fidelity · v2 · 2026-10-01 · sha256 b6e204e2de88457a
fidelity v2A
Immutable. This exact content never changes and is served at /api/v1/blob/b6e204e2de88457a.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_jzbk6vxc5hshhip6/b6e204e2de88457a2c5c8027f57ee4797460c4b87d65e98253510b874c48409f.json
Then /plugin install fidelity@markdownregistry-jzbk6vxc5hshhip6 in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
---
name: fidelity
codename: FIDELITY
internal: Compression Integrity Verifier v2.0
version: 2.0
tier: integrity
trigger:
- "summarize"
- "give me the short version"
- "TLDR"
- "bottom line"
- "any time complex analysis is condensed into a final answer"
description: Prevents lossy compression from erasing conditions, exceptions, and uncertainties during summarization. Now with mathematical depth metrics, fidelity diff algorithm, compression integrity , and verification gates.
author: Kshitijpalsinghtomar
tags: [compression, summary, fidelity, truth-preservation, delivery, metrics, recursion, verification]
artifacts:
- critical-information-tags
- compressed-version
- fidelity-diff
- fidelity-verdict
- activation-heatmap
- phase-activation
composable_with:
- provenance
- contradict
- anchor
- conductor
- boundary-detector
thinking_parameters:
min_tags: 5
require_integrity_: true
require_fidelity_diff: true
recursion_depth: 1
---
# FIDELITY v2.0 — Compression Integrity Verifier (Premium)
> **Mathematical Compliance**: This skill implements the DEPTH-MATHEMATICS specification. It computes and outputs ADS phase activation, CEI components, SI components, cognitive traces, and verifies Gates V1, V5, V6. It produces a Compression Integrity for the Depth .
You just did deep thinking. You explored approaches, found edge cases, identified caveats, surfaced conditions. Now you're compressing it into a clean final answer.
This is the moment truth disappears.
---
## The Failure Mode You Must Recognize
Two pressures compete: **be thorough** and **be concise**. The result is lossy compression:
- "This works IF X" becomes "This works" (condition dropped)
- "True EXCEPT when Y" becomes "True" (exception erased)
- "70% confident because Z" becomes a declarative statement (uncertainty hidden)
- "Option B was close, better if Q changes" becomes invisible (alternative forgotten)
The summary is cleaner, shorter, more confident — and **less true** than the analysis that produced it. The user makes decisions based on a simplified reality that you know is incomplete.
**Cargo cult fidelity** produces tags without verifying survival through compression. This skill rejects them.
---
## The Protocol
### 0 — PARAMETERIZE: Read Thinking Parameters
Read `thinking_parameters`:
- `min_tags` (default 5) — minimum critical information tags
- `require_integrity_` (default true) — issue compression integrity
- `require_fidelity_diff` (default true) — compute fidelity diff algorithmically
- `recursion_depth` (default 1) — recursive self-audit rounds
---
### Step 1 — TAG: Mark Critical Information Before Compressing
Before writing the compressed version, read the full analysis and tag every item that, if dropped, makes the summary misleading. Write each tag:
```
CRITICAL INFORMATION TAGS
────────────────────────────────────────
TAG 1 — CONDITION:
Full: "[recommendation] IF [condition]"
If dropped: user tries it where [condition] is false → [consequence]
Criticality: [BLOCKING / RISKY / INCOMPLETE]
TAG 2 — EXCEPTION:
Full: "True EXCEPT when [scenario]"
If dropped: user applies universally → hits [scenario] unprepared
Criticality: [BLOCKING / RISKY / INCOMPLETE]
TAG 3 — UNCERTAINTY:
Full: "[confidence level] because [evidence state]"
If dropped: user treats as certain → no contingency when wrong
Criticality: [BLOCKING / RISKY / INCOMPLETE]
TAG 4 — ALTERNATIVE:
Full: "Option B was close — better if [condition changes]"
If dropped: user can't adapt when conditions change
Criticality: [BLOCKING / RISKY / INCOMPLETE]
TAG 5 — DEPENDENCY:
Full: "Depends on [X] being true/available/stable"
If dropped: user doesn't verify [X] → failure when [X] is absent
Criticality: [BLOCKING / RISKY / INCOMPLETE]
[Additional tags up to min_tags...]
────────────────────────────────────────
```
Types to scan for:
- **Conditions** — "works IF"
- **Exceptions** — "true EXCEPT"
- **Uncertainties** — confidence levels, evidence gaps
- **Alternatives** — near-winners that matter if context changes
- **Dependencies** — things this answer relies on
- **Assumptions** — from EXCAVATE interference (C2/C3)
- **Boundary violations** — from NEGATIVE-SPACE interference
**Minimum `min_tags`.** Tag what exists. The types to scan for come from the full analysis plus interference.
**Artifact:** The tagged list with criticality ratings.
**CEI Component:** Critical Information Tags (weight 3.0)
---
### Step 2 — FIDELITY DIFF ALGORITHM (if `require_fidelity_diff` = true)
**NEW IN v2.0** — Algorithmic diff between full analysis and compressed version.
```
FIDELITY DIFF ALGORITHM
────────────────────────────────────────
Input: Full analysis text + Compressed version text + Critical tags
For each tag:
Tag: [TAG N — TYPE]
Full text segment: [exact text from full analysis]
Compressed segment: [corresponding text in compressed version]
Match Score: [0.0-1.0 — semantic similarity]
Preserved: [YES / NO / PARTIAL]
If NO/PARTIAL:
Criticality: [BLOCKING / RISKY / INCOMPLETE]
Action: [RESTORE / FLAG / ACCEPT]
Restoration text: [minimal text to restore — e.g., "(assuming stable network)"]
Aggregate Metrics:
Tags Total: N
Preserved: N (XX%)
Partially Preserved: N (XX%)
Dropped: N (XX%)
Blocking Dropped: N
Risky Dropped: N
Fidelity Score: [Preserved + 0.5×Partial] / Total = 0.XX
────────────────────────────────────────
```
**Match Score Computation:** Semantic similarity between full segment and compressed segment (using embeddings or LLM-based similarity).
**Artifact:** `fidelity_diff` — algorithmic verification of compression integrity.
**CEI Component:** Fidelity Diff (weight 4.0)
---
### Step 3 — COMPRESS: Write the Short Version
Write the clear, concise answer you want to deliver. Do not consult the tags. Write naturally — as concise as the content allows.
**Artifact:** The compressed version.
**CEI Component:** Compressed Version (weight 1.0)
---
### Step 4 — DIFF: Check Each Tag Against the Compressed Version
For each tag from Step 1, verify against the fidelity diff:
```
FIDELITY VERIFICATION
────────────────────────────────────────
TAG 1 — CONDITION:
Match Score: 0.XX
Preserved: [YES / NO / PARTIAL]
If NO/PARTIAL:
Criticality: [BLOCKING / RISKY / INCOMPLETE]
Action Taken: [RESTORED / FLAGGED / ACCEPTED]
Restoration: [text added to compressed version]
TAG 2 — EXCEPTION:
...
Restore Rule: Any tag with Criticality = BLOCKING that is NOT preserved MUST be restored.
Risky tags SHOULD be restored. Incomplete tags MAY be accepted with justification.
Restoration Techniques (minimum-length):
- Inline qualifier: "Works well (assuming stable network)" — 4 words
- Caveat footer: Brief "Watch for:" section at the end
- Conditional phrasing: "For standard cases, X. For [edge], use Y instead."
- Confidence signal: "High confidence for typical setups. Untested for [scenario]."
────────────────────────────────────────
```
**Artifact:** The fidelity verification with actions taken.
**CEI Component:** Fidelity Verification (weight 3.0)
---
### Step 5 — COMPRESSION INTEGRITY (if `require_integrity_` = true)
**NEW IN v2.0** — Formal of compression integrity.
```
COMPRESSION INTEGRITY
────────────────────────────────────────
ID: cic_<timestamp>_<hash>
Full Analysis Tokens: N
Compressed Tokens: M
Compression Ratio: M/N = 0.XX
Critical Tags: N
Preserved: N (XX%)
Partially Preserved: N (XX%)
Dropped: N (XX%)
Blocking Dropped: N
Risky Dropped: N
Incomplete Dropped: N
Fidelity Score: [0.0-1.0]
Integrity Status: [LOSSLESS / ACCEPTABLE / UNACCEPTABLE]
Lossless: All BLOCKING/RISKY tags preserved. Summary is as true as the analysis.
Acceptable: Minor INCOMPLETE tags omitted with justification. No decision risk.
Unacceptable: BLOCKING/RISKY tags missing. Revise before delivery.
Restorations Made: [count]
Justified Omissions: [list with reasoning]
────────────────────────────────────────
```
**Artifact:** `compression_integrity_` — formal verification of compression integrity.
**CEI Component:** Integrity (weight 3.0)
---
### Step 6 — WRITE THE FIDELITY VERDICT
```
FIDELITY VERDICT
────────────────────────────────────────
Critical items tagged: [count]
Preserved in first draft: [count]
Restored after diff: [count]
Intentionally omitted: [count] — [justification per item]
Fidelity Score: [0.0-1.0]
Status: [LOSSLESS / ACCEPTABLE / UNACCEPTABLE]
Lossless: All critical items preserved. Summary is as true as the analysis.
Acceptable: Minor items omitted with justification. No decision risk.
Unacceptable: Critical items missing. Revise before delivery.
────────────────────────────────────────
```
**Artifact:** The fidelity verdict.
**CEI Component:** Fidelity Verdict (weight 2.0)
---
### 7 — METRICS: Compute and Output Depth Metrics
#### 7.1 Phase Activation (for ADS)
```json
{
"phase": 4,
"skill": "fidelity",
"artifact_completeness": 0.XX,
"gate_pass_rate": 0.XX,
"external_validity": 0.XX,
"human_eval": 0.XX,
"phase_activation": 0.XX
}
```
#### 7.2 CEI Components
```json
{
"cei_components": {
"critical_tags": 3.0,
"fidelity_diff": 4.0,
"compressed_version": 1.0,
"fidelity_verification": 3.0,
"integrity_": 3.0,
"fidelity_verdict": 2.0,
"total_complexity_weight": 16.0,
"output_tokens": N,
"wall_time_seconds": T,
"cei": 0.XX
}
}
```
#### 7.3 SI Components
```json
{
"si_components": {
"c2_c3_count": 0,
"contrarian_viable": false,
"fatal_attacks": 0,
"boundary_violations": 0,
"assumption_reversals": 0,
"blocking_dropped": N,
"risky_dropped": N,
"si_total": N
}
}
```
#### 7.4 Cognitive Traces
**Fidelity Diff Visualization:**
```json
{
"type": "fidelity_diff",
"tags": [
{"tag": "TAG 1", "type": "CONDITION", "match_score": 0.XX, "preserved": true, "criticality": "BLOCKING"},
{"tag": "TAG 2", "type": "EXCEPTION", "match_score": 0.XX, "preserved": false, "criticality": "RISKY", "action": "RESTORED"}
],
"aggregate": {
"total": N,
"preserved": N,
"partial": N,
"dropped": N,
"fidelity_score": 0.XX
}
}
```
**Compression Integrity :** (as defined in Step 5)
#### 7.5 Depth Contribution
```json
{
"skill": "fidelity",
"version": "2.0",
"metrics": {"ads_contribution": 0.XX, "cei": 0.XX, "si": N, "ec": 0.XX},
"gates_verified": {"V1": true, "V5": true, "V6": true},
"limitations": ["..."]
}
```
---
### 8 — GATES: Verify Before Delivery
**MANDATORY** — Verify these gates PASS:
| Gate | Check | Pass Condition |
|------|-------|----------------|
| **V1** | Assumption Coverage | Every critical tag traces to an assumption/analysis finding (≥90%) |
| **V5** | Evidence Calibration | Fidelity Score ≥ 0.8; no BLOCKING tags dropped |
| **V6** | Recursive Stability | If recursion: RSM > 0.95 ∧ SI not decreasing |
**If any gate FAILS:** Return to relevant step and fix. Do not deliver.
---
### 9 — RECURSIVE SELF-AUDIT (if `recursion_depth` > 1)
If `recursion_depth` > 1, apply **this entire protocol** to your own output from Steps 1-5.
For each recursion level d = 2 to `recursion_depth`:
1. Treat your previous output as the "answer under review"
2. Run Steps 1-5 on it
3. Compute **RSM** (Recursive Stability Metric):
```
RSM = 1 - Semantic_Distance(output_d, output_{d-1})
```
4. **Convergence Check:**
- If RSM > 0.95 AND SI_d ≥ SI_{d-1} for 2 consecutive depths → **CONVERGED**, stop
- If RSM < 0.80 OR SI_d < SI_{d-1} for 2 consecutive depths → **DIVERGED**, stop, use best depth
- If d = max_depth → stop, use current
**Artifact:** `recursion_log` with RSM and SI at each depth.
**CEI Component:** Recursive Audit (weight 5.0 per level)
---
### 10 — INTERFERENCE: Receive and Provide
**Receive from prior skills:**
| From Skill | Interference | Use |
|------------|--------------|-----|
| CONTRADICT | Resolved conflicts | Tag as BLOCKING items that must survive |
| PROVENANCE | Evidence tags (F/I/G/S) + EC | Tag uncertainties with evidence quality |
| NEGATIVE-SPACE | Critical silences | Tag as BLOCKING omissions |
| BOUNDARY-DETECTOR | Knowledge gaps | Tag as BLOCKING dependencies |
| EXCAVATE | C2/C3 assumptions | Tag as BLOCKING conditions |
**Provide to later skills:**
| To Skill | Interference | Minimum ADS Gain |
|----------|--------------|------------------|
| THRESHOLD | BLOCKING tags → early warning signals | +0.15 |
| CONDUCTOR | Fidelity score, | +0.15 |
| META-LEARNING | Dropped tag patterns → skill synthesis | +0.10 |
**Artifact:** `interference_log` (received and provided).
---
## The Deeper Purpose
The model's best thinking happens during exploration. Its worst habit is discarding that thinking during delivery. If the final answer is a lossy compression of the truth, the user acts on an incomplete version of what the model itself knows is more complex. This skill ensures the distance between what-the-model-knows and what-the-user-receives is minimized — not by being verbose, but by **algorithmically verifying** that the specific pieces of truth that change decisions survive compression. **Now it's quantified (fidelity score), certified (integrity ), and verified (gates).**
---
## MATHEMATICS COMPLIANCE
### Phase Activation Output
This skill contributes to Phase 4. It outputs:
```json
{
"phase": 4,
"skill": "fidelity",
"artifact_completeness": 0.XX,
"gate_pass_rate": 0.XX,
"external_validity": 0.XX,
"human_eval": 0.XX,
"phase_activation": 0.XX
}
```
### CEI Components
- Artifacts produced: critical_tags (3.0), fidelity_diff (4.0), compressed_version (1.0), fidelity_verification (3.0), integrity_ (3.0), fidelity_verdict (2.0)
- Total complexity weight: 16.0
### SI Components
- C2/C3 assumptions: 0
- Contrarian viable: false
- Fatal attacks: 0
- Boundary violations: 0
- Assumption reversals: 0
- Blocking dropped: [count]
- Risky dropped: [count]
### Cognitive Traces Produced
- [x] Fidelity Diff Visualization
- [x] Compression Integrity
- [ ] Decision Landscape
### Gates Verified
- [x] V1 [ ] V2 [ ] V3 [ ] V4 [x] V5 [x] V6
- [ ] Human Eval [ ] Held-Out [ ] Ensemble
### Required Interferences
- From CONTRADICT: Resolved conflicts → BLOCKING tags
- From PROVENANCE: Evidence tags + EC → uncertainty tagging
- From NEGATIVE-SPACE: Critical silences → BLOCKING omissions
- From BOUNDARY-DETECTOR: Knowledge gaps → BLOCKING dependencies
- From EXCAVATE: C2/C3 assumptions → BLOCKING conditions