α
AlphaSteve
← Brain index

First-read Standard — the triage stage between scan and thesis

A first-read is a short structured document that takes a name surfaced by the daily deep-value scan and decides, in 30-60 minutes of focused work, whether it deserves a full thesis. Most first-reads end in pass or shelve. The point of the stage is to protect deep-work capacity: a kit that writes a 10-page thesis on every flagged name will produce 10-page theses on a lot of value traps. A first-read catches them at the door.

The first-read is not a thesis. It does not triangulate valuation, does not steelman consensus, does not produce a position-size recommendation. It produces a single decision with one sentence of reasoning, supported by enough work to defend the decision.

When a first-read fires

A first-read fires when the daily deep-value scan flags a candidate and the scan output picks it as a top-1-to-3 priority for the day. Selection criteria explicit in the scan output:

  1. Margin of safety visible — back-of-envelope says price is meaningfully below EPV-only floor on current numbers
  2. Moat is at least nameable — the analyst can identify the source of competitive protection in one sentence, even if not yet verified
  3. Structural setup, not headline trade — the name is dislocated for a structural reason (capital cycle, forced selling, segment misvaluation, complexity discount), not a one-day news reaction
  4. Inside circle of competence — sector and business model fall within what the kit can reasonably analyze

Names that fail any of the four criteria are noted in the scan output but not first-read. Names that meet all four are ranked by margin of safety, and the top 1-3 get first-reads.

The artifact

File location

09-Theses/<TICKER>/first-read-YYYY-MM-DD.md

The first-read lives at the ticker level — the same level as the landing file and the thesis instance folders. A ticker folder may have zero, one, or multiple first-reads over time. Multiple first-reads happen when:

  • A first-read decided "shelve" and the name resurfaces later at a different price
  • A first-read decided "pass" and material new evidence emerges (new management, segment sale, etc.)
  • A first-read decided "continue" and the resulting thesis was published; later the same name needs re-triage from a fresh angle

The newest first-read is the current state; older ones are preserved as audit trail.

Frontmatter

---
tags:
  - first-read
ticker: XYZ
name: Company Name
sector: <GICS sector>
first_read_date: YYYY-MM-DD
source_scan: YYYY-MM-DD (the daily-scan note that surfaced this name)
decision: continue | pass | shelve-with-trigger | avoid
decision_rationale: <one sentence>
shelve_trigger: <price>      # only when decision = shelve-with-trigger
shelve_horizon: <quarters>   # only when decision = shelve-with-trigger
next_action: <one phrase>    # e.g., "queue full thesis", "monitor quarterly", "drop", "filed-avoid"
---

Body — six sections, each tight

1. Why it surfaced (2-3 sentences). What signal flagged this name in today's scan. State the dislocation and the rough magnitude.

2. Business in a paragraph (4-6 sentences). What the company does, how it makes money, who buys from it, what's the recent operating trajectory. Plain-English; no jargon unless necessary.

3. Back-of-envelope valuation (1 short table + 2-3 sentences). EPV-only floor on normalized current earnings, with explicit WACC and normalization assumptions. Current price as a multiple of EPV. No full triangulation; just one number computed honestly.

4. The one thing that decides this (3-5 sentences). What is the single hinge question? Is it "does the moat hold against hyperscalers?" or "is the cycle trough actually here?" or "is the discount real or is the asset impaired?" Name it explicitly. A first-read that can't identify the hinge question hasn't done enough work to decide.

5. Top risk (2-3 sentences). What's the single biggest reason this could be a value trap? Cite specifically — leverage profile, customer concentration, regulatory exposure, accounting concern.

6. Decision (2-3 sentences). State the decision (continue / pass / shelve-with-trigger / avoid) and the one-sentence rationale. If shelve-with-trigger, name the trigger price and the horizon over which we'd revisit. avoid is reserved for names with active red-flag reasoning (governance issue, accounting concern, balance-sheet fragility) — not just "not interesting."

Length target: ~500-1,000 words total. Time cap: 60 minutes. A first-read that takes longer is being treated as a thesis and should be paused with a Backlog note.

Decision values

continue — escalate to full thesis

The name passes the four selection criteria and the back-of-envelope work reveals real conviction. The hinge question is answerable with available evidence; the top risk is manageable; the margin of safety is meaningful even after a conservative read. Next action: a Backlog item is queued and the alphasteve-thesis-builder task autonomously picks it up to write the full bundle. Tier 1 as of 2026-05-25 autonomy update (see Rules) — no user pre-approval gate. User reviews the completed thesis after the fact.

pass — drop

The work surfaced enough to know this isn't worth deeper engagement. Either the hinge question is unanswerable, the moat doesn't hold up under one paragraph of scrutiny, the apparent margin of safety dissolves on normalization, or the risk is structural rather than cyclical. The first-read documents the reasoning so future scans don't re-flag the same name without new evidence.

shelve-with-trigger — wait for a specific event

The name is potentially interesting but the price isn't there yet, or a near-term catalyst is needed to clarify the structural picture. Document the trigger explicitly: a specific price level, a specific quarterly print, a specific regulatory milestone. The name comes off the active workbench until the trigger fires.

avoid — filed with red-flag reasoning

The name has an active reason not to engage further — governance red flag, accounting concern, balance-sheet fragility, structural moat erosion, regulatory exposure. Different from pass (which means "not interesting enough"); avoid means "actively flagged as bad to own." Documented so future scans don't re-surface the name without new evidence rebutting the red flag.

Decision authority

Per Rules (updated 2026-05-25 with the autonomy doctrine):

  • Writing the first-read itself is Tier 1 (autonomous) — the scan's selection picked the name, the agent does the 60 minutes of work, the decision lands in the file.
  • All four decisions (continue, pass, shelve-with-trigger, avoid) are Tier 1 — autonomous.
  • The continue decision queues a Backlog item that the alphasteve-thesis-builder task autonomously consumes. No user pre-approval gate; the user reviews the completed thesis after the fact and can override the verdict via a revision file.

The asymmetry that remains: the agent commits to the work (scanning, triaging, building); the user retains authority over the interpretation (overriding verdicts) and over capital decisions (the thesis-builder produces a buy recommendation; the user accepts or rejects it via the portfolio-daily flow).

How first-reads propagate

To the Daily Scan log

Every first-read is logged in that day's scan output:

First-reads written today:
- XYZ: continue → queued for thesis (Backlog item P1)
- ABC: shelve-with-trigger at $32 (currently $48), 2 quarters horizon
- DEF: pass — moat thesis fails the one-paragraph test

To the Backlog

When a first-read decides continue, a Backlog item is added with:

  • Proposed thesis location: 09-Theses/<TICKER>/<TICKER> - <MMMYYYY>/
  • Source: link to the first-read
  • Why this name: one-paragraph case for deep work
  • Priority: typically P1 if the back-of-envelope conviction is strong, P2 otherwise
  • Status: awaiting approval

To the frontend

A new "Pipeline" view (or section of the Theses page) surfaces:

  • Tickers that have a first-read but no thesis instance yet (active triage)
  • Tickers shelved with a trigger (passive monitoring, similar to watchlist but at first-read depth)
  • Recent first-read decisions over the past week

To the watchlist?

No. The watchlist is for names with a published pass-with-trigger thesis verdict. A shelve-with-trigger decision from a first-read is not a watchlist entry — the work isn't deep enough to warrant the watchlist's full discipline (calibration tracker, quarterly review). Shelved names live in a separate, lighter pipeline view.

What a first-read is not

  • Not a thesis — no valuation triangulation, no consensus & gap section, no shadow matrix, no position-sizing
  • Not a screen — a screen produces a list; a first-read produces a decision per name
  • Not a placeholder for "I should look at this someday" — if a name isn't worth 60 minutes today, it isn't worth flagging
  • Not a draft thesis — a first-read that decides continue does not become the thesis; the thesis gets written fresh against the full standard

Calibration

A first-read scorecard lives in methodology-calibration alongside the thesis-level metrics. Track:

  • Hit rate of pass decisions — names we passed that subsequently appreciated >50% over 12 months (false negatives)
  • Hit rate of continue decisions — first-reads that escalated to theses with verdicts other than pass/avoid (true positives at the first-read stage)
  • Hit rate of shelve-with-trigger — triggers that fire and result in updated first-reads or full theses

After ~30-50 first-reads, the data starts to show whether the four selection criteria are well-calibrated or whether the kit is systematically missing or chasing certain patterns.

Sources

  • The first-read concept is standard practice on working research desks (sell-side and buy-side). The discipline of "fast triage before deep work" is documented in numerous practitioner accounts; see T2 and T2.
  • Buffett's "too hard" pile is the spiritual ancestor of the pass decision — names that may be good investments but aren't worth our time to figure out.
  • Greenblatt's You Can Be a Stock Market Genius (1997) describes the funnel from screen to deep work, with explicit acknowledgment that most screened names die at the first-read stage.

Linked