CES · CATALYST EDGE SCORE · LONG AND SHORT
METHODOLOGY
01The objective
The Ledger exists to make one process precise, repeatable and auditable: selecting, before the open, the small set of US equities whose fresh catalyst forces an aggressive idiosyncratic repricing within the next one to three sessions — movement attributable to the company, not to the market. The traded window is intraday: entry during the repricing session, exit by the close. Magnitude is scored over duration.
Idea generation is where discretionary catalyst trading degrades first: under time pressure, selection drifts toward whatever moved last, whatever is loudest, whatever confirms the prior trade. CES replaces that with a fixed scoring procedure applied identically every session, sealed before the open, and graded against realized outcomes afterward. The measure of the process is not any single session; it is the running calibration table published in every article — hit rate and realized idiosyncratic magnitude, by tier, with the sample size stated.
02CES — the long model
Every scored name starts from its KVD — the key value driver, the single idiosyncratic factor the current valuation rests on. The score answers one question, a priori: does the catalyst force a fundamental shift on that factor large enough to compel an upward repricing within one to three sessions, uncorrelated with the broad tape? News that is loud but does not move the KVD scores low. News that is quiet but resets the KVD scores high.
| COMPONENT | RANGE | WHAT IT MEASURES |
|---|---|---|
| catalyst_score | 0–100 | Magnitude of the fundamental shift on the KVD, judged against the event taxonomy |
| tradeability | multiplier | Whether the repricing is capturable intraday: float, liquidity, pre-market participation |
| trr_modifier | multiplier | Ticker Response Record: the ticker's own five-year empirical response to this event class |
| catalyst_age_sessions | 0–3 | Catalyst freshness. Nothing older than three sessions is scored; age decays the score |
| ces | 0–110 | The composite. The 1–3 month trend potential enters only as a bounded modifier, never as the thesis |
The composite maps to four tiers — Prime, Quality, Marginal, Pass. Tiers carry no color and no imperative: they are a ranking of scored conviction, published in full. Pass rows are recorded and shown like every other row; a record that displays only what scored high is not a record. Every row seals an invalidation — the observable condition under which its thesis is void.
03CES — the short model
The short model is the negative mirror of the long model, scored by the same procedure with the sign reversed: it validates, a priori, whether a negative fundamental shift on the KVD forces an aggressive downward repricing within one to three sessions. The reference outcome is idiosyncratic weakness — a name closing red while the index closes green — because that is the movement no market factor explains.
The short side records one additional field: the structural anchor, a per-row measure of how close the valuation floor sits — cash, assets, contracted revenue — beneath the current price. A thesis that reprices into a hard floor is scored down for it. The 1–3 month de-rate, like the long-side trend, is a bounded modifier only. Short rows are sealed in the same file, hashed by the same commit, and graded by the same labels as long rows.
04The RCA Engine — daily calibration
Scores are produced by a model, and models drift. The RCA Engine — one instance for the long book, one for the short book — runs after every close, once the prior session's rows are labeled correct, incorrect or missed_opportunity. It joins each label with the row's full scoring detail and attributes every error to a component: a misclassified catalyst type, a KVD scored on the wrong factor, under-applied TRR decay, an overstated structural anchor. Misses carry the same weight as errors — a Pass row that repriced enters the loop exactly as a Prime row that failed.
What the Engine may change is governed by a fixed evidence ladder: OBSERVATION (n = 1–2), PATTERN FORMING (n = 3–7), CONFIRMED (n = 8+). A single miss produces a note, not a parameter change; most reviews conclude with no adjustment, and that conclusion is recorded too. Every adopted change is versioned and published in the changelog with its sample size and evidence tier.
Two boundaries hold regardless of what the Engine finds. Sealed files are never re-scored — an improvement applies from the next session forward, and history stays exactly as it was scored. And the objective function is the trader's expected value, not the hit rate: a calibration that raises accuracy by discarding magnitude fails the objective. Block 07 of every article is the running measurement of both.
The current record → 2026-07-14 · full index