HC HeartCore Ventures Commercial venture house Book a call

Continuity-Governed Prompting installation

Your coding agent said it was done. The tests passed. The work wasn't there.

That failure has a name: false-green completion. Continuity-Governed Prompting structurally prevents it by requiring produced evidence before an agent can declare a task complete. Five business days. Fixed $7,500 pilot. You own everything after delivery.

Book a discovery call Read the paper

30 minutes. We map your current agent workflow and tell you whether your agents actually exhibit false-green completion. If they don't, we say so. Paper DOI: 10.5281/zenodo.20234367.

Completion standard Done = evidence trail + passing verification
design-note.md run-record.md evidence.json

A green test with no evidence is not completion.

False-green audit Measure the failure before installation
5-day install Fixed $7,500 pilot, scoped handoff
Open standard Foundation methodology, commercial service
Published null No scope-drift reduction claim
You have felt this

False-green completion is invisible to the metric everyone uses.

Verification-only success criteria cannot catch work that was never asserted by the test suite. Better prompts and stronger models do not structurally fix a completion rule that accepts a passing check as proof of work.

The agent reported the task done. CI went green. You shipped it. The work was not actually there.
An agent completed a refactor by changing nothing and passing tests that did not assert the skipped work.
A teammate spent a morning discovering that done meant the agent had quietly done nothing.
You no longer fully trust the completion claim, so you re-check everything and lose the time agents were supposed to save.
When something an agent built turns out empty, you cannot reconstruct what it actually did or why it claimed success.
What we install

A scaffold that makes completion auditable.

We install Continuity-Governed Prompting, an open methodology published by the Heart AI Foundation. Your team keeps the same agent platforms. The scaffold changes how work is framed, bounded, verified, and proven.

01

Manifest

Names exactly which files each agent task can touch.

02

Slice lock

Ties each task to current repository state so agents do not continue from stale assumptions.

03

Verification mapping

Wires named checks into the protocol before completion.

04

Evidence trio

Produces a design note, run record, and evidence JSON per completed slice.

05

Stop conditions

Requires agents to halt and report when the task, lock, manifest, or protocol disagree.

Completed slice

Every completed task produces three artifacts.

This is why a CGP-governed agent cannot false-green a task. Done is not a sentence the agent emits. Done is a produced evidence trail plus a passing check. No artifacts, no completion.

design-note.mdWhy the slice exists and what changed run-record.mdCommands run, files touched, discrepancies evidence.jsonVerification result and next recommended step
The 5-day pilot

By day 5, your team has a working scaffold and one documented slice.

Day 1

Intake and assessment criterion

We map your workflow, agent platforms, and verification commands. We agree whether the pilot measures false-green rate or evidence-trail completeness.

Days 2-3

Installation

We install the scaffold in a branch, customize the manifest, wire verification commands, and configure templates for your work units.

Day 4

First working slice

We execute one real task end-to-end with your team observing or participating. The slice produces the full evidence trio.

Day 5

Handoff and assessment

We deliver the team guide, walk through the before-versus-after assessment, and leave you with a self-operable scaffold.

Service tiers

Start with measurement. Install only when the failure is present.

The audit is the right entry point if you are unsure. The measured false-green completion rate is the deliverable, and "you do not need the install" is a valid result.

Tier 1

False-Green Completion Audit

$2,500

We measure whether your agents actually exhibit false-green completion on a representative sample of your tasks, on your stack. One week.

If the audit finds a material rate and you proceed to Tier 2 within 30 days, $1,250 credits toward the install.
Tier 2

5-Day Protocol Install

$7,500 fixed

We install the CGP scaffold in your codebase, customize it to your stack, and execute the first working slice end-to-end. Up to 10 engineers.

You own all installed artifacts after delivery.
Tier 3

Team Enablement Workshop

$5,000

Half-day live training for up to 20 engineers covering the methodology, scaffold, failure mode, and adoption playbook.

Combine with Tier 2 for $3,000 additional.
Tier 4

Quarterly Agent Governance Retainer

From $5,000/quarter

Protocol audits, false-green completion review, scaffold tuning, evidence-health checks, and quarterly reporting.

Scales by team size.
Who this is not for

This service is probably not right if your agents do not exhibit false-green completion.

About the principal

Built by the methodology author.

Dylan D. Mobley is founder of HeartCore Ventures LLC and author of Continuity-Governed Prompting. He holds an MS in Digital Forensics from Champlain College. The forensic chain-of-custody discipline applied in CGP is transferred from evidence-handling practice; the methodology is independently authored and not institutionally endorsed by Champlain.

Founder, Heart AI Foundation Preregistered benchmark Published null result Commercial implementation
What you get

A working scaffold, not an overclaim.

  • A working CGP scaffold installed in your codebase.
  • Customization to your stack, verification commands, and team practices.
  • A documented first working slice with a complete evidence trail.
  • Written team handoff guide and ownership of all installed artifacts.
What we do not promise

Scope-drift reduction. We tested it and the result was null.

We preregistered scope-drift reduction as the primary hypothesis, ran a controlled benchmark, and the result did not support it. We published the null result with a deviation note rather than bury it. We do not sell scope-drift reduction. We sell the effect the benchmark demonstrated: prevention of false-green completion and a durable evidence trail.

Open standard

We publish the methodology and the benchmark, including the result that failed.

Continuity-Governed Prompting is published by the Heart AI Foundation under CC BY 4.0. The reference scaffold is MIT-licensed. HeartCore Ventures sells installation expertise, not exclusive access to the standard.

FieldAgentic Workflow ForensicsThe open Foundation field containing CGP's preventive placement. PreregistrationOSF recordRegistered benchmark plan and attached materials. Published paperZenodo recordReliability and auditability benchmark. DOI: 10.5281/zenodo.20234367. RepositoryBenchmark code and dataPublic task specs, scripts, data, figures, and manuscript.
FAQ

Frequently asked questions

What exactly is false-green completion?

An agent declares a task complete, your verification command returns green, and the substantive work was not performed. The completion claim is false; the verification signal is green; the conjunction is the failure.

Didn't your benchmark fail?

The primary registered hypothesis, scope-drift reduction, returned a null result, and we published it. The same benchmark demonstrated a different effect clearly: across all agents, work submission rose from 79 percent under baseline to 100 percent under CGP, concentrated in agents prone to false-green completion, with 98.6 percent evidence-trail completeness.

What if our agents do not have this problem?

Then you should not buy the install, and we will tell you so. The Tier 1 Audit measures your false-green rate on your stack. Agents already at ceiling get auditability from CGP, not a reliability jump.

How is this different from CLAUDE.md, Cursor rules, or Aider conventions?

Those are scoped configuration. CGP is scoped operational discipline with an evidence-gated completion criterion. Configuration tells the agent things to remember. The scaffold makes done mean produced evidence plus passing verification.

Can my team learn this from your docs without hiring you?

Yes. The methodology, reference scaffold, implementation guide, benchmark, and preregistration are open. We sell installation expertise, training, and ongoing governance.

Are you using AI tools during the engagement?

Yes, transparently. We may use Claude Code, Cursor, Aider, and the agent platforms covered by your engagement. We will not enter your code into third-party AI tools outside your approved toolchain without written permission.

Discovery call

Tell us what your engineering team's agent workflow looks like.

We will tell you, honestly, whether your agents exhibit false-green completion and whether CGP would change your outcomes. If yes, we scope a pilot. If no, we tell you what to do instead.

Book a discovery call

Many calls end with "your agents don't have this problem." That is a valid outcome and we will say it plainly.

Transparency note

Open methodology. Separate commercial implementation.

Dylan D. Mobley is the author of Continuity-Governed Prompting and founder of both HeartCore Ventures LLC and the Heart AI Foundation. The Foundation publishes the methodology openly, including the benchmark and its null result. HeartCore Ventures provides optional commercial implementation services. The methodology may be implemented by any party under its published license terms. Foundation and HeartCore Ventures are structurally separate; the Foundation does not grant exclusive vendor status to HeartCore Ventures.