PLOTWARDEN BY COGNICLUSTER

PlotWarden keeps the development plot true.

When intent, code, evidence, and active work change at once, PlotWarden reconciles the work already in flight instead of making a team rebuild the plan by hand.

Follow one change Review the evidence

Intended end-state, demonstrated behavior, and open claims are labeled separately.

Change order — BS-2026-07-04 Illustrative scenario
Project
Black Skies
Epoch
EPOCH-2026-07-04.01
Source
Crew Combat Rules (ratified source)
Scope
Crew combat behavior and its causal consumers
Change (ratified)

CCR-2026-07-04-MORALE

Ratified rule change: morale affects card availability during combat, superseding the after-combat rule.

Ratified intent (unchanged)Approximately 10,000 simultaneous captains at full simulation speed, without time dilation.

Illustrative scenario: identifiers, timestamps, and details are invented for teaching. The scale target shown is ratified intent, not a measured result.

Why this matters

Enterprise software teams use agentic development in changing products.

For enterprise software teams, a constrained task can start correct. In a changing product, a new premise can make dependent work stale while it is still running.

  1. Premise changes
  2. Dependent work becomes stale
  3. Humans reconstruct context and re-plan
Review the evidence behind the problem

IN PLAIN TERMS

PlotWarden keeps AI coding work aligned when instructions change before results become official.

AI coding assistants can keep working from outdated instructions, producing polished work that is still wrong. In 2025, one deleted a company's production data. PlotWarden is an early system that checks proposed work against the latest approved plan before results become official.

Think of it as air-traffic control for AI coding: when instructions change while work is underway, it decides what may continue, what must stop, what needs repair, and what needs a person before the result counts. It also contains disputed information so downstream work does not trust it. An analogy for coordination, not a certification claim.

Today: a public specification, twenty control rules, dated build and evidence records, and a proving ground. What it does not yet claim is on the limits page; the full mechanism steps through on the system plot.

One change. Four different outcomes.

Workers return evidence. PlotWarden controls official state.

  • Continue
  • Retain & retest
  • Conditional repair
  • Recall & re-plan
Work itemCurrent statusWhy this outcomeBefore (ratified intent)After (updated direction)
Starbase contentART-PIPE-101 Continue No causal dependency.This work does not rely on crew-combat rules. The asset and content pipeline is independent of combat behavior. Continue as planned. No change required.
Doctrine testsDOCTRINE-TST-845 Retain & retest Timing expectations changed.The tests encode when morale applies to card availability. Assertions were valid while morale applied only after combat. Retain the tests; rerun and recalibrate them against the during-combat rule.
Combat APICOMBAT-API-V3 Conditional repair Repair only if its contract moves.The interface changes only if mid-combat morale must be exposed. The contract is stable while morale stays out of live combat. Monitor contract impact. Repair only if the ratified rule reaches the interface.
Combat tutorial scriptTUT-CMB-012 Recall & re-plan Core premise superseded.The script teaches the old morale timing throughout. The tutorial was authored against after-combat morale. Recall the script and re-plan it against the ratified rule.
Impact frontier
Trace causal relationships from the changed rule to active work; unaffected work remains outside the repair frontier.
Evidence returned
worker tool events · patches · test results · benchmark results · support sets · artifact hashes
Official outcome
A manager-owned gate checks evidence, policy, relevant authority, and the current epoch, and commits in one atomic versioned operation against the canonical graph root.

One change travels through a governed loop.

Reality can change corrective planning without silently lowering ratified intent. Every transition has a route into the public model.

  1. 01 Change or evidence arrives

    A report enters as a claim with provenance, not as an instruction.

    Open System
  2. 02 Trust and authority policy evaluates it

    Source, actor, scope, change type, environment, risk, evidence, and epoch are checked.

    Open Authority
  3. 03 Policy or a human ratifies, rejects, or quarantines it

    Human review is reserved for ambiguity, high risk, and policy exceptions.

    Open Authority
  4. 04 The impact frontier is computed

    Causal dependencies define what is affected; unknown coverage taints conservatively, and unrelated work stays outside the frontier.

    Open Reconciliation
  5. 05 Work continues, retests, repairs, or re-plans

    Current orders receive differentiated treatment instead of a blanket restart.

    Open Reconciliation
  6. 06 Workers return evidence through controlled boundaries

    Workers may act within sealed scope, but they cannot write official state.

    Open Workers
  7. 07 The decision is recorded; the epoch advances; current orders are reissued

    A manager-owned commitment gate records the result before revised work becomes current.

    Open Evidence Epoch N+1: revised sealed orders issued
Text equivalent: follow the complete loop
  1. Change or evidence arrives. A report enters as a claim with provenance, not as an instruction. Read System.
  2. Trust and authority policy evaluates it. Source, actor, scope, change type, environment, risk, evidence, and epoch are checked. Read Authority.
  3. Policy or a human ratifies, rejects, or quarantines it. Human review is reserved for ambiguity, high risk, and policy exceptions. Read Authority.
  4. The impact frontier is computed. Causal dependencies define what is affected; unknown coverage taints conservatively, and unrelated work stays outside the frontier. Read Reconciliation.
  5. Work continues, retests, repairs, or re-plans. Current orders receive differentiated treatment instead of a blanket restart. Read Reconciliation.
  6. Workers return evidence through controlled boundaries. Workers may act within sealed scope, but they cannot write official state. Read Workers.
  7. The decision is recorded; the epoch advances; current orders are reissued. A manager-owned commitment gate records the result before revised work becomes current. Read Evidence.

Final current state: Epoch N+1: revised sealed orders issued.

Claim verification and poison containment

A contradicted supporting claim is marked poisoned and traced to every known consumer. Unsafe output is quarantined, official commitment is blocked, and verification or repair work is ordered before the current epoch may advance.

Inspect evidence and containment

Four structures, one maintained plot.

Two living maps, one active plan, and one canonical evidence ledger. The ledger is the canonical mutation history; the maps and the plan are versioned, rebuildable projections that stay linked without confusing what is known, what is authorized, and what is currently in flight.

  1. 01

    Ratified Intent Graph

    What the system is ordered to become.

    Authority model
  2. 02

    Evidence-Backed Reality Graph

    What current evidence supports.

    Evidence posture
  3. 03

    Active Execution Graph

    What work is in flight and what it depends on.

    Reconciliation
  4. 04

    Immutable Evidence and Command Ledger

    Why the plot changed and who authorized it.

    Complete system

Execution boundary

Normal commands in. Evidence out. Authority stays protected.

Workers can edit, test, inspect, and invoke familiar build commands inside scoped environments. A policy router may keep light work local or move heavy builds to isolated shared capacity.

Logs, hashes, test results, artifacts, and support sets return as evidence. Workers do not merge, deploy, publish, or alter official state directly; only a manager-owned gate can commit those effects, and it validates freshness and commits in one atomic versioned operation.

Review workers and remote builds
  1. Familiar command
  2. Local or remote execution
  3. Evidence bundle
  4. Current-epoch gate

Evidence earns the right to make a claim.

We have not found the full loop documented as one system in the reviewed tools. That is a point-in-time research finding, not a claim to be first, only, or unique.

Intended
The public specification defines the governed end-state and its invariants.
Specification
Demonstrated
Dated Build Records separate observed behavior from architectural intent.
Build RecordsEvidence
Unproven
Scale, adoption, reconciliation accuracy, and comparative advantage remain open claims.
Limits

Why now / incubation

Agentic development is ubiquitous. Continuous reconciliation is not.

PlotWarden is seeking an incubator prepared to challenge the reconciliation model, sharpen its enterprise path, and connect the project with software teams running substantial parallel agentic work.

Why now
Agentic development has become normal; continuously reconciling work that goes stale in flight has not.
Why CogniCluster
Black Skies supplies a demanding proving ground, the end-state specification is public, and PlotWarden is compared continuously with current best-in-class tools.
Still unproven
Frequency, measurable reconciliation cost, adoption path, competitive uniqueness, and durability remain validation questions—not traction claims.
Incubation validates
Whether enterprise teams value this control layer, which boundaries are essential, and what evidence should earn the right to scale it.
Review the incubation case Challenge the model

Follow the argument into the evidence.

The public site separates the end-state model from dated implementation facts so each claim can be challenged on the right terms.