A new user's guide.

What TechPHA is

Process hazard analysis software for the people who sign off on it. One principle decides how everything in it behaves: the AI proposes, the engineer decides.

Nothing the AI produces enters your study on its own. Each proposal arrives tagged with a confidence value and marked as awaiting review, and it becomes part of the study only when a person accepts it. Anything the system is unsure of is flagged Engineer Review Required and stays flagged until someone confirms or corrects it.

That is not a disclaimer bolted on afterwards. You will notice it:

• Your risk matrix is yours. A new study starts with no matrix rather than a default, and the worksheet will not rank anything until you choose one. Rankings reference the criteria in force at the time, so an old study stays readable after you revise the matrix.
• The SIL band is computed from your numbers, never typed, including the case where the layers you credited already do the job and no safety instrumented function is required.
• The report names its own gaps. No facilitator named, rows unranked, a session not held, a checklist gap with no recommendation against it. The issued document says so at the front rather than leaving a reader to discover it.

The workflow

A study runs in this order. You can move around freely, but each step assumes the one before it.

1. Create the study. Title, facility, study type, the standard you work to, scope and boundaries. What you write as scope and boundaries appears in the issued report.
2. Pick your risk criteria. The matrix your organisation works to (a 5×5 arrangement is typical), with your severity and likelihood definitions. Do this early: the worksheet will not compute a risk band until a matrix is in force, and it says so on screen with a link to choose one.
3. Assemble the team. Nine roles: facilitator, scribe, process engineer, operations, instrument/control, mechanical, process safety, reviewer/approver, and observer (read-only). Whoever creates the study becomes facilitator.
4. Schedule the sessions. Date, start and end times, location, whether on site, remote or hybrid, who chaired, who scribed, and which nodes were covered. Attendance is tracked per session, so the report can state “4 of 6”.
5. Upload the drawings. PDF, DWG, DXF or TIFF. Documents and P&IDs are kept apart, because only a drawing can be analysed.
6. Let the assistant read the drawing. It extracts equipment, instruments, valves, process lines, connections and existing protective devices, each with a tag, a description and a confidence value.
7. Review what it found. This is the part that matters. Work down the list in confidence order and accept, correct or reject each item. Errors caught here are far cheaper than errors caught in the workshop.
8. Build the nodes. A node is a section of plant with a design intent, a boundary, design and operating conditions, and the equipment inside it. Nodes are proposed from the drawing and remain fully editable. You decide where the boundaries fall.
9. Raise the deviations. Guide words against parameters: no flow, more pressure, less level. The same guide word and parameter twice on one node is refused, because that is a duplicate row in the report rather than a second finding.
10. Work the worksheet. For each deviation: the cause, the consequence, the safeguards already in place, and the risk ranking. The band is computed from your matrix.
11. Run a LOPA where a scenario needs numbers. Only for the scenarios that warrant it.
12. Work the check lists. The structured prompts your procedures require in addition to guide-word analysis.
13. Issue the report. Preview it in full on any plan; export to PDF, Word or Excel on a paid one.

Every screen, and what it is for

The navigation is grouped the way a study is: the data about it, the analysis, the registers that come out, and the setup behind it. A ✦ marks screens where the assistant can help.

The safeguards register is worth a word. Keeping protective devices as records rather than free text on the worksheet is what makes “how much does this one alarm carry across the study” a question you can answer at all.

How a finding gets approved

Every entry carries a status, and the status only moves when a person moves it.

The evidence, the confidence value and the approval history are kept at every step. A reviewer or auditor can see what was proposed, what was changed and who accepted it. The trail from drawing to recommendation stays intact through review and into the report.

Confidence is guidance, not a verdict. A 94% is not a promise that the tag is right; it is the system telling you how sure it is, so you can work down the list in a sensible order.

LOPA and SIL

A LOPA continues a worksheet row; it does not start a second study. That is why the LOPA screen lists every worksheet row rather than offering a “new LOPA” button. The question at that screen is which scenarios still need one, and a list of the LOPAs that already exist cannot answer it.

You give the scenario two frequencies and credit the layers that stand between them:

• Initiating event frequency: how often the cause occurs, per year.
• Tolerable frequency: how often your criteria accept the consequence, per year. Seeded from the scenario’s severity where your risk criteria carry a target, so the number is not retyped. You can argue it up or down for that scenario.
• Independent protection layers: each with a probability of failure on demand. Eleven kinds are offered, from relief devices and rupture discs through alarms with operator response to procedures and passive mechanical protection.

The required risk reduction and the SIL band follow from those numbers:

Three behaviours are worth knowing:

• A layer you reject stays on the record and out of the arithmetic. Mark it not credited, write why, and it appears in the report with your reasoning. Deleting it would leave the next reader unable to tell a ruled-out layer from one nobody thought of.
• Layers that are not independent are flagged, rather than silently multiplied together.
• A probability outside 0 to 1 is refused rather than clamped, because clamping 1.5 to 1 would quietly accept a typo.

If either frequency is blank, no verdict is shown. A half-filled LOPA is the normal state of one being worked on, and “SIL 3” against a blank tolerable frequency would be a number invented by the absence of a number.

Check lists

Guide words ask what could deviate. A check list asks what your procedures require you to have considered regardless: the questions with no deviation attached, which is exactly why they get missed.

Nineteen standard prompts are copied onto a study the first time you open the screen, across four categories: Design, Operations, Maintenance and Emergency response. They are copied, not referenced, so the study owns what it answered: editing the standard set afterwards cannot change a check list somebody has already signed, and prompts your team adds survive the standard set moving on.

Each prompt takes one of three answers, and two of them are not the same as leaving it blank:

• Yes: considered and satisfied.
• No: a gap. This opens a field for what is missing and where.
• N/A: not applicable. A decision somebody made and can defend, which is different from a prompt nobody reached.

The screen is built around the no answers, because those are the point. A “no” that stays on the check list is a gap nobody owns, so the only way one leaves the screen is onto the recommendations register. One button, which then disappears so the same gap cannot be raised twice. Until that happens the screen counts it as unmet and not yet raised, and the issued report lists it among the study’s gaps. You can add your own prompts in any category.

What you get out

A study leaves TechPHA as a report covering the study header, scope and boundaries, the team, the sessions, the nodes, the full worksheet, the LOPA assessments, the check lists, the safeguards and the recommendations.

The preview and the exported file are built from the same study record, so the document you download cannot say something the preview did not. The preview is available on every plan. It is how you judge the output before paying for the ability to take it away.

The three formats are not one document in three wrappers:

In the spreadsheet, severities, frequencies and probabilities are written as numbers, not text. A severity exported as text sorts 1, 10, 2, which alone would make the file useless for the job it exists to do.

The report opens with what it cannot yet state. If there is no facilitator, no scribe, no matrix, unranked rows, a session not held, open parking-lot questions or an unraised check list gap, they are listed at the front. A report that quietly omits its own gaps is worse than one that names them.

What your plan includes

Capabilities by plan. Prices and the full comparison are on the pricing section.

Two things about the free plan are worth understanding. Its one study is counted lifetime, not active. Deleting it does not give the allowance back. And you can still upload a drawing on it: what a paid plan buys is the reading of the drawing, not the holding of it. A free tier you cannot put a drawing into would not be a usable tier.

If a paid plan lapses, the workspace drops to Basic and nothing is deleted. Every study, worksheet and record stays there and stays readable; what you lose is the ability to create new studies and export reports. A declined card gets a grace window rather than an immediate drop, because that is usually an expired card rather than a decision.

What it will not do

TechPHA assists a process hazard analysis. It is not a substitute for the judgement of a competent engineer, for a properly constituted HAZOP team, or for compliance with any standard, regulation or internal procedure that applies to your facility.

• Output from the AI is a proposal. It may be incomplete, may misread a drawing, and may be wrong.
• Confidence values are indicative guidance, not a warranty of correctness.
• There is no warranty that the analysis will identify every hazard, deviation, cause, consequence or safeguard.
• Nothing in the service constitutes engineering advice or certification.
• Responsibility for reviewing, correcting, approving and issuing a study rests entirely with you and the qualified people carrying it out.

You must not rely on TechPHA output as the sole basis for any safety decision, design change or regulatory submission. The terms set this out in full, and the privacy notice covers what is processed and where.

Still stuck

Write to support@techpha.com and a person will answer, or ask for a demo and we will walk a study through with you.

Start your free study