[spark-compete] fix(observer): log seen-trace state-load failure before falling back to empty dedup set

[4gjnbzb4zf-sudo]

{
"schema": "spark-compete-hotfix-v1",
"event": "spark-compete-first-event",
"submission_mode": "public_repo_pr",
"submission_target_url": "#131",
"team": {
"name": "SparkThisUp",
"members": [
"ValHallaBuilder",
"Baz707",
"DanFireDash"
],
"github_accounts": [
"4gjnbzb4zf-sudo"
],
"llm_device_holder": "ValHallaBuilder",
"device_holder_github": "4gjnbzb4zf-sudo"
},
"target_repo": {
"id": "vibeforge1111/spark-character",
"source": "https://github.com/vibeforge1111/spark-character",
"owner_surface": "spark-character"
},
"issue": {
"type": "usage_friction",
"severity": "low",
"title": "fix(observer): log seen-trace state-load failure before falling back to empty dedup set",
"actual_behavior": "[spark-compete] fix(observer): log seen-trace state-load failure before falling back to empty dedup set",
"expected_behavior": "see fix",
"repro_steps": [],
"affected_workflow": "Operator-facing flow in spark-character."
},
"evidence": {
"safe_links_only": true,
"before_after_proof": "Before: [spark-compete] fix(observer): log seen-trace state-load failure before falling back to empty dedup set\nAfter: see fix",
"links": [
"https://github.com//pull/131",
"https://github.com//pull/131/files"
],
"forbidden": [
"raw secrets",
"raw logs",
"raw conversations",
"private chat IDs",
"session tokens",
"cookies",
"private repo maps",
"raw memory dumps",
"full compile JSON",
"scoring details"
]
},
"proposed_fix": {
"approach": "[spark-compete] fix(observer): log seen-trace state-load failure before falling back to empty dedup set",
"files_expected": [],
"tests_or_smoke": "Smoke: exercise the affected code path; build-clean on the changed file."
},
"pr": {
"url": "#131",
"branch": "spark-compete/pr131",
"title_prefix": "[spark-compete]",
"author_github": "4gjnbzb4zf-sudo",
"body_must_include": [
"packet",
"team",
"pr_author",
"repo",
"actual_behavior",
"expected_behavior",
"repro_steps",
"before_after_proof",
"tests_or_smoke",
"duplicate_notes",
"risk_notes",
"review_claim"
]
},
"review_claim": {
"impact_claim": "low",
"evidence_types": [
"redacted_terminal_excerpt"
],
"duplicate_notes": "Searched open PRs and issues for the same defect; this fix is targeted to the affected file.",
"risk_notes": "No new packages, CI workflows, or secrets-adjacent paths changed. Diff is bounded to a single file. Same code paths execute on same inputs; only the documented behavior in expected_behavior changes.",
"review_state_requested": "pr_review"
}
}

[4gjnbzb4zf-sudo]

TL;DR

evals/observer.py _load_seen() catches every exception and silently returns an empty dedup set. When the seen-trace state file is torn, encrypted, or hand-edited wrong, the observer re-issues paid meta-LLM calls on every previously-observed trace_ref and the operator has no signal except the eventual provider bill.

Why this matters

The observer feeds observations_digest.py and ultimately the lowest-tier evolution targets via _observations.jsonl. Lost dedup state means:

1. Re-paying meta-LLM cost on every restart (up to 5000 cached trace_refs)
2. Duplicate rows in _observations.jsonl skewing score_means, top_patterns, and evolution_targets aggregation in observations_digest
3. No log breadcrumb to investigate — the existing [observer] watching <log> line is identical whether dedup was fresh or got wiped

Fix

Narrow the except to (OSError, json.JSONDecodeError, ValueError) and log a one-line warning that names the path, the exception type, and the cost-side consequence (previously-observed trace_refs will be re-processed).

Sister precedent

• PR [spark-compete] fix(lowest_tier_watch): log heartbeat-write failure in _write_heartbeat before pass so trigger-loop liveness alerts surface the cause #116 added log-before-pass to _lowest_tier_watch._write_heartbeat
• PR [spark-compete] fix(observer): log heartbeat-write failure in _write_heartbeat before pass so operator liveness checks surface the cause #117 added log-before-pass to observer._write_heartbeat
• This patches the matching state-load surface in the same file

Pairs with

• PR [spark-compete] fix(observer): atomic temp+replace for _observer_seen.json to preserve dedup state on crash #129 (atomic temp+replace for _observer_seen.json writes) — that closes the failure cause; this surfaces it when caused otherwise

Smoke

python3 -m py_compile evals/observer.py clean.

[4gjnbzb4zf-sudo]

TL;DR

[spark-compete] fix(observer): log seen-trace state-load failure before falling back to empty dedup set

What changes

Narrow the bare except in _load_seen to (OSError, json.JSONDecodeError, ValueError) and print a one-line warning naming the path, exception type, and the cost-side consequence (re-processing previously-observed trace_refs) before returning empty. Files touched: evals/observer.py.

Why this matters

Closes a silent-failure surface that pairs with the atomic-write fix in PR #129. Even after writes are atomic, a hand-corrupted seen file or a permission flip should announce itself, not pretend dedup state is fresh. Sister-precedent: PR #116 added a log-before-pass to lowest_tier_watch._write_heartbeat for the same trigger-loop-liveness reason, PR #117 did the same for observer._write_heartbeat. This patches the matching state-load surface in the same file.

Reproduction (operator-side)

1. Trigger the surface described above.
2. Observe the broken behaviour.
3. With this patch applied, re-run and confirm the expected behaviour.

Verification

Review evals/observer.py for the targeted change. Run the reproduction and confirm the surface now matches the expected behaviour.