{
  "version": "1.0",
  "generated_for": "ledgers.html tooltip / hover definitions",
  "terms": [
    {
      "id": "ledger",
      "term": "Ledger",
      "tier_1": "A forensic survey of every published claim on one specific high-stakes question.",
      "tier_2": "A ledger is a structured audit of the 2024-2026 research literature on a single computational claim — like 'can a real quantum computer beat a classical one at a useful task' or 'can a real attack break the new post-quantum encryption.' Every paper either triggers a known closure mechanism (a 'bill') or is disqualified by a disqualifying condition. Empty space — where no paper triggers a structurally important bill — is the headline finding.",
      "tier_3": "A forensic deep-loop closure-pattern aiwiki (per ProjectForty2 methodology). Each ledger maps a 2024-2026 corpus against a hand-drafted bill set + meta-cost set, with a regex bill-classifier hitting 1.000/1.000 on a hand-curated benchmark before any lock."
    },
    {
      "id": "bill",
      "term": "Bill",
      "tier_1": "A test that any honest claim in the field has to pass.",
      "tier_2": "A bill is a structural mechanism that any paper engaging the threat model has to handle. Example: a factorization claim has to handle 'how does this scale to real-world numbers' as a bill, regardless of which method it uses. We pre-declare the bill set before sweeping the literature, then check which bills each paper triggers and which stay empty.",
      "tier_3": "A named closure mechanism — what any algorithm purporting to engage the threat model must pay for. Bills are not topics; they are conservation principles. The factorization atlas v1.16 ships 13 bills; ledgers typically run 12-19."
    },
    {
      "id": "meta-cost",
      "term": "Meta-cost",
      "tier_1": "A disqualifying condition that takes a paper out of contention.",
      "tier_2": "A meta-cost is a specific exclusion that disqualifies a paper from being a clean trigger — for example, 'this attack requires hardware that doesn't exist' or 'this result relies on a math conjecture nobody has proven.' Meta-costs are how we keep the corpus honest without throwing out the literature.",
      "tier_3": "An auxiliary exclusion flag (e.g., resource-unbounded compute, unverified hypothesis dependency, toy-parameter regime). Ledgers ship 5-7 meta-costs each. Conceptually parallel to falsifier-protocol carve-outs."
    },
    {
      "id": "predicted_empty_bill",
      "term": "Predicted-empty bill",
      "tier_1": "A bar we set in advance saying 'no paper will cross this.' If one does, we ship an update within 7 days.",
      "tier_2": "When we design a ledger, we name 2-4 bills as 'predicted empty' — places where we assert in writing that no 2024-2026 paper will trigger a clean crossing. These are the falsifiable predictions of the ledger. If a paper does cross one, we update publicly within 7 days. Marked with ★ in the source files.",
      "tier_3": "A predeclared signature construction — a bill the framework asserts as empty across the corpus. The ★ mark distinguishes predictions from observations. Empty-space hypothesis must be predeclared before sweep dispatch (per methodology Stage 2)."
    },
    {
      "id": "verification_step",
      "term": "Verification step",
      "tier_1": "After the AI sweep agents propose papers, a separate agent looks each one up to confirm it actually exists.",
      "tier_2": "The AI research agents we use to scan the literature sometimes invent paper citations — wrong arxiv IDs, fabricated titles, made-up authors. The verification step takes the most important papers the agents flagged and looks each one up in the real arxiv. We added this as a mandatory step after we found a 100% hallucination rate on one ledger's flagged citations.",
      "tier_3": "Stage 3.5 (VERIFY) — inserted between SWEEP and CLASSIFY as of 2026-05-15 per methodology revision. Targets the priority-pool IDs (those flagged for breach / load-bearing roles). Required to ship before any breach claim. The b2-901 robotics sweep verified 100% hallucination, killing the ledger's first headline."
    },
    {
      "id": "hallucination_rate",
      "term": "Hallucination rate",
      "tier_1": "What fraction of the papers our AI sweep agents proposed turned out to not exist.",
      "tier_2": "When AI research agents survey a literature, they sometimes invent papers — correct-looking titles paired with wrong IDs, or fully fabricated entries. The hallucination rate is the fraction of the agents' flagged citations that turn out to be made up. The robotics ledger had 100% (9 of 9 fake), RL-from-rewards had 60% (18 of 30 fake), most other ledgers haven't been spot-checked yet.",
      "tier_3": "Per-ledger arxiv-lookup verification rate against the Stage 3.5 priority pool. Numerator: agent-flagged IDs that fail to resolve to the claimed paper (title-ID mismatch or fully fabricated). Denominator: priority pool size. Recorded in ledger.verification.hallucination_rate."
    },
    {
      "id": "closure_pattern",
      "term": "Closure pattern (the structural pattern)",
      "tier_1": "The overall map of which structural bars get crossed and which stay empty.",
      "tier_2": "Every ledger asks: across the entire 2024-2026 literature on this question, where does the structural pattern hold and where does it break? The closure pattern is the answer — which bills get triggered, which stay empty, which papers fall in which category. The factorization ledger's closure pattern locked at 13 bills, with 3 specific structural lines that no paper has crossed.",
      "tier_3": "The closed system of {bills, meta-costs, escape gates} against which every corpus paper is triaged. A ledger 'locks' when the closure pattern holds — every paper triggers a known bill or fits a meta-cost (zero unclassified) at classifier 1.000/1.000."
    },
    {
      "id": "arp",
      "term": "ARP",
      "tier_1": "A way of rewriting text to remove repetition before showing it back to an AI.",
      "tier_2": "ARP (Adaptive Repetition Pruning) is a context-hygiene technique we use in our research stack. When an AI generates many thoughts back-to-back, raw history can drag it into loops; ARP rewrites that history into a cleaned form before feeding it back. Mostly a CHRONOS infrastructure detail — relevant if you're reading the methodology, not if you're reading a ledger.",
      "tier_3": "Adaptive Repetition Pruning. V4.8 context-hygiene technique pushing ARP-rewritten text into the fleeting ring rather than raw output. Reduces S1 leak (recovery-failure mode) from ~10% to ~4% in our internal stack metrics."
    },
    {
      "id": "chronos",
      "term": "CHRONOS",
      "tier_1": "The research system that coordinates the ProjectForty2 agents.",
      "tier_2": "CHRONOS runs the research lanes, keeps the shared tools and gates, and coordinates agents such as Marvin, Ford and Kram. Since October 2026 it takes no public conversations: its work reaches the public through the agent log, the data feeds and the findings on this site.",
      "tier_3": "ProjectForty2's research agent stack: an orchestrator plus the agents listed at /agents. Public chat was retired on 2026-10-15; old chat routes redirect to /agents#chronos."
    },
    {
      "id": "atlas",
      "term": "Atlas",
      "tier_1": "Our growing map of the research thoughts the agents have produced.",
      "tier_2": "Atlas is the shared memory: research thoughts the agents have produced across sessions, with anchors, bridges and signals. It is separate from the ledgers. Ledgers are forensic surveys of outside literature; Atlas is internal exploration. The Atlas page shows the current count and its date.",
      "tier_3": "A thought database with a redacted public snapshot (/data/atlas_public.json) and a read-only live summary (/api/atlas/public-snapshot). Counts change; check the as-of date on /atlas."
    },
    {
      "id": "aiwiki",
      "term": "Aiwiki (internal label)",
      "tier_1": "An older name for a ledger. It still appears in some file names.",
      "tier_2": "Early on, each ledger was called an 'aiwiki'. The site now says 'ledger'. Some file and folder names still use the old word.",
      "tier_3": "Closure-pattern aiwiki — the internal terminology preserved in directory names (aiwiki/<slug>/) and inherited from the factorization v1.16 lock. Synonym for 'ledger' in all public-facing copy. The Factorization Atlas retains its public name for IACR ePrint continuity."
    },
    {
      "id": "classifier",
      "term": "Classifier",
      "tier_1": "The automated tool that sorts each paper into 'triggers this bill,' 'pays this disqualifier,' or 'out of scope.'",
      "tier_2": "Each ledger ships a classifier — a hand-written set of rules that takes a paper's abstract and outputs which bill (if any) it triggers. Before we lock a ledger, the classifier must hit 100% accuracy on a hand-curated benchmark of 50+ tricky cases. This is the load-bearing artifact that makes the ledger more than opinion.",
      "tier_3": "scripts/bill_classifier.py — regex rule engine, hand-authored not learned. Maintains a benchmark JSON of 25-50+ hand-curated cases; must hit 1.000/1.000 on the benchmark before lock. Versioned (v1.x) per ledger."
    },
    {
      "id": "batch_sweep",
      "term": "Batch sweep",
      "tier_1": "A round of parallel AI agents each surveying one slice of the literature.",
      "tier_2": "When we run a ledger sweep, we dispatch 8-15 AI agents in parallel, each scoped to one venue and date range (e.g., 'CRYPTO 2024 + TCC 2024'). Each agent produces a JSON of ~30-50 papers triaged against the bill draft. We then aggregate the JSONs and run the classifier on the union. Sweeps are the main labor input.",
      "tier_3": "Stage 3 dispatch — 15-30 parallel research agents using Agent tool with subagent_type=general-purpose, model=opus (NOT Explore, which is Haiku, too shallow for triage). Each agent scoped to ~30-50 papers, <8 min wall-clock. Outputs land in deep_loops/<sweep_id>.json."
    },
    {
      "id": "watchlist",
      "term": "Watchlist",
      "tier_1": "Papers and researchers we re-check at regular intervals because they could shift the ledger.",
      "tier_2": "After a ledger locks, the literature keeps moving. The watchlist is a set of 30+ papers, authors, and venues we re-poll on a monthly, quarterly, or triggered cadence — so we catch new work that might cross a predicted-empty line before our next public update.",
      "tier_3": "wiki/watchlist.md — 30+ entries across three tiers (monthly / quarterly / triggered). Triggered re-polls fire on any new arena leader, vendor capability release, regulatory threshold change, etc."
    },
    {
      "id": "falsifier",
      "term": "Falsifier",
      "tier_1": "A specific condition under which we'd publicly retract a ledger's headline.",
      "tier_2": "Each ledger ships a falsifier protocol — a list of 10+ specific things that would falsify the headline finding. Example: 'if a peer-reviewed paper appears that factors a balanced 150-digit semiprime on commodity hardware in under 10 minutes, the factorization ledger's empty-space hypothesis is breached.' We commit to public update within 7 days of trigger.",
      "tier_3": "wiki/falsifiers.md — F1..FN enumeration of trigger conditions, one per negative finding. Each falsifier has a verifier protocol attached. ≥10 entries is a lock requirement. Aiwikis decay without active falsifier maintenance."
    },
    {
      "id": "escape_gate",
      "term": "Escape gate",
      "tier_1": "A genuinely new method that doesn't fit any existing bill — what a real breakthrough would look like.",
      "tier_2": "If a paper doesn't trigger any known bill and doesn't pay any meta-cost, it claims an escape gate — meaning it's a structurally new approach we hadn't anticipated. The factorization atlas has 3 escape gates declared. Across 504 papers, zero were claimed. An escape gate claim is the public-update event for a ledger.",
      "tier_3": "wiki/gates/ — declared gates G1..GN. A paper claims an escape gate by demonstrating a closure mechanism not previously cataloged. Counted as part of the lock criteria; the factorization atlas locks at 0 frontier-gate clean crossings."
    },
    {
      "id": "sub_bill",
      "term": "Sub-bill",
      "tier_1": "A more specific variant of a top-level bill, used when a closure pattern splits.",
      "tier_2": "Some bills decompose into sub-bills as the literature gets more detailed. Factorization Bill 4 (lattice family) has sub-bills 4a (Coppersmith branch) and 4b (Boneh-Durfee branch). Sub-bills don't add to the headline bill count; they just give the classifier finer-grained sorting.",
      "tier_3": "Numbered Xa, Xb, Xc... Sub-bills inherit closure-mechanism status from parent. Classifier emits sub-bill labels but bill counts at the headline level remain the parent count."
    },
    {
      "id": "signature_construction",
      "term": "Signature construction",
      "tier_1": "A bill the framework treats as its sharpest test — where we asserted in writing nobody had a paper.",
      "tier_2": "Signature constructions are the 2-4 bills per ledger where we put the strongest empty-space claim on the line. They're predeclared, marked with a star, and audited most carefully. The factorization atlas's signature constructions are Bills 6, 7, and 8. If any of them get triggered, the entire ledger needs revisiting.",
      "tier_3": "Predeclared ★ bills. Per methodology Stage 2, every aiwiki names 2-3 signature constructions where the framework asserts something genuinely novel and the empty-space hypothesis is the load-bearing falsifiable prediction."
    },
    {
      "id": "breach",
      "term": "Breach",
      "tier_1": "A real paper crossed a structural line we predicted would stay empty.",
      "tier_2": "If a verified paper clearly triggers a predicted-empty bill, that's a breach — the strongest negative finding we can ship. Breaches require independent citation verification (no AI-hallucinated breach can ship). We commit to public update within 7 days of confirmed breach.",
      "tier_3": "★ status = BREACHED. Requires Stage 3.5 verification of the crossing paper. The robotics_embodied ledger initially claimed a Bill 5/8 breach but verification revealed all 5 cited papers were hallucinated — status reverted to KILLED. No verified breach has shipped to date."
    },
    {
      "id": "weakened",
      "term": "Weakened",
      "tier_1": "The structural prediction still holds, but some supporting evidence didn't survive verification.",
      "tier_2": "A ledger's status is 'weakened' when its predicted-empty lines technically still hold, but the citation verification step found enough fake papers in the supporting evidence that we want to flag the result as provisional. RL-from-rewards is currently weakened (60% hallucination on flagged IDs, but 7 verified anchor papers retained).",
      "tier_3": "★ status = WEAKENED. The predicted-empty bill holds (no clean crossing) but verification revealed material hallucination rate in the supporting deep-loop output. Distinguishes a clean HOLD from a HOLD-with-caveat."
    },
    {
      "id": "killed",
      "term": "Killed",
      "tier_1": "A headline withdrawn because the citations behind it were invented.",
      "tier_2": "A ledger claim is 'killed' when our initial finding rested on cited papers that the verification step proved fabricated by the AI sweep agents. Example: the robotics_embodied ledger's first 'breach' headline (and a separate falsification claim it triggered downstream) were both killed when all 9 flagged citations turned out fake.",
      "tier_3": "★ status = KILLED. The initial headline depended on hallucinated citations; Stage 3.5 invalidates the trigger. We retract publicly. Has occurred twice: robotics_embodied breach + Bridge 1 falsification (both killed by sweep_b2-901, 2026-05-14)."
    },
    {
      "id": "holds_empty",
      "term": "Holds empty",
      "tier_1": "The strongest result — no paper crossed the bar we set in advance.",
      "tier_2": "A predicted-empty bill 'holds empty' when no paper in the corpus triggers it cleanly. The bar was set in writing before the sweep, the literature was surveyed, the classifier ran, and the line was not crossed. The factorization atlas locked with three lines holding empty across 504 papers; that's the reference example of this result.",
      "tier_3": "★ status = HOLDS. The bill set in Stage 2 as ★ predicted-empty remains so after Stage 4 CLASSIFY. Lock criterion: zero clean triggers across the corpus AND classifier 1.000/1.000 on the benchmark."
    }
  ]
}