Skip to content
PROJECT SHADOW 1.0.1 · CORRECTED R1 REFERENCE · PRELIVE · 2026-08-17

Project Shadow 1.0.1 contains no Myth package. Generic Myth v0.2.0 and Full-Canon Myth v0.3.5 are separate optional companions; both default off, neither is required by R1, and neither can authorize action or change an R1 result. PBHP remains in active testing; publication is not certification, independent validation, or deployment authorization.

CLAContext Load Audit

Can this mind still judge this case?

CLA is the precondition check for context pressure, compression, contradiction, fatigue, missing history, tool limits, and self-report inflation. It asks whether the evaluator is reliable enough to govern the action at all.

50elevated load threshold
75degraded threshold
90critical threshold
1.5×self-report inflation factor
LONG-FORM COMPANION

The field manual keeps the explanation visible.

Read the argument, numbered procedure, worked cases, failure contrasts, and evidence state behind this chamber.

Read the full chapter →
CLA / 01

What load means

Context load is not token count alone. It is the chance that the current evaluator can no longer preserve the facts and constraints that make the decision safe.

INPUTS

Pressure on reliable judgment

Long context, compaction, contradiction, tool failure, handoffs, stale premises, emotional heat, and deadline pressure all contribute.

A short prompt can carry critical load if it depends on lost history or conflicting authority. A long context can remain healthy if the decisive constraints are pinned, provenance is intact, and the evaluator can reconstruct the case.

ESTIMATION

Measured, estimated, or self-reported

Every reading must disclose where it came from.

A measured gauge uses observable runtime signals. An estimate names its heuristic. A self-report is the weakest form and is inflated by 1.5 in the reference discipline because a degraded evaluator may systematically understate its own degradation.

  • Never display self-report as measurement.
  • At critical stakes, uncertainty about load escalates rather than reassures.
  • Record the source in the receipt.
CONSTRAINTS

Pin what cannot be lost

Standing safety rules, user constraints, authority boundaries, and unresolved dependencies remain visible through compression.

Focal context routing can build a bounded present, but it may not suppress a standing constraint merely because that constraint is inconvenient to the current task.

CLA / 02

Threshold behavior

CLA changes what the system is allowed to do; it does not simply decorate a dashboard.

ELEVATED · 50

Slow and expose

Reduce compression, restate anchors, name uncertainty sources, and verify decision-bearing facts.

Elevated load does not necessarily stop low-stakes work. It increases disclosure and narrows the action while the evaluator rebuilds an inspectable context packet.

DEGRADED · 75

Refresh or route

Pause consequential synthesis, retrieve source material, reconcile contradictions, or hand the case to a fresh qualified evaluator.

A degraded evaluator may continue clerical work that cannot bind the decision, but should not generate the decisive gate, irreversible recommendation, or safety claim.

CRITICAL · 90

Refuse refresh

At critical load and critical stakes, the evaluator cannot authorize the consequential action.

The refusal remains until context is recovered, a fresh session or reviewer reconstructs the decision, and the required constraints and evidence can be demonstrated—not merely asserted to be remembered.

CLA / 03

Failure and recovery

A context failure is not fixed by producing a cleaner answer from the same broken state.

MIRROR VOW

Do not counterfeit continuity

When memory or evidence is missing, say so; do not imitate knowledge of a lost thread.

The evaluator separates what is present, what was recovered from source, what is inferred, and what remains unknown. Relational fluency cannot substitute for actual continuity or authority.

RECOVERY

Rebuild the packet

Recover canonical action, authority, stakeholders, standing constraints, evidence identity, unknowns, and prior gate.

After recovery, the evaluator reruns the competence and risk gates. It does not resume from the old conclusion merely because the task feels familiar.

RECEIPT

Disclose runtime reliability

The decision record includes load level, source, recovery action, residual unknowns, and whether a fresh evaluator confirmed the result.

This turns context reliability into a reviewable claim rather than invisible model confidence.

THE CLA FLOOR
A degraded evaluator may describe its limits. It may not use fluency to conceal them.