is reviewed. The CLI is not MCP-compatible and does not claim MCP conformance. Future MCP tools can wrap these operations after human review without changing the evidence model.
Use Python 3.9+, Git, and a pinned JSON Schema validator; no C++ runtime changes, network client, database, execution operation, shell interpretation or output-file option. Separate proposal/review/reproduction/discrepancy schemas from findings. Every request validates the manifest, references and source identities before answering. No llms.txt: AGENTS plus the Commons README provide discovery without another overlapping root narrative.
This supports newcomer/handoff clarity and existing evidence surfaces, rather than expanding draft runtime architecture. It does not authorize experiments, production substitution seams, RFC acceptance, publication, PR merging, changes to frozen/accepted artifacts or Protocol 001 activation. Human authorization is recorded as data, never consumed as an execution capability. Self-authored tests are implementation checks, not an independent scientific review. Coverage is a curated initial index, not an exhaustive research census. Historical register statuses are explicitly dated; supplementary records do not silently revise them.
Test path/object resolution, historical identities, negatives, claim boundaries, reference failures, deterministic responses, all operations, and absence of repository mutations. Use a separate Python CI matrix; no root CMake changes.
Validation found that continuity/evidence/manifest.json records charter hash
e56d1cb5..., while actual committed charter bytes at the inspected baseline
hash to bb073baf.... The canonical register hash matches. The Commons indexes
both records and flags this mismatch; it does not repair or silently trust the
older packet manifest. Git lazy fetching and optional locks are disabled; the
implementation requires Git 2.46+ conservatively, using its
documented environment contract.