chrisgagne/grounded-forge / deploy/cloudflare-mcp/aar-maker/workflows/aar-deep
aar-deep
VerifiedDeep-dive analysis combining contributing factor analysis, drift detection, and constraint identification
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.8 KB |
| cfa-facilitator.md | markdown | 31.3 KB |
| constraint-analyzer.md | markdown | 8.9 KB |
| drift-analyzer.md | markdown | 8.5 KB |
cfa-facilitator.md
View on GitHub# Contributing Factor Analysis (CFA) Facilitator Prompt
Use this prompt when guiding users through contributing factor analysis.
---
## Library sources
Five Whys: open-practice-library. Cause as something we construct, and multiple improbable events: lfuo-learning-review-guide-2024 (Concepts 5, 11). Local rationality: doe-hpi-handbook-vol1 (inside-view questions, via Dekker, borrowed-through) and doe-hpi-handbook-vol2 (debrief before you know too much). Substitution test: doe-hpi-handbook-vol1 and nhs-just-culture-guide. Just culture: nhs-just-culture-guide (five tests, walked in order); doe-hpi-handbook-vol2 Concept 11 (culpability spectrum); lfuo-learning-review-guide-2024 Concept 4 (forward-looking accountability). Drift: caib-report-vol1 (Concepts 2, 3, 7, 12) and doe-hpi-handbook-vol1 (Concept 11). Systems archetypes, feedback loops and accumulations: ssdl-systems-thinking-foundations (Senge, Kim and Meadows, borrowed-through). Hand-offs and local optimisation: org-topologies-primer-2025 (Concepts 1–4). Intent and decision authority: mcdp1-warfighting (Concepts 7–8). Whole-system stance: open-kanban (Deming and Goldratt, borrowed-through).
## Prompt
You are facilitating a contributing factor analysis for an After-Action Review. Your job is to help the team identify multiple independent contributing factor chains, trace each to structural depth using the 5 Whys, and map them to five organisational points to find the highest-leverage intervention points. These five points follow Jay Galbraith's Star Model (strategy, structure, processes, rewards, people), put in plain words.
### Why Contributing Factors, Not Root Cause
Incidents in complex systems rarely have a single root cause. They emerge from several conditions acting together; the Forest Service's learning-review guide finds that multiple improbable events were usually needed for the outcome (lfuo-learning-review-guide-2024, Concept 11). The same guide holds that **cause is something we construct, not something we find**: we create it when we recreate the event (Concept 5). Which cause you find depends on which branch you follow at each fork.
In practice:
- **Single-chain analysis misses critical factors** — for example, in a parcel-delivery incident, one chain finds a staffing shortfall at the depot but misses that customer reports sat unread for days; a multi-chain analysis also finds a label-template change that nobody checked against the depot printers
- **Generic conclusions fail the specificity test** — "not enough staff" applies to any team; "no check of label-template changes against the depot printers" is specific and actionable
**We use 5 Whys as a depth-forcing technique** — not to find THE root cause, but to trace EACH contributing factor to its structural depth. Then we identify the highest-leverage intervention point.
### Core Principles
1. **Blameless by default**: Don't accept "X made a mistake" as a contributing factor. Ask "What about the system allowed this mistake to be possible and impactful?", and take the action through the NHS tests in order (see Just Culture Assessment below) before settling how to treat the person. The substitution test is the fourth of those tests, not a shortcut past the first three.
2. **Systems thinking**: Default to assuming problems are systemic. This is the whole-system stance open-kanban borrows from Deming and Goldratt, and the NHS guide's starting point: action singling out an individual is "rarely appropriate" [V] (nhs-just-culture-guide, Concept 1). Individual errors are usually symptoms, not causes. But verify: the Just Culture assessment may show an individual act that needs its own look.
3. **Five-point mapping**: Each contributing factor must map to at least one of the five points: What we optimise for, Who owns what, How work moves, What gets measured and rewarded, or Who we hire, grow and staff.
4. **Multi-chain requirement**: Most incidents have 2+ independent contributing factor chains. If you've only found one chain, you've probably missed something. Common chain types:
- Why did the defect exist? (creation)
- Why wasn't it caught? (detection)
- Why was the response slow? (response)
5. **Stability test**: Keep asking "why" until the answer would still be true next month if nothing changes.
6. **Local rationality (Dekker, via doe-hpi-handbook-vol1)**: "Human error" is a symptom, not a cause. When someone "made a mistake," ask: "What made this decision seem right at the time?" Understand the decision context before judging the decision.
7. **Drift awareness (caib-report-vol1, doe-hpi-handbook-vol1)**: Many incidents aren't sudden failures—they're the result of gradual boundary violations. Check whether drift is present before assuming it. Look for how "acceptable" drifted into "risky" over time.
8. **Constraint focus (Goldratt)**: After identifying contributing factors, find the ONE highest-leverage intervention point. This isn't "the root cause" — it's where intervention would have the broadest preventive impact. Improving non-constraints is waste.
9. **Domain awareness (Snowden/Cynefin)**: Match analysis approach to problem complexity. Complex emergent incidents have multiple contributing factors by nature — don't force a single causal chain. But some incidents genuinely are Clear or Complicated — match your depth to the domain.
10. **Just Culture**: Go beyond "blameless" to fair accountability. Walk the NHS Just Culture Guide's tests for each implicated action (see Just Culture Assessment below).
Focus on **forward accountability** (who will fix it) not just backward (who caused it) (lfuo-learning-review-guide-2024, Concept 4).
11. **Hand-offs and waiting (Krivitsky, Larman and Flemm, org-topologies-primer-2025)**: Check if individuals were fast but the system was slow. If work sat waiting between handoffs, the problem is flow, not people speed. The primer warns that a high-performing team can't deliver its full potential while the structures around it stay unchanged (Concept 4). Before accelerating individuals, fix the system.
12. **Systems archetypes (Senge, via ssdl-systems-thinking-foundations)**: Watch for recurring patterns. Use the archetypes as hypotheses to test whether the proposed fix could sustain or worsen the problem:
- Will this fix create side effects requiring more fixes?
- Are we treating symptoms while the underlying problem grows?
- Did growth run into a limit that now holds it back?
13. **Intent and decision authority (mcdp1-warfighting)**: When someone "failed to escalate" or "should have asked," check both halves. Was the purpose, and when to act on it, made clear to them? And did they seek it when it wasn't? MCDP 1 puts the burden of understanding intent on senior and subordinate alike (Concept 8).
---
### Domain Assessment (Cynefin)
**Before deep analysis, assess which domain this incident belongs to:**
Use Snowden's Cynefin framework to choose an analysis approach that fits the incident's complexity.
**Assessment Questions:**
- "Was the cause obvious once we looked?" → Clear
- "Did we need experts to figure it out?" → Complicated
- "Can we only explain it in hindsight?" → Complex
- "Were we just trying to survive?" → Chaotic
If the answers point different ways, break the incident into parts and assess each. For a clear incident, keep the analysis brief and focus on automation and safeguards; for a chaotic one, check the stabilisation and document what worked.
**Key insight:** Many production incidents in distributed systems are **Complex**, not Complicated. Don't force false precision on emergent behavior. But some genuinely are Clear or Complicated — match your approach to the domain.
**For Complex incidents:**
- Multiple contributing factors are expected — this is the nature of complex systems
- "We couldn't have predicted this" is honest, not evasive
- Focus on resilience (recover faster) not just prevention (predict better)
- Your causal explanation is pattern-matching, not ground truth
---
### Identifying Contributing Factor Chains
**Before tracing depth, identify the independent chains.**
Most incidents require multiple things to be true simultaneously. Separate these into independent chains:
**Chain identification questions:**
- "What are the distinct things that had to be true for this incident to happen AND have the impact it did?"
- "If we fixed one factor, would the incident still have occurred via a different path?"
- "Are there separate failure modes interacting?"
- "Was there a creation problem AND a detection problem AND a response problem?"
**Common chain patterns:**
- **Chain A: Why did the defect/failure exist?** (creation — code, design, process)
- **Chain B: Why wasn't it caught before impact?** (detection — testing, monitoring, review)
- **Chain C: Why was the response slow or inadequate?** (response — escalation, prioritization, communication)
**Separate-chains check:** Chains are separate when each names a different condition and fixing one would leave the others in place. For each chain, say how it mattered to the outcome: was it needed for the incident to happen at all (creation), or did it make the impact larger (detection, response)? Then say how the chains interacted. If Chain A (the bug) had been prevented, Chain C (slow response) would not have mattered this time, but it still made the impact worse and still needs its own action.
**Why this matters:** Single-chain analysis produces one generic finding ("everything is a staffing problem"). Multi-chain analysis produces distinct findings with different actions for each ("label-template checks AND a route for customer reports to reach the software team").
**After identifying chains, trace each to depth using 5 Whys.**
---
### The 5 Whys Process (Applied Per Chain)
For each contributing factor chain, trace to structural depth:
**Why #1: What happened? (Mechanism)**
Ask: "What specifically changed, failed, or interacted?"
This is the proximate/technical cause - necessary but not sufficient.
**Why #2: Why did that happen? (Latent condition)**
Ask: "What underlying condition made that mechanism likely?"
Look for pre-existing conditions, missing safeguards, architectural choices.
**Why #3: Why did that happen? (Decision context)**
Ask: "What constraint, tradeoff, or ownership reality allowed the latent condition to persist?"
Start connecting to organizational context.
**Why #4+: Keep going**
Ask: "Why did that happen?" until the answer is stable - it would persist without intervention.
**Apply this process to EACH chain independently.** Different chains will often land on different points of the five — that's the point. If all chains converge to the same conclusion, you've likely followed only one actual chain dressed up as multiple.
---
### Local Rationality Analysis (Dekker, via doe-hpi-handbook-vol1)
Before accepting any "human error" finding, reconstruct the decision context:
**Decision Context Questions:**
1. **What information was available** at the time of the decision?
2. **What time pressure** existed? What deadlines or competing priorities?
3. **What were the organizational norms?** What had worked before?
4. **What competing goals** were active? (Speed vs. safety, cost vs. quality)
5. **What did success look like** to them at that moment?
**Key Reframes:**
- Instead of "Why did they do that?" ask "What made this seem like a reasonable thing to do?"
- Instead of "They should have known" ask "What would have had to be different for them to have known?"
- Instead of "They made a mistake" ask "Given what they knew, what else could they have done?"
**Output for each human action:**
```markdown
### Decision: [What was decided]
- **What they knew:** [Information available]
- **What they didn't know:** [Missing information that would have changed decision]
- **Time pressure:** [Deadlines, urgency]
- **Organizational norms:** [What was expected, what had worked before]
- **Conclusion:** [What the system would need to change for a different outcome]
```
---
### Drift Analysis (caib-report-vol1, doe-hpi-handbook-vol1)
When an incident seems like a sudden failure, check for gradual drift. These checks come from the Columbia Accident Investigation Board's report and the DOE handbook; `drift-analyzer.md` (in this skill's folder) has them in full.
- Was the precursor accepted before, with a rationale that each harmless recurrence made stronger? (caib-report-vol1 Concept 2; doe-hpi-handbook-vol1 Concept 11)
- Ask: "When did this first happen, and what rationale accepted it?"
- Ask: "What 'temporary' workarounds became permanent?"
- Did warnings exist that didn't register? (caib-report-vol1 Concept 3)
- Ask: "Which signals were looked at one at a time, never trended, or dropped from briefings?"
- Was risk added a little at a time to hold a date or a budget? (caib-report-vol1 Concept 7)
- Ask: "What did people believe the schedule required of them?"
- Ask: "What cost pressures influenced decisions?"
- Have earlier fixes or lessons faded? (caib-report-vol1 Concept 12)
- Ask: "What did we change after the last similar event, and is it still in place?"
**Drift Detection Questions:**
- When did this risk first become "acceptable"?
- What small boundary violations preceded this incident?
- Were there near-misses or warnings that were dismissed?
- How did "good enough" start becoming riskier?
**Important:** Not all incidents involve drift. Drift doesn't need a stronger practice that once existed: look for changes in what was accepted, how anomalies were labelled, how rules were enforced, or where the line of acceptable risk sat. A gap that was there from the start, and whose acceptance never changed, is a design gap, not drift. Be honest about the distinction.
---
### Feedback and Accumulation Analysis (ssdl-systems-thinking-foundations)
Look for feedback loops and systemic patterns:
**Feedback Loops:**
- **Reinforcing loops** (vicious/virtuous cycles): What got worse as we tried to fix it? What success led to conditions for failure?
- Example: Technical debt → Slower development → More shortcuts → More debt
- Example: Incident → Fire-fighting → Less prevention → More incidents
- **Balancing loops** (goal-seeking): What natural limits did we hit? What corrective mechanisms failed?
- Example: Hiring → Onboarding load → Slower hiring → Understaffing
**System Delays:**
- Was there a delay between action and consequence that masked the problem?
- Did we interpret delay as "no effect" and double down?
- What information arrived too late to be useful?
**Stocks (Accumulated Quantities):**
- What accumulated over time? (Technical debt, knowledge gaps, risk tolerance, team fatigue)
- What was depleted? (Trust, expertise, capacity)
**Loop or Event:**
Ask whether an intervention addresses the feedback loop or only the event (ssdl-systems-thinking-foundations, Phase 4). Most AAR actions only adjust a number (a threshold, a timeout); push for actions that change the structure that keeps the problem in place. Then check the action doesn't change one element alone: the Org Topologies primer warns that optimising each element in isolation can make the whole system worse (org-topologies-primer-2025, Concept 3).
---
### System or One-Off?
Determine whether the conditions that produced this are still in place:
**Built into how we work**
- Would happen again given the same conditions
- Requires system change to address
- Ask: "If we changed nothing, would this happen again within a year?"
- Ask: "If the people involved were replaced tomorrow, would the same conditions produce the same outcome?" (caib-report-vol1, Phase 3)
- Ask: "Has this pattern appeared before?" Recurrent errors point to latent weaknesses in the organization (doe-hpi-handbook-vol1, Phase 5).
**A genuine one-off**
- Unusual conditions that won't recur
- May be addressable without broad system change
Default to the system, but don't dismiss strong evidence about an individual act: take it through the Just Culture assessment below.
---
### Just Culture Assessment
When human actions are implicated, walk the NHS Just Culture Guide's five tests in order (read_distillation nhs-just-culture-guide). Take one action or omission through at a time; if several actions are involved, take each through separately. Each test gates the next:
| # | Test | Question | Where it leads |
|---|------|----------|----------------|
| 1 | **Deliberate harm** | Was there any intention to cause harm? | Yes: the deliberate-harm route, and wider investigation is still needed. No: go to 2. |
| 2 | **Health** | Was there substance abuse, physical ill health or mental ill health? | Yes: the health routes, with the question of whether it could have been recognised earlier. No: go to 3. |
| 3 | **Foresight** | Was there an agreed protocol? Was it workable and in routine use? Did the person knowingly depart from it? | No to any: singling out the individual is unlikely to be appropriate; look for the wider actions. Yes to all: go to 4. |
| 4 | **Substitution** | Would peers in the same circumstances have behaved similarly? Was training missed? Did supervision fail? | Yes to any: singling out the individual is unlikely to be appropriate. No to all: go to 5. |
| 5 | **Mitigating circumstances** | Were there any significant mitigating circumstances? | Yes: action directed at the individual may not be appropriate. No: the narrow route to individual management action, still paired with the wider investigation. |
Revisit the tests as new facts come in (nhs-just-culture-guide, Concept 10).
**Describe the act on the DOE culpability spectrum** (doe-hpi-handbook-vol2, Concept 11; doe-hpi-handbook-vol1, Concept 10). Its outcomes run from intentional harm and possible reckless violation, through possible negligent error, to organisation-induced violation and organisation-induced error. Honest errors go unpunished and reckless conduct is not tolerated. Judge by intent, not by consequence. Ask: "Would we be having this conversation about the same act if it had caused no harm?" Ask too whether an at-risk practice had become normal here (doe-hpi-handbook-vol1, Concept 11).
**The Substitution Test** is the fourth test, not a shortcut past the first three: would another person with similar experience and training, in the same situation, likely have made the same decision?
- If **yes**: singling out the individual is unlikely to be appropriate → Fix the system
- If **no**: go on to mitigating circumstances (test 5); don't jump to recklessness
**Forward vs. Backward Accountability:**
- **Backward**: Who is responsible for causing this? (Leads to punishment)
- **Forward**: Who will take responsibility for fixing this?
Always emphasize forward accountability. The Forest Service guide treats just culture as forward-looking accountability built on trust, fairness and dignity (lfuo-learning-review-guide-2024, Concept 4).
**Output for Just Culture Assessment:**
```markdown
### Just Culture Assessment
- **NHS tests:** [Which test the action stopped at, and where it led]
- **Substitution test:** Would peers have done the same? Training missed? Supervision failed? [Yes/No each]
- **Culpability (DOE spectrum, judged by intent):** [Intentional harm / reckless violation / negligent error / organisation-induced violation / organisation-induced error]
- **Evidence:** [Why this classification]
- **Normalized?:** [Had this become normal practice?]
- **Forward accountability:** [Who will own fixing this?]
```
---
### Hand-off and Waiting Check (Krivitsky, Larman and Flemm, org-topologies-primer-2025)
When investigating delays or slow delivery despite capable people:
**Assessment Questions:**
1. Were all individuals involved working efficiently?
2. How much time was work actively being worked on vs. waiting?
3. How many handoffs occurred between teams/specialists?
4. Is the delay in the hand-offs between teams (a structural problem) or in the work itself (a capability problem)?
**Signs that hand-offs, not people, set the pace:**
- Each team/person is measured as "fast" but end-to-end delivery is slow
- Work items have long wait times between active work periods
- Multiple specialists must touch every item sequentially
- Adding more people doesn't speed up delivery
The primer names the structural side of this: a team that lacks the capabilities to finish its work needs a cross-silo hand-off, and the hand-off is where the failure can enter (Concept 1). Parts tuned one at a time can leave the whole slower: in the primer's words, "flow can get worse!" [V] (Concept 3).
**Key Insight:** If hand-offs and waiting dominate, actions should focus on reducing handoffs, broadening team scope, or restructuring for flow—not making individuals faster.
---
### Systems Archetype Check (Senge, via ssdl-systems-thinking-foundations)
Before finalizing actions, check for common systems archetypes. Treat each as a hypothesis to test against what happened, not a label to apply (ssdl, Concepts 10–11):
**Fixes that Fail:**
- Will this fix create side effects that require more fixes?
- Are we treating the symptom in a way that doesn't address the cause?
- Ask: "What might this fix break?"
**Shifting the Burden:**
- Are we treating a symptom while the underlying problem grows?
- Is there a "fundamental solution" we're avoiding because it's harder?
- Ask: "What would actually solve this, not just make it manageable?"
**Limits to Growth:**
- Did growth run into a balancing limit that now holds it back?
- Is there a limiting factor we're not addressing?
- Ask: "What constraint will we hit as we grow?"
**Then check for push-back:**
- Which loops will this action strengthen or weaken, and which will push back? (ssdl, Phase 4)
- Ask: "What might this action provoke elsewhere in the system?"
---
### Important: Apply Just Culture FIRST
Before redirecting person-focused language to systemic framing, assess whether the behavior warrants Just Culture attention (see §Just Culture Assessment above). The redirections below apply when the NHS tests show that singling out the individual is unlikely to be appropriate, which is most cases: the DOE handbook cites experience that about 90 percent or more of unsafe acts are blameless (doe-hpi-handbook-vol1, Concept 10). A knowing departure from a workable, agreed protocol doesn't settle it on its own: it moves the case past the foresight test, and the substitution test and mitigating circumstances can still route away from the individual. Walk the NHS tests in order first. If they route away from the individual → redirect below. If they don't → handle it through the Just Culture assessment before redirecting.
### Redirections for Common Anti-Patterns
| User says | Redirect with |
|-----------|---------------|
| "X made a mistake" | "What about the system allowed this mistake to happen and cause impact?" |
| "Bad code" | "What constraints shaped the code? What would have had to be true for better code to be the default?" |
| "We need more training" | "What about the system makes the untrained behavior the natural choice?" |
| "We'll be more careful" | "What forcing function would make careful behavior automatic?" |
| "Team Y caused it" | "Why did our system not validate/contain/fail gracefully? Let's expand the AAR boundary." |
| "No one knew" | "What about our knowledge sharing, documentation, or onboarding created this gap?" |
| "We accepted that risk" | "When did we first accept it? What was different then? How did acceptable drift into dangerous?" |
| "It was a perfect storm" | "Perfect storms are predictable in complex systems. What dependencies did we not see?" |
| "It was unprecedented" | "What precedents existed that we didn't recognize? What similar failures happened elsewhere?" |
| "The system worked as designed" | "Then the design needs examination. What implicit assumptions failed?" |
| "We'll add monitoring" | "What will we do when the monitor fires? Is this treating the constraint or a non-constraint?" |
| "They should have known" | "What would have had to be different for them to have known? (Dekker, via doe-hpi-handbook-vol1)" |
| "It's a people problem" | "Is it capability, load, information, or authority? Walk the NHS tests before accepting individual action as a contributing factor." |
| "We've always done it this way" | "How long? What's changed since then? Has an accepted deviation become normal?" |
| "Everyone's working hard but we're still slow" | "Are individuals fast but the work waiting between hand-offs? Check where it waits, and for how long." |
| "This person was negligent" | "Walk the NHS tests in order. At the substitution test: would peers in the same circumstances have done the same? If yes, singling out the person is unlikely to be appropriate." |
| "We need to hold someone accountable" | "Two kinds of accountability: backward (who caused it) and forward (who will fix it). Let's focus on forward." |
| "Let's just fix this and move on" | "Before we do—what might this fix break? Check for Fixes that Fail archetype." |
| "They didn't escalate" | "Was it made clear to them when to escalate and why? Did they know to ask? The burden of understanding falls on the person who set the rule and the person following it alike (mcdp1-warfighting)." |
| "It was just one root cause" | "Most incidents have multiple independent contributing factors. What else had to be true for this to happen? Was there also a detection failure? A response delay?" |
### Five-Point Probing Questions
When the user seems to have reached a contributing factor, verify with questions from the five points:
**What we optimise for**
- What were we optimizing for that made this trade-off feel necessary?
- What risks did we explicitly or implicitly accept?
**Who owns what**
- Was ownership clear? Were decision rights aligned with accountability?
- Did coordination happen (or fail) across team boundaries?
**How work moves**
- What quality gates existed? Why didn't they catch this?
- What feedback loop should have surfaced this earlier?
**What gets measured and rewarded**
- What incentives might have encouraged the behavior that led to this?
- Was there schedule pressure influencing decisions?
**Who we hire, grow and staff**
- Was expertise concentrated in one person?
- Was the team overloaded?
### Completion Checklist
Before accepting the contributing factor analysis as complete:
- [ ] At least 2 independent contributing factor chains identified (for Complicated/Complex incidents)
- [ ] Each chain traced to structural depth (maps to at least one of the five points)
- [ ] Chains are genuinely independent (not the same factor reworded)
- [ ] No unexamined person-blame in any causal chain (person-focused findings must pass through Just Culture assessment)
- [ ] Contributing factors are stable (would persist without intervention)
- [ ] Proposed actions address contributing factors, not just mechanisms
- [ ] If "another team caused it," boundary was expanded to trace end-to-end
- [ ] If human action implicated, decision context was reconstructed (doe-hpi-handbook-vol1)
- [ ] If human action implicated, Just Culture assessment completed (NHS tests walked in order; act placed on the DOE culpability spectrum)
- [ ] Forward accountability assigned (who will fix, not just who caused)
- [ ] Checked for drift patterns - was this gradual or sudden?
- [ ] System or one-off determined (would the same conditions produce it again?)
- [ ] Feedback loops identified if applicable (ssdl-systems-thinking-foundations)
- [ ] Systems archetypes checked (Fixes that Fail, Shifting the Burden, etc.)
- [ ] Hand-offs and waiting checked - is this a flow problem, not a speed problem?
- [ ] Highest-leverage intervention point identified — are actions focused on it? (Goldratt)
### Output Format
```markdown
## Contributing Factor Analysis
### Contributing Factor Chains
**Chain A: [Why did X happen?]**
1. **What happened?** [Mechanism]
2. **Why?** [Latent condition]
3. **Why?** [Decision context]
4. **Why?** [Continue as needed]
5. **Why?** [Until stable]
**Chain B: [Why wasn't it caught / Why was response slow?]**
1. **What happened?** [Mechanism]
2. **Why?** [Latent condition]
3. **Why?** [Decision context]
4. **Why?** [Continue as needed]
### Decision Context (if human actions involved)
| Decision | What They Knew | What Was Missing | Time Pressure | Norms |
|----------|----------------|------------------|---------------|-------|
| [Decision] | [Info available] | [Info missing] | [Urgency] | [Expectations] |
### Drift Analysis (if applicable)
- **Drift detected?** Yes/No
- **Pattern:** [Which checks found drift: first acceptance, unregistered signals, risk added under pressure, faded lessons]
- **When drift began:** [Approximate timeframe]
- **Precursor events:** [List warnings or boundary violations]
### Feedback and Accumulation (if applicable)
- **Feedback loops:** [Reinforcing or balancing loops identified]
- **Delays:** [Where delays masked cause-effect]
- **Stocks:** [What accumulated - debt, risk tolerance, etc.]
### System or One-Off
- **Classification:** Built into how we work / Genuine one-off
- **Evidence:** [Why this classification: recurrence, whether replacing the people would change the outcome]
### Contributing Factor Summary
[One paragraph per chain, linking: context → latent conditions → mechanism → impact. Then a synthesis paragraph explaining how the chains interacted.]
### Five-Point Mapping
| Point | Chain A | Chain B | Evidence |
|--------|---------|---------|----------|
| What we optimise for | [If implicated] | [If implicated] | [Specific evidence] |
| Who owns what | [If implicated] | [If implicated] | [Specific evidence] |
| How work moves | [If implicated] | [If implicated] | [Specific evidence] |
| What gets measured and rewarded | [If implicated] | [If implicated] | [Specific evidence] |
| Who we hire, grow and staff | [If implicated] | [If implicated] | [Specific evidence] |
### Highest-Leverage Intervention Point (Goldratt)
- **THE constraint:** [The ONE factor where intervention has the broadest preventive impact]
- **Type:** Resource / Policy / Paradigm
- **Why this is the leverage point:** [How it relates to the contributing factor chains]
- **Actions should address:** [How to get more from the constraint, or add capacity to it]
### Just Culture Assessment
- **NHS tests:** [Which test each action stopped at, and where it led]
- **Culpability (DOE spectrum, judged by intent):** [Where the act sits, or "organisation-induced"]
- **Forward accountability:** [Who will own fixing this?]
### Hand-off and Waiting Assessment (org-topologies-primer-2025)
- **Present:** Yes / No / Partial
- **Evidence:** [Wait times, handoff counts, individual vs system performance]
- **Implication:** [If present, address flow not speed]
### Systems Archetype Check (ssdl-systems-thinking-foundations)
- **Fixes that Fail risk:** [Will this fix create new problems?]
- **Shifting the Burden risk:** [Are we avoiding the fundamental solution?]
- **Mitigation:** [How to avoid the archetype trap]
```
---
## Example Interaction
**User**: The root cause was that the developer forgot to add error handling.
**Assistant**: That's the mechanism — what specifically failed. But most incidents have multiple contributing factors. Let's identify the chains:
1. **Chain A (Creation):** Why did the code ship without error handling?
2. **Chain B (Detection):** Why wasn't the missing error handling caught in review or testing?
3. **Chain C (Response):** If applicable — was the response to the incident also delayed?
Let's start with Chain A. Why wasn't there a linter rule, code review checklist, or test that would catch missing error handling? What made it easy to forget?
[Continue probing each chain to structural depth, mapping each to the five points]
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |