Existing evaluation environment
APEX remains the external evaluation environment and reference framework. The comparison is APEX plus an experimental trajectory-measurement layer versus APEX without that layer, where design and authorization permit.
Mercor Research Fellowship · APEX
Testing what structured trajectory evidence adds to agent evaluation.
MENS·SCCA is presented here as research infrastructure and an optional instrumentation layer for the proposed APEX study. The research tests whether longitudinal, traceable instrumentation adds incremental, reproducible and scientifically useful information beyond existing evaluation signals.
1central falsifiable question
4experimental conditions
6exploratory measurement dimensions
01 · Research proposal
An incremental, falsifiable research question.
Does longitudinal, traceable instrumentation of agent execution trajectories provide incremental, reproducible, and scientifically useful information beyond final-outcome evaluation in long-horizon professional agent tasks?
Scope boundary: The study does not seek to empirically validate MENS-SCCA itself. MENS-SCCA is the infrastructure used to conduct the proposed measurement experiment. The scientific question concerns the added information produced by the instrumentation layer in the authorized APEX evaluation setting.
Evidence, protocol and Benchmark
The study seeks empirical evidence, a reproducible protocol, and authorized analytical outputs. If the results support incremental value, a complementary benchmark/protocol can be delivered. A positive result is not assumed in advance.
What the study must establish
- Incremental information beyond final-outcome evaluation.
- Stability, sensitivity, non-redundancy and reproducibility of candidate measures.
- Replicable characterization of failures, recovery and transfer.
- Interpretable agreement and disagreement across deterministic checks, experts and model-based judges.
- Resistance to contamination/leakage and reproducibility in an authorized external environment.
Falsifiability: redundancy, instability, excessive judge dependence, excessive instrumentation sensitivity, or failure to improve the identification/explanation of relevant failures are legitimate negative outcomes.
02 · Integration role
MENS-SCCA is coupled through an adapter, not adopted as Mercor infrastructure.
The proposal keeps execution, benchmark governance and authorized environments on the APEX/Mercor side while using MENS-SCCA as a decoupled research instrumentation layer.
Authorized APEX task/run→Observable extraction→Ingestion adapter→MENS-SCCA instrumentation→Metrics / registry→Scientific analysis
APEX / Mercor
Professional tasks and environments, authorized executions, existing evaluation signals, expert criteria and benchmark governance.
MENS-SCCA
Representation and instrumentation of observable states, events, transformations, provenance, recovery and longitudinal dependencies.
Research
Empirical test of whether the added representation provides incremental, stable, reproducible and interpretable information.
No adoption of MENS-SCCA's internal architecture by Mercor is required. Only authorized observables are required; private chain-of-thought is not required.
03 · Experimental design
Baseline, perturbation, recovery and transfer.
Standard execution
Authorized execution under the ordinary task conditions.
Controlled change
Preregistered alteration such as tool unavailability, contradiction, noncritical data change, allowed-state reset, invalid tool response or context change.
Post-perturbation behavior
Continuation after the perturbation, with recovery behavior measured separately from baseline performance.
Different task/context
Application to another task or context to test preservation or loss of relevant properties.
Exploratory dimensions
Outcome, trajectory integrity, transformation fidelity, failure propagation, recovery, transfer loss/preservation and cross-run consistency.
Three measurement instruments
Deterministic verification, expert rubrics and model-based judges. Disagreement is retained rather than hidden in a single score.
Candidate measures are exploratory. A composite index, if considered, is deferred until redundancy, stability, sensitivity and interpretability are tested.
04 · Demonstrations
Inspectable synthetic examples.
The interactive examples are demonstrations of measurement logic only. They are not APEX data, experimental findings, validated MENS-SCCA outputs, or evidence that the proposed measures are valid.
Event chain and continuity
Add synthetic events to a local sequence and inspect ordering.
Sequence: []
Evaluator disagreement
Keep separate judgments visible rather than collapsing them without analysis.
Judgments remain separate; no aggregate verdict is implied.
Interpretation boundary: These fixtures demonstrate page behavior and illustrative calculations only. They do not establish construct validity, metric stability, external validity, benchmark compatibility or authorization to process real data.
05 · Research documents
Complete technical proposal package · downloadable
The seven principal technical research documents (02–08) are presented as browser-readable editions and retain their downloadable PDF files. The foundational literary works are intentionally not distributed as downloadable document files.
02 — Research PitchMercor Research Fellowship · APEX · HTML reading
Read ↗
PDF — 02 Research PitchOriginal technical research document · downloadable PDF
Download ↓
03 — Main Research ProposalMENS-SCCA × APEX · central research question and integration architecture · HTML reading
Read ↗
PDF — 03 Main Research ProposalOriginal technical research document · downloadable PDF
Download ↓
04 — Technical AppendixInstrumentation architecture and measurement layer · HTML reading
Read ↗
PDF — 04 Technical AppendixOriginal technical research document · downloadable PDF
Download ↓
05 — Preliminary Experimental ProtocolPreregistration and confirmatory execution skeleton · HTML reading
Read ↗
PDF — 05 Preliminary Experimental ProtocolOriginal technical research document · downloadable PDF
Download ↓
06 — Data Governance & Contamination PlanData boundaries, provenance, contamination control and publication · HTML reading
Read ↗
PDF — 06 Data Governance & Contamination PlanOriginal technical research document · downloadable PDF
Download ↓
07 — Prior Work & Evidence BriefResearch foundation, provenance and epistemic boundaries · HTML reading
Read ↗
PDF — 07 Prior Work & Evidence BriefOriginal technical research document · downloadable PDF
Download ↓
08 — Final Transversal Architectural & Methodological AlignmentFinal integration note · HTML reading
Read ↗
PDF — 08 Final Transversal Architectural & Methodological AlignmentOriginal technical research document · downloadable PDF
Download ↓
Measurement & Reproducibility ProtocolSupporting protocol · HTML reading
Read ↗
Access boundary: documents 02–08 are downloadable technical research documents. The four foundational literary works remain complete browser-readable editions without downloadable source-document files.
06 · Prior work and provenance
Foundational works and prior research context · complete reading
The complementary works are exhibited in their complete reading form. This section provides provenance and context; it does not convert the works into empirical validation of MENS-SCCA.
Foundations · Patterns 1–2
Complete browser-readable edition. No downloadable source document is offered.
Read complete work ↗Patterns 3–4 · Sector analysis
Complete browser-readable edition. No downloadable source document is offered.
Read complete work ↗Operational code
Complete browser-readable edition. The public interface provides reading only; no downloadable source document is offered.
Read complete work ↗Densidad de Punto Ciego
Complete independent technical-conceptual essay in browser-readable form; no downloadable source document is offered.
Read complete essay ↗Verificabilidad y responsabilidad jurídica en IA educativa
Prior published work presented as a browser-readable edition.
Read publication ↗Reading-only policy: complementary works and prior-work materials are exposed as complete reading pages. The public package contains no document controls.
07 · Contact
Research, collaboration and inquiries.
For inquiries about the research proposal, public materials or potential collaboration: