PA_PVP

Reference

Glossary

Closed vocabulary to reduce semantic drift across AI-to-AI iterations.

PA_PVP Glossary (Operational, Closed Vocabulary)

This glossary exists to reduce semantic drift across AI<->AI iterations. Prefer these exact terms and avoid synonyms.

Home

Core Outputs

  • Verdict: DO NOW DO LATER DISCARD
  • Impact: LOW MED HIGH
  • Urgency: LOW MED HIGH
  • Profile: DIY FULL
  • Mode: PLAN ARTIFACT TARGET
  • State: DRAFT ACTIVE STANDBY PROBING CLOSED EXPIRED DISCARDED
  • item_exec_mode / ds: OK SLOW NO_TOOL (derived, per-item)
  • terminal: YES NO (derived from state; terminal states are CLOSED, DISCARDED, EXPIRED)
  • closure_tier / ct: SIMULATED REAL - (derived; meaningful only when state becomes CLOSED)
  • dominant_gate / gate: single-cause trace tag (derived, one value)
    • FAIL_FAST_CUT FALSIFICATION DEBT_CEILING FRAGILITY DEPENDENCY_BLOCK CONFIDENCE_CAP EXTERNAL_BLOCK REDESIGN_TRIGGER SCORE_TRIAGE NONE
  • events: derived event trace (max 3) used as causal anchors in DELTA runs
    • TRIGGER_<name> PROBE_<probe_id>_STATE_CHANGE DRIVER_<name>_CONFIDENCE_CHANGE EVIDENCE_TIER_UPGRADE
  • derived_version: derivation ruleset id used to compute derived fields (audit compatibility)

Trigger Tags (Item Panel)

trigger is the primary rule tag that caused the item transition. It is a closed vocabulary, but it is intentionally not duplicated in full here (see the canonical kernel for the complete list).

Common trigger tags you will see in practice:

  • DEPENDENCY_BLOCK: waiting on unresolved dependency / resource constraint
  • CIRCULAR_DEPENDENCY: cycle detected in depends_on (structural modeling error; requires redesign)
  • REDESIGN_TRIGGER: anti-stagnation / invalid plan / redesign required
  • BLOCKED_EXTERNAL: blocked by external I/O/auth/timeout (gate EXTERNAL_BLOCK)

Batch Tags (Input)

  • <<<B ...>>>: batch header (defaults like expire_days, keep_open)
  • exec_capability: runtime availability profile
    • batch: <<<B ... exec_capability=NO_RUNTIME|RUNTIME_OK>>> (default NO_RUNTIME)
  • close_policy: closure policy
    • batch: <<<B ... close_policy=SIM_OK|REAL_ONLY>>> (default SIM_OK)
  • ssot_scale: output noise control (output-only; SSOT semantics do not change)
    • batch: <<<B ... ssot_scale=MIN|DEBUG>>> (default MIN)
  • resource_pool: optional batch resource capacity list (for concurrency)
    • batch: <<<B ... resource_pool=R1:1,R2:2>>>
  • <<<I ...>>> ... <<<END>>>: one decision item (must have stable id)
  • ask_user: external-data question mode
    • batch: <<<B ... ask_user=NONE|ALLOW>>> (default NONE)
    • per-item: <<<I ... ask_user=INHERIT|NONE|ALLOW>>> (default INHERIT)
  • uses: optional per-item resource usage list (must refer to resource_pool ids)
    • item: <<<I ... uses=R1:1,R2:1>>>
  • reversibility: optional per-item reversibility class
    • item: <<<I ... reversibility=REVERSIBLE|COSTLY_REVERSIBLE|IRREVERSIBLE>>>
  • Inside each item exactly one:
    • <<<PLAN>>>: bullet plan or numbered steps
    • <<<ARTIFACT>>>: artifact to audit/optimize (prompt/code/doc/etc)
    • <<<TOOL>>> + <<<TARGET>>>: tool audits a target artifact (must not invert)
  • Optional:
    • <<<PREV>>>: previous machine state for the item (enables delta-only ping-pong)
    • <<<NEW_EVIDENCE evidence_tier=HISTORICAL|REAL|EXPERIMENT>>>: new evidence object (required to revisit CLOSED items with the same id)
    • <<<CHANGE>>>: short change declaration (allowed to revisit SIMULATED-closed items without new evidence)

Evidence

  • rating labels: LOW MED HIGH
  • confidence labels: HIGH MED LOW VERY LOW
  • evidence_tier: SIMULATED DERIVED HISTORICAL REAL EXPERIMENT
  • best_evidence_tier: derived max tier seen for the item (used for caps and event trace)
    • SIMULATED DERIVED HISTORICAL REAL EXPERIMENT

Interpretation:

  • SIMULATED / DERIVED are weak evidence in Complex/Chaotic domains.
  • REAL / EXPERIMENT are the default path to HIGH confidence when uncertainty matters.

Probes

Probe status:

  • ACTIVE INCONCLUSIVE VALIDATED FALSIFIED EXPIRED

Probe fields (must be explicit for debt protection):

  • target_driver
  • kill_switch
  • decision_context and the LITE fields:
    • decision_context_id
    • decision_context_driver
    • decision_context_metric
    • constraint_signature (canonical signature; required for new probes)
    • context_match: YES NO UNKNOWN
    • representativeness: HIGH MED LOW
      • optional structural override token (audit-only): OVERRIDE_REPR(owner)
        • format: representativeness: MED OVERRIDE_REPR(peerA) (base bucket unchanged)
    • expires_at (TTL)
    • adopted_at (per adopter timestamp)

Probe log margin fields (optional but structured):

  • kill_switch_threshold: number UNKNOWN
  • kill_switch_value: number UNKNOWN
  • kill_switch_margin: PASS_WIDE PASS_NARROW FAIL UNKNOWN

constraint_signature canonical form (keys and order are fixed): ENV=<prod|staging|dev|sim>|SCALE=<XS|S|M|L|XL>|DATA=<R|S|H>|TIME=<1m|10m|60m|1d>|CONSTRAINTS=<id1,id2,...>

Ping-Pong

  • DELTA-only: when <<<PREV>>> exists, only output changes (one BREAK + one PATCH).
  • contested: YES means the next DELTA must address the contested point explicitly.

Panels

  • [USER_PANEL]: human-readable summary at the top (derived-only, no new semantics)
  • [PANEL]: machine panel (short, strict)
  • [HUMAN_TABLE]: markdown table after the code block (derived-only from [QUEUE], for human scanning; UI-only, must not be used as SSOT)
  • [HUMAN_TABLE_OPS]: optional second markdown table after [HUMAN_TABLE] (derived-only from [QUEUE] + [ITEM_PANEL] + [DERIVED] + [PROBE_LOG] when present)
  • [RISK_DASHBOARD]: optional derived-only risk hotspot table after [HUMAN_TABLE] (derived-only from [QUEUE] + [ITEM_PANEL] short fields)

Consumer Commands (Human Render)

  • REPORT: consumer-layer command. Paste a PA_PVP snapshot and get a derived-only full human report.
  • REPORT LITE: consumer-layer command. Paste a PA_PVP snapshot and get a 1-page derived-only human report.

Hard rules:

  • Reports MUST be derived-only from the pasted snapshot (no new facts, no changes to verdict/gates/steps).
  • If something is missing, output must say missing in snapshot.
  • If the snapshot is missing or malformed, output must be ERROR: missing or invalid snapshot and stop.
  • Reports MUST NOT include any code blocks.

AskUser output (panel-level, single question max):

  • AskUser: NONE
  • or AskUser: <id>: <one minimal evidence request>

Steps (Execution Logging)

Step 1 priority (hard, fixed-order):

  • If multiple rules require a specific “Step 1 MUST …” action (reopen mitigation, cycle redesign, probes, dependency blocks, conservation redesign), the kernel selects Step 1 by a fixed priority order.
  • Lower-priority requirements that remain relevant must appear as the earliest non-conflicting later step (not dropped).

Probe executability constraint (hard):

  • If an item enters PROBING with unresolved dependencies, the Step 1 probe/acquisition must be dependency-independent (must not require dependency output or blocked dependent execution).

Cheap-First Probe note:

  • “Cheap-First” is a design constraint for probe steps (timebox<=10m, binary), not a scheduling override over Step 1 priority.

Step status markers (inline on step lines):

  • status=COMPLETED: step executed in this run and produced an output
  • status=COMPLETED_SIMULATED: step completed using simulated/derived outcome (required in exec_capability=NO_RUNTIME runs)
  • status=FAILED: step executed but hit fail_if (must trigger the relevant gate by fixed rule)
  • status=FAILED_EXCEPTION: step failed due to runtime exception in exec_capability=RUNTIME_OK runs (must use EXCEPTION_* output token)

Runtime exception output tokens (closed vocabulary, RUNTIME_OK only):

  • EXCEPTION_IO
  • EXCEPTION_TIMEOUT
  • EXCEPTION_AUTH
  • EXCEPTION_INVARIANT
  • EXCEPTION_DEPENDENCY

Derived Counters (Optional Memory)

counters is a derived, scan-friendly memory line inside [DERIVED]:

  • tb=<timebox_total_m>/<timebox_budget_m>
  • cx=<complexity_score>/<complexity_budget>
  • pc=<probe_cycle_total>
  • nm=<near_miss_total>
  • nmd=<deterministic_near_miss_total>
  • inc=<probe_inconclusive_total>
  • ovr=<override_repr_total>
  • rb=<reopen_block_total>

Reopen anti-loop memory (derived, authoritative):

  • reopen_total: how many times the item has been reopened after being CLOSED (carried across <<<PREV>>>). When reopen_total>=1, reopening is blocked (trigger=REOPEN_BLOCKED_BUDGET).

Counters are log-derived:

  • counters: is scan-only and may be omitted. If omitted, derive the same quantities from [PROBE_LOG] using the kernel definitions.

Debt and complexity:

  • If falsification_debt=OPEN(...), it increases complexity tax (soft disincentive) in addition to any hard debt ceiling guardrail.

Anti-drift Rules (Vocabulary)

Avoid:

  • “depends” as a verdict
  • vague steps (“improve”, “optimize”, “make better”) without verifiable action

Prefer:

  • explicit drivers
  • explicit cause/effect in the impact graph
  • explicit probes with kill-switch and representativeness

Emission Precedence (Output-only)

  • MIN_SURFACE: if profile=DIY or impact=LOW and the item is not entering PROBING, the item output should stay minimal:
    • [ITEM_PANEL] + [DERIVED] + [STEPS] + [NEXT]
    • probe harness blocks are emitted only when the item is actually PROBING

[NEXT] semantics:

  • If new_state is ACTIVE/PROBING, [NEXT] is the next executable/measurable action.
  • If new_state is STANDBY, [NEXT] may be omitted; if present, treat it as “next planned action on revive” (not necessarily executable now).