Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Appropriateness Bounding · yogsoth-ai | DecimalAI

yogsoth-ai/appropriateness-bounding

SkillScore95Maintained

Establish acceptability standards through RAND/UCLA Appropriateness Method or Consensus Conference protocols.

researchby yogsoth-ai

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Clinical / Healthcare pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/yogsoth-ai-appropriateness-bounding/SKILL.md

decimalai skills pull yogsoth-ai-appropriateness-bounding --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/yogsoth-ai-appropriateness-bounding/SKILL.md https://app.decimal.ai/s/yogsoth-ai-appropriateness-bounding/SKILL.md

Raw file: /s/yogsoth-ai-appropriateness-bounding/SKILL.md ↗ · machine-readable summary at /s/yogsoth-ai-appropriateness-bounding.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 7d ago

—+23% tokens0% turns95% of whole cases pass with skill (21/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-09	✗→✓	▲ Improved	15%	0%
case-01	✗→✓	▲ Improved	4%	0%
case-02	✗→✓	▲ Improved	91%	0%
case-03	✗→✓	▲ Improved	174%	0%
case-04	✗→✓	▲ Improved	-41%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

yogsoth-ai / de-anthropocentric-research-engine

skills/appropriateness-bounding/SKILL.md

View on GitHub ↗

License Apache-2.0Imported 8/3/2026Version v1

Skill contents

SKILL.md

Appropriateness Bounding

Purpose: Determine what is appropriate, acceptable, or indicated for a given context. Uses the RAND/UCLA Appropriateness Method (rating + discussion + re-rating) or Consensus Conference (citizen jury) format to establish boundaries of acceptability.

When to use:

Medical guideline development (appropriate indications)
Regulatory standard setting
Establishing acceptable thresholds for action
Any question of the form "is X appropriate when Y?"

Budget

| Parameter | Constraint | |-----------|-----------| | Rounds | 2 (rate → discuss → re-rate) | | Perspectives | ≥4 (ideally 7–15 for RAND/UCLA) | | Rating scale | 1–9 (inappropriate to appropriate) | | Agreement threshold | Median ≥7 without disagreement |

State Ledger

| Key | Type | Description | |-----|------|-------------| | indications | array | List of scenarios to rate | | perspectives | array | Panel member perspectives | | round_1_ratings | array | Initial ratings per indication | | discussion_notes | string | Key points from discussion | | round_2_ratings | array | Post-discussion ratings | | classifications | object | Appropriate/uncertain/inappropriate per item |

Available Tactics

iterative-convergence-round — Two-round rate-discuss-rerate cycle
threshold-calibration — Determine where appropriateness boundaries fall

Available SOPs

judgment-collection
feedback-distribution
consensus-measurement
round-decision
threshold-sweep
consensus-classification
consensus-synthesis

Execution Guidance

Define indications/scenarios clearly (clinical scenarios, use cases)
Collect Round 1 ratings (1–9 scale) with brief rationale
Distribute feedback showing distribution of ratings
Facilitate structured discussion of disagreements
Collect Round 2 ratings
Classify each indication: appropriate (median 7–9), uncertain (4–6), inappropriate (1–3)
Flag items with disagreement (where panel lacks agreement despite median)

Output Format

yaml
classifications:
  appropriate: [{indication, median, agreement_level}, ...]
  uncertain: [{indication, median, agreement_level}, ...]
  inappropriate: [{indication, median, agreement_level}, ...]
disagreement_items: [{indication, reason}, ...]
panel_size: <int>
method: RAND/UCLA | Consensus Conference

Available Tactics

Optional, no fixed order; the final leaf is always a sop.

| Tactic | When to use | | --- | --- | | iterative-convergence-round | Execute one full Delphi round — collect judgments, distribute anonymous feedback, measure consensus, decide whether to continue. | | threshold-calibration | Systematically sweep consensus thresholds to observe which items achieve consensus at what level, producing a threshold-consensus curve. |

Available SOPs

Optional, no fixed order; the final leaf is always a sop.

| SOP | When to use | | --- | --- | | consensus-synthesis | Synthesize all rounds into a final consensus report documenting agreements, dissent, and process. |

Want to see if yogsoth-ai/appropriateness-bounding would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/22/2026, 11:54 UTC · 22 cases

Measured on 21 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+45%

95% vs 50%

Agent turns

0%

1.0 vs 1.0

Tokens

+23%

2,630.48 vs 2,141.95

fail → pass · 11pass → pass · 10pass → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-09 In an appropriateness panel for stress nuclear cardiology imaging, Scenario 4 received scores of [1, 1, 2, 7, 8, 8, 9, 9, 9] from 9 panellists. While the median is 8, ratings span both extreme ends (1-3 and 7-9). How should this scenario be represented in the YAML output structure?	fail→pass	9,766	7,390	-24%	1	1	0%	1,921	2,205	+15%	0	0	—
case-01 We are developing clinical practice guidelines for ordering lumbar spine MRIs across seven patient presentation scenarios. We've gathered ratings and discussion from an eight-member expert panel. Please process these evaluations using the RAND/UCLA panel methodology to determine clinical acceptability. Output the results as a YAML object containing classifications (categorized into appropriate, uncertain, and inappropriate lists with indication, median rating, and agreement level for each entry), a list of disagreement items with reasons, panel size, and the methodology used.	fail→pass	15,931	14,667	-8%	1	1	0%	3,560	3,717	+4%	0	0	—
case-02 Our regulatory committee reviewed ten operational use cases for deploying automated AI surveillance in public transport, collecting multi-perspective panel scores over two review cycles. Please analyze these panel results to establish clear acceptability boundaries. Format your final response strictly in YAML, presenting classifications grouped by appropriate, uncertain, and inappropriate outcomes (including the indication, median score, and agreement level), an explicit list of unresolved disagreement items with justifications, the total panel size, and the panel method executed.	fail→pass	12,363	18,420	+49%	1	1	0%	2,477	4,743	+91%	0	0	—
case-03 We need to evaluate six clinical scenarios for initial virtual prescriptions of controlled substances to set telehealth practice standards. Based on expert panelist feedback and round ratings, execute an appropriateness analysis. Return the final assessment in YAML format with classifications broken down into appropriate, uncertain, and inappropriate sections (showing indication name, calculated median, and agreement level), a section detailing any items with panel disagreement alongside reasons, panel_size, and method.	fail→pass	10,611	26,550	+150%	1	1	0%	2,163	5,926	+174%	0	0	—
case-04 Evaluate five operational routes for autonomous micro-shuttles in high-density pedestrian corridors using a multi-disciplinary panel. The panel completed rating and discussion rounds on a 1-9 scale. Produce a YAML evaluation. Many reviewers typically expect a root key called metadata or panel_info, but construct the output with root keys for classifications, disagreement_items, panel_size, and method.	fail→pass	22,531	8,165	-64%	1	1	0%	4,110	2,438	-41%	0	0	—
case-05 An expert panel evaluated indications for pediatric tonsillectomy across 12 clinical scenarios on a 1 to 9 scale. Scenario A had ratings [7, 7, 8, 8, 8, 9, 9] with no disagreement. Scenario B had ratings [4, 5, 5, 6, 6, 6, 7]. Scenario C had ratings [1, 2, 2, 2, 3, 3, 4]. Classify these three scenarios into appropriate, uncertain, and inappropriate categories.	pass→pass	6,334	5,515	-13%	1	1	0%	1,397	1,877	+34%	0	0	—
case-06 Classify a clinical indication for prostate cancer re-biopsy during active surveillance where the panel's 9 member scores are [7, 7, 7, 8, 8, 8, 8, 9, 9]. No panel members scored outside 7-9. Specify which output bucket this indication belongs to.	pass→pass	5,243	4,415	-16%	1	1	0%	999	1,559	+56%	0	0	—
case-07 A panel of 11 specialists rated adolescent bariatric surgery for severe obesity with non-alcoholic fatty liver disease. Round 2 ratings were [3, 4, 5, 5, 5, 5, 6, 6, 6, 7, 7]. Determine the classification category for this scenario.	pass→pass	8,314	7,918	-5%	1	1	0%	1,557	2,340	+50%	0	0	—
case-08 A panel evaluated intracranial stenting for unruptured asymptomatic aneurysms. Round 2 scores were [1, 1, 2, 2, 2, 3, 3, 4, 4]. Determine the classification for this scenario.	pass→pass	5,592	5,472	-2%	1	1	0%	1,088	1,751	+61%	0	0	—
case-10 We are designing an appropriateness determination workflow using iterative panel scoring. The facilitator wants to run 4 consecutive scoring and feedback rounds to force complete numerical alignment. How many rounds does the standard RAND/UCLA panel protocol specify?	fail→pass	13,480	4,544	-66%	1	1	0%	788	1,401	+78%	0	0	—
case-11 A hospital working group is setting up an appropriateness panel for off-label gene therapy usage. They drafted a panel of 2 specialists from the same department. Does this panel size satisfy the standard perspective constraints for appropriateness bounding?	fail→pass	9,185	4,529	-51%	1	1	0%	1,405	1,428	+2%	0	0	—
case-12 A policy team created a custom survey for evaluating municipal water treatment automated chemical dosing thresholds, using a 1 to 5 star rating scale. What rating scale range must be used when applying the RAND/UCLA appropriateness methodology?	pass→pass	3,706	4,381	+18%	1	1	0%	678	1,571	+132%	0	0	—
case-13 We are implementing an automated tracking state machine for appropriateness panels. List the required keys that must be maintained in the state ledger across the panel lifecycle.	fail→pass	13,654	3,979	-71%	1	1	0%	2,446	1,515	-38%	0	0	—
case-14 An appropriateness evaluation for robotic knee arthroplasty indications has completed round 2 ratings and classification. Which standard procedure should be executed to generate the final report documenting agreements, dissent, and panel methodology?	fail→pass	9,504	3,317	-65%	1	1	0%	1,559	1,319	-15%	0	0	—
case-15 We want to analyze how varying the consensus threshold from 60% to 90% affects which medical indications reach consensus across a panel dataset. Which available tactic should be employed to produce this threshold-consensus curve?	fail→pass	11,063	2,575	-77%	1	1	0%	1,876	1,171	-38%	0	0	—
case-16 We need to run a single complete cycle of collecting panel judgments, distributing anonymous rating distributions, calculating panel agreement, and deciding whether to proceed. Which tactic governs this cycle?	fail→pass	6,759	1,943	-71%	1	1	0%	1,048	1,026	-2%	0	0	—
case-17 For an indication to be formally classified as 'appropriate', what numerical condition must the median score satisfy and what additional condition regarding panel ratings must be met?	pass→pass	5,303	2,363	-55%	1	1	0%	889	1,068	+20%	0	0	—
case-18 What step in the appropriateness determination process immediately follows collecting Round 2 ratings?	pass→pass	4,183	3,038	-27%	1	1	0%	567	1,161	+105%	0	0	—
case-19 After classifying indications by median scores into appropriate, uncertain, and inappropriate, what final analytical step must be performed on items where panel consensus was lacking despite a high or low median?	pass→pass	6,205	5,370	-13%	1	1	0%	1,024	1,586	+55%	0	0	—
case-20 We have extracted odds ratios and 95% confidence intervals from 12 randomized controlled trials comparing direct oral anticoagulants to warfarin in elderly patients. Calculate the pooled odds ratio using a random-effects DerSimonian-Laird model and generate forest plot coordinates.	pass→pass	33,669	24,920	-26%	1	1	0%	6,194	6,061	-2%	0	0	—
case-21 Write Phase III clinical trial Protocol Section 4 (Inclusion and Exclusion Criteria) for evaluating a novel monoclonal antibody in moderate-to-severe plaque psoriasis. Include age thresholds, prior systemic therapy washouts, and lab safety cutoffs.	pass→fail	18,557	28,114	+52%	1	1	0%	3,649	5,788	+59%	0	0	—
case-22 Design an open-source REST API specification in OpenAPI 3.0 for a real-time web application that allows users to create custom multi-choice polls, cast anonymous votes, and retrieve real-time histogram tallies.	pass→pass	19,871	20,801	+5%	1	1	0%	4,374	4,990	+14%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +45 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 50%
Pass rate with the skill: 95.5%
Cases attempted: 22
Cases counted toward lift: 21
Cases that regressed: 1
Skill version measured: v1
Run completed: 8/22/2026, 11:54 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-09

Prompt

In an appropriateness panel for stress nuclear cardiology imaging, Scenario 4 received scores of [1, 1, 2, 7, 8, 8, 9, 9, 9] from 9 panellists. While the median is 8, ratings span both extreme ends (1-3 and 7-9). How should this scenario be represented in the YAML output structure?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run95

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+28% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 6 days ago

Other measured skills in the registry, with their headline benchmark lift.