Discovery Campaign: 50 Trials, Two Survivors

The campaign numbered its work through 50, preserved every failure, and produced two preregistered equity candidates.

• Survivors: Volatility-scaled momentum 12−1 and long-only gross profitability earned one-shot lockbox preregistrations.
• Boundary: Neither candidate has opened the 2022+ lockbox or earned broker authority.
• Ledger caveat: The 57-row JSONL has only 49 numbered registrations; it jumps from 42 to 44, while the adjacent recency-gate amendment explicitly records trial:null.

The full campaign ledger

This table preserves all numbered slots from 1 through 50 and makes the missing source row visible.

• Reading rule: Verdicts for trials 44–50 reproduce the ledger verdict rows and notes verbatim; earlier rows combine the registration with its retained report or confirm-card.
• Lineage rule: A confirm-card is counted as its own trial but does not create a new strategy lineage.

How the waves evolved

Each wave spent its failures to narrow the next question instead of recycling a good-looking chart.

• Wave 1 — broad equity battery, trials 1–15: Fifteen familiar cross-sectional and event ideas produced five confirm-cards; only momentum survived.
• Wave 2 — orthogonal equity effects, trials 16–26: Low-risk, liquidity, seasonal, composite, clustering, turnover, yield, and skew ideas added no survivor.
• Wave 2B — alternative assets, trials 27–32: Crypto funding died; crypto momentum was hindsight-contaminated; FX carry became conditional on executable IBKR-quality data.
• Wave 3A — state and web-research sweeps, trials 30 and 33–34: The regime lens found no confirmed conditional claim; web research surfaced insider clusters and pension rebalancing, neither of which survived to candidacy.
• Wave 4 — owned fundamentals, trials 35–41: Accruals, GP/A, insider clusters, 13F, and PEAD produced one survivor—fundable gross profitability—and one decisive insider refutation.
• Modern lens — trial 42: Every available canonical series was re-scored on 2017–2021 before more anomalies were allowed to advance.
• Method amendment — unnumbered between 42 and 44: New trials could advance only when the modern subwindow was positive and not materially worse.
• Wave 5 — recent-regime fundamentals, trials 44–49: COP passed but lost to GP/A; asset growth, Piotroski, EV/EBITDA, issuance, and the slow insider redesign did not fit the long-only live lane.
• Trial 50 — viral-claim decomposition: Grid harvest existed before venue costs, but range breaks, inventory beta, leverage tails, and Alpaca-like fees killed the product claim.

The modern-window lens

The 34-lineage table separates recent economic behavior from later execution-validity judgments.

• Window: 2017-01-01→2021-12-31 only; no 2022+ return value was read.
• Multiplicity: One-sided Holm controlled family-wise α=0.10 across all 34 finite rows.
• Flag meaning: improved, stable, and decayed compare the modern annualized mean with the saved full-screening mean using a ±1 Newey–West standard-error heuristic.

• Candidate reading: Momentum held stable and GP/A improved even though neither row individually survived Holm across all 34 exploratory lineages.
• Mechanical reading: Gap fade and overnight premium were the only ALIVE saved series after Holm.
• Credible reading: Their later confirm-cards prove those saved series violate frozen execution, so zero killed strategies deserved resurrection.
• Decay count: Nine rows are flagged decayed—eight equity anomaly rows plus FX carry.

Survivor confirm-card #1 — momentum

The momentum card traded headline return for an executable, volatility-controlled sleeve.

• Fundable rule: Price ≥$10, trailing 20-session ADV ≥$10m, classification-free top decile, next-open entry, real turnover, and 10/25 bp per-side costs.
• Unscaled result: +15.0%/yr net, Sharpe 0.49, and −68.5% max drawdown; 2× and 3× costs left +13.9% and +12.7%/yr.
• Registered overlay: A causal trailing-60-session 10% volatility target capped at 1× produced +6.75%/yr, Sharpe 0.55, and −27.6% drawdown.
• Fragility: Removing the best 14 months left +0.41%/yr and Sharpe 0.02; capacity was $14.9m median and $9.5m at p10.
• Decision: CONDITIONAL_PRE_REGISTER—test the exact frozen rule once in the lockbox; do not trade from screening evidence.

Survivor confirm-card #2 — gross profitability

The GP/A card preserved a smaller edge with cleaner recent evidence and meaningful modeled capacity.

• Fundable rule: Price ≥$10, ADV ≥$10m, structural non-null revenue/COR/assets eligibility, pessimistic datekey+2 sessions, staggered annual cohorts, and next-open execution.
• Result: +2.83%/yr active net excess, Sharpe 0.58, NW t 2.46, and −15.4% drawdown.
• Stress: 2× and 3× costs left +2.80% and +2.76%/yr; removing the best 14 months left +0.82%/yr with t 0.91.
• Complement: Active correlation with fundable momentum was 0.20; median binding capacity at 1% ADV was $27.3m.
• Decision: CONDITIONAL_PRE_REGISTER—pre-register GP/A alone for the same single lockbox opening.

COP versus GP/A

Cash operating profitability passed the new recency gate but did not displace the stronger registered sleeve.

• COP full window: +2.06%/yr net excess, Sharpe 0.42, and NW t 1.74.
• COP modern window: +5.00%/yr with NW t 3.11; the 2017–2021 gate passed.
• Identical-rule GP/A: +2.77%/yr, Sharpe 0.64, and NW t 2.92 in the direct head-to-head.
• Relationship: COP/GP-A correlation was 0.36; COP is a complement, not a replacement.
• Decision: Shelf COP as the first post-live incubation candidate; keep GP/A as preregistered candidate #2.

Expected false positives

Multiplicity control changes the question from “which chart looks good?” to “how often would this whole campaign fool us?”

• Naive screen: At unadjusted α=0.10 across 34 independent null rows, the global-null expectation is 3.4 false positives.
• Holm control: The modern lens instead caps the family-wise probability of at least one false ALIVE call at 10% under its assumptions.
• Not a repair tool: Holm cannot correct look-ahead, a hindsight universe, an untradeable venue, or a malformed source row; confirm-cards and execution rules do that.
• Campaign accounting: Every registered look, confirmation, diagnostic lens, invalid trial, and refusal remains in the ledger and in later deflated-statistics penalties.

Provenance

Scratchpad artifacts are session-local evidence; the repository documents are the durable interpretation and decision record.

• Ledger: [local source path] — 57 physical rows — SHA-256 c1bb3c7cfdf6b854bbee2b1d8156b1b09cfbe7b554629221b7c04b9fa65ad0f5`.
• Modern lens: [local source path] — SHA-256 5bf3fe390eadce6e0b29c7de6334b1345ad96e2011ccc26cb5af846e924229e2`.
• Equity looks: [local source path] and [local source path]
• Momentum card: [local source path] — SHA-256 3cf4efbdd82295811b7af13faedc560508faf3b3dfebf51422abf6506c5c6e23`.
• Execution refutations: gap-fade.killcard.md SHA-256 67faca0a07e8367606d02421efcbdbd5e2d2c89e7d10a2c68cb8797ae6c4bf3a; overnight-premium.killcard.md SHA-256 5dc720c535905aab3b84f15afd795937341dc76557d318dec5b74902911aeed9.
• Alternative assets: [local source path] and [local source path]
• Web sweeps: [local source path] and [local source path]
• Fundamentals: [local source path] and [local source path]
• GP/A card: [local source path] — SHA-256 490e426c5826e7bdb70fddc1d629cd787ccb2ba8a464fbffa416a328d499c47f`.
• Insider refutation: [local source path] — SHA-256 40c5bce9b7f9ba3de3d65c6eb20e04c80d75dd1c666e678c02c1f2fb6c1582f9`.
• Grid decomposition: [local source path] — SHA-256 50e5270b2a43491ccb6d9afeb9b7794928e58d99cef663ab0ec2eae8e0784c42`.
• Repository record: The Lab, End to End, both preregistrations, and the superseded free-look battery record preserve the public decision trail.