Q: Kelly CIS Pipeline — What Actually Triggers READY vs NOT-READY?

Short answer: The KB documents that the CIS loop gates on READY/NOT-READY, but the operational completeness criteria — source count minimums, synthesis depth, evidence thresholds — are not yet documented. This is a critical gap: the gate can only be as reliable as its trigger criteria.

Context

The CIS (Context → Information → Synthesis) loop is Stage 2 of Kelly's six-stage factory pipeline. It runs as a research-lead operation, potentially spawning multiple parallel sub-agents to cover different information sources simultaneously. The loop's output is a gate decision:

  • READY → advance to Planning
  • NOT-READY → keep researching or surface specific open questions

The 75-minute CSV Export case study shows CIS running fast, but doesn't document what "enough" means in terms of evidence. The Kelly handbook and tweets describe the structure of CIS but not the stopping conditions.

What the KB Has

From kelly-factory-overview and kelly-handbook-ch11-software-factory:

  • CIS has three phases: Context → Information → Synthesis
  • The research-lead produces a research-summary.md with three synthesis outputs: what we know, what we don't know, what to do next
  • READY/NOT-READY is a binary gate — pipeline does not advance until READY
  • Parallel sub-agents can cover multiple sources simultaneously
  • The TEA audit (Testing stage) has three outcome tiers — CIS has only two

The gap-analysis corpus (kelly-gas-town-gap-analysis) rates CIS as a Kelly strength but doesn't audit the completeness criteria of the READY decision itself.

What's Missing

1. Source Completeness Criteria

How many sources are required before READY is a valid decision?
- Is there a minimum number (e.g., 3+ credible sources)?
- Is there a saturation threshold (e.g., "new sources stop adding signal")?
- How are source quality and recency weighted?

The cis-pipeline concept file doesn't specify. The 75-minute example implies minimal research, but there's no documented minimum.

2. Synthesis Depth Definition

What does "good enough synthesis" look like?
- The three-output format is documented (research-summary.md: what we know / what we don't know / what to do next)
- But what distinguishes a PASS synthesis from a NOT-READY one?
- Is there a "known unknown" threshold — a point at which additional research into unknown unknowns has diminishing returns?

3. NOT-READY Trigger Criteria

What specific conditions produce NOT-READY?
- Insufficient source count?
- Conflicting information without resolution?
- Critical open questions that would block planning?
- The answer shapes how aggressive the research-lead is at advancing vs looping

4. Sub-Agent Parallelism Termination

When parallel research sub-agents are spawned, how does the research-lead know when all have reported in and synthesis is complete? Is there a "last-to-report" timeout? A "sufficient coverage" heuristic?

5. Kelly's Own CIS Practice

Kelly's tweets and the case study show rapid CIS cycles. Do they use a "best effort in time box" approach (e.g., "research for 20 minutes then synthesize regardless") or a "read until饱和" approach? The KB doesn't say.

Proposed Research Path

  1. Read Kelly's CIS loop definition — extract from kelly-handbook-ch11-software-factory.md and any CIS-specific concept files in compiled/concepts/cis-pipeline.md
  2. Check tweets for Kelly's READY criteria — kelly-tweets-pipelines.md or kelly-tweets-factory.md may have concrete examples of what triggers NOT-READY
  3. Cross-reference SuperAda's ISC criteria — SuperAda uses Ideal State Criteria per task; Kelly CIS READY criteria may map to a similar structure
  4. Extract from 75-minute case study — what was the actual evidence base that triggered READY in the CSV Export example? If this isn't documented, request it from the operator