Repository navigation
v0.3.0 — Adaptive Discovery
An update for a more natural, informed discovery conversation. Choose how you
want to collaborate, explore useful research together, and decide when your
idea is ready for the next step.
What changed
Choose A) Guide me through it, B) Work alongside me, C) Take the research and
planning lead, or D) Challenge my existing plan. These change communication and
initiative, not permissions or evidence standards.
The agent uses the 33 anchors as coverage, not a spoken sequence. It builds on
supplied context, investigates externally answerable gaps, recommends decisions,
and recognizes when enough is known for a bounded next commitment.
Research now has three explicit levels: Light Scan, Standard Research, and
Deep Research. Dedicated investigations have accepted purpose, scope, effort,
evidence standards, and return points. Quick authorized checks do not require
a new packet. Research can answer several anchors without re-questioning.
Contained cleanup can use recorded route-only self-review, with human acceptance.
Cycle review/handoff and material routing/authority repairs still require an
independent resumption check. Self-review must never be labeled independent.
Why we changed it
The owner found the opening too generic and potentially survey-like. This
revision aims for an informed, constructive exchange: helpful research at the
right moment, clear recommendations, and fewer repetitive questions. Those are
intended experience benefits, not proven productivity or adoption outcomes.
What stays the same
QDI remains the Complete starter's entrance. Human gates, evidence distinctions,
mandatory pushback, bounded authority, cleanup, and honest usage reporting stay
in force. The three DM foundations are unchanged.
Complete has 96 files; Light 14; Standard 26; Governed 34. Counts are distribution
contents, not daily reading requirements. The reusable starter and tools remain
MIT-licensed; separate explanations remain CC BY 4.0.
Verification
Ten local verification groups pass, including twenty-four injected faults,
source/download parity, reproducible archives, local link checks, and selected
written authority/routing controls. The expanded opening route stays within
the existing 4,200-word ceiling. Three fictional-project conversations informed
two contained repairs with focused retests. Two blind cross-agent resume checks
passed their declared routes; the technical review required its routed cleanup
and Blueprint beyond the core five files. All source/tool fixtures were
fictional, not real provider research. A passing text check does
not prove reliable behavior across agents. No unattended delivery, security
certification, or token-saving claim is made.
Choosing a version
v0.2.0 and its original assets
remain available. For the new conversation and research guidance, choose the
Complete starter v0.3.0 ZIP below. The three direct foundation ZIPs omit QDI;
their template content is unchanged. Do not overwrite populated project records
with blank templates. Compare instructions,
preserve accepted decisions/evidence/authority, and review the affected route.