Results War Room — Canonical Reference

One-line definition: Every company whose results land in the window appears — no rule selects the names. Each event is labelled by whether its earnings surprise and delivery multiple fell in the upper part of its own reporting cohort, recomputed per run. Descriptive only: the tradeable version failed net of cost.

1. What it is

A board of companies that have just reported, plus the ones about to. It answers a narrow question — who reported, and did the market's reaction come with real delivery money behind it — and it answers it about the past, for events that have already happened.

Two things must be kept apart, because only the second is a rule of ours:

• What appears on the board is not selected. The cohort is "companies that filed results in the window". We do not screen it, rank it, or shorten it. If a company reported, it is there.
• **What is labelled is.** Each event carries two mechanical flags describing where it sat inside that same cohort (§3). A flag is a description of an event that already occurred, not a claim about what the name will do next.

Where this stands: the board is descriptive only. Every attempt to turn the drift into a book has failed net of realistic cost (§4). The board is useful as a record of what happened around a result — not as an instruction.

2. Our variation vs. the standard technique

The standard PEAD construct sorts reported events by standardised unexpected earnings (SUE) alone and studies the subsequent drift. Ours departs on three points:

• A second, delivery-based flag. Beside the surprise we carry a delivery multiple — how far that day's delivered quantity ran above the name's own recent norm — from the NSE delivery archive. The intent is to separate a price reaction that carried settlement money from one that did not. This flag is the house addition; the surprise leg is textbook.
• Relative cuts, not fixed thresholds. Both flags are percentile cuts taken inside the current settled cohort and recomputed every run, so a "high" surprise means high relative to the companies that reported alongside it — never against a constant. There is therefore no fixed number to quote (§3).
• The board is the deliverable, not a portfolio. The classic paper ends in a long/short book. Ours ends at the board, because the book did not survive costing (§4).

3. How it works (methodology)

The cohort. Events are results filings landing in the window. Upcoming board meetings are read separately from the exchange results calendar (is_results=1) and shown as a forward list — an intimation that a company will report, never a statement about the outcome.

The two flags. For the settled events in the run:

• Surprise flag — set when the event's SUE sits in the upper part of the cohort's own SUE distribution.
• Delivery flag — set when the event's delivery multiple sits in the upper part of the cohort's own delivery distribution.

The two breakpoints are derived per run from the cohort and stored beside the snapshot (results_reactions_meta), not hard-coded. They move as the reporting season changes, which is exactly why this page quotes no number: any figure printed here would be stale by the next run and would misrepresent the rule as a constant. The live values are on the snapshot itself.

What the reader can do with them. The board's default view is everything; a reader may optionally narrow to the events where both flags are set. That narrowing is a view filter the reader chooses, not a house selection — the same rows are present either way.

What is displayed per event. The surprise, the delivery multiple, the post-event cumulative abnormal return at two horizons, whether the event has settled, and a cohort label. Where the board prints a historical cohort mean it prints it with its n, because a mean without its count is not a measurement.

Staleness. The board carries an explicit lag banner when the underlying tape is behind, so a reader can never mistake an old snapshot for today's.

4. Status, validation & honesty fence

LIVE, and descriptive only. The drift is real as a description of what the cohort did. It is not a strategy here, and this page does not present it as one:

• Every tradeable wrapper has been falsified. The PEAD book is a recorded BLOCKING failure in the strategy ledger (receipt ledger key PEAD_BOOK): the descriptive drift does not survive realistic transaction cost. See also the explosive move research notes, where the same conclusion is reached on the adjacent construct.
• The flags are not a ranking. They are two binary descriptors of a past event. They are not summed, scored, or ordered into a recommendation.
• The forward list is an intimation only. A company appearing under upcoming results has told the exchange it will hold a board meeting. Nothing about the result, the reaction, or the delivery is knowable at that point, and the board claims nothing.
• No return is promised anywhere on this surface, and no position, target or timing instruction is expressed or implied.

Promoting any part of this board to a ranked or traded claim requires its own pre-registered, leak-free study recorded in the ledger that beats the recorded numbers. Until then it stays descriptive.

5. Where it lives (code· routes· DB· timers)

• Route: /dash/results-reactions (the results reactions code), lens key results-reactions. Optional ?view=confirmed is the reader's both-flags filter.
• Reads: the internals reads code — the snapshot reader that applies the optional filter and returns the cohort count.
• Snapshot producer: the pead surface research code (breakpoints + per-row flags) on the primitives in the evlib research code. The board is a viewer over that snapshot; it computes no flag of its own.
• DB: the results_reactions table plus results_reactions_meta (the per-run breakpoints) in the research database.
• Downstream consumers: the auto analyst code (drafts a reviewable brief per event — human-reviewed, never auto-published), the season digest code, the strategy registry code, the internals pages code.
• Receipt: the surface is registered in the evidence receipt code as results-war-room; its declared data class is the delivery archive, because that is the archive the displayed number actually rests on.

6. Data & provenance

Two primary feeds and one archive, all authentic-source (Guardrail #8):

• results_calendar — NSE event-calendar API. Board-meeting intimations are filed in advance, so they are knowable at filing and forward-looking by design.
• concall_bse — BSE corporate announcements (earnings-call transcripts); the broadcast timestamp is the clock, and transcript text is stored on disk rather than in the DB.
• The NSE delivery archive — the source of the delivery multiple.

🔴 The board does not read the Screener-origin quarterly fundamentals table. That distinction is deliberate and load-bearing: naming a vendor origin on the receipt of a paid surface would attribute the number to a source it did not come from.

7. Terminology canon

• Results War Room — this board. Do not call it a screen or a shortlist; it selects nothing.
• Surprise flag / delivery flag — the two per-event descriptors. Both are cohort-relative, both are binary, neither is a rank.
• Delivery multiple — delivered quantity against the name's own recent norm. Side-blind, like all delivery measures; it says money settled, never who bought. Sibling: the dvpt notes.
• Settled — the event's post-event window has fully elapsed. Unsettled events have no complete outcome and are marked as such.
• Confirmed (view) — the reader-chosen both-flags filter. It is a view state, not a house verdict, and never a recommendation.
• PEAD — post-earnings-announcement drift, the classic phenomenon. Here it names a falsified book, not a live one.

Reader-facing definitions: the metrics glossary.

8. Decision & session history

• Board built as the surface over the PEAD event study; the study's tradeable conclusions were recorded as BLOCKING failures in the strategy ledger rather than shipped.
• Auto-analyst briefs route through the human review inbox by construction — a drafted brief is queued, never published unreviewed.
• 2026-08-31 — this page created so the Evidence Receipt has a real rule to quote for L2 instead of an explicit UNKNOWN. The one-line definition above was written to be quoted verbatim to a reader, and it states the absence of a selection rule because that is the truth about the board.

9. Open items / frozen work

• No fundable form exists, and none is being sought without a new pre-registered study.
• The snapshot is produced under the research folder, not by a timer in the automation folder — it refreshes when the study is re-run, which is why the staleness banner is load-bearing rather than cosmetic.
• Upcoming-results coverage depends on companies filing their intimation; a company that has not filed is absent, and absence is not evidence it will not report.

10. Sources of truth

• Results / falsification: the strategy ledger (PEAD_BOOK)· the explosive move research notes.
• Definitions: the metrics glossary.
• Code: the results reactions code· the internals reads code· the pead surface research code· the evlib research code· the feed manifest code.
• Index: the README· the origins notes·../../the project's running record.