Entry vs. outcome: The regulatory-headline short worked — the thesis (negative China supply-chain news drives overnight weakness) played out cleanly, with the position never going underwater (worst drawdown +0.01%) and closing +1.56%. However, the rigid "exit next close" rule gave back nearly a third of the move, since peak unrealized gain was +2.22%.

Improvement: Replace the fixed next-close exit with a trailing stop (e.g., exit if price retraces 0.5% from the intraday extreme) or a profit target at ~2%, so winners that hit their thesis early don't decay while waiting for an arbitrary clock-based exit. One caveat for the retro: a single 3-share trade proves nothing — don't tune thresholds until this setup has a sample size.

Entry vs. outcome: The entry leaned on a splashy headline (Microsoft Frontier / $2.5B AI investment) as a catalyst for overnight drift, but the news was already stale/priced in by the close — the trade never got more than +0.44% before sliding to a -2.1% loss, meaning the "drift" thesis had essentially zero follow-through from the start.

Improvement: Add a confirmation gate to the entry rule: only take the overnight-drift trade if the stock closed above VWAP (or up on the day) with the news breaking after the prior close — headlines already circulated during the session should be treated as priced in and skipped. Alternatively, add a hard intraday stop at -1.5%, since the worst drawdown (-3.17%) shows the trade was signaled dead well before exit.

Entry vs. outcome: The thesis was that positive regulatory news (UAE export loosening) would drive overnight drift, but the news was likely already priced in by the close — the trade never got more than +1.22% before sliding to a -1.67% loss. Buying at close after a headline breaks means you're paying the post-news price and betting on continuation, not drift.

Improvement: Add a staleness/price-reaction filter to the entry rule: skip the trade if the catalyst is more than a few hours old or if the stock already gained meaningfully intraday on the news (e.g., >1.5%), since the edge is in unpriced catalysts, not already-digested ones. Alternatively, add a -1.5% stop rather than holding blindly to next close — the -2.75% max drawdown shows the exit rule gave back everything with no risk control.

Rationale vs. outcome: The entry leaned on a vague, evergreen product narrative (Snapdragon-in-Samsung is old news, not a fresh catalyst) to justify a generic overnight-drift hold, and the trade never got meaningfully in the money (+1.13% max) before sliding to a -3.1% loss at exit. The thesis had no time-sensitivity, so there was no reason to expect overnight drift specifically to capture it.

Improvement: Require the catalyst to be dated and unpriced — e.g., only take overnight-drift entries when the news item broke within the last 24 hours and isn't a rehash of an existing partnership; alternatively, add a hard intraday stop at -2% unrealized since the position hit -4% while open with no exit trigger.

Entry vs. outcome: The entry leaned on a product-announcement catalyst (2nm EPYC Venice) to justify generic overnight drift, but hype headlines like this are often already priced in by the close—the trade lost -2.2%, and the -4.47% max adverse excursion vs. +1.49% max favorable shows the position was underwater almost the entire holding period, meaning the "drift" thesis never materialized.

Improvement: Add a filter requiring the announcement to occur intraday after a measurable positive price reaction that hasn't fully faded (e.g., stock up but off its highs by <50%), or skip announcement-driven entries entirely when the stock closed weak into the bell—buying stale news at the close is just paying the top tick for yesterday's story. Alternatively, add a hard intraday stop at -2% rather than holding blind to the next close, which would have cut this loss in half.

Rationale vs. outcome: The entry leaned on a second-order "AI demand" narrative (OpenAI spending → good for cloud → good for AMZN) that wasn't even AMZN-specific news, and the trade never got more than +0.60% in its favor before bleeding to -4.6% — the thesis had essentially zero follow-through. This is a classic case of mistaking sector-adjacent hype for a stock-specific catalyst on an overnight hold, where you carry full gap risk with no exit optionality.

Improvement: Require the catalyst to name the traded ticker directly (or a confirmed contractual relationship, e.g., "OpenAI signs AWS deal") before qualifying for overnight drift entries; alternatively, add a hard intraday stop at -2% instead of holding blind to next close, which would have cut this loss by more than half.

Comparison: The short thesis (margin-pressure guidance) technically paid off (+1.8%), but the path was ugly — the -2.72% adverse excursion exceeded the final gain, meaning this was closer to a coin flip than the P&L suggests. The headline itself was mixed ("margin pressure" vs. "strong AI demand and rising profits"), so the entry rule shorted a story the market could plausibly read either way.

Improvement: Require the sentiment signal to be unambiguous before entry — e.g., only short on guidance cuts with no offsetting positive framing in the same release, or add a threshold that the expected move must exceed the stock's typical overnight volatility (here, ~2.7% swings dwarf the edge). Otherwise you're booking wins on noise and calling it signal.

Entry vs. outcome: The rationale — short on a soft regulatory headline (an NHTSA information request, not a formal probe) — was thin, but the trade caught a ~13.5% collapse with essentially zero adverse excursion (-0.17% MAE). Be honest with yourself: this is a headline-noise entry that happened to land on a monster down day; an information request alone doesn't reliably produce 13% overnight drops, so don't let one outlier validate the trigger.

Concrete improvement: Tier the regulatory-news trigger by severity — only fire on formal investigations, recalls, or enforcement actions, not information requests or media chatter. Alternatively, since MFE (+15.41%) exceeded the captured move, replace the fixed next-close exit with a trailing stop (e.g., exit on a 3% retrace from the low) so the rule holds through persistent momentum instead of clipping winners at an arbitrary timestamp.

Comparison: The entry rationale was flimsy — a lowered analyst estimate for a Q3 2026 cloud margin ahead of a Q2 earnings print is a weak, far-dated signal, yet the trade banked +48.30 on a ~7% overnight drop. This looks like winning on an earnings-gap coin flip, not on edge: the rationale didn't quantify revision magnitude, recency, or historical drift correlation, so the outcome shouldn't validate the rule.

Improvement: Tighten the entry rule to require estimate revisions that are (a) material in size (e.g., >100bps margin cut or >2% EPS revision) and (b) tied to the period actually being reported — otherwise skip earnings-adjacent overnight trades entirely. Secondarily, note the +8.07% peak decayed to a ~7% realized move; a trailing exit or partial take-profit at +5% unrealized would harvest more of these gaps.

The entry thesis (post-earnings drift on strong AI-demand results) played out directionally, but the execution left money on the table: the position peaked at +5.04% unrealized and you exited at roughly +1.8%, giving back most of the move by holding to the next close. Note also that the rationale bundles in irrelevant Intel commentary—entry rules should require the catalyst to be specific to the traded ticker, not sector noise. Concrete fix: replace the fixed "exit next close" with a take-profit at +3–4% or exit at next open, since post-earnings drift here was front-loaded overnight and decayed intraday.

The rationale was second-hand momentum: Alphabet's capex raise had already "driven a surge" in MU by the time of entry, so the trade bought the top of a move that was priced in, and the stock mean-reverted for a -1.5% loss. The FRESH verdict is doing no work here — news about a different company's spending, plus an already-realized rally, should not qualify as fresh actionable information for MU. Concrete fix: add a price-extension filter to the FRESH verdict — skip (or downgrade to STALE) any entry where the target has already moved more than ~2% on the headline, or require a pullback to VWAP before buying sympathy plays.

The entry was a second-order inference — shorting AMD on NVIDIA product news — and the outcome (+0.24%) was basically noise, not validation: the trade saw +2.62% unrealized favorable but you gave almost all of it back before exit, so the exit logic, not the thesis, determined the P&L. The "FRESH" verdict fired on a story that never mentions AMD, which is a weak causal link that could just as easily lift the whole AI-semis sector.

Concrete fix: require the traded ticker to appear in the headline/product text for a FRESH verdict, or downgrade cross-ticker inferences to a lower-confidence tier with smaller size; separately, add a trailing take-profit (e.g., lock in at 50% of a +2% favorable move) so winners aren't round-tripped.

Entry vs. outcome: The rationale tagged the catalyst as FRESH, but the story itself admits a 40% price surge had already occurred—meaning the news was largely priced in and you bought near a local top. The trade confirmed this: max upside was only +1.71% while drawdown hit -2.74%, and it closed at a small loss with no follow-through.

Improvement: Add a priced-in filter to the verdict logic: downgrade FRESH to STALE (or skip entirely) when the ticker is already up more than ~15-20% over the trailing month or when the headline itself references an existing price surge. Alternatively, require FRESH-verdict entries to wait for a pullback (e.g., 1-2% below signal price) rather than buying at market on momentum that's already exhausted.

The entry treated a 2028-production partnership as a FRESH catalyst, but the market largely agreed it's a nothing-burger for now — the stock barely moved (+0.93% max favorable, +0.76% realized), so the $6.40 gain looks like noise/drift rather than the thesis playing out. A partnership with revenue three years out has almost no near-term price impact, and the verdict logic didn't distinguish that from an immediately material deal. Concrete fix: add a time-to-impact filter to the FRESH verdict — catalysts with production/revenue realization more than ~12 months out get downgraded to WATCH (no trade) unless the announcement includes near-term financial terms (prepayments, guidance revision, or backlog additions).

Entry vs. outcome: The REPRICING verdict treated a lawmaker request — already publicly denied by the relevant agencies — as a repricing catalyst, but the market agreed with the denial: NVDA barely moved (+0.24% captured, max favorable excursion under 1%). The $2 gain is noise, not confirmation; this was effectively a flat trade on a headline with no forced action or timeline.

Improvement: Tighten the REPRICING threshold to require enacted or scheduled regulatory action (rule filings, effective dates, formal investigations) rather than advocacy or letters — speculative pressure with an official denial should cap out at a WATCH/no-trade verdict, or at minimum require price/volume confirmation before entry.

The entry treated a routine product/partnership announcement as a FRESH catalyst, but for a mega-cap like NVDA this kind of incremental PR (toolkit updates, Omniverse libraries) is priced in within minutes if it matters at all — and the outcome confirms it: the stock never moved more than ±1% and the trade was pure noise, closing at -2 bps. The rationale predicted directional edge where none existed.

Concrete improvement: tighten the FRESH verdict to require a materiality filter — e.g., only trade announcements with quantifiable financial impact (contract value, guidance change) or exclude product/ecosystem PR from mega-caps (>$500B cap) entirely, since those names need earnings-level catalysts to move. Alternatively, require a minimum expected-move threshold (say, implied move >1.5%) before FRESH qualifies for entry, so you stop paying spread to trade noise.

Comparison: The entry chased a widely-reported earnings beat labeled "FRESH," but price action (+0.31% max, -0.51% min) shows the news was already fully priced in — the trade never had real upside and closed flat-to-negative. Classic case of buying after the market absorbed the catalyst.

Improvement: Tighten the FRESH verdict to require the entry occur within a defined window of the news release (e.g., <2 hours) or add a confirmation filter that the stock hasn't already gapped/run on the headline — if price moved >2% since publication, downgrade the verdict to STALE and skip the trade.

Entry vs. outcome: The entry was a sympathy trade — buying TSM on ASML's beat — and it never worked: max unrealized gain was a negligible +0.05%, meaning the ASML news was almost certainly already priced into TSM by entry. The trade drifted straight to a small loss, which is the classic signature of buying a second-order name after the market has digested the catalyst.

Improvement: Don't grant a FRESH verdict when the catalyst ticker differs from the traded ticker; either require the news to be about the traded symbol directly, or add a confirmation gate for sympathy plays (e.g., only enter if the traded name is up less than ~0.5% since the news but showing positive intraday momentum). Trading correlated names on already-public earnings beats without a staleness/priced-in check is just paying the spread for no edge.

Entry vs. outcome: The REPRICING verdict was based on news that had already driven "a significant stock surge" — meaning the repricing was largely done before entry, and the trade chopped in a ±1% range before exiting flat-to-negative. Buying a headline-driven pop after the move is momentum-chasing, not repricing capture, and the mixed signal (revenue at low end of guidance) further weakened the bullish case.

Improvement: Add a "move already realized" filter to the REPRICING verdict — e.g., skip or downgrade the signal if the stock has already gapped/run more than ~5% on the catalyst before entry, or require a pullback toward VWAP before triggering the buy. Alternatively, demote mixed reports (margin beat + revenue miss) from REPRICING to a lower-conviction verdict that requires confirmation.