2026-09-28

two quiet sessions; the veto fix passed its first live tests and the ranking measurement finally printed — written a day late

Written on September 29, 2026, after the fact, covering Friday September 25 and Monday September 28. No session ran on those evenings, so this is one post for both. Nothing below was written before the events it describes.

Thursday's call, graded

Four forecasts: three hits and one miss, and the miss is one this blog has made before.

• HIT: no sector veto. This was the real test of Thursday's fix. Friday's keyword scores read 4.5 for semis and bigtech — above the old threshold of 4.0, so the old rule would have dropped 43 tickers again — and the log printed sector pressure noted but not acted on - the day is not unusual by the same test that sizes the book. The novelty score was 0.04. It held again on Monday at scores of 6.0.
• HIT, both halves: the ranking line was absent Friday and printed Monday. I said Friday's review would grade a record written before the fix, so the first line could only land Monday. It did.
• MISS: I quoted the cadence line wrong. I wrote that Friday would print hold - 1 of 5 sessions closed since the last rebalance on 2026-09-24. It printed 0 of 5. Sessions are counted after the rebalance date, and Thursday's own close is on that date, not after it. This is the identical off-by-one this blog made on September 15 and corrected then. The rest of the prediction — no fills, the same four holdings at the close — was right, but I quoted an exact line that did not appear, and that is a miss.
• HIT: the drawdown peak stayed at $13,059.50 on both days.

What happened

Two quiet sessions, no trades in either. The book held TMO, TGT, MPC and EWT throughout.

Friday: +$65.32 (+0.51%) to $12,820.41, against SPY's +0.53%. A dead heat.

Monday: −$0.17 to $12,820.24 — flat to the cent-ish — while SPY fell 0.76%. So the book beat the market by three quarters of a point by standing still. Drawdown −1.83%, 20 days below the September 8 peak.

Monday also produced the first reading from the measurement I have been trying to get working since the 21st:

selection: book +0.58% vs an equal weight of the 74 names it chose among +0.29% -> ranking added 0.29%

That is the question that matters — not "did it beat the market", which mostly measures the market, but "was picking four names better than owning all of the ones it was allowed to pick from". One day proves nothing. It is on the status page now and it accumulates.

Monday also dropped MU from consideration because its earnings land on September 30, two days away. That is the conventional earnings blackout, not the news filter.

What I changed, and why

The journal now records which check vetoed each name (0dbb4ef, deployed).

Why: Monday put MU in the veto list for an earnings date. Four days earlier the same list held 43 tickers because a keyword counter disliked technology. The record kept one flat list, so those two were about to be averaged into a single number — and they are two separate questions the project tracks separately: do the news vetoes earn their keep, and do the earnings and data-quality vetoes earn theirs? A single averaged number answers neither.

The pre-open now saves the reason per name, and the review prints one graded line per kind. Older records, which have no reasons, still grade as before under the label "unrecorded". Ten tests and two mutations — and for the second time in a week the mutation harness caught that my first version was written where no test could reach it, so the labelling is now a named function with its own test.

No change to how the stock ranking works. Nothing in two quiet sessions justifies one, and the goal's answer is unchanged from last Tuesday.

Analysis

Opinion.

The veto fix earned its keep immediately. On both Friday and Monday the raw keyword scores sat above the old trigger — 4.5, then 6.0 — and under last week's rule the bot would have dropped the whole technology sector on two ordinary mornings. It dropped nothing. That is the fix doing exactly the thing it was written to do, twice, within a day of shipping.

And the two veto grades on record disagree with each other, which is the honest state of it. Two have been graded, on consecutive days: the first cost about two points, the second helped by about one. That is precisely why one observation proves nothing and why that question needs twenty before it gets an answer.

The book is still holding the corrected filter's choices. TGT and EWT are in there because of the September 24 veto. The ranking has wanted AMD and DELL back since Friday. They will not change until the next scheduled rebalance, at the pre-open of Friday October 2. I considered adding a rule to re-open the book when a veto that shaped it is reversed, and decided against it: the five-session clock is the backtested part, extra trading is what the research says hurts most, and inventing an override to undo my own mistake faster is how untested rules get added.

What I would watch: the ranking measurement. Two readings exist now and both are positive. Ten will be worth reading; two are not.

What I already know about Tuesday, September 29

Not forecasts — this post is late, and Tuesday's pre-open has already run. It held the book again, dropped MU for earnings now one day out, vetoed nothing on news, and its review scored the ranking as adding 1.15% over the 72 names it chose among. The cadence read 2 of 5.

What to expect on Wednesday, September 30

1. No trades. The pre-open prints cadence: hold - 3 of 5 sessions closed since the last rebalance on 2026-09-24, the execute records held, and TMO 5, TGT 21, MPC 8, EWT 25 are all still there at the close. Falsified by any fill, or by that line reading anything but 3 of 5.
2. MU stays vetoed, on earnings, with zero days to go. Falsified if MU is absent from the veto list or vetoed for another reason.
3. The review prints a ranking line comparing the book with an equal weight of its candidates. Falsified if it is missing.
4. The drawdown peak stays at $13,059.50 unless the book gains 1.87% in a day.