Resource-Aware Next-Signals — Validation Checklist

Goal

Provide a short, repeatable routine for validating the Resource-Aware Next Signals discipline against real, accumulated slice evidence, once it exists.

This is the deferred phase-5 step for the AletheIA 1.2 resource-aware track as a whole. It is the sibling of the Agent Harness Governance — Validation Checklist: the next-signals roadmap shipped docs-first as a passive watch-list; this routine is how that watch-list gets exercised against evidence instead of left to intuition.

The discipline is the same as the next-signals doc itself and the skill-evolution thesis: evidence is an input, not an authority. A single slice — however interesting — never justifies reopening the 1.2 track toward 1.3. Only cross-slice or cross-project repetition does.

When to run this

Run the routine when multiple real (non-synthetic) slice records have accumulated, such as:

• a batch of Runtime Effort Governance per-slice records emitted by real work;
• review notes from more than one comparable slice across one or more projects;
• recurring local rules that several projects independently invented for the same operational pressure (restart-package tightening, review-pause timing, slice stopping criteria).

Do not run it on a single slice, a single fixture, or one unusually hard task. The next-signals doc is explicit that those are inputs, not thresholds (resource-aware-next-signals.md §“Signals that are not enough by themselves”). Running this routine on one record proves nothing.

Step 1 — Gather comparable slices

Collect at least two real per-slice records in the runtime-effort-governance-contract shape, drawn from comparable work. For each, note:

• the task_type, classification (reversibility, blast radius, uncertainty, risk), and user_intent;
• the effort path (initial_effort → final_effort) and the escalation_reasons / stop_reason;
• whether a human_checkpoint_triggered and whether quality_floor_maintained;
• the originating project, kept as a coarse label only.

Strip secrets and vendor identifiers before storing anything. Keep the gathered set provider-agnostic.

Step 2 — Map slices to the four healthy signals

For each candidate signal from resource-aware-next-signals.md §“Healthy signals to watch for”, tally which gathered slices exhibit it:

A signal that appears in only one slice is not a hit. Record it as an input and move on.

Step 3 — Apply the threshold rule

A signal counts toward reopening the track only when it is cross-slice or cross-project, not anecdotal. Resolve each tallied signal against the rule in resource-aware-next-signals.md §“What can justify 1.3+”:

• one interesting example, one hard slice, one project’s preference, one vendor choice → input, not threshold;
• the same pattern across more than one believable slice, with comparable review language → a real signal.

If no signal clears this bar, the routine ends at keep 1.2 stable. That is the expected outcome most of the time, and it is a healthy result, not a failure.

Step 4 — Record findings

Write down, per gathered set:

• which of the four signals fired, and in how many slices / projects;
• whether each firing signal was cross-project or confined to one project;
• the outcome distribution (reinforced / no-change / other) across the slices;
• which local translations recurred, if any;
• mapping gaps — any real signal the four-signal taxonomy has no slot for. A mapping gap is itself a finding, the same way it is in the AHGE checklist.

If the optional resource-aware-signal-evidence record is in use, capture this review as one conforming record so the decision stays explicit, traceable, and reconcilable with the signal catalog.

Step 5 — Decide (with restraint)

Reopening the 1.2 track toward 1.3+ comparative evaluation is justified only when the combination named in resource-aware-next-signals.md §“What can justify 1.3+” holds:

• more than one believable real-world slice; and
• comparable review language across those slices; and
• enough structure to compare without hiding important local differences; and
• no need yet for learning-layer or auto-routing claims.

In order of discipline:

1. repeated evidence first, 2. comparative framing second, 3. benchmark packaging only after that.

If the evidence is a single anecdote, do not reopen the track. Record the observation and wait for a second comparable signal — the same posture ADR-012 names as the trigger to revisit. The reopen decision is always a human judgement; nothing in this routine reopens the track automatically.

What stays out of scope

Even with stronger signals, these stay deferred unless the evidence becomes unusually strong, per resource-aware-next-signals.md §“What should still stay deferred”:

• vendor ranking as core truth;
• auto-routing claims;
• learning-layer behavior;
• orchestration-heavy policy machinery;
• benchmark packaging before repeated evidence and comparative framing exist.

The framework stays provider-agnostic and review-oriented.