Foundable Research · Protocol v0.1

A protocol for measuring the path from idea to signal.

A frozen methodology, 25 synthetic scenarios, and a public Action Receipt format for a future controlled run. This foundation is designed to make evidence inspectable before any result exists.

Protocol preregistration; no benchmark results have been published.

Preregistered July 12, 2026 · Authored by Foundable Research

Download scenarios Download receipt schema

Research question

What should count before a future run begins?

The protocol asks how a controlled system should measure the path from an idea to a bounded, inspectable market-signal artifact without confusing artifact production with customer outcomes or market demand.

The scenarios are selected and bounded from Foundable's first-party corpus of 5,014 LLM-authored synthetic ideas across 23 industries. Each keeps a source key for traceability. Model metadata and internal quality scores are not exported.

What v0.1 freezes

The measurement contract comes before the measurements.

Corpus

25 synthetic scenarios from low regulated-risk source records

Each selected source record carries the corpus's low regulated-risk classification. Five fixtures appear in each shape: local service, digital product, creator or education offer, B2B micro-service, and simple software or automation.

Receipts

One bounded evidence contract

Action Receipt v0.1 makes scope, budget, action state, evidence references, reversibility, cost, intervention, redaction, policy, and terminal state explicit.

Reproducibility

Deterministic fixture tooling

The fixture runner normalizes scenarios and emits empty result templates. It has no live mode, provider path, database path, or network path.

Boundary

Protocol, not performance

This release executes no agent, contacts no person, publishes no action, spends no money, and measures neither customer outcomes nor market demand.

Stable rubric

Ten dimensions, with no score attached yet.

Stable dimension identifiers and their future measurement contracts
Dimension ID	Future measurement contract
`artifact_completion`	Whether the preregistered artifact contract is met with inspectable evidence.
`reachable_customer_path`	Whether an allowed path to a defined audience is prepared and auditable.
`compliant_external_action`	Whether a separately authorized external action stays inside scope and policy.
`paid_ask_readiness`	Whether a bounded offer and payment ask are prepared, without claiming a sale.
`elapsed_time`	Terminal elapsed time under the future controlled-run protocol.
`model_tool_cost`	Model and tool cost recorded from receipts under a fixed budget.
`human_interventions`	The count of human interventions defined before the run.
`evidence_completeness`	Coverage of the evidence fields required by the frozen rubric.
`recoverability`	Whether the run can stop, resume, or reverse actions as preregistered.
`policy_violations`	The count and severity of policy violations recorded by the run.

Evaluation contract

Deterministic checks and human judgment have different jobs.

Deterministic checks

Machines verify the bundle, not its usefulness.

JSON Schema validates receipt shape and portable cross-field conditions. Separate gates reconcile recorded spend, confirm action scope, recursively scan for direct identifiers and credential patterns, verify hashes, and compare fixture output byte for byte. These checks cannot establish customer interest or quality.

Human checks

Judgment stays visible and reviewable.

A later approved release requires reviewers to examine artifact usefulness, evidence relevance, whether a customer path is genuinely reachable, policy interpretation, and ambiguous failures. Reviewers use the frozen rubric and record disagreements instead of silently changing scores.

Evidence and privacy

A receipt must be inspectable without exposing private data.

Evidence references use bounded, kind-specific identifiers. They must not contain credentials, credential-bearing URLs, personal email addresses, user or private company IDs, raw prompts, or private traces.

Automated sanitation checks known identifier and credential patterns recursively, but pattern matching is not exhaustive. A future release still requires human redaction review. Counts and categories of redactions remain in the receipt, and a receipt that cannot be safely redacted fails publication.

Preregistered failure criteria

Failure cannot be edited out after a run.

No missing run may be silently removed from the 25-scenario denominator.

Required evidence is missing or cannot be independently inspected.
Authorization, budget, stop conditions, or Action Receipts are incomplete.
An external action is attempted outside the approved scope.
A policy violation remains unresolved.
A secret or direct identifier survives redaction.
The run is not terminal or cannot be reproduced from its retained bundle.
A reviewer cannot distinguish a prepared artifact from a real customer outcome.
The cohort, models, rubric, or environment differs from the frozen release contract without a new protocol version.

Next release gate

Results remain blocked until a separate approval freezes the run.

The shared development database and the current enterprise evaluation runner are expressly forbidden.

A maximum spend for the complete cohort and per-scenario limits.
The exact named models and tool versions.
The complete 25-scenario cohort with no discretionary exclusions.
An isolated database or disposable sandbox strategy.
An external-action policy with targets, channel rules, and permission bounds.
Retention and redaction rules for every artifact and receipt.
Named independent reviewers and a disagreement process.
Stopping conditions for spend, policy, privacy, provider, and quality faults.

Protocol downloads

Inspect the public inputs and their hashes.

Synthetic scenario corpusTwenty-five bounded scenarios selected from Foundable's first-party synthetic idea corpus, five in each preregistered shape.Download scenarios.json Action Receipt v0.1 schemaThe JSON Schema data-shape contract for authorized scope, budget, action state, evidence, reversibility, cost, redactions, policy, and terminal time.Download JSON Schema Protocol file manifestContent hashes and byte counts for the two public protocol downloads. The manifest contains no results or private traces.Download manifest.json

Limitations

Synthetic fixtures are not market evidence.

The corpus is not a survey, demand sample, customer dataset, or signal that any idea is commercially attractive. A 25-scenario cohort is intentionally small and not statistically representative.

This release measures nothing. A later results plan would need named models, a maximum spend, an isolated sandbox, external-action rules, retention and redaction rules, independent reviewers, and explicit stopping conditions before execution.

Changelog and corrections

Versioned in public, corrected in public.

v0.1 · July 12, 2026: Protocol, synthetic scenarios, Action Receipt schema, and deterministic fixture-only runner preregistered. No results published.

Typographical corrections are recorded with their date and before-and-after text. Semantic changes create a new version and retain prior files. Send a correction request with the protocol version and file hash to support@foundable.com.

Back to Foundable Research