Orbit Examples
Purpose: load this when matching a live issue to a known Orbit scenario. Each example is intentionally short and keeps only the details that affect classification, escalation, or required output.
Contents
- Verification gap
- Dirty-start scope risk
- State drift
- Contract missing
- Tool failure
- Parallel-loop contamination
- Sequential-loop handoff gap
- Loop-of-loops inner failure
- Pre-flight failure
- Branch-isolation squash
Scenario Table
| Example | Core input | Expected Orbit result |
|---|---|---|
A. VERIFY_GAP |
done.md exists, last progress.md entry shows failed lint, footer says DONE |
status_assessment: VERIFY_GAP, recommended_next_action: CONTINUE, handoff to Radar |
B. COMMIT_SCOPE_RISK dirty start |
dirty baseline exists and new iteration changes are scoped to fresh paths | status_assessment: COMMIT_SCOPE_RISK, next action is candidate-only staging, handoff to Guardian |
C. STATE_DRIFT |
state.env says NEXT_ITERATION=11, progress.md ends at 9 |
status_assessment: STATE_DRIFT, rebuild state from progress evidence, handoff to Builder |
D. CONTRACT_MISSING |
goal.md missing, progress.md empty, no footer evidence |
severity: P1, rebuild contract files, emit NEXUS_LOOP_STATUS: READY, handoff to Nexus |
E. TOOL_FAILURE |
runner.log shows repeated executor failure and no new iteration entry |
severity: P2, classify failures separately, bounded retry first, escalate to P1 only after budget exhaustion |
F. Cross-loop COMMIT_SCOPE_RISK |
parallel loops stage each other's files | severity: P1, unstage, isolate scopes per loop, handoff to Guardian |
| G. Sequential handoff gap | successor loop says "continue from loop-1" but has no measurable ACs or handoff checklist | severity: P1, rebuild successor contract, trigger ON_GOAL_CONTRACT_WEAK, handoff to Nexus |
| H. Loop-of-loops scoped failure | inner loop fails repeatedly while outer loop continues | severity: P1, classify inner failure independently, update outer progress, do not start dependent inner loops |
I. PREFLIGHT:FAIL |
disk below 100MB, stale lock exists, checksum valid |
stale lock auto-clears, disk failure aborts run before iteration start |
| J. Branch-isolation squash | BRANCH_ISOLATION=true, SQUASH_ON_DONE=true, verify PASS, iter branch has 8 commits |
squash to summary branch, commit once, delete iteration branch, final footer DONE |
K. CONVERGENCE_STALL |
.action-sig.log last 3 lines identical; iters report success but no net artifact change |
status_assessment: CONVERGENCE_STALL, severity: P1, persist state, inject disambiguation / re-scope, escalate to human |
L. BURN_RATE_ANOMALY budget cap |
.cost-usd sum > USD_PER_RUN_CAP; loop still running |
severity: P1, PAUSE, require explicit human resume (never auto-continue), snapshot state |
| M. Terminator-bound gap | unattended loop with only MAX_ITERATIONS set, no LOOP_TIMEOUT/USD_PER_RUN_CAP |
status_assessment: CONTRACT_MISSING, severity: P1, apply Terminator-bound gate, declare a hard bound before resuming |
Notes
- Example D preserves the
ON_GOAL_CONTRACT_WEAKtrigger because it changes contract recovery behavior. - Example G preserves sequential handoff enforcement because it distinguishes "done file exists" from "prerequisites independently verified."
- Example I preserves the
100MBpre-flight threshold and stale-lock auto-clear behavior. - Example J preserves the
BRANCH_ISOLATIONplusSQUASH_ON_DONEcompletion path. - Example K anchors semantic-stall classification to
.action-sig.logevidence (exit-code success ≠ progress). - Examples L and M preserve the hard-cap / mandatory-terminator contract added in operation-contract
v1.2.0.