Left pane = sanitized source specimen; right pane = TellTale output.

TellTale operator lane

Operator lane

SOURCE
OUTPUT
Source on the left, output on the right. The left side is the bounded artifact; the right side is the reading.
Source lane
sanitized source specimen
This pane shows a sanitized source specimen: provenance stays visible, and the transcript is explicitly partial.
Label: pathological-reset-demo
Session: pathological-session-specimen
Source artifact: reset-polluted-session-specimen.jsonl
Summary: This run is partial and should be treated as partial evidence.
Completeness: PARTIAL
Coverage note: Partial evidence: the source artifact is real, but not a clean fully alternating transcript.
Confidence: MEDIUM
Primary finding classes: EXACT_CONTENT_REPLAY, OPENAI_ITEM_ID_COLLISION, REPLAY_GENERATION, STALE_PARENT_BRANCH, TRAJECTORY_COVERAGE_GAP
Secondary warning classes: MALFORMED_REASONING_BLOCK
Primary incidents: 138
Primary finding hits: 118
Secondary warning hits: 145
Total flagged events: 242

Incident summary

  • Coverage is PARTIAL.
  • STALE_PARENT_BRANCH
  • REPLAY_GENERATION
  • REPLAY_GENERATION
  • REPLAY_GENERATION

Work log

  • Loaded report bundle
  • Rendering TellTale output
  • Ready for review / narration

Evidence posture

  • Sanitized specimen for public demonstration
  • Counts and coverage remain representative
  • Detailed raw runtime traces are intentionally withheld from the public shell

TellTale output

TellTale output

Coverage: PARTIAL
Coverage note: Partial evidence: the source artifact is real, but not a clean fully alternating transcript.
Source artifact: reset-polluted-session-specimen.jsonl
Input artifact: sanitized-public-session-bundle
Events: 1084
Primary incidents: 138
Primary finding hits: 118
Secondary warning hits: 145
Total flagged events: 242
# TellTale Report: pathological-session-specimen

> This run is partial and should be treated as partial evidence.

## 1. What was analyzed

- **Session:** pathological-session-specimen
- **Run ID:** public-demo
- **Events:** 1084
- **Primary incidents:** 138
- **Primary finding hits:** 118
- **Secondary warning hits:** 145
- **Total flagged events:** 242

## 2. Source provenance

- **Input artifact:** sanitized-public-session-bundle
- **Source artifact:** reset-polluted-session-specimen.jsonl
- **Source kinds:** session
- **Time window:** 4/19/2026, 4:03:31 AM → 4/20/2026, 10:25:20 AM

## 3. Coverage / confidence

- **Completeness:** ⚠️ PARTIAL
- **Report confidence:** MEDIUM
- **Evidence window:** 4/19/2026, 4:03:31 AM → 4/20/2026, 3:00:34 AM
- **Source window:** 4/19/2026, 4:03:31 AM → 4/20/2026, 10:25:20 AM
- **Snapshots:** 0
- **Records represented:** 724

## 4. Chronological incident summary

- **4/20/2026, 3:00:34 AM** — **coverage break**: continuity confidence drops because the evidence span no longer fully covers the source artifact.
- **4/19/2026, 4:03:31 AM** — **replay/collision onset**: repeated or stale content begins accumulating after the trust boundary weakens.

## 5. Main findings

### Finding A — continuity confidence drops across a visible coverage break
The source stops behaving like a continuous trustworthy record after a visible boundary.

### Finding B — replay/collision patterns accumulate after the break
Repeated high-specificity content and item reuse appear in ways more consistent with replay, collision, or stale-branch reuse than with normal continuation.

### Finding C — downstream state becomes mixed rather than cleanly continued
The post-break region contains enough contradictory structure to justify bounded distrust of downstream conclusions.

## 6. Evidence excerpts / trace anchors

### Structural anchors
- Anchor 1: a visible coverage break marks the point where continuity confidence drops.
- Anchor 2: replay/collision behavior appears after the break and repeats across the degraded span.
- Anchor 3: downstream state remains mixed enough that the affected region should be treated as bounded partial evidence.

*(Concrete client-facing anchors should point to stable excerpts or timestamps without exposing private runtime traces.)*

## 7. Uncertainties

- The source is partial, so we cannot claim complete historical visibility.
- The report speaks to structural trustworthiness, not user intent or root-cause certainty.
- The public demo surface intentionally withholds raw runtime traces.

## 8. Recommended interpretation / next steps

### Recommended interpretation
Treat this run as **not fully trustworthy as a continuous history**. The evidence is strong enough to justify re-checking or discarding downstream conclusions that depend on the broken region.

### Suggested next steps
1. preserve the original raw artifact bundle
2. isolate the reset boundary or suspected migration point
3. compare against a quieter control/baseline run if available
4. avoid building critical conclusions on the polluted span without further review

## Closing line

TellTale’s value here is not that it proves everything. It makes the trustworthy vs untrustworthy parts of the visible trajectory easier to separate and reason about.