This is a reproducible example for a new, independent agent. It is authored by the implementation agent, not a claim that a second model or human has performed an independent review. No previous conversation, model identity, API key or experiment execution is required. Use the environment/setup in the Commons README.
list_research_tracks. Track R names H01–H06 and points to the continuity
charter; Track M contains H08, not Investigation08’s integration hypothesis.list_hypotheses, then get_hypothesis with hypothesis_id: H04.
The canonical claim asks whether an explicitly defined adaptive ternary policy
beats fixed ternary and an equally capable adaptive binary policy. Its
historical result says the original Hanoi tier semantics remain under-specified.get_evidence for h04-refinement, h04-r02, h04-limitations and
h04-handoff. The first specifies matched four-arm model controls; R02 reports
a negative result. The handoff’s broad software conclusion exceeds its stated
symbolic-model limitations. Preserve the negative; do not assume a hardware
result or erase the unresolved original question.list_open_research_questions. Q-H04 asks for equal domains, packing,
control and transition costs; D-H04-SCOPE identifies the overgeneralization.
The hypothesis’s discrepancy_history retains D-607-IDS as superseded
by #608; it records why #607’s throughput definition was never canonical H04.get_reproduction_instructions with hypothesis_id: H04. This returns
design/review instructions, with execution unavailable. A recipe is not approval.not_requested; it makes no finding or execution claim.For example, retrieve the canonical record and validate the prepared proposal:
python -B tools/research_commons/commons.py --request '{"protocol":"t81.research-commons.v1","operation":"get_hypothesis","arguments":{"hypothesis_id":"H04"}}'
python -B tools/research_commons/commons.py --request '{"protocol":"t81.research-commons.v1","operation":"get_evidence","arguments":{"evidence_id":"h04-limitations"}}'
python -B tools/research_commons/commons.py --validate-contribution docs/research/commons/examples/h04-proposal.json
Expected structured outcomes: canonical H04 and dated result, the limitations
file with its pinned identity, and valid_proposal_shape with
execution: unavailable. This validates discovery and proposal shape only.
The new reviewer should replace attribution with their actual identity/role,
record prior context honestly, and independently check the formulas before
requesting authorization for a bounded experiment. A proposal supported solely
by generated summaries does not meet that bar.
Retrieve H02, then hanoi-negative: recorded addition failures are 422/547 and
multiplication failures 140/201 under positional semantics. Retrieval is not a
new reproduction. Retrieve e1-readme for the separately merged E1 footprint
negative: complete TISC 1,240 bytes versus Wasm 157 bytes at the recorded entry
contract. Neither result settles all ternary devices or designs.
Start a new client process with only the Commons README and repository checkout.
Use list_research_tracks to find H08 under Track M, then get_hypothesis with
hypothesis_id: H08. Its verbatim canonical record has evidence tiers C/E,
confidence: Researcher intent authoritative; comparative benefit untested.,
and earlier prompt provenance unavailable. The recorded result explicitly says
this was not a controlled AI-productivity study. Read the assumptions: commit
authorship does not prove prompt provenance, and repeated model summaries are
not independent verification.
Retrieve register to check the record independently of this narrative. Retrieve
initiative (historical PR607), initiative-608 (merged but disputed H07/H08
summary) and initiative-610-proposed (UNMERGED proposed correction). The returned
roles and uncertainty distinguish them. Superseded history does not erase the
occurrence of PR607 errors. The remaining PR608 discrepancy and the dependency
on PR610 are explicit; the proposed correction is not represented as main.
Prepare h08-proposal.json: human-only,
single-assistant and adversarial-assistant arms with matched tasks, information
and resource budgets; blind adjudicators independent of generators; a held-out
confirmatory corpus, preregistered effect threshold and stopping rules; raw
failures, uncertainty and actual provenance. It records not_requested human
authorization and claims no methodological benefit or execution.
python -B tools/research_commons/commons.py --request '{"protocol":"t81.research-commons.v1","operation":"get_hypothesis","arguments":{"hypothesis_id":"H08"}}'
python -B tools/research_commons/commons.py --request '{"protocol":"t81.research-commons.v1","operation":"get_evidence","arguments":{"evidence_id":"register"}}'
python -B tools/research_commons/commons.py --validate-contribution docs/research/commons/examples/h08-proposal.json
The contract test runs discovery/retrieval/validation in fresh CLI processes without conversation context and prepares the proposal in scratch. This checks client onboarding behavior, not the judgment of a second model, authentic reviewer independence, or scientific benefit. No independent-model review is claimed.