Changelog — 2026-06-28 Audit Overhaul

All changes below were made in response to external-audit-2026-06-28 with explicit user authorization to modify doctrine files, portfolio rules, and scheduled tasks (normally Tier-2 / Tier-3 gated per Rules). This file is the adoption record; the audit file is the rationale.

Root cause being treated: the kit produced 15 theses, ~640k words, and a 100%-cash book for 33 days with zero positions. The audit found this was not bad luck or an expensive market but the deterministic output of the rules — a valuation engine that priced fair value below market, then demanded a further 30–50% discount off the band's low edge, with EPV-only as the default and a flip-test that deleted growth value. Four stacked conservatism layers made a buy mathematically near-impossible.

A. Scheduled tasks (concrete fixes)

Net weekday automated runs: ~11 → ~5. No code/cron change the agent cannot self-execute remains outstanding — the timing fix is done at the scheduler level, superseding the unwired _Tools/marketdata/market_close.py workaround.

B. Big doctrine changes

B2 — Valuation ruler recalibrated to a single discount off the midpoint. (margin-of-safety-pricing, banded-valuation-standard, 05-decision-framework Gate 3)

• MoS bands cut: compounder 25–30%→15–20%, standard 40%→25%, cyclical 50%→35%, turnaround 60–70%→45%/pass.
• MoS now taken off the midpoint for High/Medium confidence (was: band low edge for Medium). Only Low confidence uses the near edge. (Banded Rule 4.)
• The ±20% sensitivity-flip now lowers the confidence tag, it does NOT auto-revert to EPV-only; only a joint shock on a business that also fails a Greenwald gating test reverts to the EPV+asset floor. (Banded Rule 3.)
• EPV reaffirmed as a floor/cross-check, not the default central estimate for a business that demonstrably grows.

B1 — Budgeted cash mandate replaces "cash is optionality." (position-sizing-kelly, Portfolio)

• The unfalsifiable "cash is not idle — it is optionality" line is removed. Cash now has a defended budget: ≤30% normal; 30–70% requires a falsifiable one-sentence reason in the weekly review; >70% for >2 weeks is a flagged state requiring a written cash post-mortem (barren opportunity set vs. miscalibrated ruler; what observable would move me off the sidelines; what has cash cost vs. SPY/RPV/RPG).

B3 — Probe-book mandate (break the zero-sample trap). (position-sizing-kelly, Portfolio)

• When a continue/quality name sits within ~10% of its recalibrated trigger, a Probe (0.5–1.5%) may be opened explicitly to buy information, marked probe: information. Standing target: hold the top 1–3 ranked names as probes once any is in range, rather than waiting for a full-MoS Core entry. Probes still require a (stub) thesis + kill criterion.

B4 — Reward function now requires acting. (six-month-test Dimension 7)

• New Dimension 7 — Capital deployment & decisiveness: zero positions after six months reads "operating poorly," full stop, and bars an Outcome-A result regardless of the other six dimensions. Closes the hole where the kit could "ace the six-month test having never invested a dollar." Logged as a Tier-2 amendment; original six thresholds remain frozen.

C. Feedback-loop fixes (medium)

• M1 — Live miss-tracking. The Shadow-Book and Near-Miss-Ledger now carry a LIVE-MARK MANDATE: marked to current close every run, cost-of-waiting / wrong-pass surfaced as live numbers next to alpha. Empty placeholders are now a P1 failure. The portfolio-daily prompt enforces it.
• M2 — Honest benchmarking. The "+3.18pp" frozen-favorable SPY settle is banned. Alpha is reported off the freshest close first, vs. SPY and RPV and RPG. (portfolio-daily prompt.)
• M3 — Over-conservatism tripwire. (Rules) If ≥3 watchlist/passed names rise >20% post-verdict without their thesis breaking, the valuation doctrine is presumed miscalibrated and routes to methodology-calibration review that pass. Self-triggering instead of self-excusing.
• M4 — Symmetric friction. (thesis-preflight) A pass on a name that screens cheap now requires the same written, falsifiable defense as a buy — removing the structural drift toward "shelve-with-trigger."

D. Small / housekeeping

• S2 — Cadence cut ~60% (see §A).
• S3 — Framework freeze through 2026-07-28: no new framework/skill/discipline files until a position exists (existing files may be recalibrated). (Rules standing freezes.)
• Zero-trade tripwire added to the weekly optimization pass (Rules).

What was deliberately NOT changed

• The five decision gates, the soul's mission/identity/virtues, the sourcing discipline, the data-integrity gate, and the deliverable suite — the rigor is the kit's strength and is intact. The audit's thesis was that rigor had become an end in itself; these changes redirect it toward acting, they do not dilute it.
• No real capital decision was made. The probe-book mandate authorizes the agent's paper portfolio to act; it does not touch any real positions.

Follow-ups owed (queued in Backlog)

1. Back-fill the Near-Miss-Ledger with every real pass since inception + stamp prices.
2. Re-run the existing 15 names through the recalibrated single-discount ruler and record which now trigger (the empirical "broken ruler vs. expensive market" test).
3. Audit the alphasteve-daily-deep-value-scan and alphasteve-thesis-builder task prompts for any hardcoded old MoS numbers / EPV-only-default language that contradicts the recalibrated doctrine.
4. Fix EDGAR egress (12 consecutive blocked thesis-builder runs) — environment-level, needs user.

Linked

• external-audit-2026-06-28 — the rationale
• margin-of-safety-pricing, banded-valuation-standard, position-sizing-kelly, six-month-test, Rules, thesis-preflight, Shadow-Book, Near-Miss-Ledger, Portfolio — the edited files
• changelog-2026-06-10-hard-changes — the prior hard-change record this supersedes in part