Specification Calibration System
Purpose: Use this file when running UNIFY, measuring specification effectiveness, or tuning scope heuristics.
Contents
- UNIFY loop
- logging schema
- evaluation metrics
- calibration rules
- section effectiveness
- propagation format
UNIFY Loop
RECORD -> EVALUATE -> CALIBRATE -> PROPAGATE
- run after each delivered package
- run a deeper review every
3 specs or quarterly, whichever comes first
- record to
.agents/scribe.md
RECORD Schema
Spec: [spec-id]
Scope: [Full | Standard | Lite]
Feature_Complexity: [High | Medium | Low]
Teams_Involved: [Biz+Dev+Design | Biz+Dev | Dev+Design | Biz+Design | Single]
Requirements_Count: [N]
BDD_Scenarios_Count: [N]
Sections_Used:
L0: [yes/no]
L1: [yes/no]
L2_Biz: [yes/no]
L2_Dev: [yes/no]
L2_Design: [yes/no]
L3: [yes/no]
Traceability_Completeness: [0-100%]
Cross_Team_Alignment:
biz_understanding: [High/Medium/Low]
dev_understanding: [High/Medium/Low]
design_understanding: [High/Medium/Low]
Revisions_Required: [count]
Downstream_Handoff: [Sherpa/Builder/Radar/Voyager/Canvas/None]
Evaluation Metrics
Alignment
Alignment = (Biz + Dev + Design understanding) / 3
| Result |
Interpretation |
| all High |
excellent package |
| mixed |
moderate alignment, inspect wording and coverage |
| any Low |
weak package, inspect the failing audience |
Scope Accuracy
| Range |
Interpretation |
> 0.85 |
scope heuristic is working well |
0.70-0.85 |
some over- or under-scoping |
< 0.70 |
review complexity indicators |
Evaluation Triggers
| Trigger |
Inspect |
| teams ask for clarification after delivery |
wording and audience fit |
package needs 3+ revision rounds |
scope or missing sections |
| downstream agent cannot use the package directly |
output shape and detail level |
| implementation drifts from the spec |
requirement completeness and BDD quality |
| quarterly review |
aggregate effectiveness |
Calibration Rules
- do not change heuristics before at least
3 specs exist
- maximum adjustment per cycle:
+/-0.15
- decay adjustments by
10% each quarter toward default values
- explicit user instruction overrides all calibrated defaults
Section Effectiveness Defaults
section_effectiveness:
L0_vision: 0.95
L1_requirements: 0.95
L2_biz: 0.80
L2_dev: 0.90
L2_design: 0.75
L3_bdd: 0.90
L3_traceability: 0.70
Writing Pattern Hints
| Pattern |
Typical effect |
Best fit |
problem-first L0 |
very high |
all scopes |
explicit MoSCoW in L1 |
high |
Standard, Full |
visual user flow in L2-Design |
high |
UI-heavy work |
inline technical example in L2-Dev |
high |
API-heavy work |
| BDD with concrete values |
very high |
all scopes |
| shared glossary |
medium to high |
Full or terminology-heavy work |
Propagation Format
## YYYY-MM-DD - UNIFY: [Scope x Teams]
**Specs assessed**: N
**Average alignment**: [High/Medium/Low]
**Key insight**: [insight]
**Calibration adjustment**: [old -> new]
**Apply when**: [future context]
**reusable**: true
<!-- EVOLUTION_SIGNAL
type: PATTERN
source: Scribe
date: YYYY-MM-DD
summary: [specification insight]
affects: [Scribe, Scribe, Lore]
priority: MEDIUM
reusable: true
-->
Context Library
| Context |
Best scope |
Typical strong sections |
| new product feature across 3 teams |
Full |
all L2 + full L3 |
| API extension for Biz + Dev |
Standard |
L2-Dev, L2-Biz, L3 |
| UI improvement for Design + Dev |
Standard |
L2-Design, L2-Dev, L3 |
| business-impacting bug fix |
Lite |
compact L0, key L3 |
| infra change for Dev only |
Lite |
inline L2-Dev, key L3 |
Quick UNIFY
Use this when fewer than 3 specs exist:
## Quick UNIFY
**Specs assessed**: 1
**Scope**: Standard
**Note**: [observation]
**Action**: No weight change yet
